LIST OF CONTRIBUTORS

Alexander W. Cappelen Norwegian School of Economics and

Business Administration, Bergen, Norway
Margaret Chitiga University of Pretoria, South Africa
John Cockburn Poverty and Economic Policy (PEP)
Research Network and Department of
Economics, Laval University, Québec,
Canada
James B. Davies Department of Economics, University of
Western Ontario, London, Ontario,
Canada
Bernard Decaluwe Poverty and Economic Policy (PEP)
Research Network and Department of
Economics, Laval University, Québec,
Canada
Udo Ebert Department of Economics, University of
Oldenburg, Oldenburg, Germany
Ismael Fofana Poverty and Economic Policy (PEP)
Research Network and Department of
Economics, Laval University, Québec,
Canada
Pilar Garcı´a Gómez Departament d’Economia i Empresa and
CRES, Universitat Pompeu Fabra,
Barcelona, Spain
Nigar Hashimzade Department of Economics, University of
Exeter, Exeter, UK
Michael Hoy Department of Economics, University of
Guelph, Guelph, Ontario, Canada
Peter J. Lambert University of Oregon, USA
vii
viii LIST OF CONTRIBUTORS

Ramos Mabugu University of Pretoria, South Africa

Daniel L. Millimet Southern Methodist University, USA
Gareth D. Myles University of Exeter, Exeter; and Institute
for Fiscal Studies, London, UK
Angel López Nicolás Departamento de Economı́a, Universidad
Politécnica de Cartagena, Cartagena,
Spain; and Departament d’Economia i
Empresa and CRES, Universitat Pompeu
Fabra, Barcelona, Spain
Kristian Orsini Center for Economic Studies, Katholieke
Universiteit, Leuven, Belgium
Ian Preston University College London and Institute
for Fiscal Studies, London, UK
Geoff Rowe Statistics Canada, Ottawa, Canada
Daniel Slottje FTI Consulting, Inc. and Southern
Methodist University, USA
Amedeo Spadaro PSE Paris-Jourdan Sciences
Economiques, France; FEDEA Madrid,
Spain; and Universitat de les Illes Balears,
Palma de Mallorca, Spain
Paul D. Thistle Department of Finance, University of
Nevada Las Vegas, Las Vegas, NV, USA
Georg Tillmanny Department of Economics, University of
Mainz, Mainz, Germany
Bertil Tungodden Norwegian School of Economics and
Business Administration and Chr.
Michelsen Institute, Bergen, Norway
Michael Wolfson Statistics Canada, Ottawa, Canada
INTRODUCTION

For equity, societies may wish to eliminate certain forms or manifestations

of inequality. Horizontal equity and vertical equity in the income tax are
topics which have interested me for some years. Although any shortfall from
each of these objectives can be measured in terms of unwanted inequalities,
equity per se is a different concept from equality. Equity relates to fairness,
justice and other societal norms which give expression to the best aspirations
of our collective social conscience. For example, equal access to health care
for those in equal need is an accepted norm for horizontal equity in the
health field. Vertical equity in this context means treating appropriately
differently those who have different needs. When offered the opportunity to
be Guest Editor of this volume of Research on Economic Inequality, I de-
cided to define the focus simply as ‘‘equity’’, without placing any further
restriction on topics. The papers which were ultimately included in this
volume are the ones, from among those offered, which survived a rigorous
refereeing process. Each has its own ‘‘take’’ on the concept of equity, and its
link with equality. I hope that you, the reader, will gain from reading all of
these contributions and pondering their significance.
Alexander W. Cappelen and Bertil Tungodden focus on the interdepend-
encies created by a redistributive system. In essence they ask whether, in
view of these, it can be the case that increased effort on the part of any one
individual (which, assuming no production interdependencies, would imply
increased pre-tax income for that individual), necessarily leads to increased
post-tax income (reward) for that individual. Remarkably, they show that
only lump-sum redistribution schemes have this property, and that such
schemes necessarily involve negative post-tax incomes for some, which they
interpret as a violation of the fundamental right of a person not to work.
This deep consideration of fundamentals comes with essential simplicity.
Paul D. Thistle extends a long-standing result of Lerner (1944) concerning
the welfare analysis of income distributions: ‘‘If it is impossible, on any
division of income, to discover which of any two individuals has a higher
marginal utility of income, the probable value of total satisfactions is max-
imized by dividing income evenly’’ (ibid., p. 29). Paul’s paper gives the
background, and the subsequent discussion by other contributors which has
ix
x INTRODUCTION

surrounded Lerner’s result, and goes on to weaken the restrictive assump-

tions Lerner made, in particular assuming now that the planner may have
some (if not much) information regarding the possible assignment of utility
functions to individuals. Paul shows that the fundamental idea, that of
‘‘equi-probability’’, can survive (in modified form) the introduction of more
modern concepts in distributional analysis than were around in Lerner’s
day, such as rank and generalized Lorenz dominance, and needs-based
welfare functions.
Ian Preston wrote about inequality, income tax progression and labor
supply in his Nuffield College DPhil thesis of 1989, but somehow moved on
to other things without publishing some quite important material. Ian’s
results for inequality indices which preserve the ratio dominance ordering of
underlying distributions, for example, were new to economists at the time,
but had been known to mathematicians since a 1967 Pacific Journal of
Mathematics paper by Marshall, Walkup and Wets. Ian has been persuaded
to revisit his earlier analysis for this volume. His paper ‘‘Inequality and
Income Gaps’’ inter alia tells a rounded story about the ratio and gap
dominance concepts, and associated orderings and welfare properties, which
extend the well-known Lorenz ordering in different ways.
Udo Ebert and the late and sadly missed Georg Tillmann have provided a
full and somewhat startling answer to an apparently innocuous question
about income tax design. For an income tax which intervenes in distribution
all along the income scale (in particular, having no zero bracket amount),
surely progression can remain moderate throughout? ‘‘No, it cannot’’ is the
answer, ‘‘Not if you rule out a region of negative taxes with a negative
marginal tax rate at some income values’’. The exposition is extremely clear
and the message is thought provoking.
Nigar Hashimzade and Gareth D. Myles contrast the redistributive prop-
erties of income and expenditure taxes. Heterogeneity in their model comes
through differences in skill and differences in endowments. In both a static
two-period model and a fully dynamic overlapping generations model, in-
come taxes are set and then replaced with expenditure taxes calibrated to
produce first the same welfare level, then the same level of government
revenue. Simulation exercises with various assumed degrees of correlation
between the two sources of heterogeneity indicate that, in both the static and
dynamic versions of the model, Gini coefficients for lifetime income are
(almost always) less in the case of income-based taxes than for expenditure-
based taxes.
Kristian Orsini and Amedeo Spadaro propose an index of strategic weight
within couples. The index is defined for each member as the share of
Introduction xi

resources lost, should he or she leave the household. From these weights, the
authors develop a ‘‘neutrality index’’ to determine the stance of the tax-
benefit system itself toward household formation. Then, using micro data
for Finland, Germany, Italy and the UK, the authors explore how strategic
weights are affected by the relevant tax-benefit systems, and they are able to
provide much comparative information about the family stances of the re-
spective systems. Ample discussion is included of the limitations of the study
and the future research which is now called for.
James B. Davies and Michael Hoy join the ongoing debate about the best
way to finance a health care system. They outline methods to determine the
impact on the distribution of income of a move from completely public to
totally private financing of health care, applying their methodology to a
hypothetical move of this kind – in which premia that were generated
through taxation are now based on age and gender, as well as on other
individual characteristics related to expected health care costs – for Canada.
This paper will be valued not only for its Canadian findings, but also as a
useful expository source for the sort of distributional impact analysis that
can be applied to such a question, and for its frank appraisal of the as-
sumptions which the authors needed to make to get their analysis off the
ground.
Pilar Garcı́a Gómez and Angel López Nicolás have written a paper about
equity in the utilization of health care, seeking to determine for Spain the
extent to which the National Health System reforms introduced between
1987 and 2001 have led to improved equity. The authors define their equity
ideal as ‘‘equal access given equal need for health care’’, and they compute
inequity, using concentration indices, as the component of income-related
utilization inequality which is not attributable to socio-economic needs
differences. Their careful analysis permits the tracking, over the 1987–2001
period, for each of a range of health services, of an income-related inequity
component and a second inequity component explained by the influence of
private health insurance and privately provided health care. By this, the
authors are able both to unbundle and to understand the improvements in
equity that have been attained in Spain, and to point to a range of issues
which remain for deeper study.
Michael Wolfson and Geoff Rowe’s paper, which uses the Canadian Life-
Paths microsimulation model to analyze the lifetime incidence of taxes and
transfers across Canadian worker cohorts, begins with a clear and thought-
provoking general discussion of what intergenerational fairness might mean
(for which, they argue, ‘‘there is no widely agreed concept’’). This discussion
will be found valuable per se. The empirical results the authors derive for
xii INTRODUCTION

Canada suggest that under a range of scenarios for future economic growth
and longevity, birth cohorts after those born in the 1920s and 1930s will
experience successively smaller lifetime net transfers, both cash and in kind.
These results are significantly influenced by the use of the consumer price
index to update cash transfer benefit levels and income tax thresholds and
other related parameters.
Daniel L. Millimet, Daniel Slottje and Peter J. Lambert ask a ‘‘what if?’’
question about poverty and go on to answer it rather fully. In a paper which
was handled editorially by John Bishop, Joint Managing Editor of REI, the
implications of supposing that decision makers in any country and at any
point in time tolerate a certain fixed amount of perceived poverty are ex-
plored. Differences in poverty aversion could fully account for observed
international and intertemporal variations in objective poverty, consistent
with any chosen ‘‘natural rate’’ of perceived poverty. As the authors show,
poverty aversion itself can be understood in terms of political and socio-
economic factors, and an adjustment mechanism is implied such that a
change in a relevant explanatory variable, which leads to changed poverty
aversion, would trigger a change in distribution and restoration of the nat-
ural poverty rate. The natural rate hypothesis offers a new way to under-
stand the influence of social, political and economic changes on measured
poverty. The relationship between inequality aversion and poverty aversion
is also explored with the aid of a parallel ‘‘natural rate’’ hypothesis for
subjective inequality.
John Cockburn, Ismael Fofana, Bernard Decaluwe, Ramos Mabugu and
Margaret Chitiga report on the construction and implementation of a gen-
der-sensitive macro-micro framework for South Africa. Their structure
comprises a CGE model that can deal with non-market activities and in-
formation on time use along gender lines, and a microsimulation model
based on household level data which allows the authors to account for
additional heterogeneity. The outcome is a most interesting study which
pins down the poverty impacts of trade liberalization in South Africa.

Peter J. Lambert
Editor
REDISTRIBUTION AND
MARGINAL PRODUCTIVITY
REWARD

Alexander W. Cappelen and Bertil Tungodden

ABSTRACT
A fundamental ethical question is how a redistributive system should re-
ward individual effort. Marginal productivity reward has been justified
either as a way of ensuring efficiency or as a way of respecting people’s
self-ownership. Both these arguments have their limitations. We show
that marginal productivity reward is implied by one intuitively appealing
requirement on the reward structure, which we name non-negative reward.
This result can be interpreted in one of two ways. It can be seen as a new
justification of marginal productivity reward that avoids the limitations of
the traditional arguments. Alternatively, it can be seen as a result showing
that any redistributive system that makes transfers conditional on effort,
sometimes will make the reward individuals get for their additional effort
completely conditional on others effort. Finally, we also show that no
genuine redistributive system satisfies both non-negative reward and the
liberal requirement of no forced labour.

Equity
Research on Economic Inequality, Volume 15, 1–6
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15001-8
1
2 ALEXANDER W. CAPPELEN AND BERTIL TUNGODDEN

1. INTRODUCTION

A fundamental ethical question is how a redistributive system should reward

individual effort. One prominent answer to this question is that people
should be rewarded with their marginal productivity. This answer has tra-
ditionally been given two types of justifications. First, marginal produc-
tivity reward has been justified by efficiency considerations. Economic
theory shows that deviations from marginal productivity reward create dis-
tortions that may cause Pareto-inefficiency. Second, it has been justified by
equity considerations. According to some theories of distributive justice, in
particular libertarianism (Nozick, 1974), marginal productivity reward is the
only way of respecting people’s self-ownership (see also Kolm, 2001).
Both these arguments have their limitations. The efficiency argument only
provides a justification for marginal productivity reward in situations where
there are incentive problems. For example, in situations where the Hicksian
supply of effort is inelastic, there is no efficiency reason for rewarding effort
with its marginal product. The equity argument is problematic because it relies
on some very controversial normative assumptions. Only people accepting the
basic idea of full self-ownership and the view that full self-ownership implies
marginal productivity reward would be convinced by the libertarian equity
argument. This position, however, is rejected both by utilitarians (for example,
Mirrlees, 1971; Harsanyi, 1987; Broome, 1991) and liberal egalitarians (for
example, Rawls, 1971; Fleurbaey, 1995; Moulin & Roemer, 1989).
In this paper we present a result that may be seen as an alternative jus-
tification for marginal productivity reward that avoids the limitations of the
traditional arguments. The result applies even in the absence of incentive
considerations and it relies on a much less controversial normative assump-
tion than the self-ownership argument. We show that marginal productivity
reward follows from a very appealing requirement, namely that people never
should have a reduction in their post-tax income when they increase their
effort. We name this the non-negative reward requirement. To illustrate,
consider two situations a and b, where you work harder or longer hours
in b than in a and thus have a higher total pre-tax income in b than in a. The
requirement then states that your post-tax income in b should not be lower
than your post-tax income in a.
Alternatively, the result can be seen as showing that an unavoidable con-
sequence of any redistributive system that makes transfers to an individual
conditional on his effort is that sometimes the reward individuals get for
their additional effort will be completely conditional on others effort. This
may be an obvious feature of a redistributive system in an economy where
Redistribution and Marginal Productivity Reward 3

there are interdependencies in the production technology, but the result

establishes that this is the case also when there are no interdependencies in
the production technology.
Another very appealing requirement on a redistributive system is that it
should not force anyone to work, which we name no forced labour. Our
second result, however, shows that this requirement is incompatible with
non-negative reward in any genuine redistributive system.
We present the formal framework in Section 2. In Section 3, we establish
the propositions. Section 4 provides some discussion of how to interpret the
results.

2. FORMAL FRAMEWORK

Consider a society with a population N ¼ f1; . . . ; ng; nZ2, where person i’s
effort is ei and e ¼ ðe1 ; . . . ; en Þ is the effort distribution in a particular sit-
uation e. Let O be the set of all effort distributions. We assume that all
individuals can choose between all effort levels ei 2 ½emin ; emax Þ
Rþ ; where
Rþ is the set of real non-negative numbers. The pre-tax income for each
individual i, f i : ½emin ; emax Þ ! Rþ is continuous and strictly increasing in
effort, where f i ðemin Þ ¼ 0; 8i 2 N: Note that we do not assume any inter-
dependencies in the production technology and, moreover, we do not make
any assumptions about how the choice of effort is affected by the redistrib-
utive system. Hence, each person’s pre-tax income is independent of other
people’s effort and we cover cases both with and without incentive problems.
Our object of study is a redistributive system F : O ! Rn ; where Fi(e) is
the post-taxPincome of personP i in situation e. F satisfies the balanced budget
constraint ni¼1 F i ðeÞ ¼ ni¼1 f i ðeÞ; 8e 2 O: Moreover, for F to be considered
a genuine redistributive system, we assume that at least for some e 2 O and
j 2 N; F j ðeÞaf j ðeÞ:

3. REWARDING EFFORT

Most people support some degree of redistribution, but typically also agree
that a person should be rewarded for an increase in effort. We argue that an
appealing feature of any redistributive system would be that it satisfies a
minimal reward condition saying that a person who increases his effort, and
thus increases his pre-tax income, should not experience a decrease in post-
tax income. In other words, if your effort is higher in one situation than
4 ALEXANDER W. CAPPELEN AND BERTIL TUNGODDEN

another, then your post-tax income should at least not be lower in the
situation where you exercise more effort. Formally, we can write this re-
quirement as follows:
Non-Negative Reward (NNR): For any e; e~ 2 O and j 2 N, where e~j 4ej !
F j ð~eÞ  F j ðeÞ:
One way of rewarding effort that satisfies NNR is marginal productivity
reward.
Marginal Productivity Reward (MPR): For any e; e~ 2 O and j 2 N, where
e~j aej ! F j ð~eÞ  F j ðeÞ ¼ f j ð~ej Þ  f j ðej Þ:
It turns out that the non-negative reward requirement implies that effort
is rewarded with marginal productivity, that is, it is incompatible with an-
ything else than lump-sum redistribution.
Proposition 1. A redistributive system F satisfies NNR if and only if it
satisfies MPR.
Proof. The if part is straightforward. Hence, we will only prove the only-
if part.
(i) Suppose there exist e; e~ 2 O and k 2 N such that e~k ¼ ek and
F k ð~eÞ4F k ðeÞ:
(ii) Consider a new situation e^ 2 O; where for some e>0, e^i ¼
ei þ
; 8i: P
(iii) By the continuity of fi, for a sufficiently small e, ei Þ 
i ½f i ð^
f i ðei Þo½F k ð~eÞ  F k ðeÞ: P
P the balanced budget constraint, Pi ½F i ð^eÞ  F i ðeÞ ¼
(iv) By
i f i ð^
ei Þ  f i ðei Þ: By (iii), this implies that i ½F i ð^eÞ  F i ðeÞo
½F k ð~eÞ  F k ðeÞ: By NNR, F i ð^eÞ  F i ðeÞ; 8i: Hence, it follows
that ½F k ð^eÞ  F k ðeÞo½F k ð~eÞ  F k ðeÞ:
(v) By (iv), F k ð^eÞoF k ð~eÞ: However, since e^k 4~ek ; this violates NNR.
Thus the supposition in (i) is not possible.
(vi) Consider any e; e~ 2 O and k 2 N such that e~k 4ek : We will now
show that F k ð~eÞ  F k ðeÞ ¼ f k ð~ek Þ  f k ðek Þ: Consider e^ 2 O; where
e^i ¼ ei ; 8iak and e^k ¼ e~k : By (v), F i ð^eÞ ¼ F i ðeÞ; 8iak: Hence, by
the balanced budget constraint, F k ð^eÞ  F k ðeÞ ¼ f k ð^ek Þ  f k ðek Þ:
By (iv), F k ð^eÞ ¼ F k ð~eÞ: Moreover, f k ð^ek Þ ¼ f k ð~ek Þ; and the result
follows. ’

Note that in order to establish Proposition 1, we have not required that

the post-tax income of all individuals should be positive in all situations. If
some people have negative post-tax income, however, then this may be seen
Redistribution and Marginal Productivity Reward 5

as equivalent to forcing them to work. Therefore, if we want to ensure all

individuals the right to choose not to work, then the redistributive system
should satisfy the following condition:
No-Forced Labour (NFL): For any e 2 O and j 2 N, F j ðeÞ  0:
NFL should be an appealing condition in a liberal society. However, it
turns out to be impossible to combine this condition with the non-negative
reward condition.
Proposition 2. There does not exist any redistributive system F satisfying
NNR and NFL.
Proof. Proposition 1 shows that only lump-sum taxation satisfies NNR.
However, any positive lump-sum tax violates NFL, and the result
follows. ’
Alternatively, Proposition 2 may be seen as a characterisation of liber-
tarianism, if libertarianism is interpreted as requiring that each person’s
post-tax income always should be equal to this person’s pre-tax income.
Libertarianism implies that there should be no redistribution of income in
society, and thus satisfies both NNR and NFL.

4. DISCUSSION

The underlying intuition of Proposition 1 is that any non-lump-sum redis-

tribution, that is, any system of redistribution where transfers are condi-
tional on effort creates interdependencies among the individuals in the
economy, even if there are no interdependencies in the production technol-
ogy. The existence of such fiscal interdependencies makes it impossible to
satisfy the non-negative reward requirement.
To illustrate, suppose that individual i has the pre-tax income function
f ðwi ; Li Þ ¼ wi Li ; where wi is person i’s marginal productivity and Li is person
i’s labour effort. Consider a very simple economy with only two individuals,
person 1 and person 2, where they differ in marginal productivity, i.e.,
w1 aw2 (even though the proof does not rely on this assumption). Moreover,
assume that the government redistributive policy is limited to a linear income
tax scheme, where the tax revenues are shared equally between the two
individuals in society. The post-tax incomes are then given by F 1 ¼ w1 L1
ð1  tÞ þ tððw1 L1 þ w2 L2 Þ=2Þ and F 2 ¼ w2 L2 ð1  tÞ þ tððw1 L1 þ w2 L2 Þ=2Þ: In
this case, for any positive t, person 1 may receive a lower post-tax income
when increasing his effort, if at the same time person 2 decreases his effort.
6 ALEXANDER W. CAPPELEN AND BERTIL TUNGODDEN

In other words, the linear income tax scheme creates a fiscal interdependency
between the two individuals that makes it impossible to satisfy the non-
negative reward requirement.
Proposition 1 shows that this is not only a feature of a linear tax scheme
with uniform transfers, but applies to any redistributive system that does
not rely on lump-sum redistribution. Lump-sum redistribution, however,
violates the liberal requirement of no forced labour, and thus any genuine
redistributive system faces a fundamental conflict, as reported in Proposition
2. Either sometimes it has to give some people less post-tax income when
they have increased their effort or sometimes it has to force people to work.
Finally, let us note that there is a much weaker interpretation of the idea
of non-negative reward, namely that a unilateral increase in effort by some
person never should cause a decrease in his post-tax income. This very weak
requirement does not imply marginal productivity reward and it is con-
sistent with any reasonable redistribution system. We doubt, however, that
it captures all of our moral intuitions on how to reward effort. We find the
idea that an increase in effort should imply no decrease in post-tax income,
independent of what others do, very attractive, and thus we do believe that
it is of importance to observe that lump-sum redistribution is the only
redistributive policy that has this feature.

ACKNOWLEDGMENTS

We should like to thank Peter Lambert and Peter Vallentyne for valuable
comments. The usual disclaimer applies.

REFERENCES
Broome, J. (1991). Weighing goods. London: Basil Blackwell.
Fleurbaey, M. (1995). Three solutions for the compensation problem. Journal of Economic
Theory, 6, 96–106.
Harsanyi, J. C. (1987). Bayesian decision theory and utilitarian ethics. American Economic
Review, Papers and Proceedings, 68, 223–228.
Kolm, S.-C. (2001). Modern theories of justice. Cambridge, MA: MIT Press.
Mirrlees, J. A. (1971). An exploration in the theory of optimal taxation. Review of Economic
Studies, 38, 175–208.
Moulin, H., & Roemer, J. (1989). Public ownership of the external world and private ownership
of the self. Journal of Political Economy, 97, 347–367.
Nozick, R. (1974). Anarchy, state, and Utopia. New York: Basic Books.
Rawls, J. (1971). A theory of justice. Cambridge, MA: Harvard University Press.
GENERALIZED PROBABILISTIC
EGALITARIANISM

Paul D. Thistle

ABSTRACT

For over 60 years, Lerner’s (1944) probabilistic approach to the welfare

evaluation of income distributions has aroused controversy. Lerner’s famous
theorem is that, under ignorance regarding who has which utility function,
the optimal distribution of income is completely equal. However, Lerner’s
probabilistic approach can only be applied to compare distributions with
equal means when the number of possible utility functions equals the number
of individuals in the population. Lerner’s most controversial assumption that
each assignment of utility functions to individuals is equally likely. This
paper generalizes Lerner’s probabilistic approach to the welfare analysis of
income distributions by weakening the restrictions of utilitarian welfare,
equal means, equal numbers, and equal probabilities and a homogeneous
population. We show there is a tradeoff between invariance (measurability
and comparability) and the information about the assignment of utility
functions to individuals required to evaluate expected social welfare.

1. INTRODUCTION
For over 60 years, Lerner’s (1944) probabilistic approach to the welfare
evaluation of income distributions has aroused controversy. Lerner

Equity
Research on Economic Inequality, Volume 15, 7–32
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15002-X
7
8 PAUL D. THISTLE

considered the problem of a social planner who must optimally allocate a

fixed total amount of income across a given population. The social planner
knows that each individual has a utility function with diminishing marginal
utility of income, but does not know which individual has which utility
function. The planner’s objective is to maximize expected welfare, which
Lerner takes to be the sum of expected utilities. Lerner’s famous theorem is
that, under ignorance regarding who has which utility function, the optimal
distribution of income is completely equal.
Lerner’s theorem and its interpretation have been further analyzed by a
number of researchers. Breit and Culbertson (1970) extend Lerner’s two-
person diagrammatic proof to the n-person case. However, their argument
relies on the rather special assumption that each individual has a ‘‘twin’’ with
the same utility function. Sen (1969), McManus, Walton, and Coffey (1972),
and McCain (1972) provide rigorous proofs for the n-person case. These
researchers all assume the planner has an additive utilitarian social welfare
function (SWF). Sen (1969) and McManus et al. consider the ‘‘maximin’’
objective of maximizing the lowest value of welfare given the income
distribution. Sen (1973a, 1973b) considers general, non-utilitarian SWFs.
Bennett (1981) argues that Lerner’s assumption that all assignments of utility
functions to individuals are equally likely implies that the probability of gain
from an equalizing redistribution is greater than one-half.
Lerner’s theorem has also been criticized by a number of prominent
economists, for example, Friedman (1947), Graaff (1967), Little (1957),
Musgrave (1959), and Samuelson (1964).1 Freidman, Little, and Samuelson
are critical of the invariance (measurability and comparability) requirements
Lerner imposes on the SWF. Both Lerner and his critics exhibit some
confusion on this issue. However, most of the criticism has focused on the
‘‘equi-probability’’ assumption, the assumption that all possible assignments
of utility functions to individuals are equally likely. Lerner justifies equi-
probability by assuming the planner is completely ignorant about which
individual has which utility function. For example, Musgrave (p. 108) refers
to equi-probability as an ‘‘uneasy assumption,’’ and writes ‘‘The argument,
like the principle of insufficient reason, remains inconclusive.’’ Friedman
(p. 409) writes ‘‘The analysis as given is not rigorous, primarily because of
the appeal to ‘equal ignorance’.’’ Little (p. 59) is more harshly critical,
arguing that ‘‘y from complete ignorance nothing but complete ignorance
can follow y .’’ Similarly, Graaff (p. 100) writes ‘‘From absolute ignorance
we can derive nothing but absolute ignorance.’’
Despite the controversy, Lerner’s observation that it may not be known
with certainty which individual has which utility function remains an
Generalized Probabilistic Egalitarianism 9

important insight. Moreover, this insight has not been much appreciated in
recent research on the distribution of income. Most recent work has either
explicitly or implicitly assumed that each individual’s utility function is
known to the planner. For example, it is common to define the evaluation
function directly over individuals’ incomes (e.g., Bishop, Formby, & Thistle,
1991, 1992). This formulation implicitly assumes that utility functions are
known and that they are subsumed in the functional form of the welfare
function. It is also common to assume that all individuals have the same
known utility function (cf. Lambert, 2001, pp. 87–90).
However, there are important restrictions in Lerner’s probabilistic welfare
analysis that limit its practical application. The problem originally con-
sidered by Lerner, and by most subsequent analysts, is that of allocating a
fixed amount of income. Thus, Lerner’s probabilistic approach can only
be applied to compare alternative distributions that have equal means.
Secondly, Lerner and most subsequent analysts have studied the ‘‘equal
numbers’’ case, where the number of possible utility functions equals the
number of individuals in the population. The problem is that it is not known
which individual has which utility function.
The objective of this paper is to generalize Lerner’s probabilistic approach
to the welfare analysis of income distributions by weakening the restrictions
of equal means, equal numbers, and equal probabilities. Results are obtained
for general (non-utilitarian) SWFs, utilitarian SWFs, and for ordinally
comparable individual utility functions. Some researchers have examined
different aspects of this problem. Bishop et al. (1991) and Kakwani (1984)
provide results on comparing distributions with unequal means, under the
assumptions of equal numbers and equi-probability. Sen (1969) and McCain
(1972) consider unequal numbers and unequal probabilities under the
assumptions of a utilitarian welfare function and equal means. None of
these researchers has relaxed all of the restrictions simultaneously. We also
extend these results to the case where a heterogeneous population can be
classified into distinct subpopulations. Generalizing Lerner’s probabilistic
approach allows it to be applied to a much wider range of situations.
The ‘‘equi-probability’’ assumption has been the most controversial. It
seems reasonable to believe that the central planner may have some infor-
mation regarding the possible assignments of utility functions to individuals.
If so, then some assignments of utility functions to individuals are more
likely than others. The assumption that there are an equal number of utility
functions and individuals turns out to be an important restriction. It seems
plausible that the number of possible utility functions may be (much) larger
than the population size. Then, in addition to not knowing who has which
10 PAUL D. THISTLE

utility function, it is not known which utility functions are relevant. Further,
in order to weaken the restriction of equal probabilities, we first need
to relax the assumption of equal numbers. There is then a tradeoff between
the invariance requirements of the planner’s SWF and utility assignment
information used to evaluate expected social welfare. That is, SWFs with
stronger invariance requirements use less information about which individ-
ual has which utility function.
The next section reviews Lerner’s analysis and discusses the criticisms of it.
Section 3 relaxes the equal numbers and equi-probability assumptions, and
provides the key result of the paper. Section 4 discusses efficiency-equity pref-
erences. Section 5 considers the expected welfare evaluations. Section 6 exam-
ines heterogeneous populations. Section 7 provides brief concluding remarks.

2. PROBABILISTIC EGALITARIANISM

This section reviews Lerner’s analysis and the most prominent criticisms
of it. Some of the criticisms are valid, but some of the criticisms are due to
misunderstanding, while still others are flawed. One objective of this section
is to clear up at least some of the misunderstanding. It is also important
to understand the valid criticisms, as these provide an important motivation
for the generalization proposed here.
The basic idea of Lerner’s probabilistic egalitarianism can be easily
understood. Lerner is concerned with the problem of the socially optimal
distribution of income. First, Lerner assumes that individuals’ utilities
depend only on their own income, that utilities are cardinal and comparable,
that each individual has diminishing marginal utility of income, and that the
SWF is the sum of individual utilities. Lerner assumes that the total income
is independent of the division of income. These assumptions seem unre-
markable for an analysis of income distribution. Lerner’s most controversial
assumption is that the planner does not know which individual has which
utility function. Lerner (p. 29) concludes that ‘‘y the probable value of
total satisfactions is maximized by dividing income evenly.’’
The argument for the two-person case can be seen in Fig. 1, which shows
the distribution of income and the marginal utilities. Let us call these two
people Ann and Bob. The width of the box, 0102, is equal to Ann and Bob’s
total income. The planner is presumed to know that one individual has the
utility function u1 and the other has the utility function u2, each of which is
increasing and concave. Observe that the number of utility functions is equal
to the number of individuals. If Ann has marginal utility MU1 and Bob has
Generalized Probabilistic Egalitarianism 11

MU1
MU2

E
A C

B D F

01 x1 x x* x2 02

Fig. 1. Probabilistic Egalitarianism.

marginal utility MU2, then a given distribution of income can be represented

by the distances 01x1 and 02x1. If incomes are equalized at x̄; then Ann gains
the area A+B, while Bob loses the area B. There is a net gain in the sum of
utilities equal to A. If Bob has marginal utility MU1 and Ann has marginal
utility MU2, then the same distribution of income is represented by 01x2 and
02x2. If incomes are equalized, then Ann gains the area D+E+F, while
Bob loses the area C+D+F. There is a net change of E
C, which may be
positive or negative. The planner, being ignorant of which is the true
situation, regards each as equally likely and allocates income to maximize
expected welfare, which is the expected sum of utilities. Since each situation
is equally likely, the expected change is 1/2(A)+1/2(E
C) ¼ 1/2(A+E
C),
which is positive since diminishing marginal utility implies A>C. The plan-
ner acts as if Ann and Bob both have the same utility function, 1/2u1+1/2u2,
and divides the total income equally.
Lerner, and subsequent analysts, have argued that the probability that
equalizing income will yield a welfare gain is one-half, and that this is true
for all initial income distributions. Bennett (1981) points out that, if incomes
are initially very unequal, then a welfare gain is certain. This can also be seen
in Fig. 1. The essence of Bennett’s argument is that, if the initial distribution
is sufficiently unequal, then E is larger than C. If so, there is a gain from
redistribution regardless of which individual has which utility function. This
reinforces Lerner’s argument for equalization.
12 PAUL D. THISTLE

The criticisms of Lerner’s analysis have focused on the invariance

(measurability and comparability) requirements and the equi-probability
assumption.2 Lerner is responsible for some of the confusion about
his invariance requirements. He writes (p. 24) ‘‘But we have no way
of directly comparing the well-being of different consumers.’’ and later on
the same page ‘‘y of the impossibility of measuring the satisfactions of
different consumers on the same scale.’’ This seems a clear statement that
utilities are not comparable. But on the next page he assumes ‘‘y the
satisfactions of different people are similar in the sense that they are the
same kind of thing. In other words, that it is not meaningless to say that a
satisfaction that one individual gets is greater or lesser than a satisfaction
enjoyed by somebody else’’ (italics in original). Thus, Lerner seems to be
assuming at least ordinal level comparability of utilities. Lerner (p. 25)
argues that comparability is needed since ‘‘This assumption gives meaning
to the concept of maximizing the total of satisfactions experienced by all
individuals in a society.’’ Lerner thus adopts a utilitarian SWF, so that the
underlying individual utility functions must be cardinally measurable and
fully comparable.
The confusion over the invariance requirements is compounded by the
assumption of ignorance. Lerner (p. 28) writes ‘‘There is no way of discov-
ering with certainty whether any individual’s marginal utility of income is
greater than, equal to, or less than that of any other individual.’’ This can be
(and has been) interpreted as assuming noncomparability. Since he points
out that individuals have an incentive to misrepresent their preferences,
a more reasonable interpretation is that Lerner is referring to the practical
difficulties in inferring individuals’ marginal utilities of income. This is
analogous to the informational environment in adverse selection problems
where individuals’ utility functions are private information. As Bennett
(1981, p. 165) puts it ‘‘y we can assume that in a large economy the gov-
ernment would decide not to collect and process such information simply
because of the cost involved.’’
A third possible difficulty is pointed out by Breit and Culbertson. Lerner
initially states that an equal distribution of income will maximize probable
total satisfaction. This is the result that Lerner actually proves. But later,
Lerner (p. 32) asserts ‘‘y if it is desired to maximize the total satisfaction in
a society, the rational procedure is to divide income on an equalitarian
basis.’’ Breit and Culbertson interpret the omission of the word ‘‘probable’’
as a claim of a ‘‘strong theorem’’ that an equalizing redistribution increases
total welfare for every assignment of utility functions to individuals. Lerner
(1970) asserts that this is a misinterpretation.
Generalized Probabilistic Egalitarianism 13

Breit and Culbertson (p. 440) argue that the ‘‘strong theorem’’ also holds.
They assume that every individual has a twin with the same utility function,
and that the total income of every pair of twins is equal to twice the
population average.3 Different sets of twins in general have different utility
functions. Since each pair of twins starts with the same total income,
equalizing income for each pair yields a completely equal distribution of
income. Equalizing income for each pair of twins necessarily maximizes
total welfare for the pair, and, adding up across pairs of twins, for society
as a whole. As Lerner (1970) points out, in their quest to prove that a
welfare gain is certain, Breit and Culbertson miss the point of the original
analysis. Indeed, one way to interpret Bennett’s (1981) analysis is as showing
that welfare must increase only if the initial distribution of income is
sufficiently unequal.
Little is critical of Lerner’s assumptions of cardinal and comparable
utilities and a utilitarian welfare function. Little (pp. 51–52) argues that ‘‘The
picture of a lot of separate satisfactions, each with a label tied round it, is not
convincing.’’ and the ‘‘y thesis that one could add the welfare of different
individuals to arrive at the welfare of society y has now to be abandoned.’’
Little also interprets Lerner as assuming that utilities are not comparable,
then proceeding to add utilities. This leads Little (pp. 58–59) to regard
Lerner’s argument as ‘‘paradoxical’’ and to regard the assumption of equi-
probability as ‘‘illegitimate.’’ Little’s argument is apparently based on
the view that we should be able to verify empirically whether different in-
dividuals’ utilities are comparable. But Little’s view is, at best, idiosyncratic:
‘‘So long, however, as the ‘measurement’, or estimation of satisfaction is not
objective, and there is considerable room for disagreement, then it may
sometimes be useful to speak of a hedonistic calculus – to speak, that is, of
comparing satisfactions, and differences in satisfactions’’ (p. 34).
Graaff (p. 100) also objects to Lerner’s invariance requirements, arguing
that ‘‘His concept of social welfare differs from ours in that it is based on the
sum of cardinally measurable satisfactions, and not on observable choices.’’
Graaff (pp. 33–35) views individuals’ utility functions as ordinal indicators
of preferences. But Graaff himself adopts a welfare analysis based on
Bergson–Samuelson SWFs. He recognizes that these necessarily reflect
ethical judgments. He then argues (pp. 35–40) that (cardinal) measurability
is not necessary to define a welfare function, and appears to take the
position that Bergson–Samuelson welfare functions can be defined over
noncomparable individual utility functions. This, we now know, is incorrect;
utilities must be absolutely measurable and interpersonally comparable in
order to define a Bergson–Samuelson welfare function.4
14 PAUL D. THISTLE

Friedman’s criticism is focused on the assumption of ignorance and equi-

probability: ‘‘Eliminate the assumption of ignorance and the same analysis
immediately yields a justification for inequality y . And we must clearly
be prepared to eliminate the assumption of ignorance’’ (p. 411) since it
implies equality is a more fundamental goal than maximizing social welfare.
Friedman’s argument is that, if the planner knows that Ann has the utility
function u1, then the optimal distribution gives Ann 01x in Fig. 1. Lerner
(pp. 28–29) is well aware of this argument, but his view is that whether
Ann’s utility function is u1 or u2 is ‘‘y incapable of being discovered.’’
Friedman (p. 410) then provides an alternative argument for Lerner’s
result.5 Friedman replaces equi-probability with the assumption that a
person’s income is ‘‘y statistically independent of his capacity for enjoying
it y .’’ Friedman then argues that individuals can, at least conceptually, be
classified by their capacity for satisfaction. Statistical independence implies
the average income in each class will be the same, and redistribution across
classes would invalidate statistical independence. Freidman retains Lerner’s
remaining assumptions, which imply that incomes within utility classifica-
tions should be equalized. One problem with this argument is that it
invalidly employs a law of large numbers.6 Further, if incomes are correlated
with the capacity for satisfaction, then Friedman’s argument also justifies
inequality. The reason for this is that Friedman does not distinguish
between the ex ante and ex post distributions across types. The positive
statement that the ex ante distribution is equal (or unequal) does not imply
that the normative statement that the ex post distribution should be equal.7
Samuelson attempts to avoid comparing utility functions and to avoid the
use of a utilitarian welfare function. Samuelson’s solution is to reinterpret
the problem as one of choosing an income distribution from behind a
Rawlsian veil of ignorance. Each individual is assumed to evaluate income
distributions using their own utility function; individuals’ utility functions
are private information. He argues that, if every assignment of incomes
to individuals is equally likely, then individuals’ risk aversion will lead to
unanimous preference for complete equality. This is, of course, a different
model from the one considered by Lerner.
Musgrave (pp. 108–109) comes the closest to Lerner’s position. He writes
‘‘While we cannot assume that the utility schedules of individuals are
known, the new welfare economics may have gone too far in its categorical
rejection of interpersonal utility comparisons. Such comparisons are made
continuously, and in this sense have operational meaning.’’ Musgrave then
argues that the solution to this problem is to treat individuals as if they were
identical, i.e., chose a particular utility function and assume that it applies to
Generalized Probabilistic Egalitarianism 15

all individuals. Given a utilitarian welfare function, equi-probability yields

the same outcome, applying the average utility function to all individuals.8
Taken as a whole, Lerner’s critics make three points. First, they object to
the assumption of cardinal and interpersonally comparable utility functions.
Second, they object to the utilitarian SWF. Third, they object to the
assumption of ignorance and equi-probability.

3. CONDITIONAL AND UNCONDITIONAL

EQUI-PROBABILITY

This section begins to address some of the criticisms of Lerner’s analysis.

The requirement that the number of individuals and the number of utility
function must be equal is relaxed. This allows the equi-probability assump-
tion to be generalized.
The set N ¼ f1; . . . ; ng is the fixed population, where n is the population
size. An income vector x ¼ ðx1 ; . . . ; xn Þ is a listing of each individuals’
income, where xi is the income of the ith individual. The set of attainable
income distributions is X, a compact non-empty subset of Rnþ =f0g: Mean
income for x is denoted x̄:
Each individual’s utility is assumed to be a function of her income only.
The set of possible utility functions is U ¼ fu1 ; . . . ; um g; with index set
M ¼ f1; . . . ; mg: The number of possible utility functions may be larger than
the population size, mZn.9 This allows (but does not require) each indi-
vidual to have a distinct utility function. All utility functions are assumed to
be absolutely measurable and fully interpersonally comparable.
Income distributions are evaluated via the central planner’s SWF. To
begin we assume the SWF is a Bergson–Samuelson welfare function, and is a
continuous function of individuals’ utilities.10 Several types of SWFs will be
considered to reflect different judgments on the efficiency-equity tradeoff
that the planner might make. All of the SWFs are assumed to satisfy
anonymity, or to be symmetric in utilities.
Central to Lerner’s approach is the idea that the marginal utilities of
income differ across individuals. Since the planner knows each individual’s
income, a specific assignment of n of the m possible utility functions to
individuals is required to evaluate welfare. The evaluation of welfare can be
regarded as taking place in two stages. The first step is the determination
of which n of the m utility functions are required. The second step is the
determination of the assignment of those utility functions to individuals.
16 PAUL D. THISTLE

The planner’s utility assignment information is represented by a probability

distribution over possible assignments of utility functions to individuals.
Let c ¼ ðc1 ; . . . ; cn Þ be a combination of n indices from M; let C denote
the set of all such combinations. For combinations, the order in which the
indices occur is not important. A combination simply selects n of the m
possible utility functions. The same utility function may be selected more
than once or the utility functions may all be different. There are (m+n
1)!/
n!(m
1)! such combinations.11 Let qc be the probability that the combination
c identifies the correct set of utility functions.
Given a combination c, a permutation of the indices is pðcÞ ¼
ðpðc1 Þ; . . . ; pðcn ÞÞ: Let P(c) be the set of all permutations of c. Permutations
determine the assignment of specific utility functions to specific individuals,
and are distinguished by the order of the indices. For any combination of n
different items, there are n! permutations. However, if the utility function
u1 is included n1 times, then there are n!/n1! distinct permutations. More
generally, suppose the combination c includes S distinct indices and let ncs
be the multiplicity of index s in c. Then the number of distinct permutations
in c is n!/(Psncs!). Given a combination c, the conditional probability that
p(c) is the correct permutation is denoted rp(c).
Since the planner has only probabilistic utility assignment information,
income distributions are evaluated by their expected welfare, where the
expectation is taken over the possible assignments of utility functions to
individuals. For the SWF W, expected welfare given the income vector x is:
XX
EðW jxÞ ¼ W ðupðc1 Þ ðx1 Þ; . . . ; upðcn Þ ðxn ÞÞrpðcÞ qc (1)
C PðcÞ
P P
where C and P(c) denote the sums over all combinations in C and over all
permutations in P(c).
It is not assumed that all assignments of utility functions are equally likely.
In particular, the probabilities are allowed to vary across combinations. For
example, information on aggregate economic behavior might lead the planner
to believe some combinations of utility functions are more likely than others.
However, it is assumed that all permutations of a given combination are
equally likely.

Conditional equi-probability:
If qc>0, then rp(c) ¼ (Psncs!)/n!, 8pðcÞ 2 PðcÞ and 8c 2 C:

That is, for every combination of utility functions that can occur with non-
zero probability, each permutation of that combination is equally likely to
Generalized Probabilistic Egalitarianism 17

give the correct assignment of utility functions to individuals. Also, while all
permutations of a given combination are equally likely, those probabilities
depend on the combination. Two combinations c0 and c00 will, in general,
have different multiplicities, so that rp(c0 )6¼rp(c00 ).
The assumption of conditional equi-probability is weaker than the
assumption of equal unconditional probabilities made by Lerner, and which
has been so frequently criticized. The unconditional probability of a per-
mutation is rp(c)qc, so equi-probability requires that rp(c)qc ¼ rp0 (c0 )qc0 for
8pðcÞ 2 PðcÞ; 8p0 ðc0 Þ 2 Pðc0 Þ; and 8c; c0 2 C: To see this more clearly, let
m ¼ n ¼ 2. Then the possibilities are shown in the following table:

Combination Permutation

1 2

1 1, 1 –
2 1, 2 2, 1
3 2, 2 –

Then equi-probability requires that q1 ¼ 1/2q2, q3 ¼ 1/2q2 and q1 ¼ q3 so

that the unconditional probability of each distinct permutation is 1/4. Since
combination 2 has two different permutations, it is twice as likely as com-
bination 1 or 3 to be the correct combination.
Conditional equi-probability leads to the following result.
Proposition 1. Conditional equi-probability holds iff E(W|x) is symmetric
in incomes for all x.
Proof. It is enough to illustrate the proof for n ¼ 2 and m ¼ 3. The pos-
sible combinations of utility functions and their permutations are shown
in the table:

Combination Permutations

1 2

1 1, 1 –
2 1, 2 2, 1
3 1, 3 3, 1
4 2, 2 –
5 2, 3 3, 2
6 3, 3 –
18 PAUL D. THISTLE

Then

EðW jxÞ ¼ W ðu1 ðx1 Þ; u1 ðx2 ÞÞq1

þ ½W ðu1 ðx1 Þ; u2 ðx2 ÞÞr1ð2Þ þ W ðu2 ðx1 Þ; u1 ðx2 ÞÞr2ð2Þ q2
þ ½W ðu1 ðx1 Þ; u3 ðx2 ÞÞr1ð3Þ þ W ðu3 ðx1 Þ; u1 ðx2 ÞÞr2ð3Þ q3
þ W ðu2 ðx1 Þ; u2 ðx2 ÞÞq4
þ ½W ðu2 ðx1 Þ; u3 ðx2 ÞÞr1ð5Þ þ W ðu3 ðx1 Þ; u2 ðx2 ÞÞr2ð5Þ q5
þ W ðu3 ðx1 Þ; u3 ðx2 ÞÞq6 ð2Þ

Sufficiency follows from conditional equi-probability (r1(c) ¼ r2(c)) and

symmetry of W in utilities. To prove necessity, assume x16¼x2, and that
E(W|x) is symmetric in incomes. Then Eq. (2) can be written as:

EðW jxÞ ¼ W ðu1 ðx2 Þ; u1 ðx1 ÞÞq1

þ ½W ðu1 ðx2 Þ; u2 ðx1 ÞÞr1ð2Þ þ W ðu2 ðx2 Þ; u1 ðx1 ÞÞr2ð2Þ q2
þ ½W ðu1 ðx2 Þ; u3 ðx1 ÞÞr1ð3Þ þ W ðu3 ðx2 Þ; u1 ðx1 ÞÞr2ð3Þ q3
þ W ðu2 ðx2 Þ; u2 ðx1 ÞÞq4
þ ½W ðu2 ðx2 Þ; u3 ðx1 ÞÞr1ð5Þ þ W ðu3 ðx2 Þ; u2 ðx1 ÞÞr2ð5Þ q5
þ W ðu3 ðx2 Þ; u3 ðx1 ÞÞq6 ð3Þ

Using the fact that W is symmetric in utilities, subtracting Eq. (3) from Eq.
(2) yields:

½W ðu1 ðx1 Þ; u2 ðx2 ÞÞ
W ðu1 ðx2 Þ; u2 ðx1 ÞÞðr1ð2Þ
r2ð2Þ Þq2

þ ½W ðu1 ðx1 Þ; u3 ðx2 ÞÞ
W ðu1 ðx2 Þ; u3 ðx1 ÞÞðr1ð3Þ
r2ð3Þ Þq3
þ ½W ðu2 ðx1 Þ; u3 ðx2 ÞÞ
W ðu2 ðx2 Þ; u3 ðx1 ÞÞðr1ð5Þ
r2ð5Þ Þq5 ¼ 0 ð4Þ

Since x16¼x2, each expression in brackets is non-zero. If r1(c)6¼r2(c) for any

combination c for which qc>0, then E(W|x)
E(W|x)6¼0. Therefore,
conditional equi-probability must hold. &
Sen (1969, 1973a, 1973b) and McCain (1972) also consider the possibility
that not all assignments of utility functions are equally likely. Sen and
McCain both assume that the unconditional probability that individual i’s
utility function is uj is independent of the individual. Let sij be the
unconditional probability that individual i’s utility function is uj.

Unconditional equi-probability:
sij ¼ sj, 8i 2 N:
Generalized Probabilistic Egalitarianism 19

Unconditional equi-probability is more general than the conditional

equi-probability assumption. Let CðjÞ ¼ fcjj 2 cg be the set of combinations
that select the utility function uj. For any c in C(j), there are n permutations
that assign the utility functionP uj to the ith individual. Then conditional
equi-probability implies sij ¼ cAC(j)qc/n, which is free of i.12

Proposition 2. Assume W is utilitarian. Unconditional equi-probability

holds iff E(W|x) is symmetric in incomes for all x.

Proof. Expected welfare is

XX
EðW jxÞ ¼ sij uj ðxi Þ (5)
j2M i2N

To proveP sufficiency, assume unconditional

P equi-probability and let
ūðxÞ ¼ j2M sj uj ðxÞ: Then EðW jxÞ ¼ i2N ūðxi Þ; which is symmetric in
incomes. Necessity is proved for the case m ¼ 3 and n ¼ 2; the argument
is essentially the same as in Proposition 1. Assume x16¼x2, and that
E(W|x) is symmetric in incomes. Interchange incomes and subtract to
obtain:
ðs11
s21 Þ½u1 ðx1 Þ
u1 ðx2 Þ þ ðs12
s22 Þ½u2 ðx1 Þ
u2 ðx2 Þ
þðs13
s23 Þ½u3 ðx1 Þ
u3 ðx2 Þ ð6Þ

Since x16¼x2, each expression in brackets is non-zero. If s1j6¼s2j, then

E(W|x)
E(W|x)6¼0. Therefore, unconditional equi-probability must
hold. &

That is, for a utilitarian SWF, unconditional equi-probability is equivalent

to treating each individual as if she had the average utility function, ūðxÞ:
In his analysis of general, non-utilitarian SWFs, Sen (1973a, 1973b)
assumes that each reassignment of utility functions is accompanied by a
corresponding reassignment of incomes. That is, the social planner is
assumed to know that income xj always belongs to the individual with the
utility function uj, but, in effect, does not know that individual’s name. But
symmetry of the SWF implies that individuals’ names are not relevant for
the evaluation of the income distribution. Consequently, the unconditional
probability with which individual i has the utility function uj (and income xj)
does not affect welfare, and can be left arbitrary. Lerner, and most sub-
sequent analysts, assume that the social planner knows individuals’ names
and incomes, but not their utility functions.
20 PAUL D. THISTLE

4. EFFICIENCY AND EQUITY PREFERENCE

If the SWF is consistent with the Pareto principle, then it is increasing in

utilities. Let W1 denote the set of all functions that are continuous, increas-
ing, and symmetric in their arguments. If the SWF also satisfies the principle
of transfers, then it is S-concave (Dalton, 1920; Dasgupta, Sen, & Starrett,
1973). Let W2 be the class of all continuous, increasing, S-concave functions.
Obviously, W 2  W 1 :
We will make use of the following partial orders on X. Let x~ ¼ ðx~ 1 ; . . . ; x~ n Þ
be the ordered version of x, where x~ 1  x~ 2  . . .  x~ n : Then, for any x, yAX,
x rank dominates y, denoted x4R y; iff x~ i  y~ i ; 8i 2 N; with Pat least one strict
inequality. The Lorenz curve for x is Lx ðk=nÞ ¼ ð1=nx̄Þ ki¼1 x~ i : Following
Shorrocks (1983) and Kakwani (1984), the generalized Lorenz curve for x is
the Lorenz
P curve, scaled up by mean income, G x ðk=nÞ ¼ x̄Lx ðk=nÞ ¼
ð1=nÞ ki¼1 x~ i ; k 2 N: Then x generalized Lorenz dominates y, denote x>G y,
iff Gx(k/n)ZGy(k/n), 8k 2 N; with a strict inequality for some k. That is, the
generalized Lorenz curve for x must lie nowhere below the generalized Lorenz
curve for y, and strictly above it at some point. Rank dominance implies
generalized Lorenz dominance, and a higher mean together with Lorenz
dominance implies generalized Lorenz dominance; neither converse is true.

5. EXPECTED WELFARE COMPARISONS

Following Lerner, the central planner faces the problem of evaluating

and comparing alternative income distributions and selecting the socially
optimal distribution. Since the planner does not know who has which utility
function, the planner’s objective becomes maximization of expected welfare.
Expected welfare maximization yields different objective criteria for com-
paring income distributions depending on the characteristics of individuals’
utility functions and the planner’s SWF.
We will make use of the following result.
Lemma 1.
(a) x4R y iff V(x)>V(y), 8V 2 W 1 :
(b) x4G y iff V(x)>V(y), 8V 2 W 2 :

Part (a) of the lemma states that rank dominance, x4R y; is equivalent to
unanimous preference for x by every welfare function that is consistent with
the Pareto principle. Part (b) states that generalized Lorenz dominance,
Generalized Probabilistic Egalitarianism 21

x4G y; is equivalent to unanimous preference for x by every welfare function

that is consistent with the Pareto principle and the principle of transfers. Part
(a) is proved in Saposnik (1981, 1983) and Thistle (1989). Part (b) is proved
in Shorrocks (1983) and Kakwani (1984), extending Atkinson’s (1970)
seminal result.13 The rank dominance and generalized Lorenz dominance
criteria are equivalent to first- and second-degree stochastic dominance; see
Foster and Shorrocks (1988a, 1988b), Kakwani (1984), and Thistle (1989).
Suppose the planner’s SWF is Paretian and anonymous. Then the ap-
propriate criterion for comparing income distributions is rank dominance.
If the planner’s SWF is also S-concave in utilities, then the appropriate
criterion is generalized Lorenz dominance. Let U1 be the set of all contin-
uous, increasing utility functions, and let U2 be the set of all continuous,
increasing, concave utility functions.

Proposition 3. Assume conditional equi-probability holds.

(a) If U  U1, then E(W|x)>E(W|y), 8W 2 W 1 iff x4R y:
(b) If U  U2, then E(W|x)>E(W|y), 8W 2 W 2 iff x4G y:

Proof. By Proposition 1, E(W|x) is symmetric in incomes. For (a), since

U  U1, and W is increasing in utilities, E(W|x) is increasing in incomes,
and the conclusion then follows from Lemma 1(a). For (b), since U  U 2
and W is increasing and S-concave in utilities, E(W|x) is increasing and
S-concave in incomes. The conclusion then follows from Lemma 1(b). &

Part (a) of Proposition 3 generalizes the result obtained by Bishop et al.

(1991) under the assumptions of equal numbers, equi-probability, and
a utilitarian SWF. The rank dominance criterion does not incorporate
inequality aversion. If the planner’s SWF also satisfies the principle of
transfers, then the appropriate criterion is generalized Lorenz dominance.
Part (b) of Proposition 3 generalizes the result of Kakwani (1984) under the
assumptions of equal numbers and equi-probability. Parts (a) and (b) of the
proposition allow more general interpretations of first- and second-degree
stochastic dominance, which are usually discussed in terms of a single fixed
and known utility function.
Much of the controversy surrounding Lerner’s approach has focused
on his conclusion that an egalitarian income distribution is optimal. Let
eðxÞ ¼ ðx̄; . . . ; x̄Þ be the n-vector with all incomes equal to the mean of x.

Proposition 4. If conditional equi-probability holds and U  U 2 ; then

E(W|e(x))>E(W|x), 8x6¼e(x) and 8W 2 W 2 :
22 PAUL D. THISTLE

Proof. Since eðxÞ4G x; the results follow directly from Proposition

3(b). &

This extends the results of Sen (1969, 1973a, 1973b). Sen (1973a, 1973b) as-
sumes equal numbers, which, under his assumptions, implies equi-probability.
Sen (1969) employs a more general assumption on the probabilities, but at the
cost of assuming a utilitarian welfare function.
Propositions 3 and 4 are based on the assumption that individuals’
utilities are absolutely measurable and fully comparable, a stronger
assumption than made by Lerner. These results clearly do not address
Graaff ’s, Little’s, and Samuelson’s concerns on this point. We have a sim-
ilar result for utilitarian SWFs.

Proposition 5. Assume W is utilitarian and unconditional equi-probability

holds.
(a) E(W|x)>E(W|y), 8U  U1 iff x4R y:
(b) E(W|x)>E(W|y), 8U  U2 iff x4G y:
(c) E(W|e(x))>E(W|x), 8x6¼e(x) and 8U  U2.

P
Proof. We have EðW jxÞ ¼ i2N ūðxi Þ: Parts (a) and (b) follow directly
from the equivalence of rank dominance with first-degree stochastic dom-
inance and generalized Lorenz dominance with second-degree stochastic
dominance. Since eðxÞ4G x; part (c) follows directly from part (b). &

If the invariance requirements are strengthened to ordinal level compa-

rability, then the social welfare ordering must be a lexicographic positional
dictatorship (LPD) (Gevers, 1979).14 That is, fix some assignment of utility
functions to individuals. Let u~ 1  u~ 2  :::  u~ n be the utility levels, ranked
from lowest to highest, when the income distribution is x. Similarly,
let u^ 1  u^ 2  :::  u^ n be the ranked utility levels under distribution y. Let p
be a permutation on f1; . . . ; ng; each permutation defines an LPD.15 Then x
is preferred to y by the LPD defined by p, denoted x4p y; if u~ pðiÞ ¼ u^ pðiÞ ;
i ¼ 1; :::; k
1 and u~ pðkÞ 4u^ pðkÞ : In the case where utility functions are known,
rank dominance is equivalent to unanimity among the LPDs (Thistle, 1998).
This leads to the following strong result. In particular, no restrictions on the
probabilities are necessary.

Proposition 6. Let utilities in U  U1 be ordinally comparable, and let

x4R y: Then x4p y for all permutations p with certainty.
Generalized Probabilistic Egalitarianism 23

Proof. Let p be the permutation that defines the LPD. Then:

upðc1 Þ ðx~ 1 Þ  upðc1 Þ ðy~ 1 Þ; upðc2 Þ ðx~ 2 Þ  upðc2 Þ ðy~ 2 Þ; . . . ; upðcn Þ ðx~ n Þ  upðcn Þ ðy~ n Þ (7)

with at least one strict inequality. This inequality holds for every assign-
ment of utility functions to individuals that has non-zero probability.
Therefore, x4p y with certainty. &

Rank dominance is sufficient, but not necessary, for an increase in welfare

to occur with certainty. Since the possible social welfare orders are narrowly
proscribed, the restrictions on the probabilities can be weakened. Moreover,
since the choice of p is arbitrary, the conclusion holds for all LPDs. It
should be pointed out, however, that x may be a more unequal distribution
than y, so that Proposition 6 is not an egalitarian result.

6. HETEROGENEOUS POPULATIONS

Friedman (1947, p. 410), in his critique of Lerner, argues that equi-probability

should be replaced by the assumption that a person’s income is ‘‘statistically
independent of his capacity for enjoying it.’’ Friedman then argues that in-
dividuals can be classified by their capacity for satisfaction: ‘‘Conceptually
classify the individuals by their (unknown) capacities for satisfaction. Each
such ‘satisfaction class’ will contain only individuals who have identical
capacities, i.e., have identical utility functions.’’ (p. 410, note 6). As Lambert
(2001, pp. 93–94) suggests, one possible way to classify individuals is
according to ‘‘needs,’’ that is, individuals with different needs have different
capacities for satisfaction.
More generally, individuals can be classified according to any character-
istic such that each group has a different set of possible utility functions or a
different probability distribution over the set of possible utility functions.
The characteristic used to classify individuals may or may not be relevant
for policy. If the characteristic is relevant for policy, it may or may not be
desirable to rank groups based on the characteristic.16 But so long as the
characteristic affects the evaluation of welfare, we need to analyze the joint
distribution of the characteristic and income. As Atkinson and Bourguignon
(1987) show, this leads to either group-wise or sequential procedures for
evaluation of income distributions. Atkinson and Bourguignon and the
subsequent literature assume, as in Friedman, that all individuals in a group
have the same known utility function.
24 PAUL D. THISTLE

To classify individuals according to their ‘‘capacity for satisfaction,’’ let

Nk be the set of nk individuals in group k. Then N ¼ fN 1 ; . . . ; N g g is a
partition of the population N into gZ2 groups, and G ¼ f1; . . . ; gg indexes
groups. Let yk ¼ nk/n be the population proportion of group k. We take
the composition of the population as fixed throughout the analysis.17 Let
U k  U be the set of mkZnk possible utility function for the group Nk, with
corresponding index set Mk. The sets Uk are not necessarily disjoint. That is,
we allow for the possibility that there may be individuals in different groups
that have the same utility function.
To simplify the analysis, we assume the welfare function is utilitarian and
evaluate welfare on a per capita basis. Since the planner distinguishes
between groups, the welfare function will not be symmetric for the full
population. However, there is no reason to treat different individuals within
the same group differently. The welfare function is partially symmetric in
incomes with respect to the partition N if i; j 2 N k implies that inter-
changing xi and xj does not affect welfare. As Cowell (1980, p. 523) puts it:
‘‘We have anonymity within groups but not between them.’’18
We need to modify the unconditional equi-probability condition to apply
to groups. Let skij be the unconditional probability that individual i’s utility
function is uj, where iANk and jAMk.
Group-wise unconditional equi-probability:
sijk ¼ sjk, 8i 2 N k ; 8j 2 M k ; and 8k 2 G:
That is, unconditional equi-probability holds within each group. This allows
the probabilities to differ across groups, that is, jAMh, Mk does not imply
shj ¼ skj :
This leads to the analog of Proposition 2 for utilitarian welfare functions.

Proposition 7. Assume W is utilitarian. Group-wise unconditional equi-

probability holds iff E(W|x) is partially symmetric in incomes with respect
to N for all x.
P P
Proof. Define ūk ðxÞ ¼ j2M k skij ukj ðxÞ: Then, by Proposition 2, i2N k ūk ðxi Þ
is symmetric in incomes if and only if, unconditional P equi-probability
P
holds for group k. Expected welfare is EðW jxÞ ¼ k2G yk i2N k ūk ðxi Þ;
which is partially symmetric in incomes with respect to N if, and only if,
unconditional equi-probability holds for every group, that is, group-wise
unconditional equi-probability holds. &

For a utilitarian SWF, group-wise unconditional equi-probability is equiv-

alent to treating each individual in Nk as if she had the utility function
Generalized Probabilistic Egalitarianism 25

ūk ðxÞ: Since the utility functions differ across groups, expected welfare is
partially symmetric in incomes.
Let xk be the vector of income for the individuals in group Nk. We say x
group-wise rank dominates y, denoted x4GR y; if x rank dominates y for
each group, xk 4R yk ; k ¼ 1; . . . ; g: We say x group-wise generalized Lorenz
dominates y, denoted x4GG y; if x generalize Lorenz dominates y for each
group, xk 4G yk ; k ¼ 1; . . . ; g:
Proposition 8. Assume W is utilitarian and group-wise unconditional
equi-probability holds.
(a) E(W|x)>E(W|y), 8U  U1 iff x4GR y:
(b) E(W|x)>E(W|y), 8U  U2 iff x4GG y:
(c) E(W|e(x))>E(W|x), 8x6¼e(x) and 8U  U2.

Proof. Parts (a) and (b) follow from Proposition 7 and parts 1 and 2,
respectively, of Proposition 1 in Atkinson and Bourguignon (1987).
(c) Let ek ðxk Þ ¼ ðx̄k ; . . . ; x̄k Þ be the nk-vector with all elements equal to the
mean of xk. The assumption P x6¼e(x) implies the group means x̄k cannot all
k k k
be equal. Let v ðxÞ ¼ i2NP k ū ðxÞ and observe that U  U 2 implies v is
k k k
concave. We have v ðx̄ Þ P i2N k ū ðxi Þ by part (c) of Proposition 5. Now
EðW je1 ðx1 Þ; . . . ; eg ðxg ÞÞ ¼ k2G yk vk ðx̄k Þ is concave in its arguments.
Consequently, EðW jeðxÞÞ4EðW je1 ðx1 Þ; . . . ; eg ðxg ÞÞ  EðW jxÞ: &
Proposition 8 can be applied whenever the groups are not ranked. This
extends Proposition 5 to heterogeneous populations. Alternatively, this result
extends Proposition 1 in Atkinson and Bourguignon. Finally, part (c) corrects
the problems with Friedman’s (1947) alternative proof of Lerner’s result.
We now turn to the case where the groups can be ranked on the basis of
needs as in Atkinson and Bourguignon (1987).19 We index the groups in
decreasing order of needs, so that group 1 has the greatest needs and group g
is the group with the lowest needs. The ranking of groups based on needs
imposes restrictions on the planner’s evaluation of social welfare, which in
turn impose restrictions on individuals’ utility functions.
The first restriction is that the marginal social valuation of income is positive
for each group and lower for groups with lower needs. Let sk ¼ ðsk1 ; . . . ; skmk Þ:
In terms of the ūk ; this restriction can be written as:
Needs 1.
0
(a) ūk ðxÞ40 for all x, for all sk and 8k 2 G:
0 0 0
(b) ū1 ðxÞ  ū2 ðxÞ  . . .  ūg ðxÞ40 for all x and all (s1,y, sg).
26 PAUL D. THISTLE

That is, the expected value of the marginal utility of income is positive for all
groups and is greater for groups with greater needs at all incomes. Further,
this must be true whatever the planner’s information about the assignment of
utility functions to individuals within groups. Part (a) is the Pareto principle.
Part (b) implies that a transfer from a richer individual in a group with lower
needs to a poorer individual in a group with greater needs increases welfare.
The second restriction is that the marginal social valuation of income is
decreasing in income for each group and that the between-group difference
in marginal social valuation is decreasing in income. This restriction can be
written as:
Needs 2.
00
(a) ūk ðxÞo0 for all sk and k ¼ 1, y, g.
00 00 00
(b) ū1 ðxÞ  ū2 ðxÞ  . . .  ūg ðxÞ for all x and all (s1, y, sg).

Both parts of this condition can be viewed as transfer principles. Part (a)
implies that transfers from richer to poorer individuals with the same group
increase welfare. Part (b) implies that a progressive transfer within a group
with greater needs increases welfare by more than the same transfer within a
group with lower needs. Finally, the differences among groups become less
important at higher income levels.
These conditions impose certain restrictions on the sets of possible utility
functions.
Condition MU. Let hok. Then i 2 M h ; j 2 M k implies u0i ðxÞ  u0j ðxÞ for
all x.
That is, every possible utility function in Uh has higher marginal utility of
income than every possible utility function in Uk.
Condition DMU. Let hok. Then i 2 M h ; j 2 M k implies u00i ðxÞ  u00j ðxÞ for
all x.
That is, marginal utility falls more rapidly for every possible utility func-
tion in Uh than for every possible utility function in Uk. Together, Con-
ditions MU and DMU imply that, for i 2 M h and j 2 M k ; the difference
u0i ðxÞ
u0j ðxÞ is non-negative and non-increasing in income.
This leads to the following result:
Proposition 9. Let hok.
0 0
(a) Condition MU holds iff ūh ðxÞ  ūk ðxÞ for all x and all sh, sk.
00 00
(b) Condition DMU holds iff ūh ðxÞ  ūk ðxÞ for all x and all sh, sk.
Generalized Probabilistic Egalitarianism 27

Proof. (a) Sufficiency is obvious. To see necessity, suppose that Condition

MU does not hold. Then there are indices i0 2 M h and j 0 2 M k such that
u0i0 ðx0 Þou0j0 ðx0 Þ for some income level x0. Letting shi0 ¼ 1 and let skj0 ¼ 1; we
0 0
have ūh ðx0 Þoūk ðx0 Þ: Part (b) follows from the same argument. &
Conditions MU and DMU impose strong restrictions on the sets of possible
utility functions. But recall that Atkinson and Bourguignon assume all
individuals in a group have a common known utility function. The groups
are then ranked according to the characteristics of their utility functions
based on the conditions Needs 1 and Needs 2. Under Lerner’s probabilistic
approach only the set of possible utility functions is known. Since individ-
uals can have any of the possible utility functions, these conditions must
hold for the sets of possible utility functions in order to rank groups based
on the characteristics embodied in Needs 1 and Needs 2. Assuming positive
and diminishing marginal utility, the Conditions MU and DMU are equiv-
alent to assuming that Needs 1 and Needs 2 hold for the sets of possible
utility functions. Thus, these restrictions are no stronger than those assumed
by Atkinson and Bourguignon.
Now let x1þþk ¼ ðx1 ; . . . ; xk Þ be the vector of incomes for individuals in
groups 1 though k. We say x sequentially rank dominates y, denoted x4SR y
if x1þþk 4R y1þþk for k ¼ 1; . . . ; g: Similarly, x sequentially generalized
Lorenz dominates y, denoted x4SG y if x1þþk 4G y1þþk for k ¼ 1; . . . ; g:
Since x1þþg ¼ x; sequential rank dominance and sequential generalized
Lorenz dominance imply (ordinary) rank dominance and generalized Lorenz
dominance. Also, group-wise rank dominance and generalized Lorenz dom-
inance imply sequential rank dominance and generalized Lorenz dominance.
Neither converse is true.
Proposition 10. Assume W is utilitarian and that group-wise unconditional
equi-probability holds.
(a) If Condition MU holds, then E(W|x)>E(W|y), 8U  U 1 iff x4SR y:
(b) If Conditions MU and DMU hold, then E(W|x)>E(W|y), 8U  U 2
iff x4SG y:

Proof. The results follow from Proposition 9 and Propositions 2 and 3,

respectively, in Atkinson and Bourguignon. &
This generalizes Proposition 3 to the case of a heterogeneous population
where population sub-groups can be ranked based on needs. Proposition 10
also allows a more general interpretation of sequential rank dominance
and sequential generalized Lorenz dominance, which, following Atkinson
28 PAUL D. THISTLE

and Bourguignon, are generally discussed in terms of common and known

utility functions for the members of each population group.

7. CONCLUSIONS

Lerner’s probabilistic approach to the welfare evaluation of income distri-

butions has aroused comment and controversy for over 60 years. Lerner
assumed that it is not known which individual has which utility function and
that all assignments of utility functions to individuals are equally likely.
Lerner then shows that an equalitarian division of income maximizes the
probable total satisfaction of society. Most of the criticism of Lerner’s
analysis has focused on the ‘‘equi-probability’’ assumption. Lerner’s obser-
vation that individuals’ preferences may not be known with certainty
remains an important insight.
This paper extends Lerner’s probabilistic approach to the welfare analysis of
income distributions. The restrictions of a utilitarian welfare function, equal
mean incomes, equal numbers, and equal probabilities are weakened
or eliminated. There is a tradeoff between the invariance requirements of
the planner’s SWF and the utility assignment information. This paper intro-
duces the assumption of conditional equi-probability, which requires all
permutations of a given combination to be equally likely, but allows prob-
abilities to vary across different combinations of utility functions. The general
Bergson–Samuelson welfare function has the weakest invariance requirement,
absolute measurability. For Bergson–Samuelson SWFs defined over utilities,
conditional equi-probability is equivalent to symmetry of expected welfare in
incomes. If conditional equi-probability holds, then rank dominance and
generalized Lorenz dominance can be applied as probabilistic welfare criteria.
The utilitarian welfare function is invariant to linear transformations of the
utility functions, a stronger invariance requirement. For utilitarian SWFs,
unconditional equi-probability is equivalent to symmetry in incomes. If
unconditional equi-probability holds, then rank dominance and generalized
Lorenz dominance can be applied as probabilistic welfare criteria. The LPD
has the strongest invariance requirement, and allows the probabilities to be
unrestricted. If the invariance requirement is strengthened to ordinal compa-
rability, then rank dominance implies greater welfare with certainty.
The tradeoff only becomes apparent when the number of possible utility
functions exceeds the number of individuals. In the equal numbers case,
conditional equi-probability and unconditional equi-probability both reduce
to Lerner’s assumption that all assignments of utility functions to individuals
Generalized Probabilistic Egalitarianism 29

are equally likely. Even Lerner’s equi-probability assumption is weaker than

the usual assumption that the assignment of utility functions to individuals is
known with certainty.
This paper also extends Lerner’s probabilistic approach to welfare analysis
of income distributions to heterogeneous populations. The assumptions
of equal mean incomes, equal number, and equi-probability are relaxed, but
the assumption of a utilitarian welfare function is retained. The unconditional
equi-probability condition, applied to groups within the population, is equiv-
alent to partial symmetry in incomes. That is, expected welfare is anonymous
within groups but not between groups. If group-wise unconditional equi-
probability holds, then group-wise rank dominance and generalized Lorenz
dominance can be applied as probabilistic welfare criteria. Under additional
conditions on the possible utility functions, the groups can be ranked
according to ‘‘needs.’’ Then sequential rank dominance and sequential
generalized Lorenz dominance can be applied as probabilistic welfare criteria.

NOTES
1. Meade (1945) reviews Lerner’s book, but focuses primarily on other issues.
Scitovsky (1971) also accepts Lerner’s analysis with little comment.
2. Boadway and Bruce (1984, pp. 162–163) provide a clear and convenient
summary of invariance requirements and social welfare orderings.
3. Curiously, Breit and Culbertson (p. 440, note 9) criticize Friedman (1947) for
assuming that ‘‘each individual faces another with precisely the same utility function.’’
4. See Sen (1970, 1977) and Boadway and Bruce (1984, Chapter 5).
5. Musgrave (p. 107) repeats Friedman’s argument without comment. Scitovsky
(p. 287, note 1) regards it as correcting an error in Lerner’s proof.
6. To see this, consider the two-person economy in Fig. 1. Friedman’s argument is
that statistical independence implies both Ann and Bob receive the same income.
Samuelson (p. 175) makes essentially the same point.
7. Similarly, Breit and Culbertson assume that each pair of utility twins has the
same total income, but do not prove that they should have the same total income.
8. The same point is made in Lambert (2001, pp. 91–92).
9. As McCain (1972, p. 499) points out, the assumption of a finite set of possible
utility functions is restrictive. However, the results obtained here hold for any
arbitrary finite set of utility functions, hence, by a well-known theorem of analysis,
can be extended to compact sets of utility functions.
10. The use of a Bergson–Samuelson social welfare function requires that utilities
be absolutely measurable and fully interpersonally comparable.
11. The problem of selecting the combination of utility functions can be thought
of as a problem of drawing a sample of size n from a set of m distinct objects, with
replacement and without regard to order. See Chung (1974, pp. 50–52). Sampling
with replacement allows the same utility function to be chosen multiple times.
30 PAUL D. THISTLE

12. Under unconditional equi-probability, the probability that the combination

c is correct is qc ¼ Sj2c sj :
13. This result is anticipated in Kolm (1969).
14. That is, the social welfare order is invariant to any common monotonic
(not necessarily linear) transformation of utilities.
15. The best known LPD is Sen’s (1970) ‘‘lexmin’’ order, which begins with the
worst off individual, proceeds to the next worst off in the event of a tie, etc. Lexmin is
the special case p(i) ¼ i.
16. For example, the planner may know that individuals in different geographic
areas or in different demographic groups have different ‘‘capacities for satisfaction’’
but may not consider the geographic or demographic classification important for
policy. In particular, the planner may not be willing to rank groups based on the
geographic or demographic classification.
17. See Jenkins and Lambert (1993) for an analysis of the case where the marginal
distribution of needs is not constant.
18. The definition of partial symmetry here is slightly different from that in Cowell
(1980). In the analysis here, the planner’s information or differences in needs among
groups within the population determines a specific partition of the population. Thus,
for our purposes, we want expected welfare to be partially symmetric with respect to
that specific partition. Cowell’s definition requires that the welfare function be
partially symmetric with respect to some partition of the population.
19. Bourguignon (1989) discusses the relationship between individual utility func-
tions and household utility functions. Ebert (2000) discusses the axioms underlying
Atkinson and Bourguignon’s analysis. Ok and Lambert (1999) extend Atkinson and
Bourguignon’s analysis to non-utilitarian welfare functions. See Lambert (2001,
pp. 72–77) for a lucid discussion of sequential generalized Lorenz dominance.

ACKNOWLEDGMENTS

I would like to thank Annette Brown, John Formby, Rubin Saposnik, and
especially John Bishop and Peter Lambert for helpful discussions and com-
ments on earlier versions of this paper. Much of the paper was completed
while I was on the faculty of Western Michigan University. I retain the
responsibility for any remaining errors.

REFERENCES
Atkinson, A. B. (1970). On the measurement of economic inequality. Journal of Economic
Theory, 2, 244–267.
Atkinson, A. B., & Bourguignon, F. (1987). Income distribution and differences in needs. In:
G. R. Fiewel (Ed.), Arrow and the foundations of economic theory. New York: New York
University Press.
Generalized Probabilistic Egalitarianism 31

Bennett, J. (1981). The probable gain from egalitarian redistribution. Oxford Economic Papers,
33, 165–169.
Bishop, J. A., Formby, J. P., & Thistle, P. D. (1991). Rank dominance and international
comparisons of income distributions. European Economic Review, 35, 1399–1409.
Bishop, J. A., Formby, J. P., & Thistle, P. D. (1992). Convergence of the south and non-south
income distributions, 1969–1979. American Economic Review, 82, 262–272.
Boadway, R., & Bruce, N. (1984). Welfare economics. Oxford: Basil Blackwell.
Bourguignon, F. (1989). Family size and social utility: Income distribution dominance criteria.
Journal of Econometrics, 42, 67–80.
Breit, W., & Culbertson, W. P. (1970). Distributional equity and aggregate utility: Comment.
American Economic Review, 60, 435–441.
Chung, K. L. (1974). Elementary probability theory with stochastic processes. New York:
Springer-Verlag.
Cowell, F. A. (1980). On the structure of additive inequality measures. Review of Economic
Studies, 47, 521–531.
Dalton, H. (1920). The measurement of the inequality of incomes. Economic Journal, 30,
348–361.
Dasgupta, P., Sen, A., & Starrett, D. (1973). Notes on the measurement of inequality. Journal of
Economic Theory, 6, 180–187.
Ebert, U. (2000). Sequential generalized Lorenz dominance and transfer principles. Bulletin of
Economic Research, 52, 113–122.
Foster, J., & Shorrocks, A. F. (1988a). Poverty orderings. Econometrica, 56, 173–177.
Foster, J., & Shorrocks, A. F. (1988b). Poverty orderings and social welfare. Social Choice and
Welfare, 5, 179–198.
Friedman, M. (1953). Essays in positive economics. Chicago: University of Chicago Press.
Gevers, L. (1979). On interpersonal comparability and social welfare orderings. Econometrica,
47, 75–89.
Graaff, J. (1967). Theoretical welfare economics (2e). Cambridge: Cambridge University Press.
Jenkins, S. P., & Lambert, P. J. (1993). Ranking income distributions when needs differ. Review
of Income and Wealth, 39, 337–356.
Kakwani, N. (1984). Welfare rankings of income distributions. In: R. L. Basmann, & G. F.
Rhodes, Jr. (Eds), Advances in econometrics (Vol. 3). Greenwich: JAI Press.
Kolm, S.-C. (1969). The optimal production of social justice. In: J. Margolis, & H. Guitton
(Eds), Public economics: An analysis of public production and consumption and their
relations to the private sectors. London: Macmillan.
Lambert, P. (2001). The distribution and redistribution of income (3e). Manchester: Manchester
University Press.
Lerner, A. P. (1944). The economics of control. New York: Macmillan.
Lerner, A. P. (1970). Distributional equality and aggregate utility: Reply. American Economic
Review, 60, 442–443.
Little, I. M. D. (1957). A critique of welfare economics (2e). Oxford: Oxford University Press.
McCain, R. (1972). Distributional equity and aggregate utility: Further comment. American
Economic Review, 62, 497–500.
McManus, M., Walton, G., & Coffey, R. B. (1972). Distributional equity and aggregate utility:
Further comment. American Economic Review, 62, 489–496.
Meade, J. E. (1945). Mr. Lerner on ‘The Economics of Control’. Economic Journal, 55, 47–69.
Musgrave, R. (1959). The theory of public finance. New York: McGraw-Hill.
32 PAUL D. THISTLE

Ok, E. A., & Lambert, P. J. (1999). On evaluating social welfare by sequential generalized
Lorenz dominance. Economics Letters, 63, 45–53.
Samuelson, P. A. (1964). Abba Lerner at sixty. Review of Economic Studies, 31, 169–178.
Saposnik, R. (1981). Rank dominance in income distribution. Public Choice, 36, 147–151.
Saposnik, R. (1983). On evaluating income distributions: Rank dominance, the Pareto prin-
ciple, and the Suppes-Sen Grading Principle of Justice. Public Choice, 40, 329–336.
Scitovsky, T. (1971). Welfare and competition (rev. ed.). Homewood: Richard D. Irwin.
Sen, A. K. (1969). Planners preferences: Optimality, distribution and control. In: J. Margolis &
H. Guitton (Eds), Public economics. London: Macmillan.
Sen, A. K. (1970). Collective choice and social welfare. San Francisco: Holden-Day.
Sen, A. K. (1973a). On economic inequality. New York: Macmillan.
Sen, A. K. (1973b). On ignorance and equal distribution. American Economic Review, 63,
1022–1024.
Sen, A. K. (1977). On weights and measures: Informational constraints in social welfare anal-
ysis. Econometrica, 45, 1539–1572.
Shorrocks, A. F. (1983). Ranking income distributions. Economica, 50, 3–17.
Thistle, P. D. (1989). Ranking distributions with generalized Lorenz curves. Southern Economic
Journal, 56, 1–12.
Thistle, P. D. (1998). Social structure, economic performance and Pareto optimality. Theory and
Decision, 45, 161–173.
INEQUALITY AND INCOME GAPS

Ian Preston

ABSTRACT

This paper discusses inequality orderings based explicitly on closing up of

income gaps, demonstrating the links between these and other orderings,
the classes of functions preserving the orderings and applications showing
their usefulness in comparison of economic policies.

1. INTRODUCTION
It is a truism to say that inequality is about gaps between incomes and
that reducing inequality is about closing these gaps up. Common means of
comparison between income distributions all use criteria which do show
inequality as falling when gaps close. However, explicitly asking whether
the gaps reduce throughout the whole distribution in concertina-like
fashion is a rare criterion to apply. This paper seeks to investigate the
related orderings.
The most common criteria for inequality comparison are those based on
Lorenz curves, made plausible most persuasively as indicators of inequality
by their link to progressive transfers of income. Progressive transfers are
often seen, since the arguments of Pigou (1912) and Dalton (1920), as un-
contentiously inequality reducing but this view could be challenged if there
are more than two people. A transfer from the top to the middle of the

Equity
Research on Economic Inequality, Volume 15, 33–56
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15003-1
33
34 IAN PRESTON

income distribution1 reduces inequality between the top and the middle but
increases it between the middle and the bottom.2 Regarding inequality as
having fallen overall involves giving priority to the former effect –the effect
on the gap between incomes of those involved directly in the transfer – for
which there may be good reason, but it is not obvious that it would not be
sensible to say inequality simply could not be compared.3
What convinces Dalton (1920) is the link to welfare – he is ‘‘primarily
interested, not in the distribution as such, but in the effects of the distri-
bution of income upon the distribution and total amount of economic wel-
fare, which may be derived from income (p. 348)’’. That such transfers raise
welfare is well known to be true in a typical utilitarian setting if individual
welfare depends only upon own income but if income gaps matter to in-
dividual welfare then this need not be so. This issue is taken up below and
links between economic welfare and the orderings based explicitly on gaps
are considered.
The earliest discussions of these orderings can be found outside the eco-
nomic context (for example in Marshall, Olkin, & Proschan, 1967a and
Barlow & Proschan, 1975). There are some useful papers from this period
which are less well known than perhaps they should be and a minor function
of the current paper is to bring some overlooked but highly germane mathe-
matical literature to the attention of inequality theorists. I think particularly
here of Marshall, Walkup, and Wets (1967b) which anticipates several
results of this paper.4
The major function, though, is to tell a rounded story about the ratio and
difference dominance concepts, and associated orderings and welfare pro-
perties, which extend the well-known Lorenz ordering in different ways. In
this, I am in fact taking up again some work which I engaged in some time
ago (Preston, 1989, 1990a, 1990b) and ideas which have been developed by
Moyes (1994) on the ‘‘dominance in relative differentials’’ and ‘‘dominance
in absolute differentials’’ concepts, for which he coined those terminologies,
and by Zheng (2007). The style and manner of development, in the sequel, is
intended to be somewhat in similar fashion to the way in which Rothschild
and Stiglitz’s (1973) paper developed a rounded story for the Lorenz
ordering, which had been begun by Kolm (1969) and Atkinson (1970) (see
also Dasgupta, Sen, & Starrett, 1973).
Section 2 defines the orderings and considers relations between them.
Section 3 outlines classes of functions which preserve the orderings. Section
4 considers how policies map underlying variation into distributions which
may be related according to the orderings. Section 5 concludes.
Inequality and Income Gaps 35

2. DOMINANCE ORDERINGS

2.1. Inequality

2.1.1. Orderings on Rn
It is convenient to define inequality orderings on income vectors which
have been placed in order from poorest to richest. To that end let Dn ¼
fx 2 Rn jxiþ1
xi ; for i ¼ 1; . . . ; n  1g and Dnþ ¼ fx 2 Rn jxiþ1
xi ; xi 40
for i ¼ 1; . . . ; n  1g ¼ Dn \ Rnþ denote spaces of ordered vectors.
The two crucial orderings of interest for this paper are defined by closing
up of all gaps in relative or absolute terms.

Definition 1.
(a) Say that x difference dominates y , written xhA y; iff xiþ1  xi 
yiþ1  yi for i ¼ 1,y, n1 and x; y 2 Dn :
(b) Say that x ratio dominates y , written xhR y; iff lnðxiþ1 Þ 
lnðxi Þ  lnðyiþ1 Þ  lnðyi Þ for i ¼ 1,y, n1 and x; y 2 Dnþ :

If x difference dominates y then all absolute gaps are smaller and if x ratio
dominates y then all relative gaps are smaller. Both orderings are discussed
in Marshall et al. (1967b) where they are treated as special cases of cone
orderings.5
These orderings go by different names. Moyes (1994) refers to dominance
in absolute and relative differentials. Zheng (2007) refers to absolute and
ratio differential conditions. In the absence of unanimity on any alternative
terminology, I keep to that used in Preston (1990a).
Ratio and difference dominance can obviously be nested as special cases
within a more general class of orderings requiring the closing up of gaps in
the value of any increasing function of incomes, say U. Zheng (2007) makes
this generalisation, defining a more general class of utility gap orderings. If
we choose UðxÞ ¼ lnðx þ mÞ with m 2 Rþ then we get a class of orderings
which will give ratio and difference dominance as extreme cases, in line with
the treatment of intermediate orderings in Kolm (1976a, 1976b).6
It is also useful to have definitions of transformations of vectors which do
not change inequality. In absolute terms all gaps are maintained by a par-
allel shift in a vector, called a translation, and all relative gaps by multi-
plying all incomes by a positive constant, referred to here as a rescaling. Let
en 2 Rnþ denote the vector all elements of which are unity.
36 IAN PRESTON

Definition 2.
(a) Say that x is a translation of y iff xi ¼ yi þ l; i ¼ 1; . . . ; n or
more simply x ¼ y þ len for some l 2 R and x; y 2 Dn :
(b) Say that x is a rescaling of y iff xi ¼ lyi ; i ¼ 1; . . . ; n; or more
simply x ¼ ly for some l 2 Rþ and x; y 2 Dnþ :

We can link the inequality orderings defined above to changes in income

vectors which do unambiguously close up gaps.
Definition 3.
(a) Say that x is an absolute lower end elevation of y iff, for some k
and some l40, xi ¼ yi þ l for i ¼ 1,y, k and xi ¼ yi for i ¼
k þ 1; . . . ; n with x; y 2 Dn :
(b) Say that x is a relative lower end elevation of y iff, for some k
and some l41, xi ¼ lyi for i ¼ 1,y, k and xi ¼ yi for i ¼
k þ 1; . . . ; n with x; y 2 Dnþ :

Combining absolute or relative lower end elevations with translations

or rescalings are the only ways to secure difference or ratio dominance. We
state this formally.
Theorem 1.
(a) xhA y iff x can be obtained from y by a finite series of absolute
lower end elevations and a translation.
(b) xhR y iff x can be obtained from y by a finite series of relative
lower end elevations and a rescaling.

Proof of Theorem 1.
(a) Sufficiency follows from the facts that any lower end elevation
reduces absolute gaps for i ¼ 1,y, k and leaves them unchanged
for i ¼ k þ 1; . . . ; n whereas any translation leaves absolute gaps
unchanged. To see necessity, suppose xhA y: Then x can be
obtained from y by a series of n1 absolute lower end elevations,
where the kth lower end elevation raises yi by ykþ1  yk  xkþ1 þ
xk
0 for i ¼ 1; . . . ; k; and a translation by xn  yn :
(b) The result follows from the above given that xhR y iff
lnðxÞhA lnðyÞ: &

Difference and ratio dominance are stronger inequality concepts than

those prevalent in the literature. Since Pigou (1912) and, especially, Dalton
Inequality and Income Gaps 37

(1920) it has been widely accepted that inequality is reduced by a sort of

change which cannot necessarily be reduced to changes of the sort discussed
above.

Definition 4. Say that x can be obtained from y by an (elementary) pro-

gressive transfer7 if, for some k and some l40; xk ¼ yk þ l; xkþ1 ¼
ykþ1  l and xi ¼ yi ; i ¼ 1; . . . ; k  1; k þ 2; . . . ; n and x; y 2 Dn
(Muirhead, 1903; Pigou, 1912; Dalton, 1920).

Well-known results link progressive transfers to the most widely accepted

criterion for inequality comparison, that of Lorenz dominance.
Pk
Definition 5. Say that x LorenzPdominates y, written xhL y; iff i¼1
n
½xi  yi 
0 for k ¼ 1,y, n1, i¼1 ½xi  yi  ¼ 0 and x; y 2 Dn :

The Lorenz ordering is equivalent to a relation known as majorisation

(Marshall & Olkin, 1979) which is widely used outside of economic contexts.
A famous result establishes that xhL y iff x can be obtained from y by a
finite series of progressive transfers (Hardy, Littlewood, & Pólya, 1934;
Atkinson, 1970). The Lorenz ordering can be extended to comparisons
which do not involve equal means by allowing progressive transfers to be
combined with translations and rescalings.

Definition 6.
(a) Say that x absolute Lorenz dominates y, written xhLA y; iff
xhL y þ len for some l (Shorrocks, 1983; Moyes, 1987).
(b) Say that x relative Lorenz dominates y, written xhLR y; iff
xhL ly for some l40 (Lorenz, 1905).

2.1.2. Orderings on Distributions

We can also define analogous orderings more generally on spaces of
distribution functions or, equivalently, quantile functions. Such a setting
clearly subsumes that of the earlier section, allowing for comparison of
income vectors of different dimensions but also of inequality in continuous
distributions.
Let D denote the space of nondecreasing functions from ½0; 1 to R and
Dþ denote the space of nondecreasing functions from ½0; 1 to Rþ : Let DC
denote the space of differentiable, nondecreasing functions from ½0; 1 to R
and DC þ denote the space of differentiable, nondecreasing functions from
½0; 1 to Rþ :
38 IAN PRESTON

If x is distributed according to distribution function F x : R ! ½0; 1; let

xx : ½0; 1 ! R be the corresponding quantile function, or inverse distri-
bution function, defined by xx ðpÞ ¼ supfxjF x ðxÞ  pg:
We can now define the corresponding orderings defined on distributions.
Definition 7.
(a) Say that F x hA F y iff xx ðpÞ  xy ðpÞ is nonincreasing for p 2 ½0; 1
and xx ; xy 2 D:
(b) Say that F x hR F y iff lnðxx ðpÞÞ  lnðxy ðpÞÞ is nonincreasing for
p 2 ½0; 1 and xx ; xy 2 Dþ :

Marshall et al. (1967a) and Barlow and Proschan (1975) say that F x is
star-shaped with respect to F y if xy ðF x ðxÞÞ=x is increasing. Arnold defines an
ordering identical to hR which he calls star-shaped ordering, star ordering
or simply -ordering.
Suppose that the distributions under comparison and their inverses are
differentiable with associated densities f x and f y : Then x0x ðpÞ ¼ 1=f x ðF 1
x ðpÞÞ
so that F x hA F y iff f x ðxx ðpÞÞ
f y ðxy ðpÞÞ for p 2 ½0; 1:
We can also define absolute and relative Lorenz curves using the quantile
functions (see Gastwirth, 1971):
Rq R1
LAx ðqÞ ¼ 0 xx ðpÞdp  q 0 xx ðpÞdp
Rq R1
LRx ðqÞ ¼ 0 xx ðpÞdp= 0 xx ðpÞdp

and thus define Lorenz orderings.

Definition 8.
(a) Say that F x hLA F y iff LA A
x ðqÞ
Ly ðqÞ for all q 2 ½0; 1 and xx ; xy
2 D:
(b) Say that F x hLR F y iff LR R
x ðqÞ
Ly ðqÞ for all q 2 ½0; 1 and xx ; xy
2 Dþ :

2.1.3. Relations Between Orderings

The fact that difference and ratio dominance imply but are not implied
by absolute and relative Lorenz dominance is long established.
Theorem 2. (Marshall et al., 1967a, 1967b; Jakobsson, 1976; Thon, 1987;
Arnold, 1987)
(a) If xhA y then xhLA y:
(b) If xhR y then xhLR y:
Inequality and Income Gaps 39

(c) If F x hA F y then F x hLA F y :

(d) If F x hR F y then F x hLR F y :

The point is that an absolute or relative lower end elevation can always be
implemented by a translation or rescaling followed by a finite series of
progressive transfers. A lower end elevation obviously raises mean income.
Consider a translation or rescaling which leads to the same increase in mean
income and following this by redistributing the income increase to those at
the lower end by transferring the income gains at the top end down the
distribution to those at the bottom.
It is not possible on the other hand to implement a progressive transfer by
a series of lower end elevations and translations or rescalings. A progressive
transfer anywhere other than at the extremes of the distribution raises some
income gaps at the same time as it reduces others and we have shown that
lower end elevations cannot raise income gaps.
We can illustrate the relation between these orderings by showing areas of
dominance in comparisons within the standard simplex as in Sen (1973) and
many later papers. Imagine looking down at the origin from a point along
the ray of equality in the positive orthant of income space with n ¼ 3. Any
allocation of a given total income, which we normalise to unity, can be
represented as a point in the simplex illustrated in Fig. 1. If we take an

Fig. 1. Dominance Orderings Illustrated in the Standard Simplex.

40 IAN PRESTON

arbitrary initial income vector then the six possible permutations give the
vertices of the irregular hexagonal shape shown in the figure. As is well
known, the convex hull of these six points represents the set of income
vectors which, permuted into appropriate order, Lorenz dominate the initial
income vector.
Now consider which set of points, appropriately permuted, ratio domi-
nate the initial vector. Ratios between incomes for any two individuals
are constant throughout planes containing the third axis, and these planes
intersect the simplex along rays passing through its vertices. Points of
greater equality between these two are those lying between such rays and the
bisecting ray through the same vertex. Hence the area in which all ratios are
nearer to unity is the star-shaped8 area outlined in bold within the Lorenz
hexagon. Since this fits inside the hexagon, coinciding only at its vertices, it
is diagrammatically plain that ratio dominance implies without being im-
plied by Lorenz dominance in comparisons of vectors with equal means.
Differences between any two incomes are maintained in planes which cut
the simplex along lines perpendicular to its sides. Hence, constructing, as in
the ratio-based case, an area in which all differences are diminished gives
the inverted Y-shape outlined with dots inside the hexagon. Again this fits
entirely inside the Lorenz hexagon except at its vertices. It contains however
some points inside and some outside the ratio-based star shape (the areas
shaded on the diagram), demonstrating that neither ratio nor difference
dominance implies the other.

2.2. Welfare

Dalton’s conviction that progressive transfers reduce welfare was motivated

by the recognition that if social welfare was the sum of individual utilities
which depend only on own income and do so in a concave fashion then such
transfers raise social welfare. In comparisons between vectors with equal
total income, Lorenz dominance can be identified with improvement in
utilitarian social welfare (Hardy et al., 1934; Atkinson, 1970). We can extend
this observation to comparisons involving vectors with different means by
defining a generalisation of Lorenz dominance.
Definition 9.

Pk that x generalised Lorenz dominates y , written

(a) Say xhGL y; iff
n
i¼1 ½xi  yi 
0 for k ¼ 1,y, n and x; y 2 D (Kolm, 1969;
Shorrocks, 1983).
Inequality and Income Gaps 41

Rq
(b) Say that F x hGL F y ; iff 0 ½xx ðpÞ  xy ðpÞdp
0 for all q 2 ½0; 1
and xx ; xy 2 D:

This ordering is called supermajorisation in the noneconomic context

(Marshall & Olkin, 1979). Kolm does not use the term generalised Lorenz
dominance but refers to a social preference for distributions which gener-
alised Lorenz dominate others as isophily. Generalised Lorenz dominance is
also strongly linked to utilitarian social welfare. Suppose social welfare W :
Dn ! R is the sum of individual utilities which are continuous, increasing,
concave functions of individual incomes, ui : R ! R: Clearly xhGL y im-
plies W ðxÞ
W ðyÞ: In fact W ðxÞ
W ðyÞ for all social welfare functions
with these properties iff xhGL y (Kolm, 1969; Marshall & Olkin, 1979;
Shorrocks, 1983).
What, though, if utilities are not formed in a purely self-regarding way?
Suppose individual utilities ui ðxi ; x  exi Þ depend not only on own income
but also on the gaps between own income and the incomes of others.9 For
example, suppose individual utilities have the form ui ðxi ; xÞ ¼ ui ðxi Þ þ
cðxiþ1  xi Þ where ui has the usual continuous, increasing and concave
properties but c is strongly enough decreasing. It is easy to construct
an example where a progressive transfer in the middle of the income dis-
tribution does not increase social welfare because of the harm done to
the utilities of individuals with incomes below the recipient of the transfer.
In such a context an absolute lower end elevation, however, would always
still increases social welfare because no income would fall and no gap
increase.
Lower end elevations are the natural basis for welfare comparisons cor-
responding to the difference and ratio dominance orderings.
Let us define two orderings which correspond to income changes which
should unambiguously increase social welfare even if income gaps matter to
individual well being.
Definition 10.
(a) Say that xhA y iff xiþ1  xi  yiþ1  yi ; i ¼ 1; . . . ; n  1 and
xn
yn with x; y 2 Dn :
(b) Say that xhR y iff lnðxiþ1 Þ  lnðxi Þ  lnðyiþ1 Þ  lnðyi Þ; i ¼
1; . . . ; n  1 and xn
yn for x; y 2 Dnþ :

These say that no element in the vector is reduced and either all
absolute or all relative gaps are reduced. The link to lower end elevations
is obvious.
42 IAN PRESTON

Theorem 3.
(a) xhA y iff x can be obtained from y by a finite series of absolute
lower end elevations.
(b) xhR y iff x can be obtained from y by a finite series of relative
lower end elevations.

These criteria for welfare comparison are plainly stronger than generali-
sed Lorenz dominance since all partial sums of income are obviously in-
creased by changes which raise all incomes.
Theorem 4. If xhA y or xhR y then xhGL y
We can also define similar orderings over distribution functions for which
similar links will hold.
Definition 11.
(a) Say that F x hA F y iff xx ðpÞ  xy ðpÞ is nondecreasing for p 2
½0; 1; xx ð1Þ
xy ð1Þ and xx ; xy 2 D:
(b) Say that F x hA F y iff lnðxx ðpÞÞ  lnðxy ðpÞÞ is nondecreasing for,
p 2 ½0; 1; xx ð1Þ
xy ð1Þ and xx ; xy 2 Dþ :

3. INEQUALITY AND WELFARE INDICES

We have seen that progressive transfers increase the sum of concave func-
tions of the individual elements in the vector. The class of functions defined
on the vector which are such that they rise with progressive transfers is a
wider class than this, first studied by Schur (1923) and known as Schur
convex functions. Schur convex functions are those which are said to pre-
serve the Lorenz order. A decreasing function of a Schur convex function is
said to be Schur concave. If Lorenz dominance is felt to be a convincing
criterion for judging inequality then this is the natural class of functions to
use for measuring inequality and most measures proposed for this purpose
do fall into this class (although there are notable exceptions such as the
variance of logarithms which is infamously not Schur concave, as discussed
in Foster & Ok, 1999). Similarly, if generalised Lorenz dominance is felt
to be a suitable criterion for judging social welfare improvement then
measures of social welfare ought to rise with progressive transfers and
therefore to be Schur convex, besides having other properties such as being
increasing in all incomes.
Inequality and Income Gaps 43

In this section of the paper, we discuss the classes of functions of income

vectors and distributions which preserve the orderings defined above based
on income gaps. We start by defining formally what it means for a function
to preserve an ordering.
Definition 12. Say that a function f : X ! R preserves an ordering h iff
fðxÞ
fðyÞ whenever xhy; x; y 2 X :
We consider functions which are differentiable. For this purpose we need
a notion of derivative for a mapping defined on a space of functions. Spe-
cifically, suppose f : DC ! R is a functional defined on the space of quan-
tile functions. Then we assume the existence
R of a functional derivative10
df : ½0; 1 ! R with the property that df cdp ¼ ðd=d
Þ fðx þ
ZÞj
¼0 for
x 2 DC and differentiable functions Z : ½0; 1 ! R:

3.1. Inequality

Functions which preserve the ratio and difference dominance orderings are
characterised in the following result.
Theorem 5.
n
Pk differentiable function f : D ! R preserves
(a) A Pn hA iff
i¼1 ð@=@xi ÞfðxÞ
0 for k ¼ 1,y, n1 and i¼1 ð@=@xi ÞfðxÞ ¼
0 (Marshall et al., 1967b).
n
Pk differentiable function f : Dþ ! R preserves
(b) A Pn hR iff
i¼1 xi ð@=@xi ÞfðxÞ
0 for k ¼ 1; . . . ; n  1 and i¼1 xi ð@=@xi Þ
fðxÞ ¼ 0 (Marshall et al., 1967b).
(c) RA differentiable function11 C
R 1 f : D ! R preserves hA iff
q
0 df dp
0 for qo1 and 0 df dp ¼ 0:
(d) A
R q differentiable function Rf : DC
þ ! R preserves hR iff
1
0 xdf dp
0 for qo1 and 0 xdf dp ¼ 0:

Proof of Theorem 5.
(a) If xhA y then we can get from y to x by Pa finite series of lower
end elevations and a translation. Given ki¼1 ð@=@xi ÞfðxÞ
0 for
k ¼ 1,yP , n ¼ 1, the lower end elevations all increase f and
given ni¼1 ð@=@xi ÞfðxÞ ¼
P0n the translation leaves it unaffected.
It is necessary for i¼1 ð@=@xi ÞfðxÞ ¼ 0 since, for any l,
xhA x þ len and x þ len hA x: It is necessary for
44 IAN PRESTON

Pk
i¼1 ð@=@xi ÞfðxÞ
0 as f must increase with any lower end
elevation.
(b) Obvious, noting that ratio dominance is just difference domi-
nance in logarithms.
(c) If F x hA F y then xx ¼ xy  Z where Z 2 DC : For any Z 2 DC ; we
have:

Z 1
d
fðx 
ZÞj
¼0 ¼  df Z dp
d
0

Z 1  Z q
Z q 
¼  Zð1Þ df dp þ Z0 df dp dq
0 0 0

R 1 l 2 R; F x hA F y and F y hA F x if xx ¼ xy  le;
Since, for any
we must have 0 df dp ¼ 0: Then
Z q
Z q 
d
fðx 
ZÞj
¼0 ¼ Z0 df dp dq
d
0 0

R q and for this to be positive for all increasing Z requires

0 df dp
0 for all q.
(d) Obvious, noting that ratio dominance is just difference domi-
nance in logarithms. &

The results for orderings on Rn were derived by Marshall et al. (1967b),

although the proof and the interpretation here are quite different.
The classes of indices derived here are obviously broader than that of
Schur convex functions, which are characterised by:
@ @
fðxÞ
fðxÞ
@xi @xiþ1
Functions appropriate as indices of inequality are, of course, those which
preserve the reverse orderings, "A and "R ; and indices which preserve "R
but not "LR include, for example, the variance of logarithms.

3.2. Welfare

Just as progressive transfers raise welfare as well as reducing inequality, we

consider welfare functions which are increased by lower end elevations.
Inequality and Income Gaps 45

Theorem 6.
n 
Pk differentiable function f : D ! R preserves hA iff
(a) A
i¼1 ð@=@xi ÞfðxÞ
0 for k ¼ 1,y, n 1.
n 
Pk differentiable function f : Dþ ! R preserves hR iff
(b) A
i¼1 xi ð@=@xi ÞfðxÞ
0 for k ¼ 1,y, n1 (Marshall et al.,
1967b).
C 
R q differentiable function f : D ! R preserves hA iff
(c) A
0 df dp
0 for qo1.
C 
R q differentiable function f : Dþ ! R preserves hR iff
(d) A
0 xdf dp
0 for qo1.

Proof of Theorem 6.
(a) If xhA y then we can get from y to x by a finite series of lower
end
Pk elevations. Arbitrary lower end elevations increase f iff
i¼1 ð@=@xi ÞfðxÞ
0 for k ¼ 1,y, n 1.
(b) Obvious, noting that ratio dominance is just difference domi-
nance in logarithms.
(c) If F x hA F y then xx ¼ xy þ eZð1Þ  Z where Z 2 DC : For any Z 2
DC ; we have

Z 1  Z 1
d
fðx 
½eZð1Þ  ZÞj
¼0 ¼ Zð1Þ df dp  df Z dp
d
0 0
Z q
Z q 
¼ Z0 df dp dq
0 0

Rq
For this to be positive for all increasing Z requires 0 df dp
0
for all q.
(d) Obvious, noting that ratio dominance is just difference domi-
nance in logarithms. &

Again these results are partly anticipated by Marshall et al. (1967b)

though the treatment is quite different.
The conditions for functions to preserve the welfare orderings differ only
from those to preserve the equality orderings in that the requirements
for invariance to translation and rescaling are dropped. The requirement for
nonnegativity of partial sums of derivatives remains. Functions which pre-
serve hA and hR need not be increasing in all incomes but must increase
with positive translations and scalings up of incomes, respectively.
46 IAN PRESTON

There is a tradition of linking inequality indices with measurement of

social welfare going back at least to Dalton (1920) and exemplified by Kolm
(1969), Atkinson (1970) and Blackorby and Donaldson (1978, 1980, 1984).
To expound the key results here we need to define properties of functions
implying certain sorts of response to translations and rescalings. Let e 2 DC
þ
denote the quantile function which is constant at unity.
Definition 13.
(a) A differentiable function f : Dn ! R is translatable iff there ex-
ists an increasing function g : R ! R and a function c : Dn ! R
such that f ¼ gðcðxÞÞ and cðx þ len Þ ¼ cðxÞ þ l:
(b) A differentiable function f : Dnþ ! R is homothetic iff there ex-
ists an increasing function g : R ! R and a function c : Dnþ !
R such that f ¼ gðcðxÞÞ and cðlxÞ ¼ lcðxÞ:
(c) A differentiable function f : DC ! R is translatable iff there
exists an increasing function g : R ! R and a function c : Dn !
R such that f ¼ gðcðxÞÞ and cðx þ leÞ ¼ cðxÞ þ l:
(d) A differentiable function f : DC þ ! R is homothetic iff there ex-
ists an increasing function g : R ! R and a function c : Dnþ !
R such that f ¼ gðcðxÞÞ and cðlxÞ ¼ lcðxÞ:

The notion of the equally distributed equivalent income function due to

Kolm (1969) and Atkinson (1970) is a crucial one in this literature. For
comparisons of vectors define this as w : Dn ! R by fðwðxÞen Þ ¼ fðxÞ and
for comparisons of distribution functions define it as w : DC ! R by
fðwðxx ÞeÞ ¼ fðxx Þ: Clearly w preserves the same orderings as does f. If f is
Schur convex and translatable then subtracting w from mean income gives
an index which is Schur concave and invariant to translation. If f is Schur
convex and homothetic then the proportional difference between mean in-
come and w gives an index which is Schur concave and invariant to rescaling.
In the first case we have an absolute inequality index linked to measurement
of welfare and in the latter a relative inequality index – both ideas are found
in Kolm (1969) and developed by later authors (Atkinson, 1970; Blackorby
& Donaldson, 1978, 1980, 1984).
Given that progressive transfers leave mean incomes unchanged it is
natural to construct inequality indices intended to preserve the Lorenz
ordering by comparison to mean income but this will not work if the index
is intended to preserve the orderings based on gaps. Lower end elevations
do not leave mean income unchanged. However, we can construct suitable
indices by comparison to the maximum income.
Inequality and Income Gaps 47

Theorem 7.
(a) If a differentiable function f : Dn ! R is translatable and
preserves hA then yðxÞ ¼ xn  wðxÞ preserves "A :
(b) If a differentiable function f : Dnþ ! R is homothetic and
preserves hR then yðxÞ ¼ 1  wðxÞ=xn preserves "R :
(c) If a differentiable function f : DC ! R is translatable and
preserves hA then yðxÞ ¼ xð1Þ  wðxÞ preserves "A :
(d) If a differentiable function f : DCþ ! R is homothetic and
preserves hR then yðxÞ ¼ 1  wðxÞ=xð1Þ preserves "R :

Proof of Theorem 7.
(a) If f is translatable Pn then wðx þ len Þ ¼ wðxÞ þ l so P that yðx þ
k
len Þ ¼ yðxÞPk and i¼1 ð@=@x i ÞyðxÞ ¼ 0: For kon, i¼1 ð@=@xi Þ
yðxÞ ¼  i¼1 ð@=@xi ÞfðxÞ  0:
(b) P
If f is homothetic then wðlxÞ ¼ lwðxÞ Pkso that yðlxÞ ¼ yðxÞPand
n k
i¼1 ð@=@x Þx
i i yðxÞ ¼ 0: For kon, i¼1 ð@=@xi Þxi yðxÞ ¼  i¼1
ð@=@xi Þxi fðxÞ  0:
(c) If f is translatable
R1 then wðx þ leÞ ¼ R q wðxÞ þ l so that
R q yðx þ leÞ ¼
yðxÞ and 0 dy dp ¼ 0: For qo1, 0 dy dp ¼  0 df dp  0:
(d) If
R 1 f is homothetic then wðlxÞ R q ¼ lwðxÞ soR that yðlxÞ ¼ yðxÞ and
q
0 xdy dp ¼ 0: For qo1, 0 xdy dp ¼  0 xdf dp  0: &

Thus, we have a way of constructing inequality indices preserving ratio

and difference dominance orderings from social welfare measures with
appropriate properties.
To take the simplest example, if social welfare is measured by the
homothetic
P and translatable function giving mean income fðxÞ ¼ x̄ ¼
ð1=nÞ ni¼1 xi ; which is invariant to progressive transfers and therefore yields
no interesting Schur concave inequality measure, Pn then the construction
just outlinedP gives measures x n  x̄ ¼ ð1=nÞ i¼1 ½x n  xi  and 1  x̄=xn ¼
1  ð1=nÞ ni¼1 ½xi =xn ; the mean absolute and relative shortfall from the top
income, which do preserve "A and "R ; respectively.
A social welfare function which isPtranslatable, homothetic and strictly
Schur convex is fðxÞ ¼ ð2=nðn þ 1ÞÞ ni¼1 ðn  iÞxi : The Kolm (1969) pro-
cedure
Pn gives a relative inequality index 1  wðxÞ=x̄ ¼ G ¼ ð2=nðn þ 1Þx̄Þ
i¼1 i½x i  x̄ equal to the well-known Gini coefficient. The procedure P out-
lined above gives the alternative 1  wðxÞ=xn ¼ ð2=nðn þ 1Þxn Þ ni¼1 ði  nÞ
½xi  xn  which approaches Gx̄=xn  3½1  x̄=xn  for large n. There exist,
though, functions which are Schur convex yet preserve the gap-based
48 IAN PRESTON

inequality
Pn orderings. Take the social welfare function fðxÞ ¼ ð2=nðn P þ 1ÞÞ
n
ix
i¼1 i : This is Schur concave yet 1  wðxÞ=x n ¼ ð2=nðn þ 1Þxn Þ i¼1
i½xn  xi  is still an inequality measure consistent with "R :

4. INEQUALITY-REDUCING POLICIES

Often the purpose to which the inequality orderings are to be put is to

compare outcomes under two policy regimes in which incomes are deter-
mined by different mappings from some underlying source of variation. For
such cases, general results can be derived relating the properties of these
mappings to the resulting gaps. Let us assume the underlying variation is in
a single dimension and denote the variable in question by z 2 R; distributed
according to Fz, drawn from a class of distribution functions F: The out-
comes of interest in the two regimes are denoted x and y, where x ¼
fx ðz; F z Þ and y ¼ fy ðz; F z Þ with fx : R  F ! R and fy : R  F ! R
continuous and increasing in their first arguments. Then xx ðpÞ ¼ fx ðxz ðpÞÞ
and xy ðpÞ ¼ fy ðxz ðpÞÞ:
Theorem 8.
(a) If x ¼ fx ðz; F z Þ; y ¼ fy ðz; F z Þ with fx : R  F ! R and fy :
R  F ! R continuous and increasing in their first arguments
then the following are equivalent
(i) F x hA F y for all F z 2 F:
(ii) fx ðz; F z Þ ¼ gðfy ðz; F z Þ; F z Þ for some function g : R 
F ! R such that, for all F z 2 F; gðfy ðxz ðpÞ; F z Þ;
F z Þ  fy ðxz ðpÞ; F z Þ is nonincreasing in p for all p 2
½0; 1:
(b) If x ¼ fx ðz; F z Þ; y ¼ fy ðz; F z Þ with fx : R  F ! Rþ and fy :
R  F ! Rþ continuous and increasing in their first arguments
the following are equivalent
(i) F x hR F y for all F z 2 F:
(ii) fx ðz; F z Þ ¼ gðfy ðz; F z Þ; F z Þ for some function g :
Rþ  F ! Rþ such that, for all F z 2 F;
gðfy ðxz ðpÞ; F z Þ; F z Þ=fy ðxz ðpÞ; F z Þ is nonincreasing in
p for all p 2 ½0; 1:

Proof of Theorem 8.
(a) Sufficiency of the condition in (ii) for F x hA F y for all F z 2 F is
obvious.
Inequality and Income Gaps 49

If, for some F z 2 F; fx ðz; F z Þ is not a function of fy ðz; F z Þ

then there exist p0 ; p1 2 ½0; 1 with p1 4p0 ; such that
fy ðxx ðp0 Þ; F z Þofy ðxz ðp1 Þ; F z Þ but fx ðxz ðp0 Þ; F z Þafx ðxz ðp1 Þ; F z Þ:
Then it cannot be that, for z distributed as Fz, xx ðpÞ  xy ðpÞ is
nonincreasing in p for p 2 ½p0 ; p1 :
If fx ðz; F z Þ is a function of fy ðz; F z Þ but there is an Fz and
p1 4p0 such that gðfy ðxz ðp1 Þ; F z Þ; F z Þ  fy ðxz ðp1 Þ; F z Þ4
gðfy ðxz ðp0 Þ; F z Þ; F z Þ  fy ðxz ðp0 Þ; F z Þ then again it cannot be that,
for z distributed as Fz, xx ðpÞ  xy ðpÞ is nonincreasing in p for
p 2 ½p0 ; p1 :
(b) Obvious given ratio dominance is difference dominance in log-
arithms. &

Functions g : R ! R such that g(x)/x is increasing are called star-shaped

(Bruckner & Ostrow, 1962). If the functions fx and fy, are differentiable
in z then the conditions in the theorem reduce to a comparison of deriva-
tives or elasticities – specifically, gðx; F z Þ  x is decreasing in x iff qg/qx
is less than unity and gðx; F z Þ=x is decreasing in x iff @ ln g=@ ln x is less
than unity.
If the functions fx and fy are additively or multiplicatively separated then
the role of Fz becomes irrelevant. Finding distributions Fz such that domi-
nance fails in either direction is less restricted and the theorem can be
extended to cover not only the inequality orderings based on gaps but also
Lorenz dominance. In particular, suppose F includes all distributions with
support within Z; x ¼ fx ðzÞcx ðF z Þ; y ¼ fy ðzÞcy ðF z Þ and fx ðzÞ ¼ gðfy ðzÞÞ
with g(z)/z increasing over an interval Z̄  Z then we can find a counter-
example in which there is ratio dominance in the reverse direction by
choosing Fz with support wholly within Z̄: Since ratio dominance implies
relative Lorenz dominance then this means g(z)/z falling everywhere is a
necessary condition not only for difference dominance but also for Lorenz
dominance to hold for all F z 2 F:
Theorem 9.
(a) If x ¼ fx ðzÞ þ cx ðF z Þ; y ¼ fy ðzÞ þ cy ðF z Þ with fx : R ! R and
fy : R ! R continuous and increasing and cx : F ! R and cy :
F ! R then the following are equivalent
(i) F x hA F y for all Fz with support in Z
(ii) F x hLA F y for all Fz with support in Z
(iii) fx ðzÞ ¼ gðfy ðzÞÞ for some function g : R ! R such
that g(x )x is nonincreasing in x for all xAg(Z)
50 IAN PRESTON

(b) If x ¼ fx ðzÞcx ðF z Þ; y ¼ fy ðzÞcy ðF z Þ with fx : R ! Rþ and fy :

R ! Rþ continuous and increasing and cx : F ! Rþ and cy :
F ! Rþ then the following are equivalent
(i) F x hR F y for all Fz with support in Z
(ii) F x hLR F y for all Fz with support in Z
(iii) fx ðzÞ ¼ gðfy ðzÞÞ for some function g : R ! R such
that g(x )/x is nonincreasing in x for all xAg(Z)
For cx ¼ cy ¼ 1, the second part of this theorem is the result of
Jakobsson (1976). Sufficiency of the condition on g for F x hLR F y to hold for
all Fz was recognised by Fellman (1976) and Kakwani (1977). Eichhorn,
Funke, and Richter (1984) and Arnold (1987) extend this result to drop the
assumption that fx and fy are continuous and increasing.

4.1. Applications

4.1.1. Progressive Taxation and Inequality

4.1.1.1. Fixed Pre-Tax Incomes. Suppose z denotes pre-tax incomes, as-
sumed distributed in a way unaffected by taxation. Let the tax function
be t(z) so that post-tax incomes are zt(z). By Theorem 9, the post-tax
distribution absolute Lorenz dominates and difference dominates the pre-
tax distribution whatever Fz iff t(z) is increasing in z which is to say that
marginal tax rates are everywhere positive (Moyes, 1988). This is a property
that Fei (1981) calls minimal progression.
The post-tax distribution relative Lorenz dominates and ratio dominates
the pre-tax distribution whatever Fz iff t(z)/z is increasing in z which is to
say that marginal tax rates are everywhere above average tax rates. This is
the property usually characterised as progression. The post-tax distribution
dominates the pre-tax distribution iff the tax is progressive in the sense of
taking a greater share of the incomes of the rich all the way along the
distribution. This result, proved for example in Jakobsson (1976), captures
an old idea. Seligman (1894) dates the first recorded occurrence of progres-
sive taxation to Solonic Athens and the first written recognition that it ‘‘will
lessen the disparity of fortunes’’ to Guiccardini’s sixteenth century discus-
sion of Florentine taxation, reprinted in Guicciardini (1932).12
One tax system will reduce inequality further than another, in the sense
of relative Lorenz dominance and ratio dominance, iff the elasticity of post-
tax income to pre-tax income – known as residual income progression
(Musgrave & Thin, 1948) – is everywhere lower. This is among the results
due to Jakobsson (1976).13 Suppose we have a linear tax on pre-tax income,
Inequality and Income Gaps 51

t(z) ¼ tzG so that post-tax income is z(1t)+G. Then the elasticity is

z=ðz þ G=ð1  tÞÞ so an increase in t reduces inequality.

4.1.1.2. Fixed Pre-Tax Wages. Changes in taxation change work incentives

and therefore the distribution of earnings given a fixed distribution of wages.
The inequality-reducing effect of progressive taxation is less obvious in such
a context. However, the same theorems still provide the tools for assessing
necessary and sufficient conditions for reduction in the inequality of
incomes.14
Let z now denote wages, distributed again according to F z 2 F: Let
individual hours of work be h and taxes be t(zh) so that post-tax income
is zht(zh) assuming no other source of income. Chosen hours under
the given tax system, h ¼ Z(z), may decline with wage but we assume at
least that post-tax income does not, zZ0 ðzÞð1  t0 ðzZðzÞÞ40 – the so-called
Mirrlees condition.
Then, by Theorem 9, the post-tax distribution under one tax system
Lorenz dominates and ratio dominates that under another iff the elasticity
of post-tax income to the wage is lower at each wage. This elasticity is
the product of residual income progression at zZ(z) and one plus the elas-
ticity of hours Z(z) to z (wherever Z(z)>0). Two new considerations emerge
when considering the impact of a progressive tax change. Firstly, labour
supply responses may make the distribution of earnings less equal. Sec-
ondly, even if residual income progression falls at each level of earnings,
labour supply responses could move individuals into less progressive parts
of the tax system so that residual income progression need not fall at each
wage rate.
As an example, consider the case of CES preferences with a linear tax,
t(zh) ¼ tzhG. Post-tax incomes are:

bzs ð1  tÞs
ðzð1  tÞ þ GÞ
zð1  tÞ þ bzs ð1  tÞs

The elasticity is:

zð1  tÞ z
ðs  1Þ þ
zð1  tÞ þ bzs ð1  tÞs z þ G=ð1  tÞ

An increase in t reduces the second term but increases the first and there is
no guarantee that the expression as a whole falls. Preston (1990b) shows
that this is possible for parameter values which are not unreasonable. In
52 IAN PRESTON

principle, there exist distributions of pre-tax wages such that inequality is

not reduced.

4.1.2. Immigration and Inequality

To take a slightly different example, consider the effect of immigration on
income inequality in the pre-existing resident population of a country. Sup-
pose what is now fixed is the distribution of productive abilities, zA[0,1],
distributed in the resident population of size N according to Fz.
The economy employs n(z) workers of type z to produce aRsingle type of
1
output according to a CES technology whereby output is ½ 0 znðzÞs dz1=s
with so1. This output is traded internationally at a fixed world price which
we normalise to 1. Demand for labour of type z is determined by equating
its marginal value product to the wage w(z):

Z 1 ð1=sÞ1
znðzÞs1 znðzÞs dz ¼ wðzÞ
0

Prior to immigration, this defines an equilibrium wage distribution equat-

ing demand for labour of type z to its supply, n(z) ¼ Nfz(z). For this to
involve wages increasing in z requires that we restrict attention to distri-
butions Fz such that 1+(s1)q ln fz(z)/q ln z40 everywhere.15
Now assume that there is immigration of M ¼ mN workers with ability
distributed according to distribution function Iz with density iz. The economy
reaches a new equilibrium at which the ratio of new to old wages at labour
type z is (1+miz (z)/fz(z))s1. Theorem 8 can be applied. The distribution of
wages across pre-existing resident workers is made more equal, in the sense
of ratio dominance, whatever Fz iff immigration policy is such as to guar-
antee iz(z)/fz(z) is increasing at all z so that immigrants are more concentrated
in higher earning groups than in the population already resident.

5. CONCLUSION

This paper has drawn attention to and argued a case for the interest of
inequality orderings based explicitly on closing up of income gaps. Drawing
where necessary on earlier papers, the links between these and other order-
ings have been outlined, the classes of functions preserving the orderings
have been characterised and applications have been presented showing their
use in comparison of economic policies.
Inequality and Income Gaps 53

NOTES
1. Progressive transfers are sometimes called Robin Hood transfers. The Robin
Hood of legend stole from the rich to give to the poor. No one would disagree that
that reduces inequality. He never, however, stole from the rich to give to the middle
or stole from the middle to give the poor.
2. Blum and Kalven (1953), for example, discuss income redistribution in this sort
of way.
3. One might raise a similar objection to Sen’s (1976, 1978) strengthened criterion
for comparisons of ordinal inequality for cases of more than two persons – STOIC.
Recognising this makes it clearer why he finds a link with Lorenz dominance.
4. I am myself grateful to an anonymous referee for bringing the pertinence of this
paper and its precedence in proving certain results to my attention.
5. x and y are cone ordered if the difference between them lies in a specified
convex cone. If xhA y or xhR y; for example, then the differences between the
absolute or relative gaps in the two vectors lie in the particular cone defined by the
nonnegative orthant. Majorisation, discussed below, is also a cone ordering. This is a
framework which yields useful insights but we do not adopt it here.
6. Kolm regards the view that equal absolute increases in income preserve in-
equality as ‘‘leftist’’ and the view that equal proportional increases preserve
inequality as ‘‘rightist’’. This categorisation could be questioned, particularly if we
consider the implications for views on decreases rather than increases in income. Is it
more leftist to think that equal absolute cuts in income, as through a poll tax,
preserve inequality? While it seems certainly true that the leftist would prefer a given
positive sum to be distributed through equal absolute increases than through equal
proportional ones it is not obvious that this reflects a view about how inequality
should be measured.
7. Progressive transfers are sometimes defined as any transfers from richer to
poorer individuals, not necessarily next to each other in the ordering of incomes. The
term ‘‘elementary progressive transfer’’ is from Arnold (1987).
8. The coincidence between the name of the star-shaped ordering and the shape of
the figure is purely fortuitous. The origin of the name lies in the connection with star-
shaped functions, as explained below.
9. This sort of dependence is sometimes referred to as envy but that is a somewhat
tendentious term, envy, one of the seven deadly sins, being condemned in most
ethical codes. The term envy is suggestive of a wish to bring down the incomes of
those better off. It need not be supposed that such sentiments are necessarily implied
by demoralisation arising from accentuated feelings of social inferiority – indeed the
perception that those on higher incomes deserve the high social position that one
cannot attain may be precisely the source of deterioration in psychological well being.
10. Suppose that f extends continuously to a function on the space of all con-
tinuous functions on the unit interval. Since this is a Banach space under suitable
norm k  k we could then, for example, take df to be the Fréchet derivative defined by:

kfðx þ ZÞ  fðxÞ  dfðxÞk

lim ¼ 0.
Z!0 kZk
54 IAN PRESTON

11. The assumption that the quantile function is differentiable or even continuous,
implicit in the stated domain for f, is probably stronger than needed. However,
given the use made of integration by parts in the proof, it would be necessary to
introduce a generalisation of the notion of derivative for x, and a careful treatment of
the issues involved is beyond the scope of this paper.
12. This is not to say, of course, that one can date this far back recognition of
formal criteria for inequality comparison and their relation to properties of the tax
system but Guiccardini does say, for example, in discussing the advantages of a
particular progressive tax structure that ‘‘so doing, not only shall such benefits follow
as I have said, but also the ranks of each shall be equally preserved, since we are all
citizens of the same rank, and thus all shall become truly equal as we reasonably
should be’’ (‘‘E cosı´ faccendo, non sole ne seguiranno tante utilitá e tanti beni che ho io
detto, ma ancora si conserverá equalmente el grado di ognuno, perché tutti siamo
cittadini e di uno medesimo grado, e cosı´ diventereno tutti veramente pari, come ragi-
onevolmente dobbiamo essere’’ ibid., p. 206).
13. Keen, Papapanagos, and Shorrocks (2000) extend Jakobsson’s results to cover
the case where taxes on certain ranges of income are zero.
14. We should be wary of linking changes in income inequality to welfare in this
context since individual well being depends on both income and hours.
15. If this condition were to fail then it would be appropriated to alter the defi-
nition of equilibrium rather than to assume wage might be decreasing in z. Assuming
workers of higher ability can do the jobs of less able workers then a sensible defi-
nition of equilibrium would equate demand for labour of type z and above to supply
of such labour and the equilibrium wage distribution over and around ranges of
ability where the condition fails would have flat sections. We assume away this
complication here.

ACKNOWLEDGMENTS

I am grateful for helpful comments from Tim Besley, Chris Gilbert, Terence
Gorman, Chris Harris, Peter Lambert, James Mirrlees, Stephen Nickell,
Hyun Shin, seminar participants and an anonymous referee. The paper
draws in large part upon my doctoral thesis, for the funding of which I am
grateful to the Economic and Social Research Council.

REFERENCES
Arnold, B. C. (1987). Majorization and the Lorenz order: A brief introduction. Berlin: Springer-
Verlag.
Atkinson, A. B. (1970). On the measurement of inequality. Journal of Economic Theory, 2,
244–263.
Barlow, R. E., & Proschan, F. (1975). Statistical theory of reliability and life testing: Probability
models. New York: Holt, Rinehart and Winston.
Inequality and Income Gaps 55

Blackorby, C., & Donaldson, D. (1978). Measures of relative inequality and their meaning in
terms of social welfare. Journal of Economic Theory, 18, 59–80.
Blackorby, C., & Donaldson, D. (1980). A theoretical treatment of indices of absolute in-
equality. International Economic Review, 21, 107–136.
Blackorby, C., & Donaldson, D. (1984). Ethically significant ordinal indexes of relative
inequality. In: R. L. Basmann & G. F. Rhodes (Eds), Advances in econometrics, Vol. 3:
Economic inequality: Measurement and policy (pp. 131–147). Greenwich: JAI Press.
Blum, W. J., & Kalven, H. (1953). The uneasy case for progressive taxation. Chicago: University
of Chicago Press.
Bruckner, A. M., & Ostrow, E. (1962). Some function classes related to the class of convex
functions. Pacific Journal of Mathematics, 12, 1203–1215.
Dalton, H. (1920). The measurement of the inequality of incomes. Economic Journal, 30,
348–361.
Dasgupta, P., Sen, A. K., & Starrett, D. (1973). Notes on the measurement of inequality.
Journal of Economic Theory, 6, 180–187.
Eichhorn, W., Funke, H., & Richter, W. F. (1984). Tax progression and inequality of income
distribution. Journal of Mathematical Economics, 13, 127–131.
Fei, J. C. H. (1981). Equity oriented fiscal programs. Econometrica, 49, 869–881.
Fellman, J. (1976). The effect of transformations on Lorenz curves. Econometrica, 44, 823–824.
Foster, J. E., & Ok, E. A. (1999). Lorenz dominance and the variance of logarithms. Econo-
metrica, 67, 901–907.
Gastwirth, J. L. (1971). A general definition of the Lorenz curve. Econometrica, 37, 1037–1039.
Guicciardini, F. (1932). La decima scalata. In: R. Palmarocchi (Ed.), F. Guiccardini, Opere,
Vol. VII: Dialogi e discorsi del regimmento di Firenze. Bari: G. Laterza.
Hardy, G., Littlewood, J., & Pólya, G. (1934). Inequalities. Cambridge: Cambridge University
Press.
Jakobsson, U. (1976). On the measurement of the degree of progression. Journal of Public
Economics, 5, 161–168.
Kakwani, N. C. (1977). Applications of Lorenz curves in economic analysis. Econometrica, 45,
719–727.
Keen, M., Papapanagos, H., & Shorrocks, A. (2000). Tax reform and progressivity. Economic
Journal, 110, 50–68.
Kolm, S.-C. (1969). The optimal production of social justice. In: H. Guitton & J. Margolis
(Eds), Public economics (pp. 145–200). London: Macmillan.
Kolm, S.-C. (1976a). Unequal inequalities I. Journal of Economic Theory, 12, 416–442.
Kolm, S.-C. (1976b). Unequal inequalities II. Journal of Economic Theory, 13, 82–111.
Lorenz, M. O. (1905). Methods of measuring the concentration of wealth. Publications of the
American Economic Association, 9, 209–219.
Marshall, A. W., & Olkin, I. (1979). Inequalities: Theory of majorization and its applications.
New York: Academic Press.
Marshall, A. W., Olkin, I., & Proschan, F. (1967a). Monotonicity of ratios of means and other
applications of majorization. In: O. Shisha (Ed.), Inequalities: Proceedings of a sympo-
sium (pp. 177–190). New York: Academic Press.
Marshall, A. W., Walkup, D. W., & Wets, R. J.-B. (1967b). Order-preserving functions:
Applications to majorization and order statistics. Pacific Journal of Mathematics, 23,
569–584.
Moyes, P. (1987). A new concept of Lorenz domination. Economics Letters, 23, 203–207.
56 IAN PRESTON

Moyes, P. (1988). A note on minimally progressive taxation and absolute income inequality.
Social Choice and Welfare, 5, 227–234.
Moyes, P. (1994). Inequality reducing and inequality preserving transformations of income:
Symmetric and individualistic transformations. Journal of Economic Theory, 63,
271–298.
Muirhead, R. (1903). Some methods applicable to identities and inequalities of symmetric
algebraic functions of n letters. Proceedings of the Edinburgh Mathematical Society, 21,
144–157.
Musgrave, R. A., & Thin, T. (1948). Income tax progression 1929–1948. Journal of Political
Economy, 56, 498–514.
Pigou, A. C. (1912). Wealth and welfare. London: Macmillan.
Preston, I. (1989). The redistributive effect of progressive taxation. DPhil thesis, University of
Oxford.
Preston, I. (1990a). Ratios, differences and inequality indices. IFS Working Paper W90/9.
Preston, I. (1990b). Income redistribution and labour supply specification. IFS Working Paper
W90/14.
Rothschild, M., & Stiglitz, J. (1973). Some further results on the measurement of inequality.
Journal of Economic Theory, 6, 188–204.
Schur, I. (1923). Über eine Klasse von Mittelbildungen mit Andwungen auf die Determinant-
theorie. Sitzungsber Berlin Mathematische Gesellschaft, 22, 9–20.
Seligman, E. R. A. (1894). Progressive taxation in theory and practice. Publications of the
American Economic Association, 9, 7–222.
Sen, A. K. (1973). On economic inequality. Oxford: Clarendon Press.
Sen, A. K. (1976). Welfare inequalities and Rawlsian axiomatics. Theory and Decision, 7,
243–262.
Sen, A. K. (1978). Ethical measurement of inequality: some difficulties. In: W. Krelle & A. F.
Shorrocks (Eds), Personal income distribution (pp. 81–94). Amsterdam: North-Holland.
Shorrocks, A. F. (1983). Ranking income distributions. Econometrica, 50, 3–17.
Thon, D. (1987). Redistributive properties of progressive taxation. Mathematical Social
Sciences, 14, 185–191.
Zheng, B. (2007). Utility gap dominances and inequality orderings. Social Choice and Welfare,
28, 255–280.
HOW PROGRESSIVE IS
PROGRESSIVE TAXATION? AN
AXIOMATIC ANALYSIS$

Udo Ebert and Georg Tillmanny

ABSTRACT
There is a consensus in the general public that income taxes should be
everywhere progressive. Starting from the basic properties normally re-
quired, we examine the possibilities of designing everywhere progressive
income tax schedules. An axiomatic analysis investigates the (in)con-
sistency of these requirements with further restrictions on the degree of
progression. It turns out that everywhere progressive tax schedules have
to be maximally progressive or almost proportional in some income range.

1. INTRODUCTION

In the theory of optimal income taxation the maximization of a given social

welfare function allows us to derive the form and the structure of an optimal

$
This paper is based on an unpublished discussion paper (Ebert & Tillmann, 2004) and it was
completed after Georg Tillmann’s death. Udo Ebert owes him a debt of gratitude for his
friendship and his collaboration.

Equity
Research on Economic Inequality, Volume 15, 57–71
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15004-3
57
58 UDO EBERT AND GEORG TILLMANN

tax schedule. It has turned out that these schedules are almost always
regressive for some income ranges (cf. Tresch, 2002). This is in particular
true for Mirrlees’ (1971) optimal nonlinear income tax which is regressive
even for the top incomes. On the other hand there seems to be a consensus
among voters and politicians that income taxes should be everywhere pro-
gressive and that the marginal tax rate should increase with income.1 Accord-
ing to this view, income taxation is an instrument which mainly is to be used
for redistribution and for generating a more equal income distribution.
This paper investigates the possibilities for defining everywhere or almost
everywhere progressive income taxes. The schedules considered have to
satisfy the usual properties required for income taxation. Furthermore,
several conditions on the degree of progression are introduced. Based on an
axiomatic analysis the existence of corresponding tax schedules is examined
and their general structure is described.
Musgrave and Thin (1948) have suggested four local measures of pro-
gression. We will employ residual progression: Using this measure Jakobsson
(1976) has shown that one income tax is everywhere more progressive than
another one if and only if its net income distribution Lorenz dominates
the other net income distribution for arbitrary gross income distributions.
An increase in the degree of progression implies an increase in redistribution
and yields a decrease in the level of inequality (of net income distributions).
This result nicely demonstrates the meaning and relevance of residual pro-
gression. Residual progression is therefore the proper measure whenever the
redistribution of income is to be examined.
Now the questions arise how progressive an income tax should be
and how progressive it can be. The first question is related to equity. In
the literature two general principles are suggested in order to justify and to
design taxes: the benefit principle2 and the ability-to-pay principle. Since
we do not consider the government’s expenditure here, we concentrate on
the ability-to-pay principle. Given this general framework we put lower and
upper bounds and further restrictions on the degree of progression. Thus the
analysis is normative and based on general postulates.
We then deal with the second question. At first sight it seems that we
do not face any problems when a progressive income tax is to be designed:
One can choose a tax schedule leading to constant residual elasticity for any
degree of progression. Thus progression can remain moderate all the way
along the income scale. But things are a little bit more complicated. Such a
tax schedule calls for a region of negative tax liabilities and inevitably has
a negative marginal tax rate for the lowest incomes. This property is not a
mere pathology which can be neglected since this region need not be small,
How Progressive is Progressive Taxation? 59

but it is an undesired feature of the income tax as the transfer is increasing,

and not decreasing with income. One could argue that the introduction of a
threshold could remedy this problem: The constant progression schedule
can be modified to have a zero bracket amount, i.e. zero taxes up to the
point at which the transfer becomes zero. Then indeed the negative tax rates
are avoided, but another problem arises. The tax is proportional in the
region below the threshold and no longer progressive. Thus we have to deal
with this problem in more detail.
We will impose a set of usual properties on income tax schedules and will
then perform an axiomatic analysis, which demonstrates the limits of pro-
gressive taxation. Besides continuity as a regularity condition we require
that net income is strictly positive and that the tax schedule is not a transfer
everywhere. Furthermore, we postulate that the tax liability has to be in-
creasing in income, i.e. with a taxpayer’s ability-to-pay, and we exclude rank
reversals due to taxation by bounding the marginal tax rate from above.
Finally, the tax schedule is to be progressive. These properties present a
minimal set of necessary properties and define feasible tax schedules.
Then we introduce further requirements. It turns out that – whenever one
wants to put bounds or further restrictions on the degree of progression –
one runs into difficulties. In principle, tax functions can be found for any
pattern of progression, but they do not necessarily fulfill the ability-to-pay
principle (i.e. the marginal tax rate may be negative for low incomes!). It
is impossible to design tax schedules satisfying the usual properties and
being only ‘‘moderately’’ progressive. In order to get feasible schedules
one has to admit maximal progression at some incomes. Similarly, if one
requires that the degree of progression is increasing with income, the tax has
to be almost proportional in some income ranges. Thus the analysis shows
that – given the ability-to-pay principle – further restrictions (above the
normal requirements) cannot easily be implemented.
The paper is organized as follows: Section 2 describes a minimal set of
basic properties tax schedules should satisfy. In Section 3 further require-
ments are introduced. Their implications are derived and their consistency
with the basic properties is investigated. Finally, Section 4 presents some
discussion and offers some conclusions.

2. BASIC PROPERTIES OF TAX SCHEDULES

In this section we consider the standard properties we want to impose

on income taxes. An income tax schedule is a function T : Rþ ! R
60 UDO EBERT AND GEORG TILLMANN

representing the tax liability TðX Þ of X
0 where X is taxable income.

Income is assumed to be exogenous.
Now we introduce the properties a feasible tax schedule has to satisfy.
B1. (Regularity) T is continuous for all X and continuously differenti-
able3 for all but a finite number of (strictly positive) incomes.
B2. (Positive net income; tax) TðX ÞoX for all X>0 and there is X0 such
that TðX 0 Þ40:
Continuity excludes any jump in tax liability. The restriction of differ-
entiability is typically met in the real world where piecewise differentiable or
piecewise linear income taxes tend to prevail (see e.g. Messere, 1998). The
first part of B2 guarantees that net income is always strictly positive when
gross income is strictly positive. As an implication we obtain T(0)r0. Sub-
sidies T(X)o0 are admitted, but we postulate that the tax liability is strictly
positive for some incomes.
Next we turn to equity principles and present two conditions on marginal
tax rates.
B3. (Ability-to-pay) T 0 ðX Þ
0 for all X
0:
B4. (No reranking) T 0 ðX Þ  1 for all X
0:
The principle of vertical equity requires that individuals with higher
income also have to pay higher taxes (Moyes (1988) labels the condition
‘‘minimal progression’’). According to B3, the tax liability has to be
(weakly) increasing in income. This property is related to the ability-to-pay
principle and puts the idea of the ‘‘appropriately unequal treatment of
unequals’’ (see Lambert, 2001, p. 175) in concrete forms. The principle of
horizontal equity is satisfied in our framework a priori since income is the
sole indicator of a taxpayer’s ability-to-pay. According to B4, net income
should (at least weakly) increase in gross income. Then the ranking of
incomes is not changed by taxation, a property which is also postulated
by vertical equity. Furthermore, some incentives to increase income are to
be left. It is an obvious restriction.
The properties examined up to now are standard in tax theory and have
no direct implications for the degree of progression. Therefore we finally
introduce
B5. (Progression) T(X) is progressive for all X>0.
Progression is defined in the usual way: A tax schedule T is progressive
at X>0 if and only if the average tax rate4 T̄ðX Þ ¼ TðX Þ=X is strictly
How Progressive is Progressive Taxation? 61

increasing in income at X. It is a local property and requires an increasing

tax burden (measured by the average tax rate). It should be mentioned that
it is B5 which rules out taxes with a zero bracket amount. This property is
fundamental in the following. It will not be relaxed later in the paper.
The derivative dT̄ðX Þ=dX might be used as an indicator of progression.5
But residual progression seems to be more attractive since it is related to the
intensity of the redistribution of income. Residual progression r(X) is de-
fined by the corresponding income elasticity of net income.6 It is well known
that the definition of the elasticity can be rearranged (see Lambert, 2001):

ð1  T 0 ðX ÞÞX 1  T 0 ðX Þ
rT ðX Þ ¼ ¼ (R)
X  TðX Þ 1  T̄ðX Þ

Thus residual elasticity is determined by the marginal and average tax rate
at X and indicates whether the tax is progressive. This relationship will be
used in the proofs below at various places. We have for X>0: the tax T is
progressive at X 30  rT ðX Þo1: In other words, for a progressive tax
schedule a one percent increase in gross income leads to an increase of less
than one percent

in net income. For later use we define the set R ¼ fr :
Rþ ! ½0; 1Þ
r continuous for all X but a finite number of (strictly positive)
incomes}.
The properties B1–B5 are basic properties for progressive income tax
schedules and are typically satisfied. A linear income tax represents a simple
example satisfying these properties: T 1 ðX Þ : ¼ aX  b for 0oao1 and
b>0. In the following we want to consider the set T ¼ fT : Rþ !
Rj T satisfies B1–B5}. We call T 2 T a feasible (progressive) tax schedule.

3. THE DEGREE OF PROGRESSION

Now we suggest and examine some properties which restrict the degree
of progression. In this paper we assume that the general public is interested
in redistribution by means of income taxation, but that extreme cases are to
be avoided. One extreme situation can be described by r(X)1, i.e. by a
proportional income tax T2(X): ¼ aX for 0oao1. In this case there is no
redistribution at all (measured by Lorenz dominance). We postulate that
redistribution is noticeable, i.e. there is e1>0 such that r(X)r1e1o1
for all X. The constant e1 may be arbitrarily small, but guarantees at
least a minimal degree of redistribution. The other extreme situation is
given by r(X)0. Then the only tax schedule satisfying this condition is
62 UDO EBERT AND GEORG TILLMANN

T3(X): ¼ Xb for b>0. Here redistribution is complete. Everybody ends up

with the same income b. We want to exclude this case of maximal progres-
sion, too, and introduce a lower bound e2>0 for r(X). Combining both
aspects we obtain a postulate for ‘‘moderate’’ progression which may well be
found acceptable by typical voters and politicians favoring redistribution:

P1. (Moderate progression) There are e1 and e2 such that 0oe2rr

(X)r1e1o1 for all X.

In order to examine the implications of P1 we reconsider the linear tax

T1(X) and recognize that rT 1 ð0Þ ¼ 0 and rT 1 ðX Þ goes to one when X tends to
infinity. In other words, the range of rT 1 is equal to ½0; 1Þ and therefore the
linear income tax does not satisfy P1. This result is surprising but has to be
expected in view of Proposition 1:

Proposition 1. There is no feasible tax satisfying P1.

The basic properties B1–B5 and moderate progression P1 are incon-

sistent. We obtain a typical impossibility result. At the same time we learn
that there are limits to the redistribution of income. Obviously we cannot
choose a feasible income tax for an arbitrary pattern of progression.
Therefore, in the following we want to explore the possibilities of de-
signing progressive taxes. Then we have to weaken P1 in order to get feasible
tax schedules. But at first we ‘‘invert’’ the problem: We start from a given
pattern of progression, a function of the residual elasticity rðX Þ 2 R and
ask the question if there exists any tax function T(X) leading to r(X)
(i.e. rT (X) ¼ r(X) for all X>0) and whether it fulfills the basic properties.

We establish

Proposition 2. For every r 2 R [which satisfies P1] there is a function

T : Rþþ ! R; which is continuous and piecewise differentiable, such
that rT ðX Þ ¼ rðX Þ for all X>0. If T(0), T 0 ð0Þ exist,7 TðX Þ satisfies B1–
B2, B4–B5 [and P1] and there is exactly one X0>0 such that T(X0) ¼ 0.

This result is comforting. It shows that for an arbitrary pattern of pro-

gression one can find a tax function T(X) generating it. Furthermore, T(X)
automatically satisfies all basic properties, but B3. Of course, if r 2 R
is chosen appropriately, B3 can be fulfilled along with all other properties:
e.g. rðX Þ ¼ ½ð1  aÞX =½ð1  aÞX þ b for 0oao1 and b>0 leads to the
linear income tax T1(X) which is feasible (but does not satisfy P1).
How Progressive is Progressive Taxation? 63

It is clear that – given B1–B2, B4–B5 – the properties B3 and P1 cannot be

satisfied simultaneously (cf. Proposition 1). Thus the reason for the incon-
sistency of B1–B5, and P1 is a conflict between the ability-to-pay principle
B3 and the requirement of moderate progression. There are two possibili-
ties: one can do without B3 or one can weaken P1. (By assumption pro-
gression B5 is fundamental and non-negotiable.) Ignoring B3, i.e. admitting
negative tax rates, leads to a strange type of tax schedule: one gets T(0) ¼ 0
and a subsidy for the lowest incomes X>0 which is increasing (!) with
income on some interval ð0; X̄ Þ (see Ebert and Tillmann (2004) and cf. the
discussion in the introduction). This seems to be unjust and is not too
interesting. Therefore in the following we will alter P1.
At first we still postulate a minimal degree of progression, but allow for
maximal progression:
P2. (Minimal progression) There is 0oe1o1 such that
0  rðX Þ  1 
1 o1 for all X
We know that T3(X) (as an extreme case) satisfies this property. There-
fore we can expect to obtain a possibility result:
Proposition 3.
(a) There exist feasible tax functions T(X) satisfying P2.
(b) If T(X) satisfies P2 there is X0>0 such that T(X0) ¼ 0, and8
limX !1 T 0 ðX Þ ¼ 1:

The feasible tax schedule fulfilling P2 can be described precisely: In this case
low incomes are subsidized, but as an implication of B3 the subsidy is
decreasing with income (which leads to maximal progression at X ¼ 0) and
the marginal tax rate goes to one if income tends to infinity, i.e. it is almost
one for high incomes. Example T3(X) is not the only tax schedule fulfilling
P2. There are more attractive ones, e.g.
(
aX  b for X  b=a
T 4 ðX Þ ¼ 1


X  ðb=aÞ X for X
b=a

for 0oao1, b>0, and 0oeo1. In this case we obtain rT 4 ðX Þ 2 ½0; 1  a for
XA[0, b/a] and rT 4 ðX Þ ¼
for XA[b/a,N] (i.e. P2 is satisfied). Furthermore,
T4(X)o0 for Xob/a and T 04 ðX Þ ¼ 1 
X
1 tends to one for X-N.
Thus dropping the strictly positive lower bound of r(X) allows us to get
feasible income taxes guaranteeing a minimal degree of redistribution. Next
we introduce another way of restricting the degree of progression. r(X) is by
64 UDO EBERT AND GEORG TILLMANN

definition an elasticity: An increase of gross income X by one percent in-

creases net income by r(X) percent. Progression implies that r is smaller
than unity. One can argue that r(X) should decrease or at least not increase
with income X. Therefore, assuming that the tax functions T(X) are twice
and the functions of the residual elasticity r(X) are once piecewise contin-
uously differentiable we suggest the property
P3. (Nondecreasing progression) r(X) is nonincreasing ðr0 ðX Þ  0Þ for
all X.
which reflects this idea.9 In this case the degree of redistribution does not
decrease with income. A direct implication is the requirement r(X)rr(0) for
all X, i.e. the degree of progression for the lowest income forms an upper
bound. Furthermore, this property rules out (piecewise) linearity and it has
an immediate consequence:
Lemma 1. [P3 and rT (X)>0] or decreasing progression [r0T ðX Þo0; the
strict form of P3] implies that T 0 ðX Þ strictly increases (i.e. T 00 ðX Þ40 for all
X>0).
The Lemma shows that – given B3 – the property P3 implies that T(X) is
convex. The converse is not true: Consider the tax function T 5 ðX Þ : ¼
aX  gX
for g40; 0oao1, and eo1: then T 00 ðX Þ40; but also r0T 5 ðX Þ40
for all X. Therefore the property P3 is weaker than the assumption that the
marginal tax rate is increasing.
Now we combine both ideas of refining the concept of progression: We
impose P3 and restrict the range of r again. Then we obtain the following
results:
Proposition 4.
(a) There is no feasible tax function T(X) satisfying P3 and the condition
0orT (X)o1 for all X.
(b) A feasible tax function T(X) satisfies P3 and the condition 0rrT
(X)o1 for all X if and only if there is b>0 such that T(X) ¼ T3(X) ¼
Xb.
(c) There exist feasible tax functions T(X) satisfying P3, 0orT (X) and
rT (0) ¼ 1. Then T(0) ¼ 0, T(X)>0 and T 00 ðX Þ40 for all X>0 and
limX !1 T 0 ðX Þ ¼ 1:

Part (a) shows that the ability-to-pay principle (B3) and mild progres-
sion ð0orðX Þ  rð0Þo1Þ are not compatible with each other. Therefore we
extend the range of r(X): If maximal progression is admitted (0rr(X))
How Progressive is Progressive Taxation? 65

exactly one tax schedule satisfies the conditions imposed in (b). T3(X)
implies the leveling of incomes and is not acceptable in practice.
Thus we admit r(0) ¼ 1 in (c). Then the tax function can be propor-
tional for X ¼ 0 and ‘‘nearly proportional’’ for low incomes. Indeed, this
proceeding is successful. We obtain convex tax schedules. An example
is T 6 ðX Þ : ¼ X 2 =ðX þ 1Þ: It satisfies rT 6 ðX Þ ¼ 1=ðX þ 1Þ: Furthermore, it
is interesting to note, that the marginal tax rate for this type of schedule
must go to one for high incomes though r(X) could be constant and
strictly positive for high incomes. It also turns out that incomes are never
subsidized.
Finally we get a further, simple result by inspection of Propositions 3
and 4(c):
Corollary 1. Assume that limX !1 T 0 ðX Þo1 for a feasible tax schedule
T(X). Then T(X) cannot satisfy P2 or P3.
Both properties, minimal and nonincreasing progression, require that the
marginal tax rate tends to 100% for high incomes.
In the next section we will discuss some further aspects.

4. DISCUSSION AND CONCLUSION

We have used the terms ‘‘low incomes’’ and ‘‘high incomes’’ (X-N). These
terms are vague, but it has to be stressed that the corresponding income
ranges are not necessarily negligible or irrelevant in practice. The reason
is simple: One can scale up or down any given income interval appropriately
by defining another tax schedule which possesses the same properties as
the original one. Suppose e.g. that incomes are subsidized on the interval
ð0; aÞ or that the marginal tax rate is greater than 99% on the interval (a, N)
for a given schedule T(X). Then one can define a new tax schedule TðX ^ Þ:
¼ gTðX =gÞ for g>0. In this case, the essential properties of T at X
¯^ X^ Þ ¼ T̄ðX Þ;
are inherited by T^ at X^ ¼ gX : One obtains T^ 0 ðX^ Þ ¼ T 0 ðX Þ; Tð
^
and rT^ ðX Þ ¼ rT ðX Þ: If e.g. g ¼ 10:000; it means that X measures income
in 10.000h. Thus the results presented above are also relevant for tax
policy since we can find tax schedules used in practice which possess these
properties.
Finally we comment on the property limX !1 T 0 ðX Þ ¼ 1 implied in Pro-
positions 3 and 4(c). If the marginal tax rate tends to one for X-N,
one would expect that the tax schedule is highly progressive. This need
not be the case. Consider e.g. the tax schedule T 7 ðX Þ : ¼ ðX þ 1Þ  ðX þ 1Þ

66 UDO EBERT AND GEORG TILLMANN

for 0oeo1. It is feasible and we get

1

X ðX þ 1Þ
1
T 07 ðX Þ  ; rT ðX Þ ¼ ; and r0T 7 ðX Þ  0.
ðX þ 1Þ1
7
½ðX þ 1Þ
 1

Therefore we obtain T 07 ð1Þ ¼ 1; but rT 7 ð1Þ ¼
: Though the marginal

tax rate goes to one, any (nonextreme) degree of progression e, 0oeo1, can
be attained in the limit.
The above analysis demonstrates that a great variety of feasible tax
schedules exists. But whenever one wants to restrict the pattern of progres-
sion further, problems arise. Taxes with an easy-to-understand degree of
moderate progression have an undesired property (negative tax rates). Once
we seek to rule out this feature, in fact no schedule survives: A feasible tax
schedule cannot satisfy the property ‘‘moderate progression’’ (P1). There-
fore we relax conditions, one by one, to investigate the sort of taxes which
are after all possible. It turns out that one has to admit the possibility of
maximal progression: Then a feasible tax schedule fulfilling P2 subsidizes
low incomes and taxes all other incomes. We find maximal progression at an
income of zero since the subsidy is (absolutely) decreasing with income.
Similarly, if it is required that residual progression does not increase with
income (P3), we have to admit the possibility that the tax schedule is pro-
portional at X ¼ 0 (leaving aside the particular situation in which we get
T 3 ðX Þ ¼ X  b). In this case all incomes are taxed. The income tax is con-
vex in income. Moreover, P2 and P3 imply that the marginal tax rate tends
to one for high incomes. Most of the restrictions on the degree of progres-
sion, which have been considered in this paper, are problematic at the
boundaries of the income range.

NOTES
1. See Roemer (1999) for a political economy of progressive income taxation and
Lambert (2001) for a discussion of the ‘progressive principle’ and its justification.
2. Ebert and Tillmann (2007) derive tax schedules by means of the benefit principle.
3. We assume that in X ¼ 0 the right limit exists and that there is nZ0 such that in
X i ; i ¼ 0; :::; n; the left and the right limits exist.
4. Below we assume that the limit T̄ð0Þ exists.
5. This measure has also been suggested in Lambert (1984) as appropriate for
exploring the revenue responsiveness properties of a progressive income tax as
incomes grow.
How Progressive is Progressive Taxation? 67

6. rT (X) may be discontinuous at a finite number of (strictly positive) incomes.

Again we then consider the left and right limits.
7. In most cases the limits Tð0Þ and T 0 ð0Þ will exist, but it is well known that there
are functions f (X), differentiable for all X>0, for which these limits do not exist
(e.g. f (X) ¼ sin(1/X)). For the rest of the paper we exclude these cases.
8. We have to assume that limX !1 T 0 ðX Þ exists.
9. This condition is necessary and sufficient for an equiproportionate growth in all
incomes to reduce after-tax inequality (see Moyes, 1989).

ACKNOWLEDGMENT
Helpful comments and suggestions by Peter Lambert and a referee are
gratefully acknowledged.

REFERENCES
Dieudonné, J. A. (1969). Foundations of modern analysis. New York: Academic Press.
Ebert, U., & Tillmann, G. (2004). How progressive is progressive taxation? Wirtschaftswis-
senschaftliche Diskussionsbeiträge V-257-04. University of Oldenburg, Oldenburg.
Ebert, U., & Tillmann, G. (2007). Distribution-neutral provision of public goods. Social Choice
and Welfare, forthcoming.
Jakobsson, U. (1976). On the measurement of the degree of progression. Journal of Public
Economics, 5, 161–168.
Lambert, P. J. (1984). Non-equiproportionate income growth, inequality, and the income tax.
Public Finance, 39, 104–118.
Lambert, P. J. (2001). The distribution and redistribution of income (3rd ed.). Manchester:
Manchester University Press.
Messere, K. C. (Ed.) (1998). The tax system in industrialized countries. Oxford: Oxford
University Press.
Mirrlees, J. A. (1971). An exploration in the theory of optimum income taxation. Review of
Economic Studies, 38, 175–208.
Moyes, P. (1988). A note on minimally progressive taxation and absolute income inequality.
Social Choice and Welfare, 5, 227–234.
Moyes, P. (1989). Equiproportionate growth of incomes and after-tax inequality. Bulletin of
Economic Research, 41, 287–294.
Musgrave, R. A., & Thin, T. (1948). Progressive taxation in an inflationary economy. Journal of
Political Economy, 56, 498–514.
Roemer, J. E. (1999). The democratic political economy of progressive income taxation.
Econometrica, 67, 1–19.
Tresch, R. W. (2002). Public finance, a normative theory. Amsterdam: Academic Press.
68 UDO EBERT AND GEORG TILLMANN

APPENDIX

At first we prove some technical results used later on.

Lemma A.
(1) If Tð0Þ ¼ 0 and T 0 ð0Þ exists, then T̄ð0Þ exists and T̄ð0Þ ¼ T 0 ð0Þ:
(2) Let TðX Þ satisfy B2, B4, and B5 and let T̄ð0Þ exist.
(a) If Tð0Þ ¼ 0 then
(i) 1oT 0 ð0Þ  1 ) rT ð0Þ ¼ 1
(ii) ½T 0 ð0Þ ¼ 1 and rT ð0Þ exists ) rT ð0Þ 2 ½0; 1
(b) Tð0Þo0 ) rT ð0Þ ¼ 0
(3) If T 0 ð1Þ exists, then T̄ð1Þ exists and T̄ð1Þ ¼ T 0 ð1Þ:
(4) T 0 ð1Þo1 ) rT ð1Þ ¼ 1:

Proof.
(1) T 0 ð0Þ ¼ limX !0 ðTðX Þ  Tð0ÞÞ=ðX  0Þ ¼ limX !0 T̄ðX Þ ¼ T̄ð0Þ:
(2) (a) (i) If T 0 ð0Þ ¼ 1 we get TðX Þ ¼ Tð0Þ þ T 0 ð0ÞX þ T 00 ðZÞX 2 =2 with
Z 2 ð0; X Þ by Taylor’s formula and then TðX Þ ¼ X þ T 00 ðZÞX 2 =2:
B2 yields that T 00 ðZÞo0: Then T is not progressive. Suppose now
that 1oT 0 ð0Þo1: Then

1  T 0 ðX Þ
rT ð0Þ ¼ lim rT ðX Þ ¼ lim
X !0 X !0 1  T̄ðX Þ

1  lim T 0 ðX Þ
¼ ¼1 by ð1Þ
1  lim T 0 ðX Þ

(a) (ii) Suppose that T 0 ð0Þ ¼ 1: For T 0 ðX Þo0 we have

04T 0 ðX Þ4T̄ðX Þ: This implies that T̄ð0Þ ¼ limX !0 T̄ðX Þ ¼ 1:
Since rT ð0Þ exists by assumption, the limit limX !0 rT ðX Þ ¼
limX !0 ð1  T 0 ðX ÞÞ=ð1  T̄ðX ÞÞ exists, too. We obtain 0  rT ð0Þ
 1: Some examples in Ebert and Tillmann (2004) demonstrate
that for every a 2 ½0; 1 there is TðX Þ such that rT ð0Þ ¼ a:
(b) Now suppose that Tð0Þ ¼ : Ao0:
(i) If T 0 ð0Þ 2 R; the numerator 1  T 0 ð0Þ is finite and we get
ð1  T 0 ð0ÞÞ
rT ð0Þ ¼ lim ¼0
X !0 ð1  TðX Þ=X Þ

since TðX Þ=X ! 1

How Progressive is Progressive Taxation? 69

(ii) Suppose that T 0 ð0Þ ¼  1: We define GðX Þ : ¼ TðX Þ þ A:

Then Gð0Þ ¼ 0 and G 0 ðX Þ ¼ T 0 ðX Þ: Taylor’s formula
yields Gð0Þ ¼ GðX Þ þ G0 ðX Þð0  X Þ þ G00 ðZÞX 2 =2 with Z 2
ð0; X Þ and thus GðX Þ þ G 00 ðZÞX 2 =2 ¼ G 0 ðX ÞX :
If G 0 ð0Þ ¼ þ1 then G 00 ðX Þo0 in a small neighborhood
of 0. Then 0  G 0 ðX ÞX ¼ GðX Þ þ G 00 ðZÞX 2 =2oGðX Þ and
GðX Þ ! 0 implies that G 0 ðX ÞX ! 0: If G0 ð0Þ ¼ 1 then
G 00 ðX Þ40 in a neighborhood of 0 and 0
G0 ðX ÞX ¼
GðX Þ þ G 00 ðZÞX 2 =24GðX Þ:
GðX Þ ! 0 implies that G 0 ðX ÞX ! 0: Thus always limX !0
0
G ðX ÞX ¼ 0: Now limX !0 rT ðX Þ ¼ limX !0 ðð1  T 0 ðX ÞÞX Þ=
ðX  TðX ÞÞ ¼ limX !0 ðX  G0 ðX ÞX Þ=ðX  TðX ÞÞ ¼ 0 since
limX !0 ðX  TðX ÞÞ ¼ A40:
(3) Assume that limX !0 T 0 ðX Þ ¼ : B exists. Then for all e40 there exists
X(e) such that B 
 T 0 ðX Þ  B þ
for all X
X ð
Þ: On the other
Rhand T̄ðX Þ ¼ ðTðX ð
ÞÞ þ ½TðX Þ  TðX ð
ÞÞÞ=X and TðX Þ  TðX ð
ÞÞ ¼
X 0
X ð
Þ T ðZÞ dZ:
Then we obtain

ðX  X ð
ÞÞ TðX Þ  TðX ð
ÞÞ

ðB 
Þ 
X X
ðX  X ð
ÞÞ
 ðB þ
Þ and lim T̄ðX Þ ¼ B
X X !1

(4) Since T̄ð1Þ ¼ T 0 ð1Þ by (3) we obtain limX !1 rT ðX Þ ¼ ð1  lim

T 0 ðX ÞÞ=ð1  lim T̄ðX ÞÞ ¼ 1:

Proof of Proposition 1. B1, B2, B4, and B5 allow us to apply Lemma A (2).
rT ð0Þ ¼ 0 or rT ð0Þ ¼ 1 and P1 lead to a contradiction. Thus T 0 ð0Þ ¼ 1
which contradicts B3. &
Proof of Proposition 2.
(i) Existence: Let rðX Þ 2 R: If there is a (tax) function T(X) leading to
the residual progression rðX Þ; it must satisfy (cf. the definition (R)):

rðX Þ
T 0 ðX Þ ¼ 1  rðX Þ þ TðX Þ (*)
X
It is a linear inhomogeneous first-order differential equation. As
1r(X) and r(X)/X are continuous for all X>0 there is a (up to a
constant) unique solution T(X) (Dieudonné, 1969).
70 UDO EBERT AND GEORG TILLMANN

(ii) T(X) is differentiable for all X>0.

(iii) Now assume that rðX Þ is continuous in the intervals ½0; X 1 Þ;
ðX 1 ; X 2 Þ; . . . ; ðX n ; 1Þ: There is always a (up to a constant) unique
solution of (*) in each interval. These solutions can be combined
such that the resulting tax function T(X) is continuous and almost
everywhere differentiable. Then B1 is satisfied if we assume that
Tð0Þ and T 0 ð0Þ exist.
(iv) There is a unique solution T(X) s.t. TðZ 0 Þ ¼ T 0 for every ðZ 0 ; T 0 Þ:
If T 0 ¼ Z 0 the solution is T 1 ðX Þ ¼ X : It satisfies T 1 ð0Þ ¼ 0:
Now choose T 0 s.t. 0oT 0 oZ 0 : Then the solution T 2 ðX Þ fulfills
T 2 ðX ÞoX for all X>0 since T 1 and T 2 must not intersect. This
proves B2.
(v) Rearranging (R) we obtain T 0 ðX ÞX  TðX Þ ¼ ð1  rðX ÞÞ
ðX  TðX ÞÞ:
Since rðX Þo1 and TðX ÞoX ; the function TðX Þ is progressive B5.
(vi) We rewrite (R) again and obtain X ð1  T 0 ðX ÞÞ ¼ rðX ÞðX  TðX ÞÞ:
Then rðX Þ ¼ 0 implies T 0 ðX Þ ¼ 1: rðX Þ40 and B2 yield
0
T ðX Þo1: Thus we get B4.
(vii) There is X 0 40 such that TðX 0 Þ ¼ 0: Suppose TðX Þ40 for all X>0,
(*) yields that T 0 ðX Þ40: B2 then implies that Tð0Þ ¼ 0: If T 0 ð0Þ exists,
we obtain T 0 ð0Þ ¼ 1  rð0Þ þ rð0ÞT̄ð0Þ or ðT 0 ð0Þ  1Þð1  rð0ÞÞ ¼ 0:
For 0  T 0 ð0Þ  1 we get T 0 ð0Þ ¼ 1 or rð0Þ ¼ 1; i.e. in each case
rð0Þ ¼ 1 in view of Lemma A (2) (a) (i). Then T cannot be progressive.
(viii) There exists at most one X0 such that TðX 0 Þ ¼ 0: (*) implies that
T 0 ðX 0 Þ ¼ 1  rðX 0 Þ ¼ 0 and T 0 ðX Þ41  rðX Þ for X 4X 0 :
(ix) If P1 is imposed, B3 cannot be satisfied because of Proposition 1.
&

Proof of Proposition 3. See Proposition 2 and example T 4 ðX Þ:

(
aX  b for X  b=a
T 4 ðX Þ ¼ 1


X  ðb=aÞ X for X
b=a

If T 0 ð1Þ exists, then T̄ð1Þ ¼ T 0 ð1Þ: Suppose that T 0 ð1Þo1; then

rT ð1Þ ¼ 1 by Lemma A (4), a contradiction. Thus T 0 ð1Þ ¼ 1:
Proof of Lemma 1. Differentiate (*):
   0
00 0 TðX Þ  X TðX Þ
T ðX Þ ¼ r ðX Þ þ rðX Þ
X X
Then T00 (X)>0 for [r0 (X)r0 and rðX Þ40] or r0 (X)o0. &
How Progressive is Progressive Taxation? 71

Proof of Proposition 4.
(a) Since rT ð0Þ40 we get T 0 ð0Þ ¼ 1 by Lemma A (2) (a) (ii), a con-
tradiction to B3.
(b) If rT ð0Þ ¼ 0 then r0T ð0Þ ¼ 0 and thus TðX Þ ¼ X  b is the only func-
tion satisfying (*).
(c) Example T 6 ðX Þ guarantees existence. If rT ð0Þ40 then Tð0Þ ¼ 0 by
Lemma A (2) (b). B3 implies T 0 ðX Þ
0: The conditions 0orT ðX Þ and
r0T ðX Þ  0 imply T 00 ðX Þ40 (Lemma 1). Thus TðX Þ40 for X 40:
Since rT ð1Þo1 we obtain T 0 ð1Þ ¼ 1 by Lemma A (4). &
INEQUALITY AND THE CHOICE OF
THE PERSONAL TAX BASE

Nigar Hashimzade and Gareth D. Myles

ABSTRACT

It is possible to employ either income or expenditure as the base for

personal taxation. A considerable literature has developed that investi-
gates the relative efficiency of these bases. The answer is usually in favor
of the expenditure tax since it does not distort the choice between con-
sumption and saving. In contrast, the literature is almost silent on the
relative equity of the two bases. We investigate the redistributive con-
sequences of the choice in models with two sources of heterogeneity: skill
in employment and lump-sum endowment. The Gini coefficient is used to
measure the degree of equity achieved by the tax bases in static and
dynamic settings. Income taxes and expenditure taxes that generate equal
welfare or equal revenue are compared. In the static economy the income
tax leads to lower inequality except when skill and endowment are neg-
atively correlated. Inequality is always lower with the income tax in the
dynamic economy. These results support the choice of income as the base
for personal taxation if reduction in inequality is a priority of policy.

1. INTRODUCTION

Equity will feature high on any list of the properties that a good tax system
should possess. Horizontal equity simply requires that equals are treated

Equity
Research on Economic Inequality, Volume 15, 73–97
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15005-5
73
74 NIGAR HASHIMZADE AND GARETH D. MYLES

equally. The interpretation of vertical equity is more open to debate. The

idea that vertical equity implies redistribution from the fortunate to the less
fortunate is generally accepted. What is disputed is how to define the
fortunate, and the extent of justifiable redistribution. Since Mirrlees (1971),
the theory of taxation has focused on the possession of a high level
of (unobserved) skill as the characteristic of good fortune and has viewed
(observed) income as the market signal of skill. Within this literature
attention has focused on the link between the redistributive motives of the
government and the structure of the optimal tax function (see Hashimzade
& Myles, 2007). The choice of the personal tax base from an equity pers-
pective has received little attention.
The argument for employing expenditure, rather than income, as the base
for taxation has a long and distinguished history. There are claims that
it can be traced back to Hobbes (1660) (see Batina & Ihori, 2000), but
certainly both Mill (1888) and Ramsey (1927) argued that saving should be
exempt from taxation – precisely the feature that distinguishes an expendi-
ture base from an income base. Some of the strongest arguments in favor of
an expenditure tax were made by Kaldor (1955) and the Meade (1978)
review of taxation in the UK. The use of an expenditure tax was also
proposed in the US (see US Treasury Department, 1942). Despite this,
expenditure taxation has been adopted in just two countries (India and
Sri Lanka), and then only briefly. Most academic discussion has focused on
the efficiency benefit of expenditure taxation, with little said on the equity
aspects. One exception is Kaldor (1955), but his discussion of ‘‘taxable
capacity’’ does not correspond well to modern concepts of tax theory. The
contribution of this paper is to provide an assessment of the relative success
of expenditure taxation and income taxation in achieving equity objectives.
Why might an expenditure tax be preferred to an income tax? At least
three reasons are normally cited. First, an income tax distorts the choice
between consumption and saving but an expenditure tax does not. Second,
an expenditure tax allows integration of the tax treatment of the personal
and corporate sectors. Third, an expenditure tax removes the need to dis-
tinguish between capital gains and income. The second and third points may
have relevance from an administrative perspective. From the perspective of
economic analysis the major point is the first. If income is the basis for
taxation then the return to saving is taxed. The taxation of income from
saving raises the relative price of future consumption. This provides a
disincentive to save and, it is generally claimed, reduces the aggregate
level of saving and hence of national income. If expenditure forms the
base for taxation, savings are only taxed when expenditures are made. This
Inequality and the Choice of the Personal Tax Base 75

eliminates the distortion in the saving decision. This point is developed

further in Auerbach (2006).
Economic analysis of the choice of tax base has employed a range of
models to address this efficiency argument. The starting point for under-
standing much of the analysis is the result of Chamley (1986) and Judd
(1985) that the long-run tax on capital should be zero. The inefficiency of the
tax on capital income is emphasized by the welfare calculations of Chamley
(1981) and his observation that the replacement of a capital tax by a lump-
sum tax leads to an increase in consumption and welfare. These results point
to an efficiency advantage for the expenditure tax rather than the income
tax. Additional results have been obtained from simulations using overlap-
ping generations economies. Altig, Auerbach, Kotlikoff, Smetters, and
Walliser (2001) employ the Auerbach–Kotlikoff model (overlapping gener-
ations with each consumer living 55 years) and show that both a flat tax
and a consumption tax raise national income compared to a proportional
income tax. Furthermore, the consumption tax raises national income by
more. Endogenous growth models have also displayed the property that
a switch from an income tax to an expenditure tax raises the growth rate
(see the survey in Myles, 2000).
The literature has much less to say on the equity implications on the
choice between the tax bases. Comments have been made about the morality
issues (an income base taxes what you put into the economy, an expenditure
base taxes what you take out) but this is not what normally determines
the choice between tax instruments. One approach to an assessment of the
equity implication has been to calculate the effects of reform using data
from tax filing. Feenberg, Mitrusi, and Poterba (1997) contrast income taxes
and retail sales tax with various exemptions and find that the average tax
burden rises for most of the low income groups when a retail sales tax is
used. However, such analysis does not take account of re-optimization by
consumers or equilibrium adjustments.
The theory of equity requires taxes to compensate for differences in life-
time utility generated as the consequence of unchangeable characteristics
that are not the result of economic choices. For example, the Mirrlees (1971)
model of income taxation summarized such characteristics in the level
of skill. Having a higher level of skill raises economic opportunities and
therefore puts high skill consumers in a potentially better situation – but
it remains a choice whether to take advantage of the opportunity. The
first-best tax system involves a lump-sum tax levied on the value of these
unchangeable characteristics to provide compensation for those less fortu-
nate in the allocation (i.e. those with lower skill in the Mirrlees’s model).
76 NIGAR HASHIMZADE AND GARETH D. MYLES

From this perspective an income tax can be viewed as an approximation to

the first-best lump-sum tax on earning potential. The potential for greater
earned income is not the only source of inequality.
Inequality can also be due to unearned wealth (such as bequests). It
is worth noting that unearned wealth is not subject to tax when income is
chosen as the tax base. In contrast, an expenditure tax does tax this wealth
when it is consumed. The taxes, therefore, have different impacts on the two
sources of inequality and it may be thought that the expenditure tax will
be more redistributive since it taxes both sources (earned and unearned) of
finance for consumption. Some theoretical work on these lines has been
undertaken by Correira (2005) who combines the two sources of inequality.
Her main result is to show that an increase in the consumption tax with a
corresponding reduction in the income tax raises efficiency.
What we do in this paper is build upon the recognition of the two sources
of inequality and their interaction with the choice of tax base in achieving
redistribution. This is undertaken by extending the standard model of in-
come taxation to incorporate variation in initial wealth and variation in skill
across the population. We then contrast the success of the income tax to that
of the expenditure tax in reducing inequality. There is clearly an open
question here about which will be the most successful. The income tax does
not tax initial wealth but the expenditure tax is a blunter tool when it comes
to taxing skill. A priori, there is apparently a trade-off between the relative
benefits of the two instruments.
It is helpful at this point to describe how we implement the analysis. The
idea of judging the redistributive success of alternative tax instruments is
clearly open to a number of potential interpretations. To make the idea
concrete a framework for coherent comparison has to be developed. What
we choose to do is to measure inequality using the Gini coefficient applied to
lifetime income. We then contrast the value of the Gini achieved by income
taxation to that achieved by expenditure taxation. To ensure comparability
we make these comparisons at both equal levels of welfare and at equal
levels of government revenue. That is, we set the income tax rate, compute
the level of welfare (or revenue) and then find the expenditure tax rate that
generates the same level of welfare (or revenue). The Gini coefficients are
then computed and contrasted. These comparisons are made in both a static
economy and a dynamic economy. It is worth noting that government
revenue is distributed as a lump-sum grant to consumers and that the
grant is the only income for the poorest households. Holding revenue con-
stant is therefore equivalent to holding welfare constant for a Rawlsian
social welfare function. Since we make our welfare comparison using the
Inequality and the Choice of the Personal Tax Base 77

utilitarian criterion these two experiments span the range of values for the
importance given to equity.
The key element in the dynamic economy is the process through which
skill is transmitted from parent to child. We choose to model this directly via
the probability that a high-skill parent has a high-skill child (and similarly
for a low-skilled parent). An extended approach could consider the edu-
cation decision as standing between innate ability (transmitted probabilis-
tically from parent to child) and labor market skill. The level of education is
determined partly by parental wealth, partly by incentives, and partly by
government policy as in Blumkin and Sadka (2005). The first of these factors
will typically enhance inequalities in ability, whereas education policies are
invariably redistributive (see Hanushek, Leung, & Yilmaz, 2003). Incentives
can work in either direction. Recent evidence (Jäntti et al. (2005)) shows
there are substantial earnings persistence across generations, which suggests
that the redistributive effect of education is limited. In any case, the prob-
abilities we use can be understood as a reduced form representation of this
set of interactions. From this perspective we use a range of transmission
probabilities in order to encompass the possible mappings between wealth,
education, and skill.
The second section of the paper presents the comparison in a static
economy which is an extension of the standard Mirrlees’s framework.
Section 3 contrasts the two taxes in an overlapping generations model with
bequests. Conclusions are given in Section 4.

2. STATIC ECONOMY

This section contrasts the success at achieving equity of the income and
expenditure taxes in a static economy. The economy has two periods and a
population of consumers who differ in income and initial endowment. In the
first period each consumer makes a labor supply decision and allocates in-
come between consumption and saving. Consumers are retired in the second
period and finance consumption from saving. The economy is static, but the
fact that saving plays a key role in smoothing consumption across the life-
cycle allows the effect of income and expenditure taxes to be distinguished.

2.1. Model

Consumers are differentiated by two characteristics: initial endowment and

skill in employment. The level of skill is measured by the wage rate received.
78 NIGAR HASHIMZADE AND GARETH D. MYLES

Table 1. Types and Labeling.

eL eH

wL LL LH
wH HL HH

The endowment of consumer h is denoted eh and the wage received per unit
of labor supply is denoted wh. A consumer is described by the pair {eh, wh}.
The initial endowment can take one of two values, eL and eH, with
eLoeH. eL is called the low endowment and eH the high endowment. The
level of skill can also take two values. The low skill level is wL and the high
skill level is wH, with wLowH. The economy, therefore, has four types of
consumer. The labeling of these types is summarized in Table 1.
The population size is fixed so it is the proportion of each type that is
relevant for measuring welfare and inequality. Let ph denote the proportion
P of
population that is of type h, hA{LL, LH, HL, HH}. By definition h ph ¼ 1:
Using the labeling of types we have
X
X 2
s2e ¼ ph e2h
ph e h (1)

Similarly,
X
X 2
s2w ¼ ph w2h
ph wh (2)

and
X X X
sew ¼ ph eh wh
ph e h ph wh (3)

The correlation between endowment and skill plays a key role in the
interpretation of our results. The correlation coefficient is defined by r ¼
sew/sesw.
Each consumer lives for two periods. They work during the first period
of life and are retired in the second period. In the absence of taxation the
first- and second-period budget constraints for a consumer of type h are

x1h þ sh ¼ ‘h wh þ eh (4)

and

x2h ¼ ð1 þ rÞsh (5)

Inequality and the Choice of the Personal Tax Base 79

where ‘h is labor supply, sh is saving, and r is the (fixed) interest rate. These
per-period budget constraints combine to give the lifetime budget constraint

x2h
x1h þ ¼ ‘ h w h þ eh (6)
1þr

The labor supply and consumption choices are made to maximize the utility
function

Uðx1h ; x2h ; ‘h Þ ¼ a lnðx1h Þ þ ð1
aÞ lnð1
‘h Þ þ d lnðx2h Þ (7)

This specification of utility assumes that all consumers have the same pre-
ferences, so we abstract from the issue of capabilities affecting inequality
(Foster & Sen, 1997). We adopt a specific functional form to permit the
numerical comparison of the tax bases.
The degree of inequality in the population is measured by using the Gini
coefficient applied to the discounted value of lifetime income,
1 XX
G ¼1
minfI j ; I k g (8)
H 2m j k

where I j  ‘j wj þ ej þ rsj =ð1 þ rÞ and m is mean income. We interpret one

tax system as being more successful in reducing inequality than an alter-
native system if it generates a lower value of the Gini coefficient. We are not
the first to relate income taxation to economic indices. For example, Kanbur
and Keen (1989) consider how the income tax should be chosen to minimize
the value of a poverty or inequality measure. What has not been analyzed
previously is how income and expenditure taxes perform as determined,
in our case, by the value of the Gini with income taxation relative to the
Gini with expenditure taxation. Using the population proportions the Gini
coefficient can be written as
1 XX
G ¼1
Hpj pk minfI j ; I k g
H 2I j k
1 XX (9)
¼1
p p minfI j ; I k g
HI j k j k

P
where I ¼ j pj I j :
From this point onward let the low wage be given by wL ¼ 0 and the high
wage by wH ¼ w. Also, we consider only tax systems that are linear. Hence,
80 NIGAR HASHIMZADE AND GARETH D. MYLES

there is a constant marginal rate of tax and a common lump-sum subsidy for
all consumers. This applies to the income tax and the expenditure tax.

2.2. Income Tax

The introduction of income taxation modifies the budget constraints in the

two periods of life to

x1h þ sh ¼ eh þ wh ‘h ð1
tÞ þ g (10)

and

x2h ¼ ð1 þ rð1
tÞÞsh þ g (11)

where t is the tax rate and g the lump-sum grant. Note that the endowment
is not taxed since it is not income and that the treatment of interest income
in Eq. (11) reduces the return to saving. Combining these two budget con-
straints provides the lifetime budget constraint
x2h 2 þ rð1
tÞ
x1h þ ¼ eh þ wh ‘h ð1
tÞ þ g (12)
1 þ rð1
tÞ 1 þ rð1
tÞ

The lifetime budget constraint reveals how second-period incomes are dis-
counted at the net-of-tax rate of interest, and hence that the relative price of
second-period consumption is increased. This change in relative price dis-
torts the allocation of consumption across the lifecycle. Using this budget
constraint it is now possible to derive the optimal choices of each individual.
Consider first an individual of type LL or LH who has a low level of skill.
Since wh ¼ 0 for h 2 fLL; LHg it must be that ‘h is also zero. The consump-
tion choices of a low-skill consumer then solve the optimization problem
x2h 2 þ rð1
tÞ
max U h ¼ a lnðx1h Þ þ d lnðx2h Þ s:t: x1h þ ¼ eh þ g
fx1h ;x2h g 1 þ rð1
tÞ 1 þ rð1
tÞ
(13)

This optimization has the solution for the consumption levels in the two
periods of life
 
1 a 2 þ rð1
tÞ
xh ¼ eh þ g (14)
aþd 1 þ rð1
tÞ
Inequality and the Choice of the Personal Tax Base 81

 
d 2 þ rð1
tÞ
x2h ¼ ð1 þ rð1
tÞÞ eh þ g (15)
aþd 1 þ rð1
tÞ

and the solution for saving

 
d g d
sh ¼ eh þ ð2 þ rð1
tÞÞ
1 (16)
aþd 1 þ rð1
tÞ a þ d

The high-skill consumers will choose to supply a strictly positive amount of

labor for tax rates below some strictly positive cut-off point. All the nu-
merical computations that follow are for values below the cut-off. Hence,
while recognizing that corner solutions can arise, we consider only interior
solutions for high-skill consumers. For the consumers with wh ¼ w40 this
implies optimal choices are derived from

max U h ¼ a lnðx1h Þ þ ð1
aÞ lnð1
‘h Þ þ d lnðx2h Þ s:t: Eq: ð12Þ (17)

fx1h ;x2h ;‘h g

The resulting levels of consumption and labor supply are

 
1 a 2 þ rð1
tÞ
xh ¼ eh þ wð1
tÞ þ g (18)
1þd 1 þ rð1
tÞ

 
d 2 þ rð1
tÞ
x2h ¼ ð1 þ rð1
tÞÞ eh þ wð1
tÞ þ g (19)
1þd 1 þ rð1
tÞ

 
1
a eh g 2 þ rð1
tÞ
‘h ¼ 1
1þ þ (20)
1þd wð1
tÞ wð1
tÞ 1 þ rð1
tÞ

and the quantity of saving is

 
d 2 þ rð1
tÞ g
sh ¼ eh þ wð1
tÞ þ g
(21)
1þd 1 þ rð1
tÞ 1 þ rð1
tÞ

The tax policy is assumed to be purely redistributive so the revenue

raised by the government is returned as a lump-sum transfer to consumers.
All consumers receive the transfer regardless of income or endowment.
Denote the transfer by g. The value of the transfer is calculated from the
82 NIGAR HASHIMZADE AND GARETH D. MYLES

government’s two-period budget constraint

" #
g X r X
gþ ¼t ph w‘h þ p sh (22)
1þr h
1þr h h

In equilibrium g is obtained by taking into account the dependence of the

optimal choices of the consumers on the tax and transfer.

2.3. Expenditure Tax

With expenditure taxation the budget constraints in the two periods of life
become
x1h ð1 þ tÞ þ sh ¼ eh þ wh ‘h þ g (23)

x2h ð1 þ tÞ ¼ ð1 þ rÞsh þ g (24)

where t is the constant rate of expenditure taxation. Notice how the ex-
penditure tax treats the endowment and labor income symmetrically, and the
fact that income from saving is not taxed except as expenditure on con-
sumption. The allocation of consumption across time is not distorted by the
tax. Combining these two constraints into the lifetime budget constraint gives
 
1 x2h 1 2þr
xh þ ¼ eh þ wh ‘h þ g (25)
1þr 1þt 1þr

The lifetime budget constraint reflects the distortion introduced into the
consumer choice between leisure and consumption.
The optimal labor supply of an individual with a low skill level remains
‘h ¼ 0: The choices of consumption and savings are given by
 
a 1 2þr
x1h ¼ eh þ g (26)
a þ d1 þ t 1þr
 
d 1þr 2þr
x2h ¼ eh þ g (27)
a þ d1 þ t 1þr

and
 
d 2þr g
sh ¼ eh þ g
(28)
aþd 1þr 1þr
Inequality and the Choice of the Personal Tax Base 83

We again choose to remain within the range of parameter values for

which the labor supply of an individual with a high skill level is strictly
positive. Consumption, labor supply, and savings are then
 
1 a 2þr
xh ¼ eh þ wh þ g (29)
ða þ dÞð1 þ tÞ þ 1
a 1þr

 
dð1 þ rÞ 2þr
x2h ¼ eh þ wh þ g (30)
ða þ dÞð1 þ tÞ þ 1
a 1þr

 
1
a eh g 2þr
‘h ¼ 1
1þ þ (31)
ða þ dÞð1 þ tÞ þ 1
a wh wh 1 þ r

and
 
dð1 þ tÞ 2þr g
sh ¼ eh þ wh þ g
(32)
ða þ dÞð1 þ tÞ þ 1
a 1þr 1þr

The value of the transfer to every consumer is computed from the gov-
ernment budget constraint which, for expenditure taxation, is given by

g X  x2

gþ ¼t ph x1h þ h (33)
1þr h
1þr

2.4. Contrast

The intention is to contrast the success of the alternative tax bases at

achieving redistribution. As noted in Section 1, we need to be careful in the
way we conduct the comparison in order for the results to be meaningful.
The process adopted is to set the income tax at a fixed rate and compute the
level of welfare this generates. The expenditure tax is then derived that leads
to the same level of welfare. The government budget constraint implies the
equilibrium transfers for these tax rates which allows the value of the Gini
coefficient to be calculated. This provides a comparison of the redistribu-
tion achieved for income and expenditure tax bases at an equal welfare level.
The exercise is then repeated for a pair of taxes that generate identical levels
of government revenue (and through the government budget constraint
provide an identical value of the lump-sum transfer).
84 NIGAR HASHIMZADE AND GARETH D. MYLES

The second important aspect is to ensure that we make the comparison

for a sufficiently wide range of the underlying parameters. Recall that the
economy has both skill and endowment differences between consumers.
Numerically testing a range of specifications revealed that the parameter
that distinguishes different cases is the coefficient of correlation between
skill and endowment. We therefore conduct our equal welfare (and equal
revenue) comparisons for the full range of values of the correlation coeffi-
cient between
1 and 1.
The details of our calculations are as follows. We consider two values of
the income tax rate (t ¼ 0.1 and t ¼ 0.3). We set the low wage and low
endowment level at 0. The high endowment is set at e ¼ 1. For each tax rate
we consider high wages of w ¼ 1 and w ¼ 5. We set the probability of the
wage–endowment pairs (0,0) and (1,1) at p and the probability of the pairs
(0,1) and (1,0) at q. The population variances of the wage and of the
endowment are (p+q)2 and the correlation between the two is 1
4q. Hence
q ¼ 0 gives perfect positive correlation between endowment and skill, and
q ¼ 1/2 gives perfect negative correlation. By varying q between 0 and 1/2
we are then able to cover the range of correlation coefficients.
The contrast between the two tax bases is illustrated in Figs. 1 and 2 for
w ¼ 1. The correlation coefficient is measured on the horizontal axis and the
value of the Gini coefficient on the vertical axis. There is a single curve for
the income tax (GI) and two curves (GEW and GER) for the expenditure tax.
GEW is the value of the Gini at the same welfare level as achieved by the
income tax and GER is the value of the Gini at the same revenue level as the

0.6

0.5

0.4
GI
0.3 GEW
GER
0.2

0.1

0
-1 -0.5 0 0.5 1

Fig. 1. t ¼ 0.1, e ¼ 1, w ¼ 1.
Inequality and the Choice of the Personal Tax Base 85

0.5

0.4

0.3 GI

GEW
0.2
GER

0.1

0
-1 -0.5 0 0.5 1

Fig. 2. t ¼ 0.3, e ¼ 1, w ¼ 1.

income tax. The results show that when e ¼ w ¼ 1 the expenditure tax
generates a lower value of the Gini coefficient than the income tax (and
hence achieves an equilibrium with less inequality) when there is a nega-
tive correlation between endowment and skill. When the correlation
becomes sufficiently positive the income tax generates a lower Gini coeffi-
cient. There is very little difference between the comparison with equal
welfare and that with equal revenue. For these parameter values the choice
between the two tax bases is dependent upon the value of the correlation
coefficient.
The outcome of the comparison is different when eow. This is illustrated
in Figs. 3 and 4 for w ¼ 5. In these cases the income tax produces a lower
value of the Gini index for the entire range of values for the correlation
between skill and endowment. The reason for this change in outcome is that
the increase in labor income relative to endowment income permits the
income tax to be more successful at redistributing since it is levied on an
increased proportion of total income.
The results show that the relative success of the two tax bases is dependent
upon the relative values of labor income and initial endowment and the
coefficient of correlation between these values. It is an interesting observa-
tion that the expenditure base only achieves a lower value of the Gini
coefficient when there is negative correlation – in all other cases the income
base is preferable. The relative performance of the income tax is better with
positive correlation because in this case a higher average tax is levied on
those who are both skilled and receive a high endowment. Hence the tax on
income is a good proxy for a tax on the unobserved endowment.
86 NIGAR HASHIMZADE AND GARETH D. MYLES

0.5

0.45

0.4
GI
0.35 GEW

GER
0.3

0.25

0.2
-1 -0.5 0 0.5 1

Fig. 3. t ¼ 0.1, e ¼ 1, w ¼ 5.

0.5

0.4

GI

0.3 GEW

GER

0.2

0.1
-1 -0.5 0 0.5 1

Fig. 4. t ¼ 0.3, e ¼ 1, w ¼ 5.

It is likely that the empirical evidence would determine that the cor-
relation is positive in practice, thus providing a preference for the income
base. Researching the evidence would also reveal that in practical terms
initial endowments invariably arise from bequests. To argue that these
interactions are adequately captured within the static model would seem
to be pushing its interpretation too far. Instead, a better approach is to
Inequality and the Choice of the Personal Tax Base 87

model bequests explicitly by adopting an intertemporal model that embodies

a bequest motive.

3. DYNAMIC ECONOMY

The analysis of the static economy has demonstrated how the correlation
between endowment and skill affects the choice between the tax bases. In
practice non-earned initial endowments arise primarily from bequests which,
in turn, depend on the earning capacity of predecessors. This implies an
endogenous correlation between skill and endowment related to the trans-
mission mechanism of skill between generations. The static model captures
some of the consequences of inequality but it does not reflect the fact that the
endowments and skills are linked via the choices of dynasties of households.
It is therefore necessary to study a dynamic economy in which parents
choose to leave bequests to their children. This allows the accumulation of
wealth over time, the development of inequality, and the formation of an
endogenous intertemporal link between skill and endowment. What is key in
this economy is the mechanism by which skills are transmitted between
generations. We now repeat the comparison of the tax bases in a dynamic
economy where the transmission mechanism can be made explicit.

3.1. Model

We adopt an infinite-horizon economy that is populated by heterogenous

agents. The agents have identical preferences but differ in skill and endow-
ment. Each agent lives two periods. In the first period an agent receives an
endowment and labor income which he divides between consumption and
saving. In the second period he divides his savings between consumption
and bequest. The bequest becomes the endowment of his descendant. There
is no population growth, and the total size of the population is normalized
to unity, with equal proportions of young and old agents in every time
period. With government intervention the agents pay tax and receive a
transfer in every period. We consider two tax schemes: an income tax levied
on labor income and interest income, and an expenditure tax levied on
consumption in every period. For simplicity, wages and the interest rate
are exogenously fixed. The wage for skilled workers is normalized to unity,
and that for unskilled workers is normalized to zero, so that in equilibrium
unskilled workers do not supply labor.
88 NIGAR HASHIMZADE AND GARETH D. MYLES

The preferences of a consumer are described by the lifetime utility

function
uðÞ ¼ Uðx1 ; ‘Þ þ dV ðx2 ; bÞ (34)

where
Uðx1 ; ‘Þ ¼ a ln x1 þ ð1
aÞ lnð1
‘Þ (35)

and
V ðx2 ; bÞ ¼ b ln x2 þ ð1
bÞ ln b (36)

with x1 and x2 consumption in the first and in the second period of life,
respectively, ‘ labor supply, and b the bequest.
An infinitely lived government collects taxes and redistributes the reve-
nues equally among all agents in every period. We assume that the gov-
ernment can commit to a policy of a constant tax rate and a constant
transfer. There is no borrowing constraint on the government.
From the solution to each agent’s optimization problem we can express
the bequest as a function of endowment. Because the bequest becomes the
endowment of the next generation in a given dynasty, this function can be
viewed as a law of motion for the endowment. The functional form of the
law of motion depends on whether the bequestor is skilled or unskilled.
Therefore, in every generation the law of motion of the endowment switches
randomly between two regimes. We assume the probability that the de-
scendant of a skilled worker is skilled is equal to pss, and that the probability
the descendant of an unskilled worker is unskilled is equal to puu. The
process of these random switches is a two-state Markov chain with the
transition matrix
" #
pss 1
puu
P¼ (37)
1
pss puu

The process is ergodic and irreducible if pss o1; puu o1; and pss þ puu 40;
with ergodic probabilities
2 3
1
puu
" #
62
p
p 7 p1
6 ss uu 7
p¼6 1
p 7 (38)
4 ss 5 p2
2
pss
puu
Inequality and the Choice of the Personal Tax Base 89

45ο
eu es e

Fig. 5. Convergence to Steady State Bequests.

The ergodic probabilities can be interpreted as unconditional probabilities

of being in each regime (see Hamilton, 1994, Chapter 22). Hence, in the long
run on average p1 agents are skilled and p2 ¼ 1
p1 are unskilled. Any initial
distribution of endowments in the long run converges to a bimodal distri-
bution, with peaks at the stationary points of the two regimes. This is
illustrated in Fig. 5 where eu and es are the long-run bequests of the unskilled
and skilled, respectively. The government will run an ‘‘on average’’ balanced
budget if it computes the amount of transfer taking p1 skilled and p2 un-
skilled agents, with corresponding stationary endowments, as the tax base.

3.2. Taxation

With an income tax the first- and second-period budget constraints of an

agent with endowment e are
x1 þ s ¼ e þ w‘ð1
tÞ þ g (39)

x2 þ b ¼ sð1 þ rð1
tÞÞ þ g (40)

90 NIGAR HASHIMZADE AND GARETH D. MYLES

The wage of unskilled agents is set at zero so they supply no labor. The
optimal choices of an unskilled agent are
  
1 a 1
xu ¼ eþg 1þ ,
aþd 1 þ rð1
tÞ
  
db 1
x2u ¼ eþg 1þ ½1 þ rð1
tÞ ð41Þ
aþd 1 þ rð1
tÞ

‘u ¼ 0 (42)

d g
su ¼ x1u
(43)
a 1 þ rð1
tÞ

  
dð1
bÞ 1
bu ¼ eþg 1þ ½1 þ rð1
tÞ (44)
aþd 1 þ rð1
tÞ

The solution to the optimization problem of a skilled agent is

  
s a 1
x1 ¼ e þ wð1
tÞ þ g 1 þ (45)
1þd 1 þ rð1
tÞ

  
db 1
xs2 ¼ e þ wð1
tÞ þ g 1 þ ½1 þ rð1
tÞ (46)
1þd 1 þ rð1
tÞ

a xs1
‘s ¼ 1
(47)
1
a wð1
tÞ

d g
ss ¼ xs1
(48)
a 1 þ rð1
tÞ

  
s dð1
bÞ 1
b ¼ e þ wð1
tÞ þ g 1 þ ½1 þ rð1
tÞ (49)
1þd 1 þ rð1
tÞ

The long-run average tax revenue is

TR ¼ t½p1 ðw‘s þ rss Þ þ p2 rsu  (50)

Inequality and the Choice of the Personal Tax Base 91

and the transfer is computed from the government budget constraint

TR ¼ g. The budget need not balance every period so we are implicitly
assuming that the government can borrow and lend at the rate of interest r.
With the expenditure tax the first- and second-period budget constraints
of an agent with endowment e are

x1 ð1 þ tÞ þ s ¼ e þ w‘ þ g (51)

x2 ð1 þ tÞ þ b ¼ sð1 þ rÞ þ g (52)
Using the fact that unskilled agents supply no labor,
  
a 1 1
x1u ¼ eþg 1þ (53)
a þ d1 þ t 1þr

  
db 1 1
x2u ¼ eþg 1þ (54)
a þ d1 þ t 1þr

  
dð1
bÞ 1
bu ¼ eþg 1þ ð1 þ rÞ (55)
aþd 1þr
The solution to the optimization problem of a skilled agent is
  
a 1 1
x1s ¼ eþwþg 1þ (56)
1 þd1þt 1þr

  
db 1 1
x2s ¼ eþwþg 1þ ð1 þ rÞ (57)
1 þ d1 þ t 1þr

a xs1 ð1 þ tÞ
‘s ¼ 1
(58)
1
a w
  
dð1
bÞ 1
bs ¼ eþwþg 1þ ð1 þ rÞ (59)
1þd 1þr
The long-run average tax revenue is

TR ¼ t½p1 ðx1s þ x2s Þ þ p2 ðx1u þ x2u Þ (60)

and the transfer is computed from TR ¼ g.

92 NIGAR HASHIMZADE AND GARETH D. MYLES

3.3. Contrast

This section contrasts the two tax bases. This is done using two different
approaches. First, we consider the dynamic evolution of the economy be-
ginning from an arbitrary assignment of initial endowments. Second, we
analyze the instantaneous stationary equilibrium with population propor-
tions equal to the ergodic probabilities. In both cases we focus upon the
value of the Gini coefficient using equal welfare and equal revenue com-
parisons.
Figs. 6–9 depict the value of the Gini coefficient for income taxation and
the two Gini coefficients for expenditure taxation: one expenditure Gini is at
the same welfare level as for the income tax, the other Gini for expenditure is
at the same government revenue level. Two income tax rates are considered
(t ¼ 0.2 and t ¼ 0.4) and two different probabilities for a high-skill parent to
have a high-skill offspring (pss ¼ 0.2 and pss ¼ 0.8). The ergodic probabil-
ities that these generate imply long-run average proportions for high-skill of
5/13 and 5/7, respectively.
These simulations compute the Gini coefficient for 100 generations of 100
families, with zero initial endowment and a uniform distribution of skills
(0 or 1 with equal probability) for the families in the first generation. We
plot only from generation 10 since by this point the effect of the assumptions
on initial distribution has disappeared. In every period the agents (families)
choose their optimal consumption, leisure, and bequest given their endow-
ment and skills (wage income). The bequest becomes the endowment of the
agent’s offspring, whose skill is determined randomly, according to Eq. (34).

0.375

0.35

0.325

0.3

0.275
GI GER GEW

Fig. 6. t ¼ 0.2, pss ¼ 0.2, puu ¼ 0.5.

Inequality and the Choice of the Personal Tax Base 93

0.35

0.3

0.25

0.2

0.15
GI GER GEW

Fig. 7. t ¼ 0.4, pss ¼ 0.2, puu ¼ 0.5.

0.275

0.225

0.175

0.125

0.075
GI GER GEW

Fig. 8. t ¼ 0.2, pss ¼ 0.8, puu ¼ 0.5.

The economy does not reach a steady state since there is always randomness
in the ability of offspring.
What is observed in all the figures is that the Gini for income taxation is
on average below the two Ginis for expenditure taxation. Increasing t and
reducing pss emphasizes this effect. The results confirm the observation
made in the static setting that the income tax leads to a lower value of the
Gini. It should be observed that in this model for pss ¼ 0.8 there is a positive
correlation between wage income and endowment driven by the fact that
94 NIGAR HASHIMZADE AND GARETH D. MYLES

0.25

0.2

0.15

0.1

0.05
GI GER GEW

Fig. 9. t ¼ 0.4, pss ¼ 0.8, puu ¼ 0.5.

0.7 0.55

GI
GI
0.6 GER
GER 0.45
GEW
GEW
0.5
0.35
0.4

0.25
0.3

0.2 0.15
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Fig. 10. pss ¼ 0.2.

high-skill parents leave a higher bequest and are more likely to have high-
skill offspring. In contrast, for pss ¼ 0.2 the correlation between wage
income and endowment is negative. In all cases in the long-run equilibrium
the average endowment of both skilled and unskilled is less than the wage of
the skilled. Hence, the outcome in the dynamic economy (lower Gini with
the income tax) is consistent with the one in the static economy.
Inequality and the Choice of the Personal Tax Base 95

0.7
0.5 GI
0.6 GI GER
GER GEW
0.4
0.5 GEW

0.3
0.4

0.3 0.2

0.2 0.1

0.1 0
0 0.2 0.4 0.6 0.8 1 0 0.2 0.4 0.6 0.8 1

Fig. 11. pss ¼ 0.8.

The results for the cross-section, ‘‘stationary’’ analysis confirm the ob-
servations from the dynamic process. In Figs. 10 and 11 we plot the Gini for
an economy with the (instantaneous) proportion of skilled agents equal to
p1, the ergodic probability, or the long-run average proportion of skilled, for
a fixed pss, and puu varying from 0.01 to 0.99. In every case the Gini for the
income tax is below the two Ginis for the expenditure tax. This emphasizes
that the few cases in which the expenditure base is observed to produce
a lower Gini than the income base during the dynamic evolution are con-
sequences of particular random realizations of the economy. The long-run
stationary outcome confirms that the expected position is for the income
base to ensure a lower value of the Gini.

4. CONCLUSIONS

The intention of the paper was to contrast the relative success of alternative
bases for personal taxation. In a static model with inequality arising from
skill in employment and from initial endowment the income base performed
better in all cases considered if there was positive correlation between the
sources of inequality. The expenditure base only bettered the income base
when there was negative correlation and a low level of income from
employment. These results were strengthened in the dynamic model. The
96 NIGAR HASHIMZADE AND GARETH D. MYLES

income base performed better except for a small number of realizations

of the economy, and was clearly better in the long-run equilibrium. If the
choice over the tax base rests on the reduction of inequality these results
provide evidence in favor of an income base.
It seems natural to question the extent to which policy recommendations
can be drawn from these stylized models. Both models capture the fact that
inequality of income has two dimensions – earned and unearned – and the
dynamic model also involves accumulation of inequality over time through
the role of bequests. We would agree that the static model is limited by the
lack of transmission of inequality across generations. For this reason we
prefer to focus upon the outcome of the dynamic model. Reassuringly, the
results of the dynamic model not only support those from the static model
but are actually more decisive. The income tax performed better than the
expenditure tax for all the parameter combinations considered (many of
which have not been reported in the paper). The dynamic model was
simplified by the assumption of a fixed interest rate but this can be ration-
alized by assuming a small open economy or a constant marginal product
for capital. The advantage remains that it avoided intermixing issues of
redistribution and dynamic inefficiency. We therefore feel that our con-
clusions on the advantage of the income tax are robust.

ACKNOWLEDGMENTS
The paper has emerged from discussions of the Mirrlees Review of UK
Taxation. We wish to thank members of the Review, especially Richard
Blundell, Stephen Bond, and James Mirrlees. The paper has also benefited
from discussion at seminars in Exeter, Lancaster, and Manchester. The
comments of the three anonymous referees and the editor Peter Lambert
have also been helpful.

REFERENCES
Altig, D., Auerbach, A., Kotlikoff, L., Smetters, K., & Walliser, J. (2001). Simulating tax
reform in the United States. American Economic Review, 91, 574–595.
Auerbach, A. J. (2006). The choice between income and consumption taxes: A primer.
University of California.
Batina, R. G., & Ihori, I. (2000). Consumption tax policy and the taxation of capital income.
Oxford: Oxford University Press.
Inequality and the Choice of the Personal Tax Base 97

Blumkin, T., & Sadka, E. (2005). Income taxation with intergenerational mobility: Can higher
inequality lead to less progression? European Economic Review, 49, 1915–1925.
Chamley, C. (1981). The welfare cost of capital income taxation in a growing economy. Journal
of Political Economy, 89, 468–496.
Chamley, C. (1986). Optimal taxation of capital income in a general equilibrium with infinite
lives. Econometrica, 54, 607–622.
Correira, I. H. (2005). Consumption taxes and redistribution. CEPR Discussion Paper No.
5280.
Feenberg, D., Mitrusi, A., & Poterba, J. (1997). Distributional effects of adopting a national
retail sales tax. In: J. Poterba (Ed.), Tax policy in the economy (Vol. 11).
Foster, J. E., & Sen, A. (1997). On economic inequality. Oxford: Clarendon Paperbacks.
Hamilton, J. D. (1994). Time series analysis. Princeton: Princeton University Press.
Hanushek, E. A., Leung, C. K. Y., & Yilmaz, K. (2003). Redistribution through education and
other transfer mechanisms. Journal of Monetary Economics, 50, 1719–1750.
Hashimzade, N., & Myles, G. D. (2007). The structure of the optimal income tax in the quasi-
linear model. International Journal of Economic Theory, 3, 5–33.
Hobbes, T. (1660). Leviathan (reprinted by Penguin Classics, 1985).
Jäntti, M., Bratsberg, B., Roed, K., Raaum, O., Naylor, R., Österbacka, E., Björklund, A., &
Eriksson, T. (2005). American exceptionalism in a new light: A comparison of inter-
generational earnings mobility in the Nordic countries, the United Kingdom and the
United States. Memo. 34/2005, University of Oslo.
Judd, K. L. (1985). Redistributive taxation in a simple perfect foresight model. Journal of Public
Economics, 28, 59–83.
Kaldor, N. (1955). An expenditure tax. London: George Allen and Unwin.
Kanbur, R., & Keen, M. (1989). Poverty, incentives, and linear income taxation. In: A. Dilnot,
& I. Walker (Eds), The economics of social security. Oxford: Clarendon Press.
Meade, J. E. (1978). The structure and reform of direct taxation. London: George Allen and
Unwin.
Mill, J. S. (1888). Principles of political economy. New York: Appleton.
Mirrlees, J. A. (1971). An exploration in the theory of optimum income taxation. Review of
Economic Studies, 38, 175–208.
Myles, G. D. (2000). Taxation and economic growth. Fiscal Studies, 21, 141–168.
Ramsey, F. P. (1927). A contribution to the theory of taxation. Economic Journal, 37, 47–61.
US Treasury Department (1942). Proposal for a ‘‘consumption expenditure tax’’,
www.taxhistory.org/Civilization/Documents/Spending/hst9369/9369-1.htm
STRATEGIC WEIGHT WITHIN
COUPLES: A MICROSIMULATION
APPROACH

Kristian Orsini and Amedeo Spadaro

ABSTRACT
Individual strategic weight plays an important role in the intra-household
allocation of resources; however, empirical studies invariably find such
weight difficult to define in a plausible and computable way, given the
available data. This paper proposes a framework for the calculation of
household members’ strategic weight that can be easily computed using a
microsimulation model. The index proposed for each member as the share
of resources the household would lose should he or she abandon it. The
causes of strategic weight differentials are analysed in four EU countries
with significantly different employment structure and tax-benefit systems
(Finland, Germany, Italy and the United Kingdom), using EUROMOD,
an integrated EU-15 microsimulation model.

1. INTRODUCTION

What advantage is there in individualising income tax or social benefits, as

opposed to splitting or pooling them? Does it matter whether family and

Equity
Research on Economic Inequality, Volume 15, 99–132
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15006-7
99
100 KRISTIAN ORSINI AND AMEDEO SPADARO

other personal circumstances are taken into account when calculating in-
work benefits or tax credits? What are the likely consequences of each policy
option in terms of personal income and welfare distribution, as opposed to
household distribution? How does redistribution policy affect the household
decision-making process and the welfare of individuals within families? With
regard to reforms of the tax or redistribution system, much of the economic
and political debate has focused on such questions in all European countries
over the past three decades.
Economists have for long been ill-equipped to tackle these issues, insofar
as they have become accustomed to treat households as if they were indi-
viduals, and to use household data in a similar fashion. The need to analyse
policy impacts at an individual level forced researchers to propose alterna-
tives to the unitary model, in order to explicitly take into account the ex-
istence of various decision-makers whose preferences quite likely differ.
One broad class of model represents multi-person household behaviour in
a non-cooperative framework.1 Such models show that if negotiation be-
tween spouses is viewed as a repeated game, non-cooperative behaviour may
occur if household members have divergent interests which cannot be rec-
onciled. Contributions belonging to this family of models and inspired by
the marriage market models of Becker (1974), such as those made by
Grossbard-Shechtman (1984) or Grossbard-Shechtman and Neuman
(1988), clearly show how strategic weight is related to individuals’ relative
income and conditions within the marriage market.
Other types of models start from the a priori assumption that spouses
know each other’s preferences well and that they exploit the gains to be had
from cooperation during their long-term relationship as a couple.2 The na-
ture of the bargaining process is thus cooperative, and game-theoretic sup-
port (the Folk theorem) is provided for Pareto efficiency and appears to be a
natural extension of the unitary setting. Models of this type focus on effi-
cient intra-household outcomes by employing an explicitly axiomatic ap-
proach to bargaining solutions, such as Nash bargaining, and by specifying
outside options for each individual in the household (Manser & Brown,
1980; McElroy & Horney, 1981; Haddad & Kanbur, 1994; Konrad &
Lommerud, 2000; Lundberg & Pollak, 1993).
A further type of model simply takes for granted that the equilibrium
outcome is Pareto efficient, without taking into account any bargaining
rules. This is the case of the collective model (Chiappori, 1988, 1992;
Bourguignon, Browning, Chiappori, & Lechène, 1993)3 in which the spouses
engage in a bargaining process which not only affects their behaviour, but
also each spouse’s well being. The principal appeal of the collective model is
Strategic Weight Within Couples 101

that it provides a single framework for analysing the decision-making pro-

cess and intra-household allocations, and economic behaviour and distri-
bution are therefore analysed within a single and theoretically consistent
framework. As a result, several empirical studies have explicitly adopted the
collective framework to analyse the labour supply and welfare distribution
effects of reforms of the tax-benefit system.4 The contribution made by these
papers is significant, particularly in the field of welfare evaluation, where the
unitary approach remains fundamentally unchallenged.
In both cooperative and non-cooperative models, the difficulties in com-
puting the strategic weights of each of the spouses continue to be the key
issue. The lack of available data, theoretical restrictions and dependence on
the tax-benefit schedule5 (which assigns a different implicit weight to each
household member on the basis of various economic and socio-demographic
characteristics) make it hard to define a criterion for computing strategic
weight.
Without wishing to question the validity of the solutions proposed by
various authors6 (which basically rely on estimation or calibration proce-
dures), in this paper we propose an alternative, highly intuitive, approach;
this is based on the hypothesis that each spouse’s strategic weight is pro-
portional to the share of resources lost by the household if he or she aban-
dons it. The underlying rationale is that individual control over money is
important for the decision-making process within the household and the
subsequent distribution of resources and welfare; thus, a substantial body of
literature suggests that an individual’s strategic weight within the household
is related to his/her contribution to its financial resources (see Browning,
Bourguignon, Chiappori, & Lechène, 1994; Phipps & Burton, 1992, 1993; or
Blumber (1988)).
The essential idea is similar to that contributed to game theory literature
by Shapley (1953). His index (the Shapley value) captures the importance of
adding a player to the winning coalition of a game (and thereby determining
his strategic weight).7 Similarly, we propose an index aimed at capturing the
strategic importance of each of the individuals in a given household by
removing him or her from the coalition represented by the marriage.
However, this is not totally symmetric to the computation of the Shapley
value, for several reasons. Firstly, even if we assume that household mem-
bers play a cooperative game and that the surplus is shared in accordance
with the index we propose, the outside options are not identical for those
individuals entering a coalition (e.g. marrying) as for those leaving it (e.g.
divorcing). Secondly, we do not construct any type of bargaining game
(whether cooperative or non-cooperative). We simply claim that this index
102 KRISTIAN ORSINI AND AMEDEO SPADARO

may approximate the ex ante strategic weights in any ‘‘household game’’.

For example, it can be used as a proxy of individual strategic weight for a
policy evaluation exercise in a collective framework. It would therefore be
possible, in a discrete labour supply model, to compute a set of alternative
strategic weights and to use these variables as determinants of private con-
sumption shares, without having to rely on a calibration method. Such an
approach would be consistent with the view that a household member’s
strategic weight is an endogenous variable, partially determined by his or
her behaviour (but also by the ‘‘caring’’ or ‘‘egoistic’’ preferences of the
partner).
Additionally, and we believe this is our most important contribution, our
index permits a comparative analysis of the performance of redistribution
systems in equalising or disequalising the strategic weight of household
members, both within and across countries. These two aspects have crucial
implications with regard to evaluating redistribution policies. Interestingly,
this index may also reveal social planners’ preferences with respect to family
policy and intra-household resource allocation.
The computation of this strategic weight index is based on microsimu-
lation techniques, since it intrinsically relies on a counterfactual premise.
Microsimulation models are powerful instruments whose analytical poten-
tial in the various fields of economic research have not yet been fully ex-
plored (Bourguignon & Spadaro, 2006). The advantage of using a
microsimulation model lies in its capacity to fully describe the current eco-
nomic situation, as well as potential counterfactuals, thereby capturing the
complex effects of taxes and benefits.
The paper is devoted to the computation of the strategic weight of each
household member. Specifically, we examine how strategic weight differen-
tials depend on household characteristics, employment patterns (which to
some extent reflect individual preferences) and the tax-benefit systems
(which, by contrast, represent social preferences). To this end, we consider
four European countries with profoundly different tax-benefit systems:
Finland, Germany, Italy and the United Kingdom.
The paper is structured in the following way. Section 2 introduces our
definition of household members’ strategic weight. Section 3 describes the
data selection and EUROMOD, the microsimulation model used to derive
strategic weights. Section 4 presents some results regarding the attitude of
social planners toward the family, as inferred from the tax-benefit system.
Section 5 analyses strategic weight differentials, focusing in particular on the
role of the labour market and of the tax-benefit systems. Section 6 presents
the conclusions reached.
Strategic Weight Within Couples 103

2. DETERMINING INDIVIDUAL STRATEGIC WEIGHT

In what follows, we assume that households simply exist because, for what-
ever reason, it is convenient for individuals to form them. Let us assume that
no public good is at stake and that agents’ behaviour is purely egoistic,8
thus, they will form part of the household only as long as this continues to
be a ‘‘convenient strategy’’. In other terms, household members will not
accept ‘‘commanding’’ a share of resources lower than their marginal con-
tribution to overall household welfare. The ‘‘strategic weight’’ of each in-
dividual within the household is hence determined by a hypothetical
counterfactual, corresponding to the share of resources that would be lost if
he or she were to ‘‘withdraw’’ from the household.
In formal terms, the weight of an individual i may be defined as:
YDðnÞ
YDðn
iÞ
li ¼ (1)
YDðnÞ
where YD(n) and YD(n
i) represent household disposable income, with and
without household member i. Logically, the strategic weight of an individual
depends on two major factors: his/her own original income and the weight
assigned to him/her by the tax-benefit system. Since disposable income may
be divided into gross income GY() and net transfers NT(), we have that:
GYðnÞ þ NTðnÞ
ðGYðn
iÞ þ NTðn
iÞÞ
li ¼ (2)
YDðnÞ
or simply:
l i ¼ mi þ t i (3)
where:
GYðnÞ
GYðn
iÞ
mi ¼
YDðnÞ
(4)
NTðnÞ
NTðn
iÞ
ti ¼
YDðnÞ
Normalising the indexes with respect to their sum permits a better com-
parison of the strategic weight of each household member relative to the
others. Thus, the strategic weight of member i can be computed as:
li
li ¼ (5)
P
nk
lk
k¼1
104 KRISTIAN ORSINI AND AMEDEO SPADARO

The following relation also holds:

li ¼ mi þ ti (6)
where
Pnk both right-hand-side terms have also been normalised with respect to
k¼1 lk :
The earlier decomposition allows us to capture the weight that a tax-
benefit system assigns to each household member, given their prevailing
roles in society in terms of age and gender.
Alternatively, the proposed index can be employed as a mechanism for
revealing social planners’ preferences with regard to household formation. If
the tax-benefit system is perfectly neutral with respect
P to householdPsize, li
can be reduced to the income share of person i and li ¼ 1. Thus, li>1
indicates a system which favours household formation, as the share of dis-
posable income lost by the household
P if a member leaves exceeds his/her
gross income, and vice versa for lio1. For confirmation, let us take the
following example9: the disposable income of each individual is
(1
t)Y+bna, where (1
t)Y is the net income and bna is a subsidy equal
to b times household size raised to a. Hence, assuming that all household
members have the same net income:
ð1
tÞY þ b½naþ1
ðn
1Þaþ1 
li ¼ (7)
n½ð1
tÞY þ bna 
If a ¼ 0, then li ¼ 1/n, which corresponds to a neutral tax-benefit system.
On the other hand, if the parameter a is larger than 0, the tax-benefit system
favours, overall, household formation. We can therefore define the sum of
the unstandardised li as a neutrality index; an index close to 1 means that
the tax-benefit system approaches neutrality with respect to family size (and
composition), while an index lower than 1 implies a tax-benefit system which
discriminates against families. In turn, an index greater than 1 implies a pro-
family tax-benefit system.
Obviously, the approach proposed has several shortcomings. The treat-
ment of children, for example, is not fully satisfactory. The possibility of
abandoning the household is an option available to adult household mem-
bers, but not to children, especially younger ones; thus, it is not possible to
compute the strategic weight of children. Further research should probably
address the issue of how parents bargain over their children. The latter may
be viewed as a type of public good into which both parents invest resources
and subsequently bargain over their respective shares of the ensuing residual
income. A priori, it seems likely that the parent who is most likely to obtain
custody would in some way ‘‘incorporate’’ the children’s strategic weight
Strategic Weight Within Couples 105

into his or her own weight. In the following analysis we will adopt this
approach and assume that children will follow the mother in the case of her
leaving the household.10
Secondly, we have not considered the possibility of behavioural reactions.
When one member leaves the household, the other may decide to increase
his/her labour supply; alternatively, the entry of a new member may cause a
corresponding decrease. This hypothesis is extremely fragile, given that la-
bour supply behaviour is a crucial element in the analysis of outside options.
The explicit inclusion of behavioural reaction would require the definition of
a full model of household members’ labour supply and the game they play
to allocate household resources. This would be extremely complicated and is
beyond the scope of the present paper.
Thirdly, we ignore the role of alimony and child support, which is usually
established by the courts or agreed between the spouses in the case of
divorce. Once more, this decision is debatable, since in theory it means
neglecting an important influence on the threat point of each household
member. Unfortunately, however, such information is lacking in the micro-
datasets available.
Our intuition is that public goods, behavioural reactions and alimony
legislation are all likely to reduce strategic weight differentials; the strategic
weight we assume may therefore be considered as a special case of a more
complex and realistic rule which takes the neglected factors mentioned
above into account.

3. DATA SELECTION AND MICROSIMULATION

SOFTWARE

As explained in the previous section, the index is based on a counterfactual

situation, represented by the effects upon disposable income of one of the
household members leaving it (either alone or with the children). These
counterfactuals are simulated using EUROMOD, an EU-wide integrated
microsimulation model which permits the simulation of the tax system and
most benefits unrelated to previous employment history (principally family
benefits, housing allowances and income support).11
The present paper focuses on four EU countries, namely Finland,
Germany, Italy and the UK; these were selected in order to permit us to
analyse a sufficiently large variety of tax-benefit systems and social models,
106 KRISTIAN ORSINI AND AMEDEO SPADARO

with different gender distributions of market and home production roles

(see Esping-Andersen, 1990, 1999).
Data for Finland are provided by the Income Distribution Survey, which
contains a combination of register data and information gathered through
interviews by Statistics Finland. The dataset refers to 1998 and contains
detailed socioeconomic information for 25,010 individuals resident in 9,345
households. German data come from the German Socioeconomic Panel
(GSOEP), established by the German Institute for Economic Research
(DIW) in 1984. Unlike in Finland, only interviews are used to collect the
annual data. The 1998 dataset supplies information regarding 18,772 indi-
viduals in 7,677 households. Italian data are collected every two years by the
Survey of Household Income and Wealth (SHIW), conducted by the Bank
of Italy. In this paper we use the 1995 dataset, which provides information
about 23,924 individuals living in 8,135 households. Finally, data for the
UK come from the Family Expenditure Survey, and are produced by the
Office for National Statistics. It collects information for 15,586 individuals
and 6,797 households over the period 1995–1996.
For each country, we selected a subsample of married and cohabiting
adult couples (i.e. aged at least 18) with and without children, irrespective of
their activity status. Children were defined as single persons aged under 30
and living with their parents. This very broad definition was intended to
avoid the exclusion of a significant number of households with grown-up
children in Italy. For the sake of simplicity, we excluded single parents and
three-generation households. Table 1 shows the sample size, before and after
this selection, for the four countries. The proportion of individuals in the
sample subsequently included in the subsample varies from 71.6% in Italy to
59.9% in Finland, which is in fact the country with the greatest proportion
of single households.
Table 2 offers some descriptive statistics for the subsamples in the four
countries considered. Since only heterosexual couples were selected, the
number of females is identical to the number of males. Average age appears
to be very similar across the panel, with females being approximately two
years younger than their male partners. With regard to the proportion of
males and females in employment, significant variation is apparent across
the different ‘‘social models’’. Finland’s male employment rate is almost
10% higher than that for Italy and the UK. However, it is in the female
employment rate that differences are most striking: in Finland the rate of
female employment is almost twice than that of Italy, while Germany and
the UK occupy an intermediate position. It should be remembered that the
above data refer to a period from the mid- to late 90s, and that female
Strategic Weight Within Couples 107

Table 1. Weighted Sample Before and After Selection.

Finland Germany Italy UK

Before selection
No. of individuals 5,086,139 78,956,258 57,206,842 57,443,762
No. of households 2,355,000 32,289,963 19,816,115 24,490,138
After selection
No. of individuals 3,046,674 57,934,344 40,976,950 39,245,363
No. of households 992,192 19,507,731 12,470,477 13,304,952
Share of total sample
Individuals 59.9 73.4 71.6 68.3
Households 42.1 60.4 62.9 54.3

Source: Authors’ calculations, based on EUROMOD.

employment rates have significantly increased in recent years in all countries

except Finland. Regarding household typologies, it is noticeable that child-
less households are the dominant household form in all countries except
Italy.12 Indeed, Italy is characterised by a particularly high incidence of
households with grown-up children. Finland, Germany and the UK have
similar shares of households with one and two children. Finland and Italy,
moreover, have a significant proportion of households with three or more
children (above 11%).

4. SOCIAL PREFERENCES AND THE NEUTRALITY

INDEX
Before examining the distribution of strategic weight per se, it is interesting
to analyse the distribution of the neutrality index, as defined in Section 2.
Thus, the tax-benefit system awards a ‘‘family bonus’’ when the neutrality
index is greater than 1, but applies a ‘‘family penalty’’ when this figure
is lower than 1. In the first case the tax-benefit system is obviously
‘‘pro-family’’, whereas in the second case it is ‘‘anti-family’’.
Table 3 shows the average neutrality index: with respect to the four
household typologies analysed (i.e. couples without children, couples
with one child, couples with two children and couples with three or more
children), Italy dominates Finland, which in turn dominates the UK, which
in turn dominates Germany. Italy is in fact the only country which appears
to have a slightly pro-family tax-benefit system. In the case of households
 108
Table 2. Descriptive Statistics (Weighted).
Finland Germany Italy UK

KRISTIAN ORSINI AND AMEDEO SPADARO

Males Females Males Females Males Females Males Females

No. of adult individuals 989,338 989,338 17,487,514 17,481,694 12,467,897 12,467,897 13,303,374 13,303,374
Average age 49.8 47.5 50.1 47.4 50.6 46.8 48.4 45.9
% adults in employment 74.5 69.7 66.7 49.0 65.8 35.8 64.4 53.5
% secondary education 35.09 36.82 39.9 40.0 58.8 54.5 71.5 72.1
% tertiary education 29.57 30.24 33.1 21.8 7.3 6.0 22.2 22.7
% no children 43.9 53.8 28.7 48.8
% one child 22.2 20.5 27.6 20.2
% two children 21.8 19.4 32.6 21.9
% three or more children 12.1 6.3 11.1 9.2

Source: Authors’ calculations, based on EUROMOD.

Strategic Weight Within Couples 109

Table 3. Average Strategic Weight by Number of Children.

Finland Germany Italy UK

Couples without children 0.958 0.936 1.022 0.945

Couples with children
One child 0.928 0.857 1.015 0.904
Two children 0.911 0.822 1.037 0.861
Three or more children 0.889 0.739 1.077 0.845

Source: Authors’ calculations, based on EUROMOD.

with three or more children, the sum of the strategic weight is 0.739 for
Germany, 0.845 for the UK, 0.889 for Finland and 1.077 for Italy. For
couples without children the differences are slightly more contained, and the
tax-benefit system approaches neutrality. Germany and the UK are again
the two systems which are furthest from neutrality (0.936 and 0.945, re-
spectively), while Finland and Italy are closer (in absolute value) to neu-
trality: the sum of strategic weight is 1.022 for Italy and 0.958 for Finland.
The magnitude of the family ‘‘penalty’’ therefore increases with household
size; it is greatest for households with numerous children and lowest for
childless households. Italy displays the opposite trend, as households with
numerous children apparently receive a larger family ‘‘bonus’’ than house-
holds with few children.
Obviously, differences in the neutrality index do not only result from the
tax-benefit system, but may well be produced by demographic factors or
differences in employment levels. Therefore, the information contained in
Table 3 is further disaggregated in Table 4. In the latter, we focus exclusively
on working-age households and disaggregate the previous figures by female
employment status (i.e. in employment or not in employment), arguably one
of the most significant factors affecting gender-based strategic weight differ-
ences. Germany still appears to have the most anti-family tax-benefit sys-
tem. In particular, the family ‘‘penalty’’ appears to be extremely consistent
in households with children where the mother is not in employment. Finland
and the UK display identical features, while Italy is again an outlier having
slightly pro-family tax-benefit system. Family ‘‘neutrality’’ is observed for
childless couples in which the female is in employment: Italy and the UK
approach unity, while Germany and Finland continue to penalise such
couples.
One last issue, of great political and also policy relevance, is how the level
of the penalty or bonus varies as a function of income. Fig. 1 shows the sum
110 KRISTIAN ORSINI AND AMEDEO SPADARO

Table 4. Average Strategic Weight by Female Employment Status

(Working-Age Households).
Finland Germany Italy UK

Couples without children

Female partner not in employment 0.942 0.866 1.035 0.881
Female partner in employment 0.953 0.971 1.001 0.993
Couples with children
Female partner not in employment 0.818 0.745 1.085 0.794
Female partner in employment 0.927 0.869 0.977 0.918

Source: Authors’ calculations, based on EUROMOD.

1.2 1.2
Finland Germany
1.1 1.1

1 1
0.9 0.9

0.8 0.8

0.7 0.7

0.6 0.6
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Couples without children Couples without children
Couples with children Couples with children

1.2 1.2
Italy United Kingdom
1.1 1.1

1 1
0.9 0.9

0.8 0.8

0.7 0.7
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10
Couples without children Couples without children
Couples with children Couples with children

Fig. 1. Neutrality Index by Disposable Income Deciles (Couples With and Without
Children).

of female and male strategic weights as a function of disposable income.

Interestingly, all the systems studied tend towards neutrality as income in-
creases. In Finland, Germany and the UK, however, the approach is bot-
tom-up: at lower income levels the tax-benefit system produces a higher
Strategic Weight Within Couples 111

family penalty, which is reduced as income increases. The size of the penalty,
moreover, is higher in the case of households with children. This situation
could be interpreted as providing support for the somewhat conservative
view that the welfare state (in particular, income support) encourages family
disruption and lone motherhood and therefore contributes to social
instability.
However, what does the anti-family or pro-family nature of the tax-
benefit system imply in terms of the strategic weight of the household
members? Are the benefits of the family ‘‘bonus’’ equally shared, or does the
system award extra strategic weight to one of the household members? In
other words: does increasing family as a whole necessarily mean increasing
the welfare of each member? Furthermore, which elements of the tax-benefit
system affect the distribution of strategic weights? These questions will be
discussed in the following section.

5. STRATEGIC WEIGHT

In the following analysis, individual strategic weights have been standard-

ised with respect to their total value, as defined in Eq. (5). This facilitates
comparison not only across household members, but also across countries.
The average male-female strategic weight differential appears to be lowest
in Germany and highest in Italy (the normalised strategic weight for females
and males is 0.506 and 0.494, respectively, in Germany, compared to 0.354
and 0.646 in Italy). The case of Italy is broadly in line with our expectations,
given the differential in male and female employment rates and, therefore, in
access to primary income. The results are more surprising for Germany.
Male employment rates are similar in Germany and the UK, whereas the
British female employment rate is higher than the German one. Neverthe-
less, the relative strategic weight for German females (0.506) is always higher
than that for British women (0.406), and even their Finnish counterparts
(0.471), despite the fact that the latter have significantly higher employment
rates.
Table 4 shows the normalised average strategic weight for females and
males, disaggregated for households with children and households with one,
two and three or more children. It is immediately apparent that the strategic
weight of females without children is quite similar in Germany, the UK and
even Italy (varying from 0.344 to 0.369). In households with children, how-
ever, the pattern is extremely different. Having one, two or three or more
children raises the strategic weight of German mothers to 0.484, 0.547 and
112 KRISTIAN ORSINI AND AMEDEO SPADARO

0.674, respectively. In these countries, it is clear that other aspects of the

system play a role at least as important as the employment rate in explaining
gender-based strategic weight differentials. At the opposite extreme is Italy,
where the presence of children does not appear to be an influence, while
Finland and the UK are located between these two extremes.
Average strategic weight differentials, however, tend to be somewhat un-
informative, given the fundamental heterogeneity of employment statuses
and earning capacities in the households sampled. An interesting question
concerns the pattern of strategic weight differentials with respect to total
income. Figs. 2 and 3 show, respectively, the pattern of strategic weights, by
household disposable income, in households without and with children. For
couples without children the profile is remarkably flat; this is surprising, as
the share of female employment may be expected to rise in line with in-
creasing disposable income. Germany and the UK display a very similar
pattern; the strategic weight of males is always between 0.6 and 0.7, while
that of females ranges from 0.3 to 0.4. In Finland the gender-based strategic
weight differential is far more contained (approximately 0.45 for females
and 0.55 for males), although this gap widens in the higher income deciles.
Italy, on the other hand, displays a highly atypical pattern: the strategic
weight for females is initially very low (0.27 in the first income decile), then

0.8 Finland 0.8 Germany

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 Males 0.3
0.2 0.2 Males
Females
0.1 0.1 Females
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

0.8 Italy 0.8 United Kingdom

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 Males 0.2 Males
0.1 Females 0.1 Females
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Fig. 2. Strategic Weight by Household Disposable Income Decile (Households

Without Children).
Strategic Weight Within Couples 113

0.8 Finland 0.8 Germany

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 Males 0.3
0.2 0.2 Males
Females
0.1 0.1 Females
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

0.8 Italy 0.8 United Kingdom

0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 Males 0.2 Males
0.1 Females 0.1 Females
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Fig. 3. Strategic Weight by Household Disposable Income Decile (Households

With Children).

increases in the second and third decile, to fall again in the fourth decile.
Finally, from the fifth decile onwards it increases more or less linearly.
Within the higher income deciles, however, the gender-based strategic
weight gap approaches the levels of the other countries studied.
When observing couples with children, inter-country differences become
more evident. Finland, Germany and the UK show a typical X pattern: in
the lower deciles, mothers have a higher strategic weight, which is then
progressively reduced, while the strategic weight of fathers increases sym-
metrically over the whole range. What does vary across the countries ob-
served is the crossing point i.e. the point where fathers’ strategic weight
surpasses that of mothers. In the UK this occurs in the second decile, in
Finland in the fifth decile and in Germany in the seventh decile. In general,
the differences appear to be largely contained, even more so than in the case
of households without children. This evidence suggests that the tax-benefit
system more than compensates for the lower employment rate typically
experienced by mothers (with the exception of Finland, where the employ-
ment rate gap of mothers tends to be less extreme).
Two features are worthy of comment: the extremely high strategic weight
of mothers in the lowest decile in Germany, and the heterogeneous pattern
114 KRISTIAN ORSINI AND AMEDEO SPADARO

of strategic weight differentials in Italy. The first is probably a result of the

generous income support benefits for lone mothers. In the second, the pat-
tern is once more somewhat difficult to explain, but may be related to the
specific characteristics (e.g. age or labour market participation) of house-
holds in the different income deciles.
A priori, demographic variables, labour market participation and the tax-
benefit system all contribute to shaping the pattern of strategic weights. In
order to more accurately separate the role of market and state institutions in
determining strategic weight differentials, it is useful (for households of
working age only) to look at the influence of female employment status
upon strategic weight differentials. Obviously, differences in male employ-
ment rates are also significant, but these tend to principally affect retirement
behaviour; within primary working age, male employment rates are quite
similar. In the following analysis we therefore concentrate on the crucial role
of female employment.

5.1. Female Employment

Table 5 shows the standardised strategic weight for females and males of
working age (20–60), disaggregating this information by the employment
status of the female partner. It is immediately noticeable that, in general, the
gender-based strategic weight gap for childless couples in which the female
works tends on average to be both quite small and fairly similar across
countries, ranging from 0.434 (United Kingdom) to 0.475 (Italy). The lower
strategic weight of working female spouses is probably due to gender differ-
ences in working hours as well as in hourly wages, which still penalise
women.
However, in childless households in which female partners do not work,
their strategic weight falls to 0.183 in Italy, 0.215 in the United Kingdom,
0.288 in Germany and 0.362 in Finland. These differences across countries
are not only a result of the various tax-benefit systems: some of the greater
strategic weight enjoyed by, for example, Finnish women could reflect their
earlier entry into the labour market and, consequently, access to contrib-
utory benefits (particularly pensions).
In households with children, on the other hand, even if the female partner
does not work, the differential between the strategic weight of the male and
female partners is much narrower, mainly due to the tax and benefit en-
titlements children generate. The greatest strategic weight once more cor-
responds to Italian males (0.759), whereas in Germany the index for mothers
 Strategic Weight Within Couples
Table 5. Normalised Average Strategic Weight by Female Partner Employment Status and Presence of
Children.
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

Couples without children

Female partner not in employment 0.638 0.362 0.712 0.288 0.817 0.183 0.785 0.215
Female partner in employment 0.549 0.451 0.550 0.450 0.525 0.475 0.566 0.434
Couples with children
Female partner not in employment 0.557 0.443 0.478 0.522 0.759 0.241 0.655 0.345
Female partner in employment 0.505 0.495 0.456 0.544 0.518 0.482 0.521 0.479

Source: Authors’ calculations, based on EUROMOD.

115
116 KRISTIAN ORSINI AND AMEDEO SPADARO

outweighs that of fathers even when the former do not work (0.522 and
0.478 for females and males, respectively). The UK and Finland occupy an
intermediate position.
In the case of working mothers, the gap is even narrower than that for
childless couples. Here, the additional strategic weight produced by children
(which we assign to mothers) makes the distribution almost egalitarian. In
Germany, the strategic weight of mothers again outweighs that of fathers,
but in the remaining countries the figure falls fractionally short of 0.5.
It can thus be observed that amongst working-age couples in which the
female partner is in employment, the distribution of strategic weight tends to
be quite egalitarian. This does not mean that the tax-benefit system plays an
insignificant role: it may well compensate for shorter average working hours
and lower average wages among mothers. The impact of net transfers is
therefore equally important for non-working-age families, working-age
families in which the mother is in employment and working-age families in
which the female does not work. In the next section we will specifically
address the issue of net public transfers.

5.2. Net Public Transfers

Using the framework established above, normalised strategic weights may

be decomposed, for both males and females, into a market component
(original income) and a public transfers component (net transfers).13 Table 6
presents this decomposition for the four countries studied.
Italy is remarkable for the significant role of net transfers in defining
strategic weights within childless households. This is unsurprising: as chil-
dren tend to stay longer with their families than in the rest of Europe, and
family formation tends to be considerably delayed, childless households are
on average older than in the other European countries considered, and thus
they display a higher share of transfers related to old age. Net transfers here
are positive (on average) for both female and male spouses, although the size
of the transfer tends to reinforce the strategic weight differential of original
income. Again, this was foreseeable: since retirement benefits are employ-
ment-related, they tend to reproduce strategic weight differential patterns
similar to those generated by original income. This also appears to be the
case in Finland, while in Germany and the UK net transfers tend to have a
very small average effect. At least for the UK, this could be partially ex-
plained by the fact that the primary source of income for many retired
Strategic Weight Within Couples 117

people lies in private rather than public pensions, and thus reflects original
income.
Net transfers also tend to be negative for families with children. The
different age structure of households with and without children is likely to
have an influence, since adults in households with children tend to be active
in the labour market. Germany is noticeable for the significant role of net
transfers in determining strategic weight; in particular, it appears that taxes
strongly reduce the relative strategic weight of males and increase that of
females. At the other extreme is Italy, where public transfers apparently play
only a marginal role in the case of households with children, and relative
strategic weight is basically determined by market incomes. This is also
consistent with the Italian welfare state model, which is strongly biased
towards pensions, with only minor child-related benefits and income sup-
port schemes. Only in households with three or more children is the strategic
weight of mothers increased through the tax-benefit system. The United
Kingdom and Finland, on the other hand, display similar patterns for
households with children: in both cases transfers reduce the relative strategic
weight of males to increase that of. Once again, this is probably due to the
interaction between employment and earning differentials and progressive
taxation.
As in the previous section, it is possible to analyse the role of transfers and
original income across income deciles. Figs. 4 and 5 show the profile of
strategic weights, by income decile, before and after public transfers to
households without and with children, respectively.
For each decile, strategic weight has been decomposed into market and
net transfer components. The figures show how strategic weight is modified
by net transfers: the dotted line represents strategic weight calculated using
gross income, whereas the solid line represents strategic weight calculated
using disposable income i.e. gross income plus net public transfers. An ex-
amination of Fig. 4 reveals that the pattern is quite similar across countries
i.e. public transfers ‘‘harmonise’’ strategic weight. Strategic weight calcu-
lated using gross incomes tends to increase with income decile, while public
transfers increase strategic weight differentials in lower deciles and reduce
them in upper deciles. The decile in which the effect changes varies across
countries: in Italy, for example, net transfers are positive for both men and
women up to the ninth decile; this household typology is on average older
than its counterparts in the other countries analysed. In the UK, on the
other hand, the effect is reversed in the fifth and sixth deciles, probably due
to the lesser role of public old age benefits. In Finland and Germany the
switch occurs between the sixth and the eighth decile. From a gender-based
118 KRISTIAN ORSINI AND AMEDEO SPADARO

Finland, Males Finland, Females

1 1
0.9 Disposable income 0.9 Disposable income
0.8 Original income 0.8 Original income
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Germany, Males Germany, Females

1 1
0.9 Disposable income 0.9 Disposable income
0.8 Original income 0.8 Original income
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Italy, Males Italy, Females

1 1
0.9 0.9 Disposable income
0.8 0.8 Original income
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 Disposable income 0.2
0.1 Original income 0.1
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

United Kingdom, Males United Kingdom, Females

1 1
0.9 0.9 Disposable income
0.8 0.8 Original income
0.7 0.7
0.6 0.6
0.5 0.5
0.4 0.4
0.3 0.3
0.2 Disposable income 0.2
0.1 Original income 0.1
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Fig. 4. Strategic Weights Computed for Gross Market Income and Disposable
Income, by Household Disposable Income Decile (Households Without Children).
Strategic Weight Within Couples 119

Finland, Males Finland, Females

1.8 1.8
1.6 Disposable income 1.6 Disposable income
1.4 Original income
1.4 Original income
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Germany, Males Germany, Females

1.8 1.8
1.6 Disposable income 1.6 Disposable income
1.4 Original income 1.4 Original income
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Italy, Males Italy, Females

1.8 1.8
1.6 Disposable income 1.6 Disposable income
1.4 Original income
1.4 Original income
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

United Kingdom, Males United Kingdom, Females

1.8 1.8
1.6 Disposable income 1.6 Disposable income
1.4 Original income
1.4 Original income
1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
1 2 3 4 5 6 7 8 9 10 1 2 3 4 5 6 7 8 9 10

Fig. 5. Strategic Weights Computed for Gross Market Income and Disposable
Income, by Household Disposable Income Decile (Households With Children).
120 KRISTIAN ORSINI AND AMEDEO SPADARO

perspective, net public transfers have an ambiguous effect. As Table 6

shows, strategic weight is increased for both males and females, although
(except in the UK) the increase is greater for the former, in both absolute
and relative terms. The decile patterns show that net public transfers con-
stantly increase strategic weight differentials based on original income in
Finland and in Germany. Original labour market differences are thus du-
plicated through employment-related benefits. In the case of Italy, however,
net transfers tend to increase strategic weight differentials based on original
income in the very bottom and top deciles. In the UK, by contrast, net
transfers reduce gender-based strategic weight differentials in the two lowest
income deciles and increase them in the rest of the distribution.
Fig. 5 depicts a similar pattern to Fig. 4, with some exceptions. For males
in Finland and Germany, net public transfers decrease, on average, their
strategic weight over the whole income range. The same applies to the UK,
starting from the second income decile (in the first income decile positive net
transfers still have a positive effect on the strategic weight of males). The
extent of the reduction is considerable in Germany and the UK, especially in

Table 6. Average Impact of Net Transfers on Standardised Strategic

Weight (by Number of Children).
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

No children 0.574 0.426 0.631 0.369 0.655 0.345 0.641 0.359

Original income 0.374 0.325 0.509 0.331 0.373 0.206 0.612 0.332
Transfers 0.200 0.101 0.122 0.039 0.282 0.138 0.029 0.026
% change 53.554 31.058 23.919 11.717 75.610 67.123 4.814 7.905
One child 0.525 0.475 0.515 0.485 0.626 0.374 0.567 0.433
Original income 0.721 0.536 1.036 0.431 0.665 0.393 0.826 0.432
Transfers
0.196
0.062
0.521 0.054
0.039
0.019
0.258 0.000
% change
27.158
11.466
50.259 12.564
5.795
4.777
31.283 0.054
Two children 0.512 0.488 0.453 0.547 0.645 0.355 0.583 0.417
Original income 0.793 0.497 1.240 0.386 0.762 0.371 0.974 0.347
Transfers
0.281
0.009
0.787 0.161
0.116
0.016
0.391 0.069
% change
35.474
1.771
63.487 41.782
15.278
4.294
40.113 19.999
Three or more children 0.473 0.527 0.326 0.674 0.632 0.368 0.522 0.478
Original income 0.745 0.408 1.173 0.344 0.731 0.324 0.803 0.278
Transfers
0.272 0.120
0.847 0.330
0.099 0.044
0.282 0.200
% change
36.552 29.317
72.195 95.860
13.586 13.681
35.062 72.037

Source: Authors’ calculations, based on EUROMOD.

Strategic Weight Within Couples 121

the lowest deciles. This is probably related to the generous income assistance
available for single-earner households; if based solely on original income,
males would have an extremely high strategic weight. However, access to
income support and generous supplements for children reduces the loss of
income that the household would experience if the male partner were to
leave. In Finland and Italy the reduction in males’ strategic weight is much
lower. In Finland this is mainly due to the presence of a second earner in the
household, while in the case of Italy, the lack of a safety net reduces the
equilibrating effect of public transfers upon strategic weight. This also ex-
plains why the strategic weight for males is considerably higher in the lowest
deciles (i.e. in the part of the distribution characterised by lower female
employment rates).
With respect to childless couples, the effect of transfers switches from
positive to negative earlier in the distribution. This is once more due to the
different age structure of households with children. The presence of children
also explains why strategic weight is so markedly increased by public trans-
fers: lone mothers have access to significantly greater income resources than
single males (although of course their needs are much greater). This explains
why in the lowest income deciles the strategic weight of females is raised
above that of males.
In the following section we will attempt to confirm the explanations
offered above by examining the effect of each transfer component sepa-
rately.

5.3. Decomposing Net Transfers

In this section we will explore in detail how various instruments contained in

tax-benefit systems affect intra-household strategic weight differentials,
making intensive use of the microsimulation model. Instruments have been
classified into broad groups: (i) taxes and social security contributions, (ii)
income support and housing benefits, (iii) family benefits, (iv) old age and
sickness benefits, and (v) unemployment benefits. For each group of meas-
ures we simulate the strategic weight that would result if these did not exist;
this allows us to estimate the specific contribution of each element of the tax-
benefit system. The analysis is, once more, performed for households both
with and without children.
Tables 7 and 8 display the difference between the baseline strategic weight
and the strategic weight resulting from the removal of the specific instru-
ment. The results are again disaggregated for households with and without
122 KRISTIAN ORSINI AND AMEDEO SPADARO

Table 7. Average Impact on Individual Strategic Weight of Different

Instruments, by Income Decile (Households Without Children).
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

Taxes/SSC
1 0.00 0.00
0.04 0.04
0.02 0.02 0.00 0.00
2 0.00 0.00
0.02 0.02 0.02
0.02
0.01 0.01
3 0.00 0.00
0.01 0.01
0.06 0.06
0.01 0.01
4 0.00 0.00
0.01 0.01
0.04 0.04
0.02 0.02
5 0.00 0.00
0.02 0.02
0.01 0.01
0.01 0.01
6 0.00 0.00 0.00 0.00
0.02 0.02
0.03 0.03
7 0.00 0.00
0.02 0.02
0.04 0.04
0.01 0.01
8 0.00 0.00
0.02 0.02
0.02 0.02
0.02 0.02
9 0.00 0.00
0.01 0.01 0.00 0.00
0.01 0.01
10 0.00 0.00
0.03 0.03
0.03 0.03
0.01 0.01

Housing/S.A. benefits
1
0.01 0.01
0.07 0.07
0.09 0.09
0.04 0.04
2 0.00 0.00
0.03 0.03
0.02 0.02
0.03 0.03
3 0.00 0.00
0.01 0.01
0.07 0.07
0.01 0.01
4
0.01 0.01 0.00 0.00
0.02 0.02 0.00 0.00
5 0.00 0.00
0.01 0.01 0.01
0.01
0.01 0.01
6 0.00 0.00 0.01
0.01 0.00 0.00
0.01 0.01
7 0.00 0.00
0.01 0.01
0.02 0.02 0.00 0.00
8 0.00 0.00
0.01 0.01
0.01 0.01
0.01 0.01
9 0.01
0.01 0.00 0.00 0.01
0.01 0.00 0.00
10 0.03
0.03 0.00 0.00 0.00 0.00 0.00 0.00
Family benefits
1 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
2 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
3 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
4 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
5 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
6 0.00 0.00 0.01 0.00 0.00 0.00 0.00 0.00
7 0.00 0.00 0.00 0.01 0.00 0.00 0.00 0.00
8 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
9 0.00 0.00 0.00 0.00 0.01 0.00 0.00 0.00
10 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00

Old age/sickness benefits

1 0.01
0.01 0.09
0.09 0.24
0.24 0.02
0.02
2 0.01
0.01 0.12
0.12 0.18
0.18 0.04
0.04
3 0.01
0.01 0.15
0.15 0.16
0.16 0.01
0.01
4 0.01
0.01 0.09
0.09 0.19
0.19 0.00 0.00
5 0.01
0.01 0.10
0.10 0.18
0.18 0.01
0.01
Strategic Weight Within Couples 123

Table 7. (Continued )
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

6 0.00 0.00 0.08
0.08 0.16
0.16 0.00 0.00

7 0.01
0.01 0.04
0.04 0.13
0.13 0.00 0.00
8 0.00 0.00 0.04
0.04 0.10
0.10 0.00 0.00
9 0.00 0.00 0.04
0.04 0.07
0.07 0.00 0.00
10 0.03
0.03 0.02
0.02 0.03
0.03 0.00 0.00
Unemployment benefits
1 0.00 0.00
0.01 0.01 0.02
0.02 0.00 0.00
2 0.00 0.00 0.00 0.00 0.03
0.03 0.00 0.00
3 0.01
0.01 0.00 0.00
0.03 0.03 0.00 0.00
4
0.01 0.01 0.02
0.02
0.01 0.01 0.00 0.00
5 0.00 0.00 0.00 0.00 0.02
0.02 0.00 0.00
6 0.00 0.00 0.03
0.03 0.00 0.00
0.01 0.01
7 0.00 0.00
0.01 0.01
0.01 0.01 0.00 0.00
8 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
9
0.01 0.01 0.00 0.00 0.01
0.01 0.00 0.00
10 0.03
0.03 0.00 0.00 0.00 0.00 0.00 0.00

Note: Figures are in italics for absolute changes in the range (0, 0.1], underlined for absolute
changes in the range (0.1, 0.4] and in bold for absolute changes in the range (0.4, N).
Source: Authors’ calculations, based on EUROMOD.

children. Different fonts have been used for the figures to facilitate their
reading; this makes immediately clear which instruments play a significant
role in reshaping intra-household strategic weight differentials.
For childless families, the situation is relatively simple; since such house-
holds tend, on average, to be older the main benefits they receive are pen-
sions, and thus the tax system has only a marginal influence upon strategic
weight, particularly in the case of Finland. When it does play a significant
role it is the female partner who is principally favoured. In both Germany
and in Italy, the existence of family-based provisions in the tax system (e.g.
joint taxation or deductions for dependent spouses) tends to increase the
income loss that the household would experience if the female spouse were
to leave, although this effect is rather modest.
Income support and housing benefits, by contrast, are important. Their
effect is to reduce the income loss that female partners would suffer if their
male partners were to leave the household (in Germany and the UK) or,
alternatively, to reduce the income to which employed partners (in this case
male) having a dependent spouse or child are entitled. As all such benefits
124 KRISTIAN ORSINI AND AMEDEO SPADARO

Table 8. Average Impact on Individual Strategic Weight of Different

Instruments, by Income Decile (Households With Children).
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

Taxes/SSC
1
0.07 0.07
0.11 0.11
0.03 0.03
0.02 0.02
2
0.03 0.03
0.20 0.20
0.04 0.04
0.02 0.02
3
0.03 0.03
0.13 0.13
0.06 0.06
0.05 0.05
4
0.02 0.02
0.19 0.19
0.04 0.04
0.05 0.05
5
0.02 0.02
0.12 0.12
0.04 0.04
0.04 0.04
6
0.02 0.02
0.10 0.10
0.03 0.03
0.06 0.06
7 0.00 0.00
0.09 0.09
0.02 0.02
0.04 0.04
8
0.02 0.02
0.08 0.08
0.02 0.02
0.02 0.02
9 0.00 0.00
0.07 0.07
0.02 0.02
0.03 0.03
10 0.02
0.02
0.06 0.06
0.03 0.03
0.02 0.02
Housing/S.A. benefits
1
0.08 0.08
0.18 0.18 0.01
0.01 0.04
0.04
2
0.05 0.05
0.22 0.22 0.00 0.00
0.04 0.04
3
0.04 0.04
0.17 0.17
0.01 0.01
0.05 0.05
4
0.02 0.02
0.19 0.19 0.01
0.01
0.06 0.06
5
0.02 0.02
0.11 0.11
0.01 0.01
0.04 0.04
6
0.01 0.01
0.09 0.09
0.01 0.01
0.05 0.05
7 0.00 0.00
0.07 0.07 0.00 0.00
0.03 0.03
8
0.01 0.01
0.06 0.06 0.00 0.00
0.01 0.01
9 0.01
0.01
0.01 0.01 0.00 0.00
0.02 0.02
10 0.05
0.05 0.00 0.00 0.00 0.00
0.01 0.01

Family benefits
1
0.02 0.02
0.08 0.08
0.05 0.05 0.03
0.03
2
0.04 0.04
0.07 0.07
0.06 0.06 0.02
0.02
3
0.03 0.03
0.04 0.04
0.06 0.06
0.04 0.04
4
0.02 0.02
0.05 0.05
0.02 0.02
0.03 0.03
5
0.03 0.03
0.03 0.03
0.02 0.02
0.03 0.03
6
0.02 0.02
0.02 0.02
0.02 0.02
0.04 0.04
7
0.02 0.02
0.03 0.03
0.01 0.01
0.03 0.03
8
0.03 0.03
0.01 0.01
0.01 0.01
0.01 0.01
9 0.00 0.00
0.01 0.01 0.00 0.00
0.02 0.02
10 0.02
0.02 0.00 0.00 0.00 0.00
0.01 0.01

Old age/sickness benefits

1 0.00 0.00 0.02
0.02 0.04
0.04 0.01
0.01
2 0.01
0.01 0.01
0.01 0.04
0.04 0.04
0.04
3 0.01
0.01 0.02
0.02 0.03
0.03 0.01
0.01
Strategic Weight Within Couples 125

Table 8. (Continued )
Finland Germany Italy United Kingdom

Male Female Male Female Male Female Male Female

4 0.01
0.01 0.01
0.01 0.05
0.05 0.01
0.01

5 0.01
0.01 0.02
0.02 0.06
0.06 0.01
0.01
6 0.00 0.00 0.01
0.01 0.05
0.05
0.02 0.02
7 0.01
0.01 0.02
0.02 0.06
0.06 0.00 0.00
8
0.01 0.01 0.02
0.02 0.06
0.06 0.01
0.01
9 0.01
0.01 0.00 0.00 0.05
0.05
0.01 0.01
10 0.02
0.02 0.00 0.00 0.02
0.02 0.00 0.00

Unemployment benefits
1 0.02
0.02 0.01
0.01 0.02
0.02 0.00 0.00
2 0.02
0.02 0.01
0.01 0.02
0.02 0.02
0.02
3 0.00 0.00 0.03
0.03
0.01 0.01 0.01
0.01
4 0.02
0.02 0.01
0.01 0.01
0.01 0.01
0.01
5 0.01
0.01 0.02
0.02
0.01 0.01 0.01
0.01
6 0.00 0.00 0.01
0.01
0.01 0.01
0.02 0.02
7 0.00 0.00 0.01
0.01 0.00 0.00
0.01 0.01
8
0.01 0.01 0.03
0.03 0.00 0.00 0.00 0.00
9 0.01
0.01 0.00 0.00 0.00 0.00
0.01 0.01
10 0.02
0.02 0.01
0.01 0.00 0.00 0.00 0.00

Note: Figures are in italics for absolute changes in the range (0, 0.1], underlined for absolute
changes in the range (0.1, 0.4] and in bold for absolute changes in the range (0.4, N).
Source: Authors’ calculations, based on EUROMOD.

are means-tested, their effect is strongest in lower income deciles; in higher

income deciles, as female partners have access to private sources of income
(from either current or previous employment) they are ineligible for means-
tested benefits.
Finally, public pension systems tend to reproduce inequalities in access to
private sources of income produced throughout working life. Since childless
households are mainly older households, retirement income has a strong
influence upon strategic weights. This effect of pensions is particularly
strong in Italy, given the old age bias of the Italian welfare system. The same
is true (to a lesser extent) in Germany. In the UK, public pensions mainly
affect the lower income deciles, since higher income households have largely
opted out of the state system in order to join private pension schemes.
Finally, the effect of pensions on the strategic weight of males and females in
Finland appears to be much less biased than in the other countries. This is
mainly due to the combination of a flat rate universal old age allowance,
126 KRISTIAN ORSINI AND AMEDEO SPADARO

coupled with the historically high participation of females in the labour

market.
In households with children the situation is much more complex, and here
the tax system plays an important role. In Germany, the prevailing joint tax
system, characterised by significant tax deductions for dependent children
(for higher income groups), significantly increases the strategic weight of the
female partner. This effect is highest in the second income decile and de-
creases progressively as household income (and the income of the female
partner) increases. The Italian and British tax systems are basically indi-
vidualised, but elements of family-based taxation remain in the case of the
child tax credit and in the deductions for married couples (dependent
spouse). In the UK there also exists a special tax credit for lone parents.
These elements tend to increase the strategic weight of the female spouse. As
the income of mothers increases, however, the tax advantages (i.e. deduc-
tions for dependent spouse) for couples with children either disappear or
become relatively less important. Finally, although the tax system in Fin-
land is totally individualised, it nevertheless reduces the strategic weight of
male partners, especially in the lowest income decile. This is mainly due to
the fact that taxation reduces the weight of the private resources of male
spouses to the ‘‘advantage’’ of non-working females.
Housing benefits and social assistance (as income support) are particu-
larly important in the case of households with children. Most income sup-
port schemes, in fact, include fairly generous child-related supplements as
well as special allowances for lone parents. The most generous income sup-
port scheme is clearly the German one. The Finnish benefit system is also
quite generous, but the number of recipients (as well as the size of the
transfer) is smaller, given the higher employment rates for both males and
females. In the case of the UK, we have included Family Credit (a tax credit)
within aggregate income support and housing benefit: this explains why
the effect of these instruments tends to be significant in all income deciles.
Finally, Italy lacks a well-developed system of income support; employed
workers are entitled to social transfers if their wages are below a certain
threshold, but the effect of this scheme is extremely limited.
A similar scheme also provides income supplements to employed parents
of dependent children. These child benefits are quite strictly means-tested
and their effect is significant only in the lower income deciles. Family income
support is also means-tested in the case of Germany, and higher income
groups usually prefer the more favourable tax deduction scheme.
In Finland and in the UK, on the other hand, child benefit is universal,
although its impact is moderate over the whole range of the distribution.
Strategic Weight Within Couples 127

What seems a priori illogical is the negative effect on women’s strategic

weight of removing family benefits in the lowest income deciles. This is
in fact largely explained by interactions between the tax and benefit systems
i.e. cutting off child benefit leads to increased housing benefit and income
support.
The above also demonstrates the limitations of our approach: in highly
complex welfare states it is impossible to measure the precise impact of a
specific instrument, especially since every instrument has been designed
taking into account the other instruments which exist in the tax-benefit
system. However, this paper by no means attempts to measure the overall
impact of this system upon the individual strategic weight of spouses; in-
stead, the aim is simply to demonstrate the diverse effects of particular
elements of the tax-benefit system upon the strategic weight of partners
within a family.

6. CONCLUSIONS

Employing a highly intuitive concept of intra-household strategic weight,

based on microsimulation techniques, we have computed the strategic
weight of each spouse and examined its dependence on the tax-benefit sys-
tem in four European countries between which that system differs greatly.
The results show that these differences play an important role in determin-
ing such strategic weights. We believe our proposed index may be of great
utility in the empirical evaluation of redistribution systems, as we have
shown. Naturally, we understand and accept the limitations discussed in
Section 2. For example, the framework adopted is completely static; that is
to say, when calculating the strategic weight of one of the partners, we did
not consider the possibility that the other partner may adjust his/her be-
haviour in the labour market, and nor did we consider the role of household
production or of public goods. Moreover, the decision to ‘‘assign’’ children
to mothers may be questionable. Traditionally, however, children are as-
signed to female spouses on the basis of socially dominant gender roles, and
women’s ‘‘control’’ over children may well compensate for their lower stra-
tegic weight with sole regard to income (Lundberg & Pollak, 1993). It is
important to note, however, that strategic weight should not be interpreted
as a sharing rule, but instead as simply one factor among several which may
affect the intra-household distribution of part or all of its resources. With
regard to a sharing rule, it would be reasonable to assume that part of total
household income is used to purchase non-private goods and services, and
128 KRISTIAN ORSINI AND AMEDEO SPADARO

that only a residual share is allocated in accordance with strategic weight

differentials (see Chen & Wolley, 2001).
Bearing such limitations in mind, our approach provides further clarifi-
cation of how social and individual preferences interact to determine stra-
tegic weight within the household, and of the distribution of strategic weight
differentials.
Individual preferences principally affect individuals’ labour supply strat-
egies, which play a significant role in determining earning capacity and,
consequently, strategic weight differentials. However, differences in male
and female employment rates are only one of the factors affecting strategic
weights within a couple. Net transfers, both positive and negative, also play
a significant role in reshaping strategic weight differentials. While some
measures are largely neutral, others tend to reduce existing inequalities or
exacerbate intra-household strategic weight differentials.
On this point, it is interesting to note that a pro-family tax-benefit system
might have ambiguous effects on the welfare of individuals within the fam-
ily. If such a system is intended to reduce the outside options of the ‘‘weak’’
partner, the effect on the individuals within the household might be far from
desirable. This is particularly evident in the Italian case: the pro-family
system implicitly encourages family formation and preservation, but does so
at the cost of reducing the strategic weight of the weak partner.
At the other extreme is Germany, whose system unexpectedly appears to
penalise family stability. However, the policies which reduce incentives for
family preservation simultaneously produce a more equitable distribution of
strategic weight, and possibly of resources.
This effect is particularly important for the lowest deciles, where strategic
weight differentials are most important and where the unequal distribution
of resources (influenced by the unequal strategic weights of the partners)
may perversely affect individual welfare, producing poor individuals within
non-poor families, for example. However, it is also in the lowest income
deciles where public policies may significantly adjust strategic weights. In
higher income deciles the strategic weight of spouses tends to be far more
equal; individual preferences (and the distribution of human capital and
talents) are the principal factors shaping strategic weight differentials in
higher income groups. The marginal role of public policies is also demon-
strated by the fact that the distribution of strategic weight is similar within
highly diverse institutional situations.
The index presented in this paper may therefore be employed as a
straightforward tool to analyse the impact of tax-benefit systems on relative
strategic weight, to compare their effect across countries and to assess the
Strategic Weight Within Couples 129

impact of tax-benefit reforms which may affect differently the strategic

weight of individuals within the household.
More ambitiously, this index could be used as a starting point for the
elaboration of a more realistic sharing rule i.e. one which takes into account
dynamic strategies e.g. individual responses to the threat of family breakup,
as Rubinstein (1982) remarks, adult control over younger children and
economies of scale in the purchase of public goods and services.
Empirically, it would be interesting to discover if there exist natural
experiments within countries which could be used to validate the index
proposed here or if empirical regularities may be found to suggest that
couples in which, say, the female’s strategic weight is high, adapt their
behaviour, and that such changes are to be expected. This and other related
topics must, however, be left for future research.

NOTES
1. See, for example, Leuthold (1968), Ashworth and Ulph (1981), Bourguignon
(1984), Chen and Wolley (2001), Rubinstein (1982) and Binmore (1985). See Donni
(2006) for an extensive review of non-cooperative models and their properties.
2. Lundberg and Pollak (2003) have shown, however, that if current decisions
affect spouses’ future strategic weight, then inefficient outcomes are possible. For a
discussion, see also Lundberg and Pollak (1994), Ott (1992) and Donni (2006).
3. See Vermeulen (2002) for a complete survey.
4. See the special issue of the Review of Economics of the Household (Vol. 4,
No. 2, June 2006) on the collective model and its application to the evaluation of
tax reforms.
5. This last point is highly relevant to policy analysis. Several papers suggest a
correlation between the tax-benefit system and the strategic weight. In a study by
Beblo, Beninger and Laisney (2003), to cite merely one of these, the calibrated
strategic weight is then regressed (together with other demographic variables) on the
ratio of the earnings potentials of the spouses (i.e. the average disposable income
when switching from 0 to 40 h, given the alternative labour supply strategies avail-
able to the partner). The coefficients of the regression are then used to predict the
strategic weight under alternative scenarios. A change in the tax-benefit system
would in fact alter the earnings potentials and hence the strategic weight.
6. On the contrary, we wish to stress the importance of conducting further re-
search in these directions.
7. The Shapley value has been also applied to the decomposition of inequality by
Shorrocks (1999) and Sastre and Trannoy (2002).
8. We will discuss the implications of this strong hypothesis at the end of this
section.
9. This example has been suggested by an anonymous referee who we are logically
and unfortunately unable to acknowledge.
130 KRISTIAN ORSINI AND AMEDEO SPADARO

10. Alternatively, children can be seen as a commitment mechanism. They have

long-term implications, not least because they generate a liability to child support
which is not modelled here in any way.
11. For a detailed description of EUROMOD, see Sutherland (2001).
12. It should be noted that childless households could be comprised by younger
couples, as well as older couples whose children have already left the household.
13. Replacement incomes in this case have been treated as net transfers; arguably,
however, they could be considered as deferred wages.

ACKNOWLEDGMENTS

This is a completely revised version of the paper ‘‘Sharing resources within

the household: a multi-country microsimulation analysis of the determi-
nants of intra-household ‘strategic weight’ differentials and their distribu-
tional outcomes’’, PSE Working Paper No. 2005-04.
We are grateful to Vincenzo Atella, Dani Cardona, André Decoster,
Jacques Lecacheux, Holly Sutherland and Luc Gladiateur for their in-depth
comments. We are indebted to four anonymous referees for their useful
comments. All errors or omissions are entirely the responsibility of the au-
thors. This paper was written as part of the MICRESA (Micro Analysis of
the European Social Agenda) project, financed by the Improving Human
Potential programme of the European Commission (SERD-2001-00099).
For the countries considered in this paper, EUROMOD relies on the fol-
lowing microdata: the Income Distribution Survey supplied by Statistics
Finland; the German Socio-Economic Panel Study made available by DIW;
the Survey of Household Income and Wealth, provided by the Bank of
Italy; and the Family Expenditure Survey, supplied by the British Office for
National Statistics. We are indebted to our present and former colleagues of
the EUROMOD team. Amedeo Spadaro gratefully acknowledges financial
support from the Spanish Government – MCYT (SEJ2005-08783-C04-03).

REFERENCES
Ashworth, J., & Ulph, D. (1981). Household models. In: B. Charles (Ed.), Taxation and labour
supply (pp. 117–133). London: Allen and Unwin.
Beblo, M., Beninger, D., & Laisney, F. (2003). Family tax splitting: A microsimulation of its
potential labour supply and intra-household welfare effects in Germany. ZEW Discussion
Paper, No. 03-32, Manheim.
Becker, G. (1974). A theory of marriage: Part II. The Journal of Political Economy, 82, S11–S26.
Strategic Weight Within Couples 131

Binmore, K. (1985). Bargaining and coalitions. In: A. Roth (Ed.), Game-theoretic models of
bargaining (pp. 259–304). Cambridge: CUP.
Blumber, R. L. (1988). Income under female versus male control: Hypotheses from a theory of
gender stratification and data from the third world. Journal of Family Issues, 9(1), 51–84.
Bourguignon, F. (1984). Rationalité individuelle ou rationalité stratégique: le cas de l’offre
familiale de travail. Revue Économique, 35(1), 147–162.
Bourguignon, F., Browning, M., Chiappori, P.-A., & Lechène, V. (1993). Intrahousehold al-
location of consumption: Some evidence on French data. Annales d’Economie et
Statistique, 29, 137–156.
Bourguignon, F., & Spadaro, A. (2006). Microsimulation as a tool for evaluating redistribution
policies. Journal of Economic Inequality, 4(1), 77–106.
Browning, M., Bourguignon, F., Chiappori, P.-A., & Lechène, V. (1994). Income and out-
comes: A structural model of intrahousehold allocation. The Journal of Political Econ-
omy, 102(6), 1067–1096.
Chen, Z., & Wolley, F. (2001). A Cournot-Nash model of family decision making. The Eco-
nomic Journal, 111(October), 722–748.
Chiappori, P.-A. (1988). Rational household labor supply. Econometrica, 56(January), 63–89.
Chiappori, P.-A. (1992). Collective labor supply and welfare. The Journal of Political Economy,
100(June), 437–467.
Donni, O. (2006). Les modéles non-coopératifs d’offre familiale de travail: Théorie et évidence.
L’Actualité Economique: Revue D0 analyse Économique, 82, 181–206.
Esping-Andersen, G. (1990). The three worlds of welfare capitalism. Princeton, NJ: Princeton
University Press.
Esping-Andersen, G. (1999). Social foundations of postindustrial economies. New York: Oxford
University Press.
Grossbard-Shechtman, A. (1984). A theory of allocation of time in markets for labor and
marriage. Economic Journal, 94, 863–882.
Grossbard-Shechtman, S., & Neuman, S. (1988). Women’s labor supply and marital choice.
Journal of Political Economy, 96, 1294–1302.
Haddad, L., & Kanbur, R. (1994). Are better-off households more unequal or less unequal.
Oxford Economic Papers, 46, 445–458.
Konrad, K., & Lommerud, K. (2000). The bargaining family revisited. Canadian Journal of
Economics, 33, 471–487.
Leuthold, J. (1968). An empirical study of formula income transfers and the work decision of
the poor. Journal of Human Resources, 3, 312–323.
Lundberg, S., & Pollak, R. (1993). Separate spheres bargaining and the marriage market. The
Journal of Political Economy, 101(December), 988–1010.
Lundberg, S., & Pollak, R. (1994). Noncooperative bargaining models of marriage. The
American Economic Review, 84(2), 132–137.
Lundberg, S., & Pollak, R. (2003). Efficiency in marriage. Review of Economics of the House-
hold, 1, 153–167.
Manser, M., & Brown, M. (1980). Marriage and household decision-making: A bargaining
analysis. International Economic Review, 21, 31–44.
McElroy, M., & Horney, M. (1981). Nash-bargained household decisions: Towards a gener-
alisation of the theory of demand. International Economic Review, 22, 333–349.
Ott, N. (1992). Intrafamily bargaining and household decisions. Berlin/Heidelberg: Springer-
Verlag.
132 KRISTIAN ORSINI AND AMEDEO SPADARO

Phipps, S., & Burton, P. (1992). What is mine is yours? The influence of male and female income
on patterns of household expenditure. Discussion Paper 92-12, Department of Economics,
Dalhousie University.
Phipps, S., & Burton, P. (1993). Collective models of household behaviour: Implications for
economics policy. Paper presented at the Status of Women Economic Equality Work-
shop, Ottawa, 29–30 November.
Rubinstein, A. (1982). Perfect equilibrium in a bargaining model. Econometrica, 50, 97–109.
Sastre, M., &, Trannoy, A. (2002). Shapley inequality decomposition by factor components: Some
methodological issues. Published in Moyes, P., Seidl, C., & Shorrocks, A. (2002).
Shapley, L. S. (1953). A value for n-person games. In: H. W. Kuhn, & A. W. Tucker (Eds),
Contributions to the theory of games (Vol. II, pp. 307–317). Princeton: Princeton
University Press.
Shorrocks, A. F. (1999). Decomposition procedures for distributional analysis: A unified frame-
work based on the Shapley value. Unpublished paper, University of Essex.
Sutherland, H. (2001). Final report EUROMOD: An integrated European tax-benefit model.
EUROMOD working paper EM9/01.
Vermeulen, F. (2002). Collective household models: Principles and main results. Journal of
Economic Surveys, 16, 533–564.
PROGRESSIVITY IMPLICATIONS
OF PUBLIC HEALTH INSURANCE
FUNDING IN CANADA

James B. Davies and Michael Hoy

ABSTRACT
We adopt a standard distributional impact methodology, based on
Atkinson’s cost of inequality approach, to estimate the degree of implicit
redistribution created through public funding of health insurance in
Canada. The first stage of the exercise is to determine the public health
insurance benefits received by families of various age and composition and
to add these to measured after-tax incomes. In our base case, which uses
the Atkinson Mean Logarithmic Deviation as inequality index, we find
that accounting for public health insurance benefits implies a reduction in
inequality equivalent to 2.4% of per capita income. We then model the
implications of moving to a hypothetical fully privatized system while
proportionately refunding to individuals the tax revenues saved in doing
so. This would give rise to a further 2.4% equivalent per capita income
reduction resulting from increased inequality in the distribution of after-
tax income. Thus, for this scenario, moving from public financing of
health insurance in Canada to a fully privatized system implies an overall
increase in inequality equivalent to a loss of 4.8% of per capita income.
This corresponds to an increase of about 25% in existing inequality. Not

Equity
Research on Economic Inequality, Volume 15, 133–167
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15007-9
133
134 JAMES B. DAVIES AND MICHAEL HOY

surprisingly, the impact of publicly financed health insurance in reducing

inequality is strongest for the elderly.

1. INTRODUCTION

In recent years, there has been considerable debate about the best way to
finance health care. At the same time, there have also been some important
changes in funding practices. As noted in Wagstaff et al. (1999, p. 269), ‘‘It is
well known that during the last decade or so, there has been a shift in many
OECD countries away from public sources of finance (for health care) to
private sources’’. One of the principle arguments for doing so is to reduce
the burden of taxation. If the tax system is progressive, however, it is also
important to measure the equity benefits of publicly funded health care in
order to be able to make an informed choice regarding the tradeoff between
reduced taxes, with the accompanying efficiency benefit to the economy, and
any welfare loss due to reduced equity. Our goal in this paper is to measure
this latter cost using the Canadian health care and income tax systems as an
example.
More specifically, in this paper we study the distributional impact of a
hypothetical move from completely public to totally private financing of
health care, applying our methods to Canada. In a private system, premiums
would be based on age and gender, as well as on other individual charac-
teristics related to expected health care costs. In a public system, premiums
tend to be more uniform or even absent. In Canada, in particular, essentially
all funding comes from the government’s general tax revenues (see Aba,
Goodman, & Mintz, 2002, p. 2). Here, we do a standard distributional impact
analysis of switching between these radically different methods of funding.
We adopt some simplifying assumptions which we believe, for the most
part, lead to fairly conservative estimates of the reduction in inequality
generated by a public health care system in comparison to a private system.
For example, we assume that premiums would only depend on age and
gender under a private system, which may lead to underestimation of the
fees that would be paid by poor people, who may on average have higher
health risks, or others with pre-existing health problems. One feature, how-
ever, that may lead to an over-estimate of the equalizing impact of a public
system is that impact analysis does not allow for behavioural changes.1
After a move to a private system, lower income households may opt for
lower levels of coverage and lower premiums. In contrast, we assume they
Progressivity Implications of Public Health Insurance Funding in Canada 135

would have to pay the full actuarial cost of unchanged coverage, perhaps
tending to exaggerate the disequalizing impact of moving to a private
system. There are also equity effects that could in principle be incorporated
in impact analysis but which are not included in our calculations. For
example, although it is a matter of dispute, some studies indicate that lower
income people receive less care than higher income groups in public systems.
If so, then they could perhaps get the same level of coverage under a private
system with lower premiums than we assign them in our calculations. We
neglect this aspect in part because the evidence on inequities of this type
is mixed. Also, such inequities are not a necessary aspect of a public
health care system, but are consequences of the methods of delivery and the
information available to individuals.2
In Canada, public health care is funded largely out of general tax revenues
and is shared by provincial and federal governments. The Canada Health
Act goes so far as to forbid explicit user charges for medically necessary
doctor and hospital services. In 1977, the federal government altered the EPF
(Established Programs Financing), which was used to transfer funds to
provinces based specifically upon postsecondary education and health care
expenditures, in favour of a block grant covering both areas in a way to be
determined by the provinces. Further changes were made in 1996–1997 when
the EPF was abandoned altogether in favour of the CHST (Canada Health
and Social Transfer) which pooled even more programs into a block grant.3
This complicates the question of assigning these tax costs to determine the
distributional impact of the required ‘‘additional’’ tax collections needed to
finance a publicly provided health care system. If all taxes rise together to
create the needed revenue, the incidence may be approximately proportional
to income – in line with the well-known result that the tax system as a whole
typically has roughly proportional incidence (see e.g., Gillespie, 1980;
Pechman, 1985). In contrast, if a payroll tax is earmarked to partially fund
health care, or if personal income tax is the marginal source of funds, the
distributional impact may deviate from proportionality. An additional
complication is that in a world of private health insurance, it is possible that
premia would qualify for income tax deductions or credits. This could
increase the effective regressivity of the funding burden; that is if the value of
this tax relief rose more than in proportion to income. Also, in health care
systems that are largely privatized, some provision of health insurance
is usually provided to low income individuals (e.g., Medicaid in the US).
Reducing the extent to which individuals qualify for such benefits, however,
can actually lead to increased costs of health coverage due to the shift of
care from visits to doctor’s offices to more expensive visits to emergency
136 JAMES B. DAVIES AND MICHAEL HOY

departments and increased hospitalizations. Hence, the expected savings

in tax-financed health provision from increased privatization may not be
experienced to the extent expected (e.g., see Butler, Johnson, & Rimsza,
2006). In any case, the distribution of the funding burden is likely to be more
regressive under the case of private insurance than for publicly provided
insurance, where premiums depend on level of coverage and personal char-
acteristics rather than earnings or income.
We assess the benefit of publicly provided health care for each family
to be the avoidance of private health insurance premiums that would be
required for the same level of health care based on age and gender, net of
additional tax burdens imposed to fund the system. This is equivalent to
assuming that insurance premiums in a hypothetical private health insur-
ance plan would be based only on age and gender and that individuals
would all purchase the same coverage (i.e., for the same health benefits) as is
currently provided in the public system. These and other assumptions are
discussed in detail in the paper.
It is perhaps not surprising that we find that inequality in disposable in-
come is lower under a public health care system than under a private system.
But we should want to know more than this. Is the increased welfare resulting
from reduced inequality of sufficient magnitude to be a serious consideration
in deciding whether to maintain a publicly provided health care system?
To answer such questions we must make some value judgements, and we
must be clear about how much importance is placed on equality. We do this
by adopting the framework of utilitarian social welfare functions (swf ’s).
We illustrate the approach by using the additive iso-elastic family of swf ’s,
which generate the well-known Atkinson inequality indices. The Atkinson
indices allow one to vary the degree of inequality aversion, and to translate
inequality changes readily into equivalent changes in per capita income
through their ‘‘cost of inequality’’ interpretation. This gives us a flexible tool
that makes the welfare impact of inequality changes comparable with those of
efficiency changes.
It is well established that the cost of providing public health care is par-
ticularly high for those in older age groups. Therefore, we consider separately
the subgroup of the population for those with family heads aged 65 and older
and perform a decomposition analysis by age group (families with heads
aged under 65 and families with heads aged 65+). We do indeed find that the
equality enhancing effect of public health care provision is especially strong
for the older age group. The effects on between-group inequality levels
depend very much on the particular inequality index and decomposition
method chosen. These results prove interesting in a number of respects.
Progressivity Implications of Public Health Insurance Funding in Canada 137

Overall, our computations indicate that public health care, as funded in

Canada, reduces inequality considerably.
The remainder of the paper is organized as follows. The next section
outlines our methods. Section 3 presents results for the Canadian case. In
Section 4 we then discuss those results, considering their limitations and also
how they might differ under other assumptions and in other countries.
Conclusions are provided in Section 5.

2. METHODOLOGY

Due to the progressivity of the Canadian income tax system, after-tax

incomes are distributed more equally than are before-tax incomes.4 Many
government transfers, such as old age security and welfare payments, are also
progressive. The incidence of other government programs, such as health
care and education, however, is less well studied due to measurement prob-
lems as well as lack of appropriate data.5 The report ‘‘National Health
Expenditures in Canada 1975–1994’’ (1996) provides data on public health
expenditures on individuals by age and gender allowing us to ascertain what
the relevant private health insurance premiums would have to be to cover
people of different ages and genders. We assume that, in the absence of
public health insurance, individuals would be charged premiums for private
coverage based on their age and gender and that these premiums would equal
the relevant age- and gender-based expenditures. There are several important
assumptions implicit in this approach which we list below and then discuss.
1. In the absence of publicly provided health care, people would purchase
the same type of private insurance coverage regardless of their income
and this coverage level would be equal to that required to provide existing
levels of care.
2. This approach presumes that the cost of health insurance would be equal
to the amount that is required to finance existing levels of care at existing
delivery costs as generated by the public health care system.
3. This approach also presumes that insurance premiums would not depend
on any factors other than age and gender.
While it is necessary to make simplifying assumptions to make progress
here, it must be recognized that the above are strong assumptions. In the case
of the first assumption noted above, it is not likely that everyone would wish
to purchase the same level of coverage in a private insurance scenario, and
even if that were so it is not likely that the coverage level chosen would just
138 JAMES B. DAVIES AND MICHAEL HOY

happen to coincide with the particular services provided by the existing

public health system.6 As for the second assumption, it is possible that
private health insurance could be linked to efficiencies in the delivery of
health care which could lead to lower costs, although the opposite is also a
possibility. Finally, considering assumption 3, private insurers are likely to
use more than age and gender as categorization variables. Pre-existing health
problems, behavioural traits such as smoking, and results of diagnostic tests
(e.g., persistent high blood pressure) are likely to influence health insurance
premiums. We will discuss the impact of adopting more general assumptions
in the last section of the paper. The assumptions we have adopted allow for a
clear basis on which to develop benchmark comparisons.
If there were no public health insurance system, then substantial tax
revenues would be freed up and we need to account for this. In 1994, for
example, public health expenditures in Canada were $52 billion. This
represents 49% of the aggregate personal income taxes paid according to the
Survey of Consumer Finance (1994), which we use in the calculations
reported later in this paper. In those calculations we assume that in the
absence of a public health care system, income taxes would be reduced by
equal proportionate reductions in the amount paid across all income
groups.7 This type of tax cut preserves the existing degree of liability
progression, and hence has been referred to as an LP-neutral tax cut. Since
higher income individuals generally have higher tax liabilities both in
absolute terms and as a fraction of pre-tax income, an LP-neutral tax cut
leads to an increase in relative inequality. As an alternative, consider a tax
cut that returns the same percentage increase in post-tax income to everyone
(i.e., one that preserves residual progression, termed an RP-tax cut). Pfahler
(1984) showed that, given any existing progressive (strictly convex) income
tax schedule, an LP-neutral tax cut leads to a less equal distribution than
would an RP-neutral tax cut (see also Lambert, 2001, pp. 219–224).
Moreover, if in addition the existing pre-tax income distribution is positively
skewed, then the majority would favour an RP-neutral tax cut. Thus,
we recognize that our suggested tax-cut, being an LP-neutral one, is not
the only one that could be considered. Moreover, our choice may exaggerate
the inequality implications of moving to a private health insurance scheme.
We return to this point again in the concluding section.
In order to assess the relative effects on economic welfare and inequality
under a public and hypothetical private health insurance scenario, we use
Atkinson’s (1970) cost of inequality framework.8 Let Yede be the ‘‘equally
distributed equivalent income’’: that is, the income that if received by
all would generate the same social welfare as the actual income distribution
Progressivity Implications of Public Health Insurance Funding in Canada 139

Y1, y , Yn. The gap between Yede and mean income, m, is an appealing
measure of the ‘‘cost’’ of inequality as it reflects the fraction of income that
could be eliminated while maintaining the same level of welfare as obtained
by the existing distribution of income provided incomes were equally
distributed. Thus, the difference in this measure for our two income dis-
tributions that are generated by different (hypothetical) health care systems
provides a useful measure of the relative efficiency of the two systems
in generating welfare. Moreover, this measure provides for a useful com-
parison with any measures of market inefficiency for either or both systems.
Thus, the Atkinson inequality measure, defined as


Y ede
A¼1
(1)
m
represents the percent reduction in mean income that could be allowed if
income were to be distributed equally without reducing social welfare. This
measure varies between 0 (complete equality) and 1 (complete inequality –
one person holds all the income). It can be used to assess the impact of
inequality on welfare from the viewpoint of widely varying social welfare
functions. Atkinson illustrated with the additive (Utilitarian) social welfare
function with iso-elastic utility function; that is, U(y) defined by
yð1
eÞ
UðyÞ ¼ ea1; e40
1
e
¼ lnðyÞ e ¼ 1 ð2Þ
whose properties are so familiar to economists in other contexts. Here, the
parameter e reflects aversion to inequality, and ranges from 0 (complete
insensitivity to inequality) to N (sensitivity to inequality is so strong it
approaches the Rawlsian case). Applying Eq. (2) in a Utilitarian social
welfare framework, we get the associated inequality measures:
"
 #1=ð1
eÞ
1 X yi ð1
eÞ
Ae ¼ 1
ea1; e40 (3)
n m

In the case of e ¼ 1 we obtain the particular member of the Atkinson family

which isPordinally equivalent to the so-called mean logarithmic deviation MLD
½
ð1=nÞ lnðyi =mÞ: Thus, we also refer to this particular member of the
Atkinson family as the AMLD measure, with formula given below:
 X

1 y
AMLD  A1 ¼ 1
exp ln i (4)
n m
140 JAMES B. DAVIES AND MICHAEL HOY

In the calculations reported in this paper we use the Atkinson measure,

which is attractive because of its explicit ethical basis, flexibility, and ready
interpretation in welfare terms.9 In order to illustrate how this framework
helps us to assess inequality in welfare terms, consider the following example.
Suppose the AMLD index of some income distribution A were 0.18 while for
some other distribution B, with the same per capita income level, the degree
of inequality were 0.13. One would say that in A the equivalent of 18% of
income per capita is effectively ‘‘lost’’ due to inequality, while the ‘‘cost of
inequality’’ is only 13% in the case of distribution B. Thus, moving from A to
B is equivalent in welfare terms to a 5% increase in per capita income.
Measured inequality, and the cost of inequality, of course depend on how
averse one is to inequality. In terms of the Atkinson measures we illustrate
this point with a simple two-person case using parameter values e ¼ 1 and
e ¼ 2 in Fig. 1. Consider the initial income distribution of y1 ¼ $18,000 and
y2 ¼ $54,000, implying an average income of $36,000. This pair of incomes

Y2

.
(18000, 54000)

e=1

e=2
Y2 + Y1 = 72,000

Y1

27,000 36,000
31,180

Fig. 1. The Equally Distributed Equivalent Income Per Capita is $31,180 for e ¼ 1
and $27,000 for e ¼ 2. Thus, the Per Capita Cost of Inequality is $4,820 (or 13.4%
of Total Income) for e ¼ 1 and $9,000 (orP25% of Total Income) for e ¼ 2.
The Underlying Welfare FunctionPis W ¼ i ðyð1
eÞ i =ð1
eÞÞ for e40; ea1 and
W ¼ i lnðyi Þ for e ¼ 1.
Progressivity Implications of Public Health Insurance Funding in Canada 141

represents approximately the average after-tax incomes of the poorest 50%

and the richest 50% of the population (see Table 1) in Canada in 1994. For
parameter value e ¼ 1, the measure of inequality is A1 ¼ 0.134 while for
e ¼ 2 it is A2 ¼ 0.25. As illustrated in the graph, the same level of social
welfare could, if incomes were equally distributed, be obtained with average
income level of $31,180 for the case of e ¼ 1 (i.e., 13.4% less income) and
with average income of $27,000 (i.e., 25% less income) for the case of e ¼ 2.
The cost of inequality is thus regarded as higher if one has higher inequality
aversion.10
While it is not necessary to take the social contractarian view of the
Utilitarian Welfare framework in order to justify use of the Atkinson index,
we do find it helpful to refer to the risk measurement literature in order to
help judge what are reasonable values for the parameter e. The iso-elastic
form of U is often used to estimate the degree of relative risk aversion.
Estimates vary across individuals and studies, but values of e ¼ 1 to e ¼ 2
are generally considered reasonable.11 Since we do not wish to exaggerate the
importance of inequality, we avoid the use of higher values in our calcu-
lations, and focus mainly on the results for e ¼ 1. We also indicate outcomes
for the even more conservative view of inequality implied by e ¼ 0.5.
We ascertain the benefit from public health care in reducing the degree of
inequality in ‘‘net’’ incomes as follows. First, we find the degree of income
inequality in measured after-tax incomes using the Atkinson indices noted
above. Then we determine the actuarial value of the health insurance
coverage, or ‘‘health insurance benefit’’, that the public system provides for
each family, based on the number, age, and gender of its members. (See the
next section for more details.) We add this benefit to their after-tax incomes
in order to generate our net income value in the presence of public health
care. Recognizing that in the absence of a public health care system indi-
viduals would be required to pay privately for heath insurance but would
avoid having to pay for health care through the tax system, we then deter-
mine the distribution of net after-tax income that would result if income tax
payments were reduced by 49% across all families. Since higher income
families pay higher income taxes, this change in taxes will lead to greater
benefits for higher income families than for lower income families and so
increase measured inequality. The degree of inequality in this hypothetical
income distribution (with no public health insurance) relative to that in
existing net incomes (i.e., including public health care benefits) provides a
measure of the effect of public health care funding in reducing inequality.
Further complications could arise in measuring the impact of how gov-
ernments may alter the tax system in moving to a world of private health
142 JAMES B. DAVIES AND MICHAEL HOY

Table 1. Decile Means – 1994 Family Incomes, Taxes, and Health Care
Benefits.
Decile After-Tax Income Tax Gross Health Net Health Care
Income Paid Care Benefit Benefita

All ages
1 7,157 166 2,119 2,037
2 13,204 626 4,399 4,091
3 17,786 1,607 4,404 3,614
4 23,126 3,001 5,133 3,658
5 28,427 4,999 4,397 1,941
6 33,968 7,015 4,209 762
7 40,373 9,440 4,233
405
8 47,876 12,396 4,239
1,852
9 58,514 16,504 4,460
3,650
10 87,086 30,567 4,822
10,198
Top 5% 102,075 38,723 5,072
13,956
Top 1% 146,493 70,364 5,384
29,191
Overall 35,750 8,632 4,241 0
Under age 65
1 6,434 114 1,492 1,435
2 13,479 1,046 1,867 1,353
3 19,728 2,509 2,202 969
4 25,850 4,589 2,472 217
5 31,312 6,470 2,657
522
6 37,027 8,555 2,957
1,247
7 43,263 10,937 3,237
2,137
8 50,689 13,625 3,493
3,202
9 61,382 17,962 3,849
4,977
10 89,874 31,919 4,356
11,328
Top 5% 104,896 40,400 4,560
15,292
Top 1% 149,352 73,390 4,613
31,449
Overall 37,903 9,772 2,858
1,944

Age 65 and older

1 10,630 145 7,408 7,337
2 12,993 124 7,397 7,336
3 14,483 400 7,435 7,239
4 16,685 932 8,587 8,129
5 20,170 1,052 11,067 10,550
6 23,120 1,298 11,644 11,006
7 27,028 2,517 11,535 10,298
8 32,428 4,197 11,890 9,828
9 41,398 7,491 12,253 8,572
10 66,675 19,448 12,314 2,758
Progressivity Implications of Public Health Insurance Funding in Canada 143

Table 1. (Continued )
Decile After-Tax Income Tax Gross Health Net Health Care
Income Paid Care Benefit Benefita

Top 5% 80,548 26,738 12,581
557

Top 1% 125,541 53,335 12,995
13,212
Overall 26,553 3,757 10,153 8,306

Sources: Columns 2 and 3: Statistics Canada (Household Surveys Division) – Economic Fam-
ilies, 1994 Incomes; column 4: Health Canada. National Health Expenditures in Canada. Policy
and Consultation Branch – 1996; column 5: author’s calculation.
a
Net health care benefit is equal to gross health care benefit less 49% of income tax paid.

care. For example, it is possible that governments would provide income tax
deductions or partial credits for health insurance premia. Such tax relief
could either increase or reduce the disequalizing impact of switching to the
private system depending on patterns of expenditure on health insurance.
(If this spending did not vary with income, credits would be equalizing and
deductions could be equalizing as well if marginal tax rates did not rise too
sharply with income. Once expenditures are allowed to rise with income,
results become less clear-cut.) Also, there is likely to be significant variation
in such spending even among families at the same income level and with the
same demographic structure. Loomis and Revier (1988) outline a method for
analysing the impact of selective tax or subsidy measures that could be
applied to analyse the impact of tax relief for health insurance expenditures
when there is this real-world heterogeneity.
In addition to being sensitive to assumptions noted earlier in this section,
our results depend on two further important aspects. First, in an actual
privatization of health care the tax relief might not be distributed as we have
assumed. Second, a tax reduction of this size can be expected to have
incentive effects on labour supply and other income generating activities.
Thus, both the distribution of before- and after-tax money income would
likely change as a result of the tax reduction. We discuss how our results
might differ if these factors were taken into account later.

3. RESULTS
We perform two experiments. Our first experiment is to consider how in-
cluding the monetized value of public health provision changes the measured
level of inequality. We do this by determining the health insurance premiums
144 JAMES B. DAVIES AND MICHAEL HOY

that are inferred for each family based on the number, age, and gender of
family members.12 We will refer to this as the gross health benefit HBi for
each family, i. HBi is, under our assumptions, the actuarially fair private
insurance premium that would have to be paid in order to obtain access to
the existing level of health care services. Let Y Bi represent before-tax income,
YA A B
i be after-tax income, and Ti be income tax paid (i.e., Y i ¼ Y i
T i ).
In each of the Tables 2–4 and 6–8, the first set of data on incomes is simply
based on Y A i ; measured after-tax incomes. The second set of columns
is based on after-tax incomes plus the imputed value of health benefits
YA i þ HBi ; which we will refer to as income under public health, and so
reflects the impact of accounting for these benefits when measuring income
inequality. Comparing these two sets of results allows us to infer how
much more equally distributed disposable income is if one accounts for the
benefits of having one’s health insurance premiums effectively paid. Unless
there is a sufficiently strong and positive relationship between family income
and imputed health care benefits, the degree of inequality will be less once
these benefits are included in income. This effect occurs because an equal
absolute increase in income leads to a reduction in relative inequality.13
Our second experiment is to compare the income statistics, and in par-
ticular the values of inequality indices, of the incomes Y Ai þ HBi ; which more
accurately reflect real or full disposable income in the presence of the publicly
financed health insurance plan, with a set of hypothetical incomes reflecting
the scenario should the public system be replaced entirely by private health
insurance. In the third set of columns, we model these hypothetical incomes
should the public insurance system be replaced entirely by a private one
(hence HBi is not added to after-tax incomes since individuals must purchase
their health coverage out of their disposable income). However, since
the government does not have to finance the public health system, a
proportion (49%) of income tax is assumed to be rebated to every family
according to the LP-neutral tax-cut approach previously discussed. So the
statistics provided in this third set of columns are based on Y A i þ 0:49T i ;
which we refer to as income under private health.
In Table 1, we see that the gross health care benefits are in fact some-
what positively correlated with income. Note, for example, that families in
the lowest decile on average receive the smallest health care benefit. This
occurs because small families (and single individuals in particular) are over-
represented in the lowest decile. However, a somewhat smaller absolute in-
crease in income to a low income earner than to a higher income earner can
still reduce relative inequality if that transfer represents a larger fraction of
original income. This is demonstrated in the second set of columns of Table 2
 Progressivity Implications of Public Health Insurance Funding in Canada
Table 2. Family Incomes, 1994, All Ages.
Decile After-Tax Income After-Tax Income Under Public Health After-Tax Income Under Private Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 7,157 2.00 2.00 9,277 2.32 2.32 7,239 1.81 1.81

2 13,204 3.69 5.70 17,603 4.40 6.72 13,511 3.38 5.19
3 17,786 4.97 10.67 22,189 5.55 12.27 18,575 4.64 9.83
4 23,126 6.47 17.14 28,258 7.07 19.34 24,600 6.15 15.99
5 28,427 7.95 25.09 32,824 8.20 27.54 30,884 7.72 23.71
6 33,968 9.50 34.59 38,177 9.55 37.09 37,415 9.36 33.06
7 40,373 11.29 45.89 44,606 11.15 48.24 45,012 11.26 44.32
8 47,876 13.39 59.28 52,115 13.03 61.27 53,967 13.49 57.81
9 58,514 16.37 75.64 62,973 15.75 77.02 66,623 16.66 74.47
10 87,086 24.36 100.00 91,908 22.98 100.00 102,106 25.53 100.00
Top 5% 102,075 14.29 107,147 13.41 121,102 15.15
Top 1% 146,493 4.11 151,877 3.81 181,068 4.54
Overall 35,750 39,992 39,992
Gini 0.354 0.322 0.374
CV 0.681 0.627 0.742
Atkinson 0.205 0.181 0.229
(e ¼ 1)
Atkinson 0.102 0.089 0.114
(e ¼ 0.5)
Atkinson 0.449 0.396 0.480
(e ¼ 2)
Theil 0.203 0.176 0.231

145
 146
Table 3. Family Incomes, 1994, Heads Aged Under 65.
Decile After-Tax Income After-Tax Income Under Public Health After-Tax Income Under Private Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 6,434 1.70 1.70 7,926 1.94 1.94 6,490 1.52 1.52

2 13,479 3.56 5.25 15,346 3.77 5.71 13,993 3.28 4.80
3 19,728 5.21 10.46 21,930 5.38 11.09 20,961 4.91 9.71
4 25,850 6.82 17.28 28,322 6.95 18.04 28,105 6.58 16.29
5 31,312 8.26 25.54 33,969 8.33 26.37 34,491 8.07 24.36
6 37,027 9.77 35.31 39,984 9.81 36.18 41,231 9.66 34.02
7 43,263 11.41 46.72 46,500 11.41 47.59 48,637 11.39 45.41

JAMES B. DAVIES AND MICHAEL HOY

8 50,689 13.37 60.10 54,182 13.29 60.88 57,384 13.44 58.84
9 61,382 16.19 76.29 65,231 16.00 76.88 70,208 16.44 75.28
10 89,874 23.71 100.00 94,230 23.12 100.00 105,558 24.72 100.00

Top 5% 104,896 13.84 109,456 13.43 124,747 14.61

Top 1% 149,352 3.95 153,966 3.78 185,414 4.35
Overall 37,903 40,761 42,704
Gini 0.349 0.336 0.366
CV 0.661 0.639 0.715
Atkinson 0.209 0.197 0.230
(e ¼ 1)
Atkinson 0.101 0.095 0.112
(e ¼ 0.5)
Atkinson 0.480 0.433 0.509
(e ¼ 2)
Theil 0.198 0.187 0.222
 Progressivity Implications of Public Health Insurance Funding in Canada
Table 4. Family Incomes, 1994, Heads Aged 65 and Over.
Decile After-Tax Income After-Tax Income Under Public After-Tax Income Under Private
Health Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 10,630 4.01 4.01 18,038 4.92 4.92 10,701 3.78 3.78

2 12,993 4.89 8.90 20,389 5.55 10.47 13,054 4.59 8.37
3 14,483 5.45 14.35 21,918 5.97 16.44 14,680 5.17 13.53
4 16,685 6.28 20.63 25,272 6.88 23.32 17,143 6.03 19.57
5 20,170 7.60 28.23 31,236 8.52 31.84 20,687 7.29 26.86
6 23,120 8.71 36.95 34,764 9.48 41.32 23,758 8.37 35.23
7 27,028 10.19 47.14 38,563 10.52 51.84 28,265 9.97 45.19
8 32,428 12.20 59.34 44,318 12.06 63.90 34,490 12.13 57.33
9 41,398 15.58 74.92 53,651 14.61 78.51 45,079 15.87 73.19
10 66,675 25.08 100.00 78,989 21.49 100.00 76,231 26.81 100.00
Top 5% 80,548 15.22 93,129 12.73 93,686 16.55
Top 1% 125,541 4.78 138,536 3.82 151,748 5.41
Overall 26,553 36,706 28,399
Gini 0.318 0.260 0.341
CV 0.687 0.542 0.776
Atkinson (e ¼ 1) 0.148 0.104 0.170
Atkinson 0.080 0.055 0.094
(e ¼ 0.5)
Atkinson (e ¼ 2) 0.257 0.184 0.288
Theil 0.175 0.118 0.208

147
148 JAMES B. DAVIES AND MICHAEL HOY

[see column entitled ‘‘After-tax Income under Public Health’’] where we see
that the share of income for all deciles 1 through 6 is higher when public
health care benefits are included in income. We find that the cumulative share
of income is higher throughout the distribution (i.e., we have Lorenz
dominance), which implies that all relative inequality indices must agree that
accounting for imputed health care benefits reduces measured income
inequality. Note that the AMLD measure indicates an inequality level of
0.205 for (unadjusted) after-tax incomes but falls to 0.181 when (imputed)
government-provided health care benefits are included. Thus, the perceived
cost of inequality falls by 2.4% of per capita incomes if one includes health
care benefits in income. This drop is quantitatively important as it eliminates
more than 10% of the cost of inequality in ‘‘measured’’ after-tax incomes.
The way that public health care benefits are funded has an important effect
on the distributional changes that one would expect in a move to private
health insurance. For example, the average income tax paid by families in
the first two deciles is only $166 and $626, respectively, while in the tenth
(highest income) decile the amount is $30,567. Therefore, if one were to deduct
the share of income tax (49%) that is required to fund public health care from
each family’s tax bill, the result would be a more unequal distribution of after-
tax incomes. The quantitative impact of this exercise is reported in the third set
of columns of Table 2 [After-tax Income under Private Health]. We see that
(hypothetical) after-tax incomes are higher for each decile, as expected, since
tax payments are reduced, but the decile shares and the inequality measures
indicate that the distribution is less equal than existing after-tax incomes with
or without the adjustment to include publicly provided health care.
To determine the distributional implications of moving to a privately
funded health insurance system, we compare the values in the second set of
columns of Table 2 [After-tax Income under Public Health] to those in the
third [After-tax Income under Private Health]. Note that incomes in the
second set of columns, Y A i þ HBi ; are inflated relative to measured after-tax
incomes in order to reflect the monetary benefit of publicly provided health
care, while incomes in the third set of columns, Y A i þ 0:49T i ; are inflated
relative to measured after-tax incomes because the equivalent amount of
money used to finance publicly provided health care is ‘‘given back’’ to
individuals via hypothetical tax cuts. Thus, since average incomes are the
same in the second and third sets of columns, the link between the inequality
analysis and welfare analysis as discussed in the second section of this paper
is consistent.14
All the inequality values are higher for the private health care scenario
than under existing public health care. The AMLD measure (i.e., the
Progressivity Implications of Public Health Insurance Funding in Canada 149

Atkinson measure with parameter value e ¼ 1) indicates that if one were to

move from the existing public health care system to a purely private one, the
cost of inequality would rise by the equivalent of 4.8% of per capita income
(i.e., from 18.1% of income to 22.9%), an increase of about 25% of existing
inequality. Since the AMLD measure embodies a fairly conservative attitude
concerning the cost of inequality, this is a substantial difference. It suggests
that the equity benefit from funding health care benefits through the tax
system rather than relying entirely on a private health insurance system is
equivalent to finding a way to raise per capita or total income by 4.8%.
From our data sources, health care expenditures represent $4,241 per family
and average income is $35,750 per family, implying that the overall impor-
tance of health care expenditure is approximately 12% of income. Achieving
an equity benefit equivalent to a 4.8% increase in income through allocation
of 12% of income seems quite impressive.
In Table 2, we also provide results using the Atkinson inequality index with
parameter values e ¼ 0.5 and e ¼ 2. The corresponding cost of inequality that
would be generated by moving to a private health care system would be
equivalent to a per capita income loss of 2.5 and 8.4%, respectively. These
comparisons show how the value of a public health care system in reducing
inequality depends (naturally) on how averse one is to inequality.
It is well documented that health care spending is higher for older indi-
viduals. Therefore, we separated our sample of families into those with heads
less than 65 years of age and those 65 years of age or older and did the same
analysis as above separately on these groups. The results are reported in
Tables 3 and 4. Comparing Tables 3 and 4 we see, as expected, that gross
health care benefits are substantially higher for families headed by older
individuals. Including public health care benefits increases average real
income for the age group 65 years and over from $26,553 to $36,706 while
only from $37,903 to $40,761 for families in the under 65 years age group.
Another way of seeing this point is to look at the net health care benefit from
the public system computed in Table 1 for the average family in the under 65
group (
$1,944) with that for the average family in the 65 years and older
group (+$8,306). Thus, public funding of health care represents a substantial
transfer of income to older families from younger families.
Also note that after-tax incomes are more equally distributed for families
with heads 65 years of age and older. Despite this fact, the equality enhancing
implications of publicly provided health care are especially effective for the
65+ age group, for which inequality according to the AMLD is less due to
publicly provided health care by the difference 0.066 (i.e., 0.170–0.104),
equivalent to 6.6% of per capita incomes of this group. For the remaining
150 JAMES B. DAVIES AND MICHAEL HOY

families (with heads under age 65), the equality enhancing effect of public
health care is the equivalent of 3.3% of per capita income (i.e., 0.230–0.197).
Given the differential impact of public financing of health care on the
‘‘young’’ and ‘‘old’’ age groups, we perform two common types of decom-
position analysis on our results. The first is based on the generalized entropy
measures of inequality Ea. This family of inequality measures depends on
the degree of inequality aversion, a:

 
a 
1 P yi
Ea ¼
1 aa0; 1
Naða
1Þ m


1X m
E0 ¼ ln a¼0 (5)
N i yi



1 X yi y
E1 ¼ ln i a¼1
N i m m

For the values ao1, the entropy measure Ea is ordinally equivalent to the
Atkinson measure Ae, where a ¼ 1
e. The two are related according to the
following relationship:
Ae ¼ 1
½ða2
aÞE a þ 11=a ao1; aa0
(6)
A1 ¼ 1
expð
E a Þ a¼0

Among relative inequality measures, only the family of generalized entropy

measures (and scalar multiples of these) are additively decomposable (see
Shorrocks, 1984). This property requires that, if one partitions the population
into subgroups (indexed k ¼ 1, 2, y , K), then overall inequality, I, can be
expressed as a weighted sum of the inequality values calculated for subgroups,
Ik, Pplus a term representing ‘‘between-group’’ inequality, IB (i.e.,
I ¼ k wk I k þ I B ). If the weights wk are the population shares of the sub-
groups, then only E0 (i.e., a ¼ 0) satisfies this complete set of properties.
The results of our decomposition for the generalized entropy measure E0
are reported in Table 5. The between-group term is reported as a percentage
of the overall value of inequality. This value is reduced from 3.9 to 0.0005%
when including the public health benefits in income. Moving to the private
insurance scenario, the between-group measure of inequality is 5% of overall
inequality. These results reflect the fact that mean income differences between
the age groups are lowest when using after-tax incomes that include public
health benefits ($36,706 for the ‘‘old’’ versus $40,761 for the ‘‘young’’) and
highest under private health insurance ($28,299 for the ‘‘old’’ versus $42,704
for the ‘‘young’’). In fact, it is a property of E0 that its between-group
Progressivity Implications of Public Health Insurance Funding in Canada 151

Table 5. Inequality Decomposition by Age Group.

Income Group Decomposition Based Decomposition Based on Atkinson
Definition on Entropy Measures Measures

Mean E0 A0.5 (equiv A1 (equiv A2 (equiv

income income) income) income)

After-tax All 35,750 0.229 0.102 0.205 0.449

income (32,103) (28,421) (19,698)
Ageo65 37,903 0.234 0.101 0.209 0.480
(34,075) (29,981) (19,709)
AgeZ65 26,553 0.160 0.080 0.148 0.257
(24,428) (22,623) (19,729)
Between 3.9% 3.88% 2.24% 0.094%

After-tax All 39,992 0.199 0.089 0.181 0.397

income (36,433) (32,753) (24,155)
under Ageo65 40,761 0.219 0.095 0.197 0.433
public (36,889) (32,731) (23,111)
health AgeZ65 36,706 0.109 0.055 0.104 0.184
(34,687) (32,888) (29,952)
Between 0.005% 1.08% 0.11% 1.6%

After-tax All 39,992 0.260 0.114 0.229 0.480

income (35,433) (30,833) (20,795)
under Ageo65 42.704 0.261 0.112 0.230 0.509
private (37,921) (32,882) (20,957)
health AgeZ65 28,399 0.186 0.094 0.170 0.228
(25,729) (23,571) (21,924)
Between 5.0% 3.88% 3.09% 1.85%

Note: Between means between group inequality according to the index approach for E0 and
according to the cost of inequality approach (i.e., applicable for the Atkinson based decom-
position approach). Values are in percentage of overall inequality terms.

inequality value, E B0 ; is based

P on the ratios of the overall mean income to the
subgroups means ðE B0 ¼ k wk lnðm=mk ÞÞ:
Although the Atkinson measures do not lend themselves to additive dec-
omposability based on the index measures of inequality per se, they can be
used to decompose inequality in another manner (see Blackorby, Donaldson,
& Auersperg, 1981; Lambert, 2001, pp. 113–114). In this case, one measures
between-group inequality by comparing the equally distributed equivalent
incomes of the subgroups rather than their mean incomes. Let xk represent
the equally distributed equivalent income for subgroup k (and x for the
population as a whole), as defined implicitly in Eq. (1). The cost of inequality
152 JAMES B. DAVIES AND MICHAEL HOY

is then Ck P¼ mk
xk for subgroup k and C ¼ m
x overall. It follows that

C ¼ C B þ k wk C k where CB is the ‘‘between-group cost of inequality’’ and
the weights wk are the population weights. We report (in Table 5) the
percentage of overall inequality that CB accounts for (i.e., ðC B =CÞ  100).
The corresponding index valueP of the between-group inequality, IB, for the
Atkinson measure is I B ¼ C B = k wk xk :
As when we used the entropy-based decomposition approach, we see a
substantial decrease in between-group inequality when comparing after-tax
income plus public health benefit either to measured after-tax incomes or the
after-tax incomes associated with a private health system for the cases where
the inequality aversion parameter is e ¼ 0.5 and e ¼ 1 (the AMLD measure).
The reason for this pattern, however, is not the same as for the entropy-based
comparison where reduced between-group inequality was due to a reduced
difference of mean income for the young to the old. The Atkinson-based
decomposition is driven by the differences in equally distributed equivalent
incomes of the groups rather than mean incomes. In fact, in the case of e ¼ 1,
the equality enhancing effect of including public health benefits in after-tax
income is so strong for the old that their equally distributed equivalent
income becomes higher than that for the young even though the mean in-
come of the young is higher by about 10%. So between-group inequality in
the case of the measure A1 is due to the older group having a higher equally
distributed equivalent income rather than a lower mean income, a reversal
relative to the cases when E0 or A0.5 is used as the basis of comparison.
The above discussion sets up well the pattern-breaking decomposition
analysis based on the measure A2. In this case, the value of between-group
inequality is substantially higher when adding public health benefits to
measured after-tax income (from 0.094% of the overall cost of inequality to
1.6% as a result of including public health benefits). The reason this happens
is that including benefits from public health does more to reduce inequality
for the old. The inequality measure A2 is sufficiently sensitive to this advan-
tage that the equally distributed equivalent income is much higher for the old
relative to that of the young ($29,952 versus $23,111). In a sense, from a
between-group perspective, the public health system increases inequality by
making the individuals in the old age group too well off relative to the young.
However, it is important to note that the overall level of inequality is still at
its lowest in these comparisons for the definition of income that includes
public health benefits. Also, part of the reason the equally distributed equiv-
alent income is so much higher for the old when public health benefits are
included is that the Atkinson measure of inequality with e ¼ 2 places so
much weight on the lowest incomes (see Chiu, 2007, for details). Even for
Progressivity Implications of Public Health Insurance Funding in Canada 153

after-tax incomes not including public health benefits, mean income for the
lowest decile of the distribution for the old is higher than for the young
(see Table 1). The inequality measure A2 places enough weight on this aspect
that equally distributed equivalent income is even slightly higher for the
old relative to the young when using measured after-tax incomes ($19,729
versus $19,709).
One important criticism of using families as the sample unit is that indi-
viduals in families of different size and composition but with the same family
income do not experience the same standard of living. Thus, one needs to
adjust incomes for family size. This cannot be done simply by dividing family
income by the number of individuals in the family. We must account for
economies of scale in meeting certain economic needs, such as housing, for
the family members. The standard method uses household equivalence scales,
which allow one to assign an equivalent per capita income to each member of
the family based on family income and composition. We applied the OECD
equivalence scales to the families in our sample and generated equivalent per
capita after-tax incomes.15 In doing this we follow standard practice and
assume that economies of scale within the household do not apply to health
benefits. Therefore, we do not apply the household weights used to derive
equivalized incomes to the gross health benefits included in the second set of
columns of Tables 6–8. Instead, we added the unadjusted per capita value of
gross health benefits for the family to each person’s income after the
equivalization procedure was applied to after-tax incomes. The reason for
making the calculations this way is that families would in fact require this
amount of money to purchase their health insurance privately and no
economies of scale would be relevant to this expenditure.16 The net income
[After-tax Income under Private Health] in the third set of columns of Tables
6–8 is computed by three steps: (a) adding to the after-tax incomes of column
set 1 the 49% of tax paid to cover public health care costs, (b) deducting the
amount required to pay for health insurance privately, and then, (c) applying
the equivalizing procedure to get per capita equivalent incomes.
Although we adjust family income according to an equivalence scale
which is not simply the inverse of the number of family members, except for
that part of income which is the imputed health benefit from publicly pro-
vided health insurance, in computing inequality statistics we weight indi-
viduals equally regardless of family type that they belong. However, as
Ebert (1997, 1999) has established, this approach for weighting incomes and
individuals across different family types creates problems (paradoxes) when
assessing welfare implications of redistributions of income between different
types of family units. His proposed method of weighting individuals by the
 154
Table 6. Equivalent Incomes, 1994, All Ages.
Decile After-Tax Equivalent Income After-Tax Equivalent Income Under After-Tax Equivalent Income under
Public Health Private Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 5,675 2.96 2.96 6,751 3.22 3.22 4,298 2.23 2.23

2 9,680 5.03 7.99 11,187 5.33 8.55 8,107 4.20 6.43
3 12,015 6.25 14.24 14,282 6.81 15.36 9,934 5.15 11.57
4 13,981 7.27 21.51 16,351 7.79 23.16 12,127 6.28 17.86
5 15,986 8.32 29.84 17,899 8.54 31.70 15,085 7.82 25.67
6 18,176 9.46 39.29 19,896 9.48 41.18 17,882 9.26 34.94
7 20,858 10.85 50.14 22,525 10.74 51.92 21,162 10.97 45.90

JAMES B. DAVIES AND MICHAEL HOY

8 24,235 12.61 62.76 25,926 12.36 64.28 25,209 13.06 58.97
9 29,129 15.15 77.90 30,797 14.68 78.96 31,156 16.14 75.10
10 42,465 22.10 100.00 44,126 21.04 100.00 48,063 24.90 100.00
Top 5% 49,797 12.97 51,499 12.29 57,551 14.92
Top 1% 73,024 3.80 74,868 3.57 89,214 4.63
Overall 19,219 20,973 19,301
Gini 0.292 0.269 0.350
CV 0.575 0.537 0.719
Atkinson 0.138 0.125 0.211
(e ¼ 1)
Atkinson 0.069 0.062 0.105
(e ¼ 0.5)
Atkinson 0.304 0.270 0.507
(e ¼ 2)
Theil 0.141 0.126 0.213
 Progressivity Implications of Public Health Insurance Funding in Canada
Table 7. Equivalent Incomes, 1994, Heads Aged Under 65.
Decile After-Tax Equivalent Income After-Tax Equivalent Income Under After-Tax Equivalent Income under
Public Health Private Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 5,377 2.78 2.78 6,371 3.12 3.12 4,089 2.02 2.02

2 9,263 4.79 7.57 10,272 5.02 8.14 8,212 4.05 6.06
3 11,782 6.10 13.67 12,806 6.27 14.41 11,161 5.50 11.57
4 14,040 7.26 20.93 15,082 7.38 21.79 13,833 6.82 18.39
5 16,242 8.41 29.33 17,305 8.47 30.26 16,513 8.15 26.53
6 18,490 9.56 38.89 19,610 9.59 39.85 19,069 9.39 35.93
7 21,218 10.97 49.86 22,346 10.93 50.79 22,337 11.01 46.94
8 24,618 12.74 62.60 25,789 12.62 63.41 26,374 13.01 59.95
9 29,555 15.29 77.90 30,789 15.07 78.48 32,273 15.92 75.87
10 42,775 22.10 100.00 44,017 21.52 100.00 48,993 24.13 100.00
Top 5% 49,998 12.94 51,250 12.55 58,395 14.40
Top 1% 72,570 3.76 73,844 3.62 89,479 4.42
Overall 19,334 20,437 20,283
Gini 0.299 0.285 0.340
CV 0.578 0.554 0.679
Atkinson (e ¼ 1) 0.144 0.133 0.192
Atkinson 0.072 0.066 0.095
(e ¼ 0.5)
Atkinson (e ¼ 2) 0.320 0.284 0.493
Theil 0.146 0.134 0.192

155
 156
Table 8. Equivalent Incomes, 1994, Heads Aged 65 and Over.
Decile After-Tax Equivalent Income After-Tax Equivalent Income Under After-Tax Equivalent Income under
Public Health Private Health

Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative Income ($) Share (%) Cumulative
share (%) share (%) share (%)

1 9,260 5.02 5.02 14,675 5.99 5.99 3,075 2.40 2.40

2 11,630 6.30 11.32 18,016 7.35 13.34 4,600 3.59 5.99
3 12,775 6.92 18.24 19,283 7.87 21.20 5,543 4.33 10.32
4 13,802 7.48 25.72 20,244 8.26 29.46 6,815 5.32 15.63
5 14,874 8.06 33.77 21,449 8.75 38.21 8,002 6.24 21.88

JAMES B. DAVIES AND MICHAEL HOY

6 16,376 8.91 42.69 22,301 9.14 47.35 10,127 7.94 29.81
7 18,499 10.00 52.68 24,303 9.89 57.24 12,734 9.91 39.73
8 21,569 11.67 64.35 27,407 11.16 68.40 16,298 12.70 52.42
9 25,978 14.07 78.41 31,844 12.98 81.38 21,707 16.93 69.35
10 39,855 21.59 100.00 45,655 18.62 100.00 39,285 30.65 100.00
Top 5% 47,851 12.99 53,815 11.00 49,639 19.41
Top 1% 75,863 4.12 81,907 3.35 87,102 6.81
Overall 18,459 24,515 12,815
Gini 0.241 0.179 0.415
CV 0.549 0.416 1.001
Atkinson (e ¼ 1) 0.092 0.057 0.266
Atkinson 0.049 0.030 0.144
(e ¼ 0.5)
Atkinson (e ¼ 2) 0.175 0.105 0.500
Theil 0.108 0.066 0.316
Progressivity Implications of Public Health Insurance Funding in Canada 157

inverse of the equivalence scale for the family type to which they belong
avoids such problems. However, our approach is justified by the fact that
our problem here involves implicit redistribution through health benefits
that are not subject to adjustments from the household equivalence scales.
We see in Table 6 that the inequality indices for equivalized incomes are
generally less than if one uses family incomes. However, the order of mag-
nitude in the reduction of the AMLD index resulting from inclusion of public
health care benefits, 0.138 to 0.125 (i.e., approximately a 10% reduction) is
similar to that found when using family incomes. The incomes in the third set
of columns [After-tax Income under Private Health] are the equivalized in-
come values generated from after-tax incomes if taxes were reduced in order
to reflect the absence of public health expenditures but individuals had to
purchase health insurance privately. Unlike the case when using family in-
comes rather than equivalent incomes, the most relevant comparison using
equivalized incomes is between the first and third set of columns (rather than
the second and third). This follows because in Tables 6–8, we have deducted
the health care costs that are presumed to result from private insurance when
there is no public insurance. Since these expenditures are not subject to
economies of scale, we did not want to apply the equivalizing procedure for
such funds. Therefore, rather than comparing the inequality indices in col-
umn sets 3 to 2 in order to infer how important public health care is in
equalizing income and promoting social welfare, one should make the com-
parison between column sets 3 and 1. This still leads to a substantial differ-
ence in the value of inequality with and without publicly financed health care.
For example, the cost of inequality implied by the AMLD index would rise
from 13.8% of per capita income to 21.1% in the absence of public health
insurance. One must be cautious, however, in making these comparisons and
extending them to the notion of social welfare because of the equivalizing
procedure.17

4. DISCUSSION

In this section, the implications of various aspects of our methodology and

assumptions are considered. Perhaps, the most critical assumptions we have
made have to do with our method of imputing the value of public health
insurance to individuals. Also, we realize that the issues of financing through
taxation rather than private insurance premiums are not the only important
equity issues concerning the delivery of health care services, be they funded
publicly or privately. Another very relevant issue is the relationship between
158 JAMES B. DAVIES AND MICHAEL HOY

income and the take-up rate of services available. Several studies have
investigated whether the extent to which individuals use (or are offered)
public health care services varies by income. We discuss this issue as well,
albeit rather briefly, in this section.
A common method of assessing equity in health care provision is to
determine relative usage of health care by individuals of different incomes.
For example, Dunlop, Coyte, and McIsaac (2000), using data from the same
survey as this study uses, found that individuals from lower income groups
were more likely to be frequent users of primary care physicians in Canada,
but that after adjusting for differences in health need, those with lower
incomes and fewer years of schooling were less likely to visit specialists. If
low income individuals make less intensive use of health care facilities and
programs, then at least from the perspective of usage the provision of health
care may be inequitable in that the incidence of benefits is regressive. How-
ever, for public health care provision to be regressive in an overall (financial)
sense, it must be that the benefits to lower income individuals are a lower
fraction of their income than for higher income individuals rather than
simply being of lower absolute amount. The opposite relationship between
income and health care usage is, of course, also possible. For example, Barer,
Manga, Shillington, and Siegel (1982) found that hospital use was highest for
the lowest income class and relatively stable for all other income classes, thus
indicating a progressive incidence of this aspect of publicly provided medical
care. In terms of overall incidence, a study by van Doorslaer et al. (2000)
investigating the equity in the delivery of health care concludes that for most
European countries, ‘‘y (even) y after controlling for the fact that also the
need for health care tends to be more concentrated at the bottom end of the
income distribution, little evidence of an inequitable overall health care dis-
tribution emerges’’. See also the paper in this volume by Gómez and Nicolás
(2007) that provides an analysis regarding utilitsation by income under pub-
lic and private health insurance scenarios for Spain and thus represents a
useful complement to our research.
Although the above-mentioned types of studies address important equity
issues concerning the incidence on the take-up rate of public health care for
people with different incomes, our focus on distributional implications is
concerned with the relative financial costs of public versus private health
insurance. These two different perspectives are complementary to under-
standing the overall equity effects of health care provision.
Another very useful approach is to analyse how different sources of fund-
ing (e.g., social insurance, employee and employer contributions, indirect
taxes, direct taxes, private insurance costs, etc.) for health care have different
Progressivity Implications of Public Health Insurance Funding in Canada 159

distributional implications. In a series of papers, including Wagstaff and van

Doorslaer (1997), Wagstaff et al. (1999), and van Doorslaer et al. (1999), it
has been demonstrated how different systems of financing health care are
more or less regressive. These studies show how different combinations of
public and private sources of funds for health care imply different degrees of
progressivity and regressivity in the system compared to a (hypothetical)
funding system that elicits payments in proportion to an individual’s income.
This basic approach also allows for a decomposition of inequality into
components of vertical progression, horizontal equity and re-ranking, based
on the seminal work of Aronson, Johnson, and Lambert (1994). This work is
also complementary to our own, which studies the distributional impact of
the benefits of public health care as well as funding effects. For Canada,
however, the particular source of tax funds used to finance the public system
is not at all clear, even from the perspective of identifying federal versus
provincial sources.18 We have presumed that tax cuts made possible by
eliminating the public health care system would occur through the income
tax and would be done by a proportionate decrease in all tax rates. This
represents a certain amount of tax flattening, although not a move entirely to
a flat rate tax system. We believe this is a plausible scenario given the trend
towards the flattening of income tax systems, as documented in Davies
and Hoy (2002). Moreover, the province of Alberta recently introduced
provincial tax reforms moving it to a flat rate tax system (see McMillan, 2000
for details).
In contrast, one could argue that general taxes, including provincial sales
taxes and the federal GST (goods and services tax), may be reduced. As
noted earlier, such taxes tend to be distributionally neutral and so any
reduction would have no impact on inequality. Thus, one could imagine a
possible range of inequality impacts from tax reduction that range from zero
to our estimates as worthy of consideration. This is one reason we computed
separately the effects from considering the benefits of the public health
system and the implied tax reductions when measuring the relative equity
implications of public versus private systems.
There is an indirect effect of replacing public with private insurance in
regard to the relative cost and coverage of health care between low and high
income individuals that deserves more discussion. If a menu of policies with
different coverage levels and costs were offered under private insurance, as
one would expect, then on average low income individuals would tend to
purchase lower coverage health insurance plans with associated lower
costs.19 If these lower coverage and lower cost policies provided a lower level
of health care services (and expenditures) than the existing public system,
160 JAMES B. DAVIES AND MICHAEL HOY

then the subjective value of the public system could, for low income indi-
viduals, be less than is implied by the expenditures we have used for them.20
Another way of dealing with this issue is to recognize that forcing all in-
dividuals into the same health care coverage, as is implicitly done with a
public system, creates an efficiency loss with some people holding more
health insurance than they would like while others hold less. Such efficiency
effects could be measured and compared to our equity effects. Of course,
there are other efficiency issues involving asymmetric information which
may make one or the other system more efficient. One such issue is that
adverse selection may well reduce the efficiency of a private system while the
enforced pooling of a public system can avoid this pitfall (see Hoy, 2006 for
a more elaborate discussion of this point).
Even if, in a fully privatized system, individuals did purchase the same
insurance coverage as exists in the public system, private insurance premiums
could differ from the public health care costs because in a private system
costs of delivery would be different. These costs could be lower, due to more
incentives arising from competition to reduce costs, or could be higher if
certain economies of scale are lost as a result of privatization and/or the
presence of advertising and other marketing costs are sufficiently high.
Another factor that may lead to higher costs resulting from privatization is
that the public health care system is effectively a monopsonist in terms of its
labour input and this advantage is lost in a private system.
Another aspect of the implicit pricing and valuation of insurance premiums
made in this paper is that private insurance premiums are assumed to depend
only on the age and gender of family members. Private insurance companies
selling individual polices, however, would tend to use finer information con-
cerning the health status of individuals in order to categorize them into risk
classes with appropriately differentiated premiums. Taking account of such
differences would imply more inequality under a system of private health
insurance. Adverse distributional impacts of categorical discrimination in
general terms are demonstrated by Hoy (1984), and Bossert and Fleurbaey
(2002), while for the particular use of information from genetic tests, see Hoy
and Lambert (2000) and Hoy and Ruse (2005). Data limitations concerning
individuals’ health status do not allow us to take such possible behaviour by
insurers into account but doing so could well lead to substantially higher
levels of inequality in the hypothetical private health care scenario. However,
this effect may be attenuated if health insurance is largely provided through
employment.
The introduction of such a sizable (49%) reduction in income taxes would
likely lead to incentive effects with respect to labour supply and other
Progressivity Implications of Public Health Insurance Funding in Canada 161

income earning decisions. Thus, one might observe changes in the before-
tax, and hence also the after-tax, distributions of income. These changes
could lead to either a more or less equally distributed before-tax and/or
after-tax income distribution. To take account of such incentive effects is
beyond the scope of this paper. However, the methodology that we have
used allows one to compare the equity effects of public financing of health
care to the efficiency effects of tax reductions,21 such as those found by the
study of Fullerton and Rogers (1993) measuring the efficiency effects of the
1986 tax reforms (TRA86) in the US.

5. CONCLUSION
In this paper, we apply a standard distributional impact analysis to measure
the implications for Canada of moving from its existing public system of
health care to a fully privatized system based on health insurance premiums
assessed according to age and gender. The implication of including the health
care benefits received through the public system, as quantified by implied
health insurance premiums that would have to be paid under a private sys-
tem, is an increase in equality that in our base case is equivalent to an increase
of 2.4% of per capita incomes. Reduced government expenditures from
moving to a private health care system would allow for a reduction in taxes
equivalent to 49% of income tax paid. If the government used these savings
to reduce income taxes and made the cuts proportional to individuals’ tax
payments, as we assumed in our calculations, then the overall reduction in the
degree of inequality in after-tax incomes under the public system in our base
case is equivalent to an increase of 4.8% of per capita incomes. Of course, tax
cuts could be implemented differently. If the tax cuts were made in a neutral
fashion with respect to the after-tax distribution of income, then the lower
value of a 2.4% gain in per capita incomes should be associated with the
public health care system. We suggest that tax cuts may well be made in a
manner that has a distributional impact in between these two extremes, and
so suggest that the range of 2.4–4.8% in equivalent per capita income is a
reasonable one to infer regarding the distributional impact of the public heath
care system. Since for our sample (Canada, 1994) health care expenditures
represent 12% of income, the equity benefit of using income taxes to finance
health care in Canada rather than rely on private insurance is substantial for
any number in this range.
There are many provisos and caveats to this conclusion. First and foremost,
like any normative study involving the distribution of income, one must choose
162 JAMES B. DAVIES AND MICHAEL HOY

a norm or ethical view on how important income inequality is to society. We

believe we have been quite conservative in our choice of ethical standard,
adopting the Atkinson measure with a moderate aversion to nequality, to
generate our base case alluded to above. Another important caveat is that we
have adopted an impact analysis which, although quite common for measuring
the distributional impact of public policies, has the disadvantage of assuming
that individuals’ behaviour would be the same under the two systems.
It is extremely unlikely that Canada would move to a purely private health
care system, at least in the near future. How relevant are our calculations in
light of this? Whether a specific private–public mixed system would generate a
large increase in inequity depends very much on its specific design features. A
mixed system with a sufficiently high level of health care provided free of
charge to all (or just to low income individuals) might well not be substan-
tially worse from a distributional perspective than the existing public system
and could even be better. In contrast, bringing some health care expenditures,
like prescription drugs, under the umbrella of publicly funded care may have
important equity benefits. One could use the same methodology here to assess
how this would affect inequality and social welfare.
An argument sometimes advanced in favour of moving towards private
funding of health care, besides wanting to lower the overall tax burden, is
that private health care can lead to efficiency through competition and
greater variety of choices in what is included in health care plans. Again, we
do not dispute such claims but rather emphasize that our analysis suggests
that publicly provided health care, which is funded out of income taxes, has
a strong progressive effect on equality and that this effect should not be
ignored. Proposals to replace the system, either partially or fully, with a
private alternative must demonstrate that there is a very strong efficiency
rationale for doing so, before a balanced assessment of efficiency and equity
considerations could sanction such a change.

NOTES
1. Other examples of impact studies include Duclos and Tabi (1998), who analyse
the implications of several social policies on income inequality; Countryman (1999),
who analyses the distributional implications of the Canadian unemployment system;
Davies and Hoy (2002), who compare the distributional effects of hypothetical flat
rate tax schemes to an existing graduated rate tax scheme.
2. For studies concerning these issues, see Barer et al. (1982), Dunlop et al. (2000),
and Van Doorslaer et al. (2000).
Progressivity Implications of Public Health Insurance Funding in Canada 163

3. The CAP (Canada Assistance Plan), which was a cost-sharing arrangement

between provincial and federal governments for supporting social assistance and
welfare programs, was also rolled into the CHST, thus delinking further sources of
funding with health expenditures (see Snoddon, 1998 for details).
4. See Duclos and Tabi (1998), and Davidson and Duclos (1997) for studies of this
effect. We do not perform statistical inference as is done in the latter of these studies.
5. Good examples of studies concerning the equity implications of publicly
subsidized post-secondary education are Mehmet (1978) and Meng and Sentance
(1982) and see Barer et al. (1982) for a study on publicly funded health care.
6. Of particular relevance to our study is that many studies show that higher
income individuals typically purchase higher levels of health care in a private market
(see, for example, Blomqvist & Carter, 1997); i.e., that health care is a normal good.
7. The National Accounts indicate that federal and provincial governments
received $98.5 billion of personal income tax revenue in 1994. The public health
expenditures of $52 billion are 53% of this total. We have also done our calculations
assuming that all income tax payments fall by 53%, instead of the 49% figure
implied by the SCF, and find that the results are very similar to those reported here.
8. See also Kolm (1969, pp. 186–187) who first proposed this and other indices as
measures of relative injustice per dollar of social income.
9. Note that the choice of utility function being iso-elastic is necessary if one
wants to satisfy the desirable property that the percentage cost of inequality, i.e., A in
Eq. (1), be invariant to equal proportionate changes in all incomes (see Lambert
(2001), pp. 98–99). For further discussion on this family of indices, see Jenkins
(1991).
10. We use the three parameter values e ¼ 2, e ¼ 1, and e ¼ 0.5 in our calcula-
tions. For this simple example, the cost of inequality is 6.7% for the case of e ¼ 0.5
(i.e., the same level of welfare can be achieved with average incomes of $33,588 if it
were equally distributed).
11. A number of studies have tried to infer policy makers’ e from studies of
government policies. As reported by Lambert (2001, Chapter 5), the values thus
obtained range from about 1.4 to 2.0. A wider range of values is obtained using
alternative methods. Stern (1977) surveys a variety of estimation methods and
reports values of e ranging from 0.4 to 10.0. Blake (1996) even argues on the basis of
studies of relative risk aversion that e could lie in the range 7.9–47.1. See also Palsson
(1996) for a discussion of interpersonal differences in degree of risk aversion.
12. For example, in 1994 the value (average expenditure) of public health care for a
single female in the age category 65+ was $7,382 while for a family of four, including
a male and female adult in the age category 45–64 ($1,546 and $1,591, respectively)
and two children in the age category 0–14 ($592 each), the value is $4321 for the
family. See Table 22B of National Health Expenditures in Canada: 1975–1994 (1996).
13. A simple and extreme example illustrates this point. If family A and B have
incomes of $30,000 and $60,000, respectively, but then each family receives an equal
absolute increase of $100,000, then the distribution becomes $130,000 and $160,000,
respectively, which reflects less relative inequality (i.e., the ratio of family incomes
goes from 2:1 to approximately 1.23:1).
14. There is one proviso here, however, and that is the implicit assumption often
made in such welfare comparisons that the utility functions of different individuals
164 JAMES B. DAVIES AND MICHAEL HOY

are the same. This is not necessarily an innocuous assumption in the context of heath
care provision and financing.
15. The particular OECD scale we adopted uses a weight of 1 for a single adult
family; for a family of size two add 0.7 to the weight if the second member is an adult
and 0.5 if the second member is a child; add 0.5 to the family weight for each
additional family member, whether that member is a child or an adult. So, for
example, a family of size 4 (2 adults, 2 children) with family income level $70,000
implies a per capita equivalized income of $70,000/2.7 ¼ $25,926 for each person in
the family. There are, however, many variations to this approach (e.g., see Figini,
1998, p. 5, footnote 3).
16. See Steckmest (1996), Harding (1995), and Smeeding et al. (1993) for a
discussion of this point.
17. Note in particular that the average income is not the same in these two
columns (i.e., column sets 2 and 3) when using equivalized incomes, as it was when
comparing family incomes. If our study involved only income transfers (between
family types) that were fully subject to the economies of scale implicitly recognized
by the family equivalence scales, we could avoid this difficulty by adopting Ebert’s
(1997) suggested procedure of weighting individuals accordingly, as noted above.
Also, see Decoster and Ooghe (2003) who apply these different approaches (weight-
ing with individuals, equivalent individuals, and not weighting at all) to Belgian data
and find that ‘‘using the number of equivalent individuals as weights y leads to
fanciful results with respect to the choice of equivalence scale’’ (ibid, p. 193).
18. With the exceptions of Quebec, and more recently Alberta and Ontario, the
basis for provincial taxation of income is the use of a given fraction of individual’s
federal income tax payable. This ties the two sources together and reduces the
concern for individuals as to whether tax cuts will come from one level of govern-
ment or the other or both.
19. See Blomqvist and Carter (1997) for an analysis of this question and also a
review and critique of previous empirical studies relating income to demand for
health care.
20. However, private choices may not reflect true subjective values in the presence of
borrowing constraints. Low income individuals could prefer higher levels of health care
than they would purchase in any given period but could be prevented from spending
more due to borrowing constraints. Thus, it is not clear by how much, if at all, our
calculations may overstate the benefits of public health care to lower income individ-
uals and hence imply greater redistributive effects of the system than actually exist.
21. For example, Fullerton and Rogers (1993) compute the gain from replacing
entirely the federal personal income tax with non-distortionary lump-sum taxes to be
2.02% of lifetime income or, when discounting welfare gains for all generations,
0.68% of lifetime income.

ACKNOWLEDGMENTS

We thank an anonymous referee and the editor, Peter Lambert, for very
helpful remarks. As usual, of course, any errors or limitations of this work
Progressivity Implications of Public Health Insurance Funding in Canada 165

are the responsibility of the authors. Both authors thank SSHRC for fund-
ing and a succession of first-rate research assistants; namely, Jeremy Lise,
Laura Pearson, and Warren Goodlet.

REFERENCES
Aba, S., Goodman, W., & Mintz, J. (2002). Funding public provision of private health: The case
for a copayment contribution through the tax system. C.D. Howe Institute Commentary
No. 163.
Aronson, J. R., Johnson, P., & Lambert, P. J. (1994). Redistributive effects and unequal tax
treatment. Economic Journal, 104, 262–270.
Atkinson, A. (1970). On the measurement of inequality. Journal of Economic Theory, 2,
244–263.
Barer, M. L., Manga, P., Shillington, E. R., & Siegel, G. C. (1982). Income class and hospital use
in Ontario. Ontario Economic Council, Occasional Paper No. 14.
Blackorby, C., Donaldson, D., & Auersperg, M. (1981). A new procedure for the measurement
of inequality within and among population subgroups. The Canadian Journal of
Economics, 14(4), 665–685.
Blake, D. (1996). Efficiency, risk aversion and portfolio insurance: An analysis of financial
asset portfolios held by investors in the United Kingdom. The Economic Journal, 106,
1175–1192.
Blomqvist, A. G. & Carter, R. A. L. (1997). Is health care really a luxury? Journal of Health
Economics, 16, 207–229.
Bossert, W., & Fleurbaey, M. (2002). Equitable insurance premium schemes. Social Choice and
Welfare, 19, 113–125.
Butler, R. J., Johnson, W. G., & Rimsza, M. E. (2006). Medicaid disenrollment costs: Maricopa
County, AZ. Pediatrics, forthcoming.
Chiu, H. (2007). Intersecting Lorenz curves, the degree of downside inequality aversion, and tax
reforms. Social Choice and Welfare, forthcoming.
Countryman, G. J. (1999). The effect of unemployment benefits on income inequality in the
Canadian provinces. Canadian Public Policy, 25, 539–556.
Davidson, R., & Duclos, J.-Y. (1997). Statistical inference for the measurement of the incidence
of taxes and transfers. Econometrica, 65, 1453–1465.
Davies, J. B., & Hoy, M. (2002). Flat rate taxes and inequality measurement. Journal of Public
Economics, 84, 33–46.
Decoster, A. & Ooghe, E. (2003). Weighting with individuals, equivalent individuals or
not weighting at all: Does it matter empirically? Research on Economic Inequality, 9,
173–190.
Duclos, J.-Y., & Tabi, M. (1998). Inégalité et redistribtuion du revenu, avec une application au
Canada. Unpublished mimeo.
Dunlop, S., Coyte, P. C., & McIsaac, W. (2000). Socio-economic status and the utilisation of
physicians’ services: Results from the Canadian National Population Health Survey.
Social Science & Medicine, 51, 123–133.
Ebert, U. (1997). Social welfare when needs differ: An axiomatic approach. Economica, 64,
233–244.
166 JAMES B. DAVIES AND MICHAEL HOY

Ebert, U. (1999). Using equivalent income of equivalent adults to rank income distributions.
Social Choice and Welfare, 16, 233–258.
Figini, P. (1998). Inequality measures, equivalence scales and adjustment for household size and
composition. LIS Working Paper No. 185.
Fullerton, D., & Rogers, D. L. (1993). Who bears the lifetime tax burden? Washington DC:
Brookings Institution.
Gillespie, W. I. (1980). The redistribution of income in Canada, The Carleton Library. Ottawa,
Canada: Gage Publishing Ltd.
Gómez, P. G., & Nicolás, A. L. (2007). Public and private health insurance and the utilitisation
of health care in Spain. In: P. J. Lambert (Ed.), Research on economic inequality (Vol. 15,
pp. 169–195). Oxford, UK: JAI.
Harding, A. (1995). Health, education and housing outlays: Their impact on income distributions.
STINMOD Discussion Paper No. 7.
Health Canada (1996). National health expenditures in Canada 1975–1994. Ottawa, Canada:
Minister of Supply and Services Canada.
Hoy, M. (1984). The impact of imperfectly categorizing risk on income inequality and social
welfare. Canadian Journal of Economics, 39, 177–206.
Hoy, M. (2006). Risk classification and social welfare. The Geneva Papers on Risk and
Insurance: Issues and Practice, 31, 245–269.
Hoy, M., & Lambert, P. J. (2000). Genetic screening and price discrimination in insurance
markets. The Geneva Papers on Risk and Insurance Theory, 25, 103–130.
Hoy, M., & Ruse, M. (2005). Regulating genetic information in insurance markets. Risk
Management and Insurance Review, 8(2), 211–237.
Jenkins, S. (1991). The measurement of income inequality. In: L. Osberg (Ed.), Economic
inequality and poverty: International perspectives. Armonk, New York and London:
M. E. Sharpe.
Kolm, S.-C. (1969). The optimal production of social justice. In: J. Margolis & H. Guitton
(Eds), Public economics. London: Macmillan.
Lambert, P. (2001). The distribution and redistribution of income: A mathematical analysis
(3rd ed.). Manchester and New York: Manchester University Press.
Loomis, J. B., & Revier, C. F. (1988). Measuring regressivity of excise taxes: A buyers index.
Public Finance Quarterly, 16, 301–314.
McMillan, M. L. (2000). Alberta’s single-rate tax: Some implications and alternatives. Canadian
Tax Journal, 48(4), 1019–1052.
Mehmet, O. (1978). Who benefits from the Ontario University System. Toronto: Ontario
Economic Council.
Meng, R. & Sentance, J. (1982). Canadian universities, who benefits and who pays? The
Canadian Journal of Higher Education, 12, 47–58.
Palsson, A.-M. (1996). Does the degree of relative risk aversion vary with household charac-
teristics? Journal of Economic Psychology, 17, 771–787.
Pechman, J. A. (1985). Who paid the taxes, 1966–1985. Washington: Brookings Institution.
Pfahler, W. (1984). Linear income tax cuts: Distributional effects, social preferences and rev-
enue elasticities. Journal of Public Economics, 24, 381–388.
Shorrocks, A. F. (1984). Inequality decomposition by population subgroups. Econometrica, 52,
1369–1386.
Smeeding, T., Saunders, P., Coder, J., Jenkins, S., Fritzell, J., Aldi, J., Hagenaars, M., Hauser,
R., & Wolfson, M. (1993). Poverty, inequality, and family living standards. Impact
Progressivity Implications of Public Health Insurance Funding in Canada 167

across seven nations. The effect of noncash subsidies for health, education and housing.
Review of Income and Wealth, 39(3), 229–258.
Snoddon, T. R. (1998). The impact of the CHST on interprovincial redistribution in Canada.
Canadian Public Policy, XXIV(1), 49–70.
Steckmest, E. (1996). Noncash benefits and income distribution. SNF-prosjekt nr. 2395,
Sosialokonomisk Institutt, Universitetet I, Oslo.
Stern, N. (1977). The marginal valuation of income. In: M. J. Artis & A. R. Nobay (Eds),
Essays in economic analysis. Cambridge: Cambridge University Press.
Van Doorslaer, E., Wagstaff, A., Van der Burg, H., Christiansen, T., & Citoni, G. (1999). The
redistributive effect of health care finance in twelve OECD countries. Journal of Health
Economics, 18, 291–313.
Van Doorslaer, E., Wagstaff, A., Van der Burg, H., Christiansen, T., & De Graeve, D. (2000).
Equity in the delivery of health care in Europe and the US. Journal of Health Economics,
19, 553–583.
Wagstaff, A., & Van Doorslaer, E. (1997). Progressivity, horizontal equity and reranking in
health care finance: A decomposition analysis for the Netherlands. Journal of Health
Economics, 16, 499–516.
Wagstaff, A., Van Doorslaer, E., Van der Burg, H., Calonge, S., & Christiansen, T. (1999).
Equity in the finance of health care: Some further international comparisons. Journal of
Health Economics, 18, 263–290.
PUBLIC AND PRIVATE HEALTH
INSURANCE AND THE
UTILISATION OF HEALTH CARE
IN SPAIN

Pilar Garcı́a Gómez and Angel López Nicolás

ABSTRACT

This paper reports an analysis of the evolution of equity in the utilisation

of health care in Spain over the period 1987–2001, a time span covering
the development of the modern Spanish National Health System. Our
measures of utilisation are the probabilities of visiting a doctor, using
emergency services and being hospitalised. For these three measures, we
obtain indices of horizontal inequity from microeconometric models of
utilisation that exploit the individual information in the Spanish National
Health Surveys of 1987 and 2001. We find that by 2001, the system had
improved insofar as differences in income no longer lead to differences in
utilisation given the same level of need. However, tenure of private health
insurance leads to differences in utilisation given the same level of need,
and its contribution to inequity has increased over time, both because
insurance is more concentrated among the rich and because the elasticity
of utilisation for the three services has also increased.

Equity
Research on Economic Inequality, Volume 15, 169–195
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15008-0
169
170 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

1. INTRODUCTION

Spanish society has undergone a major overhaul in the three decades that
have elapsed since the death of Franco. The transformation from dictator-
ship to democracy and the devolution of government to the regions have
combined with the sheer passage of time to transform an obsolete public
sector into one comparable to that of other developed countries. The health
care system is one of the areas where reforms have been far reaching; and in
this paper, we aim to evaluate the change between 1987 and 2001 in one of
the indicators that serve to assess its performance: the existence and degree
of inequities in health care utilisation. The choice of these two dates is
motivated by the fact that the most comprehensive package of reforms
for the health care system was systematised and put forward by the 1986
General Health Act. This suggests 1987 as an obvious baseline period.
Fourteen years later, the main features of the new health care system – tax-
funded universal insurance and a modernised primary care network – had
been fully developed across Spain, so 2001 is an appropriate period to
compare with the baseline. One of the goals stated in the General Health
Act is the elimination of socio-economic health inequalities in the utilisation
of health care (Article 3). Although this objective is stated in terms of in-
equalities, our outcome of interest is equity in utilisation, defined as equal
access given equal need for health care. In line with the recent literature
(Van Doorslaer, Koolman, & Jones, 2004; Van Doorslaer, Koolman, &
Puffer, 2002), our methodology consists in calculating indices of socio-
economic inequality in health care utilisation and subsequently decom-
posing these indices into the part that can be attributed to socio-economic
differences in need and the part that can be attributed to other socio-
economic differences. According to the criterion expressed above, the latter
component of inequality is interpreted as an index of inequity.
Our empirical application uses data from the 1987 and 2001 editions of
the National Health Survey or Encuesta Nacional de Salud (CIS, 1987,
2001). The comparison of two cross sections of the Spanish population has a
limited ability to reflect the causal effect of a multi-faceted package of re-
forms. Nevertheless, our contention is that the implementation of these
reforms should change the joint distribution of utilisation and socio-
economic characteristics after controlling for health care needs, and in fact
our results show that by 2001, the system had improved insofar as differ-
ences in income no longer lead to differences in utilisation given the same
level of need. However, tenure of private health insurance (PHI) leads to
differences in utilisation given the same level of need, and its contribution
Public and Private Health Insurance and the Utilisation of Health Care in Spain 171

to inequity has increased over time, both because insurance is more con-
centrated among the rich and because the responsiveness of utilisation to
private insurance has also increased.
Section 2 presents the main characteristics of the health system and the
reforms that have taken place in the recent past and provides a brief review
of previous relevant studies. Section 3 presents the methodology that we
adopt for the measurement of inequities in health care utilisation and the
explanation of their changes over time. Section 4 presents the empirical
results and Section 5 discusses the implications of our results.

2. THE TRANSITION OF THE SPANISH HEALTH

CARE SYSTEM AND PREVIOUS LITERATURE ON
INEQUITIES IN UTILISATION

2.1. Institutional Changes and the Role of Private Insurance

At the end of the dictatorship in 1975, the Spanish health system was based
on a social security scheme paid by employers and employees and comple-
mented by a network of health care centres owned by different organisa-
tions. One of the characterising features of the pre-democratic system was a
strong bias towards hospital care. While the 1970s had witnessed the cre-
ation of a public network of modern hospitals, primary and preventive
services in the public network were underdeveloped: general practitioners
were typically available for two and a half hours per day at isolated outlets
which lacked administrative and diagnostic support (EOHCS, 2000). The
arrival of democracy unleashed the latent demand for a better health
care system and important legislative and managerial changes ensued. The
Ministry of Health was created in 1977 and the 1978 Constitution con-
secrated public coverage for all citizens. Momentum gathered after 1983
when the government initiated a series of reforms to integrate the different
networks. In 1986, the General Health Act transformed the social security
system into a National Health System.
Thus, two main structural reforms with a potential impact on socio-eco-
nomic inequalities in utilisation of health care occurred during the period
studied in this paper. Firstly, the system was finally consolidated as a tax-
funded universal coverage National Health System within which individuals
are entitled to a comprehensive set of benefits including not only primary
and specialised inpatient and outpatient care, but also subsidised medicines
with zero co-payments for specific groups such as pensioners or disabled
172 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

persons and reduced co-payments for drugs for chronic diseases including
AIDS. Secondly, primary care has been totally reformed by replacing the
obsolete outlets mentioned above with team-based practices staffed by doc-
tors and nurses who have received specific training in family medicine and
whose activities not only include curative care, but also preventive care,
health promotion, follow-up of patients and services targeted to particular
population groups such as the mentally ill and drug users. Although by 1987
the whole of the population was covered by the National Health System, the
material implementation of the primary care reform (i.e. the replacement of
the obsolete outlets with modern primary care centres) all over Spain was
slow: while it was planned as far back as 1984 and turned into law in 1986,
only 50% of the population was covered by the modern centres in 1992 and
the proportion reached 81% by 2000 (EOHCS, 2000). This is in fact the
most important reform to have taken place during the period under study.
For these reasons, it seems appropriate to evaluate the change between 1987
and 2001.
In this study, we intend to pay special attention to the role of PHI as a
determinant of inequities in health care. The concern about the equity
effects of PHI is partly justified by the fact that expenditure on PHI has
received public subsidies in the form of tax bonuses. Prior to 1999, the
subsidy operated via personal income tax: individuals received a 15% rebate
on insurance premiums (and on any other expenditure on health care).
Currently, it operates via corporate tax: premiums are considered tax-free
in-kind salary and companies can deduct from profits the cost of collec-
tive policies (thus obtaining a 35% tax bonus on their cost). The current
fiscal treatment of PHI is explained by the recent history of the health
system, since prior to the reforms discussed above some groups, notably
the self-employed, were excluded from public insurance. In this context,
the subsidies fulfilled a similar role to the tax deductions for the purchase
of (principal) private insurance in the US, since they facilitated access to
the only form of health insurance available to a group of the population.
This would also explain why, unlike in systems such as the Canadian, PHI
carriers are allowed to offer all – or any subset – of the comprehensive
range of services covered by public insurance. Coupled with the existence
of universal public insurance, this confers on PHI the nature of a duplicate
in the Spanish system (it is noteworthy that insurance for services not
covered by the public scheme – i.e. services for which PHI has a supple-
mentary role1 such as dental care – are marketed separately from other
policies). A similar situation is found in, among other countries, the UK,
Portugal, Greece and Italy.
Public and Private Health Insurance and the Utilisation of Health Care in Spain 173

The services covered by PHI are mostly provided by Preferred Provider

Organisation (PPO) type networks. The professionals within these networks
receive a discounted fee for service from the insurer, and are allowed to run
consultancies in the public network. To a lesser extent, there is also some
vertical integration between insurers and providers in the form of Health
Maintenance Organisations whose staff are paid on a full-time salaried basis.
There are some aspects of the Spanish system for which the services of the
typical duplicate PHI policy differ from the equivalent services in the public
scheme. Firstly, whereas the public scheme requires patients to visit a GP
before being referred to a specialist, PHI policies allow patients to resort to a
specialist in the private network without a GP referral. Secondly, the choice
of provider, particularly for outpatient services, is wider under private cov-
erage. As for inpatient services, hospital amenities tend to be superior under
private coverage (e.g. individual hospital rooms). These two differences
constitute the marketing strong points for PHI policies. In contrast, these
features are somehow offset by the fact that, thirdly, outpatient medical
treatments prescribed by a doctor in the public network are heavily sub-
sidised (the co-payments vary from 40 to 0%) whereas prescriptions by
doctors visited under private coverage are not. And fourthly, private insur-
ance policies cover some but not all hospital expenses. According to OCU
(1997), many companies limit to 30 days the number of hospital days that
will be paid for within a given year, with stricter limits to the number of days
in intensive care. The OCU report also reveals that the public sector covers a
much more comprehensive list of treatments than any of the policies offered
by the private sector. The subsidies for PHI might induce undesired effects
in terms of (in)equity, because PHI alters the patterns of utilisation, as
shown by Rodrı́guez and Stoyanova (2004). Moreover, for the particular
case of specialist visits, Jones, Koolman, and Van Doorslaer (2007) and Van
Doorslaer et al. (2002) have obtained evidence that supports the notion that
PHI in Spain actually generates pro-rich inequity in utilisation.
After presenting these institutional features, it is important to consider the
way in which socio-economic factors can lead to unequal utilisation between
two individuals with the same need for health care, and how the changes in
the Spanish system might have altered the underlying mechanism. In 1987,
public health care offered a relatively poor alternative to private health care,
so individuals naturally sought private care when need arose. Private care
was paid out of pocket or was covered by a private insurance policy, so,
prior to 1987, we can expect rich and/or privately insured individuals to be
more likely to use health care than poor individuals who are otherwise equal
in terms of need. The gradual improvement in the public network would
174 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

lead us to expect such socio-economic differences to disappear by 2001.

However, a key feature of private care is the possibility of visiting a spe-
cialist without needing a GP referral. In the reformed Spanish system of
2001, this would lead us to expect a negative effect on the probability of
using a GP and a positive effect on the probability of using a specialist for
both income and PHI, because for a given level of need, poorer/not privately
insured individuals would be more likely to use the services of a GP than the
services of a specialist.

2.2. Previous Studies for Spain

Apart from the studies cited above, there is a growing body of literature on
the evaluation of the reforms in the Spanish National Health System since
the Health Act of 1986 in terms of inequities in utilisation. The pioneering
work of Rodrı́guez, Calonge, and Reñé (1993) offered evidence, with data
from 1987, on the degree of inequity in public health care consumption as
measured by the expenditure devoted to doctor visits and hospitalisations in
the public network. A similar method was followed by Abásolo Alessón
(1998) with data for 1993. More recently, Urbanos (1999, 2001a, 2001b) has
considered the dynamics of inequity and analysed data for 1987, 1993, 1995
and 1997 within a unified methodological framework. Urbanos considers
actual consumption data (number of visits and inpatient days) as well as an
expenditure aggregate, and her results suggest a decrease in inequity during
the period 1993–1995. Moreover, for 1997, she finds that the inequity indices
for visits to GPs and specialists and inpatient days are not statistically sig-
nificant. In contrast, she finds that there is a significant degree of pro-rich
inequity in emergency visits and that individuals without PHI are likely
to over-utilise public health care outlets, especially GP visits. Abásolo,
Manning, and Jones (2001) test the existence of equity in the utilisation
of public-sector GPs in Spain in 1993 by analysing whether different socio-
economic characteristics influence the probability that the individual visited
a public-sector GP during a two-week period. They find that individuals
with PHI were less likely to visit a GP. Van Doorslaer et al. (2002) find
a significant degree of pro-rich inequity in specialist visits and pro-poor
inequality in GP visits using data from the Spanish sample of the 1996 wave
of the European Community Household Panel (ECHP). Van Doorslaer et
al. (2004) and Jones et al. (2007) again find that there is a significant degree
of pro-poor inequity in both the probability of visiting and the conditional
number of visits to a GP, whereas there is pro-rich inequity in both the
Public and Private Health Insurance and the Utilisation of Health Care in Spain 175

probability of contacting a specialist and the conditional number of visits.

Masseria, Koolman, and Van Doorslaer (2004) obtain point estimates that
would suggest evidence of pro-rich inequity in hospital admissions using
data from the ECHP, but the null hypothesis of no statistical significance
cannot be rejected from these estimates.
This paper contributes to the existing literature on a series of fronts. First,
unlike Rodrı́guez et al. (1993), Urbanos (1999, 2001a, 2001b) and Abásolo
et al. (2001), we do not restrict the analysis to publicly provided health care.
The reason is that, as discussed above, privately provided health care and
PHI have received public subsidies during the period considered. Secondly,
most of the existing studies do not address the equity effects of PHI, and this
paper offers some methodological advantages with respect to those that do
so, such as Van Doorslaer et al. (2002), which will be discussed later on. A
third contribution is that we use two comparable health surveys with rich
information on health status spanning 14 years since the General Health
Act. Despite the obvious limitations of all before–after evaluations, this is a
plausible empirical strategy to approximate the effects of the evolution of
the system on equity.

3. METHODS

3.1. Measuring and Decomposing Inequalities in Health Care Utilisation

The operational concept of inequity used in the recent literature is

socio-economic inequality in utilisation not justified by socio-economic
inequalities in need. Therefore, it is necessary to compute measures of socio-
economic inequality in utilisation, decompose these measures and sub-
sequently decide which components might be justified by the unequal needs.
The literature on health inequalities has recently adopted a standard tool for
the measurement of socio-economic inequalities in health or health care
utilisation: the concentration index (CI) (Wagstaff, Van Doorslaer, & Paci,
1989). The CI has a similar interpretation to the more familiar Gini index
for pure inequality. In fact, the two inequality measures differ in that the
ranking variable is a measure of socio-economic status (usually income) (CI)
rather than health/utilisation (Gini). The CI ranges between –1 and 1. A
value of –1 would mean that all health/health care utilisation is concentrated
in the poorest person, whereas a value of 1 would result if all health/
utilisation were concentrated in the richest person. If health/utilisation is
equally distributed over income in the sense that the pth percentage of the
176 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

population ranked by income has exactly the pth percentage of total health/
utilisation for any p then the CI is equal to zero.2
Suppose we are interested in calculating the CI for a measure of health
care utilisation on income using individual data from the population of
interest. Let yi denote a measure of utilisation for the ith individual,
i ¼ 1, 2, y, N, and R0i denote the cumulative proportion of the population
ranked by income up to the ith individual (their ‘‘relative income rank’’).
The CI of utilisation on income can be written in terms of the covariance
between the measure of utilisation and the relative income rank (see e.g. Van
Doorslaer & Jones, 2003):


2
CI ¼ covðyi ; R0i Þ (1)
ȳ
where ȳ ¼ Eðyi Þ:
We consider three types of health care utilisation: visits to doctors, use of
emergency services and hospitalisations. For each of these services, our
measure of utilisation consists in the probability of utilisation at least once
within a given time period. In the case of visits to doctors the time period is
15 days, whereas for the other two services the time period is one year. For
2001, we will also consider separately the probabilities of having visited a GP
or a specialist, since the survey provides information on the speciality of the
doctor in the last visit, so we will analyse five types of health care services.
While the health surveys offer information on the number of events for each
of the three services, we abstain from considering measures of equity in the
number of events. This is motivated by the fact that the distributions for the
numbers of events are concentrated on 0 and 1. For instance, less than 5%
(6% for 2001) of individuals report more than one visit to the doctor and less
than 2% (1% for 2001) report more than two. The case of hospitalisations is
even more extreme in this respect, as only for 2001 do we find individuals
reporting more than one event, and these individuals make up less than 2%
of the sample. Furthermore, the studies that have considered both the
probability of contact and the conditional number of events have found that,
where there are inequities, these operate in the same direction for both
dimensions of utilisation (Van Doorslaer et al., 2004).
For each of the three types of health care, we specify a linear probability
model (LPM) in the following way:
yji ¼ 1 ðindividual i reports 1 or more episodes of health care jÞ
X j
¼ aj þ bk xki þ
ji ð2Þ
k
Public and Private Health Insurance and the Utilisation of Health Care in Spain 177

where 1(
) is the indicator function, j ¼ 1,2,3 (in the case of 1987) and
j ¼ 1,2,3,4,5 (in the case of 2001) refer to doctor visits, hospitalisations and
emergency visits (in the case of 1987) and doctor visits, hospitalisations,
emergency visits, GP visits and specialist visits (in the case of 2001) and i is
an individual subscript.
It follows that the probability of utilisation of service j by individual i,
P(yji ¼ 1), can be written as
X j
Pðyji ¼ 1Þ ¼ aj þ bk xki (3)
k

Our choice for the LPM is justified on the grounds that the linearity in
parameters is particularly useful for our purposes of decomposing inequal-
ities in the probability of utilisation (this property has been exploited by
Masseria et al. (2004) in their study of inequity in the utilisation of inpatient
services). Moreover, the well-known limitation of the LPM model produc-
ing predicted probabilities outside the [0,1] interval is irrelevant in our case,
since we are only interested in the parameters of the model, which are
consistently estimated by OLS. Further, we have verified that a comparison
of the marginal effects of the LPM with those implied by a probit model
does not reveal substantive differences.
As shown by Wagstaff, Van Doorslaer, and Watanabe (2003), if the
probability of utilisation is described by Eq. (3), then an inequality index for
the probability of utilisation is given by
X
j x̄k  X j
j
CI ¼ bk j CI0k ¼ Zk CI0k (4)
k P̄ k k

The term in brackets is the elasticity of P with respect to xk evaluated at the

population means and CI0 k denotes the concentration index of xk against
income. Moreover, if we denote the elasticity of the probability of utilisation
with respect to the explanatory variable xk as

bjk x̄k
Zjk  (5)
P̄j
then we can rewrite the decomposition in such a way that the CI is simply a
weighted sum of the inequality in each of its determinants, with the weights
equal to the elasticities, as expressed in the last part of Eq. (4). As mentioned
by Van Doorslaer and Koolman (2004), the decomposition in Eq. (4) clar-
ifies how each correlate xk of utilisation contributes to total income-related
utilisation inequality: this contribution is the product of two factors: (i) its
178 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

impact on utilisation (as measured by the elasticity) and (ii) how unequally it
is distributed over income (as measured by the CI).
Measures of horizontal inequity are easily obtained from the decompo-
sition of income-related inequality in the utilisation presented in Eq. (4)
(Van Doorslaer et al., 2004; Gravelle, 2003). All that is required is an
agreement on what variables in the model of utilisation can be considered as
legitimate determinants of unequal utilisation from a normative point of
view – we shall call these variables need variables – and what variables do
not satisfy this condition, i.e. non-need variables. The first group typically
includes demographics, marital status and health indicators. The non-need
variables will comprise variables that do not reflect need for health care once
the need variables are controlled for. This group typically includes markers
of socio-economic status. In our case, we use a measure of income and an
indicator of tenure of PHI.
Let w denote the (1  L) vector of need variables and z denote the (1  P)
vector of non-need variables. The vector of explanatory variables in the
models for utilisation is simply the concatenation of w and z, i.e.
x ¼ (w,z) ¼ (w1, w2, y, wL, z1, z2, y, zP) ¼ (x1, x2, y, xL+1, xL+2, y,
xK) and its dimension is (1  K), with K ¼ L+P. An index of horizontal
inequity in utilisation for service j, HIj, is given by the part of the overall CI
for service j that can be attributed to the non-need variables that determine
the utilisation of that service. From Eq. (4), HIj is defined as

X
K X
L
HI j ¼ Zjk CI0k ¼ CIj  Zjk CI0k (6)
k¼Lþ1 k¼1

This method differs in an important way from the method of ‘‘indirect

standardisation’’ by Wagstaff and Van Doorslaer (1996). The method of
indirect standardisation consists in first computing the CI of actual utili-
sation and then deducting from it the CI of predicted utilisation, where
predicted utilisation is obtained from the estimation of an econometric
model for utilisation as a function of need variables. This procedure has
been criticised on the grounds that the omission of variables which, despite
not qualifying as need indicators from a normative point of view, are nev-
ertheless associated with utilisation, may lead to biased estimation
(Schokkaert & Van de Voorde, 2004; Gravelle, 2003). This is particularly
relevant for the purposes of this study. Since we wish to evaluate the impact
of PHI on utilisation, and since PHI tenure is strongly associated with
income and other socio-economic characteristics, omission of income – a
non-need variable – from the utilisation equation may lead to biased
Public and Private Health Insurance and the Utilisation of Health Care in Spain 179

estimates for the impact of PHI. The existing studies for the case of Spain
mostly rely on the indirect standardisation method. Indeed, only Van
Doorslaer et al. (2004) and Van Doorslaer, Masseria, and Koolman for the
OECD Health Equity Group (2006) use the method discussed above, but
their analysis does not consider the effect of PHI.
In relation to the point discussed in the previous paragraph, we must note
that the literature on utilisation generally treats PHI as an endogenous
variable (see Vera-Hernández, 1999, for the case of Spain). This is motivated
by the recognition that unobserved factors that affect the purchase of PHI
are correlated with unobserved factors that affect utilisation. Our steps to
address this issue consist in enriching the specification for utilisation with a
wide set of health status indicators in an attempt to capture all relevant risk
factors. This should purge the estimate for the effect of PHI of biases arising
from the omission of the utilisation equations of health factors that simul-
taneously drive the propensity to purchase PHI. We subsequently test this
assumption.

3.2. Decomposing Inequity Over Time

The previous section shows how horizontal inequity in utilisation can be

expressed as the contribution of non-need variables to an index of socio-
economic inequality in utilisation. It is then straightforward to use the ap-
proach proposed by Wagstaff et al. (2003) in order to decompose the
difference in inequity between two periods. The method is a derivation of
the well-known Oaxaca decomposition whereby the difference between the
CIs of the population at period t and period t1 can be written as
X
K X
K
DHIj ¼ HIjt  HIjt1 ¼ Zkt ðCI0kt  CI0kt1 Þ þ CIkt1 ðZkt  Zkt1 Þ
k¼Lþ1 k¼Lþ1
(7)
Then, the contribution of any particular variable, say income, within the
vector z to the difference in inequity is given by
DHIjincome ¼ Zincome;t ðCI0income;t  CI0income;t1 Þ
þCI0income;t1 ðZincome;t  Zincome;t1 Þ ð8Þ
In practice, we shall compute the difference in inequity (and contri-
butions towards this difference) between 2001 and 1987. Moreover, in
order to assess the relative importance of the inequality versus the health
180 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

elasticity component in the contribution of each non-need variable

in z, we also compute the relative excess elasticity compared with year
1987, i.e. (Zk2001Z  k1987)/|Z
 k1987|, and the relative excess inequality,
ðCI0k2001  CI0k1987 Þ=CI0k1987 ; for k ¼ L+1, y, K.

3.3. Statistical Inference

Many of the statistics that we are going to report are non-linear functions of
the data whose sampling distributions are hard to obtain. For this reason,
we have used bootstrapping methods in order to derive standard errors. The
bootstrap estimates for standard errors are computed following the five-step
approach used by Van Doorslaer and Koolman (2004). The number of
replications has been set to 500.

3.4. Data and Variable Definitions

We use the 2001 and the 1987 editions of the Encuesta Nacional de Salud or
ENS (CIS, 1987, 2001). These are nationwide surveys that collect informa-
tion on health and socio-economic characteristics of individuals. The sur-
veys contain separate samples for adults (16+) and children. The analysis
in this paper is based on the adult samples. The sampling scheme is a
multi-stage stratified process whereby primary strata are autonomous com-
munities (2001 edition) or provinces (1987 edition). Within primary strata,
substrata are defined according to residence area population size. Within
substrata, municipalities (primary sampling units) and sections (secondary
sampling units) are selected according to a proportional random sampling
scheme. Finally, individuals are randomly selected from the sections. The
survey documentation includes weighting factors that correct for the fact
that the number of observations within the primary strata is not propor-
tional to actual population. We use these weights whenever a nationwide
statistic is computed. The information contained in the data files does not
allow the identification of all the primary sampling units (because muni-
cipalities with a population below 100,000 are not identified). Similarly,
information about the secondary sampling units is omitted, so it is impos-
sible to control for cluster effects at either the municipality level or the
section level.
The ranking variable is equivalised total monthly income earned by the
household (hereafter, income). In the ENS, this is measured as a categorical
Public and Private Health Insurance and the Utilisation of Health Care in Spain 181

variable with 12 response categories in 1987 and 6 response categories in

2001. In order to obtain a continuous measure for income and also over-
come the fact that for both editions there is a substantial proportion of item
non-response, we specify an interval regression model using a wide range of
explanatory variables referring to both the respondent and the head of
household. These variables are the relationship between interviewee and
head of household, education of head of household, occupation of head of
household, employment status of head of household, tenure of PHI, age and
gender of the head of household and regional dummies. Except for the
upper quantiles, the distributions for the predictions of income compare well
with data from the Continuous Household Expenditure Survey (Encuesta
Continua de Presupuestos Familiares or ECPF) of 1987 and data from the
Spanish sample of the 2001 wave of the ECHP. The evolution of income
inequality as measured by the Gini index also compares well with external
sources.3
The initial 1987 ENS sample included 29,647 individuals. From the initial
sample, five observations were dropped as income could not be predicted,
and after deletion of those not responding to one of the relevant questions
the final sample contains 29,185 observations in the visits to doctor esti-
mation, 28,849 in hospitalisation and 29,122 in use of emergency services.
In turn, the initial 2001 ENS sample included 21,067 individuals from all
the autonomous communities, although the observations from Ceuta and
Melilla were dropped as there were no individuals from these two regions
in the 1987 sample. From the remaining 20,748, after deletion of those
not responding to one of the relevant questions the final sample con-
tains 20,644 in the visits to doctor estimation, 20,635 in hospitalisa-
tion, 20,636 in emergency visits, 20,644 in GP visits and 20,644 in specialist
visits.

4. EMPIRICAL RESULTS

As discussed in Section 3.1, we specify and estimate LPM for the probability
of visiting a doctor during the last fortnight, hospitalisation over the last 12
months and emergency services utilisation over the last 12 months. The
explanatory variables in the models are: (i) the logarithm of equivalent
household income; (ii) 14 age–gender categories corresponding to age
groups 16–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59,
60–64, 65–69, 70–74, 75–79 and 80+ for men and women (the omitted
category corresponds to women aged between 16 and 19); (iii) three marital
182 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

status categories: never married or divorced, married, and widowed (never

married or divorced is the omitted category); (iv) five categories of self-
assessed health: very good (omitted category), good, fair, bad, and very bad;
(v) six chronic illnesses: cholesterol, high blood pressure, diabetes, bronchitis
or asthma, heart diseases and allergy; (vi) whether daily activities or leisure
had been limited by any of the chronic diseases in the last 12 months;
(vii) whether daily activities or leisure had been limited because of pain in
the last two weeks; (viii) whether the individual had had to stay in bed for
more than half a day in the last two weeks; (ix) whether the individual had
had an accident in the last year and (x) tenure of PHI. Table A1 shows
summary statistics of the variables included in the model.
Table A2 contains the parameter estimates for the equations correspond-
ing to each of the services by OLS. Table A3 reports the results for the tests,
based on Hausman (1978) and outlined in Wooldridge (2003), of the as-
sumption of exogeneity of PHI in the specified utilisation equations. These
tests require the use of excluded instruments for PHI and, in line with
previous studies (Vera-Hernández, 1999; Jones et al., 2007), we use occu-
pational characteristics of the head of the household as variables that affect
the purchase of PHI but do not directly affect the respondent’s utilisation of
health care. The first line in each of the 1987 and 2001 panels in Table A3
reports the test for the null hypothesis of no significance of these instruments
in the reduced form equation for PHI. The remaining statistics are the
t-values for the tests of significance of the reduced form equation for PHI
residual in each of the utilisation equations (i.e. tests for the null of ex-
ogeneity of PHI in these equations). All these statistics are robust to the
in-built heteroscedasticity of the LPM. As the figures in the table reveal, the
P-values for the null hypothesis are above 5% level except in the case of
emergency visits in 2001. This suggests that the inferences for emergency
visits should be treated with caution.
The estimates in Table A2 conform to a priori expectations. Men are in
general less likely to use health care than women of the same age, and,
conditional on the wide set of health indicators, the probability of utilisation
does not increase with age. Interestingly, in 1987, women aged 20–29 were
more likely to have used hospital care than women in any other female age
group (all else held equal). However, in 2001, the age group of women most
likely to have used a hospital is 25–34. This reflects the delay in fertility
decisions of Spanish women over the last two decades. Income significantly
and positively affected the probability of utilisation of any doctor in 1987
and specialists in 2001. In contrast, income negatively affected the prob-
ability of visiting GPs in 2001. The estimates associated with the health
Public and Private Health Insurance and the Utilisation of Health Care in Spain 183

variables (all actually reflecting some form of bad health) have the expected
(positive) sign and many are statistically significant. For instance, all else
held equal, an individual who self-assesses his/her health as poor is about
20% more likely to have visited a doctor in either of the two periods, and
14% (20%) more likely to have used a hospital in 1987 (2001). In 1987,
private insurance is positively and significantly associated with hospital use
(all else held equal). In 2001, it is positively associated with – in addition to
hospitalisations – the use of emergency services and specialists, but nega-
tively associated with the use of GPs. In fact, an individual with private
insurance was about 5% more likely (1% less likely) to visit a specialist (GP)
than an individual with the same observed characteristics without private
insurance.
The estimates for the models permit the calculation of the inequality
measures presented in Tables 1A and 1B. Note that in both 1987 and 2001,
the utilisation of the three types of services (visits to doctors, emergencies
and hospitalisations) have unequal utilisation distributions which are all
pro-poor. The concentration indices are statistically significant and the
point estimates are greater for 2001, revealing that the degree of pro-poor
inequality is exacerbated over time. Fig. 1 presents the contribution of each
group of variables to the overall CI. These figures reveal that a very large
portion of the CI is explained by need, which is concentrated among
the poor.
The second row of Table 1A presents the inequity measure for each of the
services as defined in Section 3.1. For each of the services, HI (inequity
index) is the part of the CI (inequality index) explained by income and
tenure of PHI (i.e. the non-need variables in our specifications for the
probability of utilisation).
Note that in 1987, the HI indices for total visits and hospitalisations
reveal a significant degree of pro-rich inequity. In these cases, both income
and tenure of PHI contribute positively to the HI index. This means, in
1987, that while overall utilisation is concentrated among the poor, rich
individuals and/or individuals who enjoyed PHI (who tend to be richer than
average) had more chances of using these health services than poor indi-
viduals and/or individuals without PHI at the same level of need. In contrast,
the HI indices for the three services are statistically not different from zero
in 2001, implying that for a given level of need, there are neither pro-rich nor
pro-poor differences in the chances of utilisation explained by income or
insurance status.
In order to analyse the changes over time for these indices in more detail,
it is useful to isolate the sources of their changes. As discussed in Section 3.2,
 184
Table 1A. Concentration Indices and Inequity Indices and Changes Over Time.
1987 2001

Visits Hospital Emergency visits Visits Hospital Emergency visits GP visits Specialist visits

PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

CI 0.0626 0.0342 0.0219 0.0959 0.0847 0.0465 0.1478 0.0121
HI 0.0146 0.0246 0.001 0.0002 0.0281 0.0065 0.0479 0.0991
Income 0.0115* 0.0125 0.0011 0.0102 0.0078 0.0182 0.0439 0.0602
PHI 0.0031 0.0121 0.0021 0.0099 0.0203 0.0117 0.0039 0.0388

Note: Values significantly different from zero (at Po0.05) in bold typeface. *At Po0.10.

Table 1B. Changes Over Time (2001–1987).

Total Visits Hospital Emergency Visits

CI2001–CI1987 0.0333 0.0504* 0.0246

HI2001–HI1987 0.0149 0.0035 0.0055
Relative excess elasticity income 2.0125 0.2870 20.1116
Relative excess elasticity PHI 1.8760 0.4902 6.1485
Relative excess inequality income 0.1293
Relative excess inequality PHI 0.1141

Note: Values significantly different from zero (at Po0.05) in bold typeface. *At Po0.10.
Public and Private Health Insurance and the Utilisation of Health Care in Spain 185
1987 1987 1987 2001 2001 2001 2001 2001

Specialist visit

GP visit

Emergency

Hospital

Visits

Emergency

Hospital

Visits

−0.2 −0.15 −0.1 −0.05 0 0.05 0.1 0.15

Log income Demographics Marital status Need variables Private insurance

Fig. 1. Contributions to Concentration Indices.

the contribution of each covariate to the index is given by the product of the
elasticity of the probability of utilisation and the CI of the covariate. So, it
might be the case that the impact of income, say, on the chances of using a
particular service does not change but income becomes better distributed.
This would lead, ceteris paribus, to a reduction in the contribution of
income to the degree of pro-rich inequality in the chances of utilisation.
Table 1B presents the relevant decompositions for the two non-need co-
variates that we have used in the specification. The table offers a clear
indication of the direction in which the relevant magnitudes have evolved
over time. First note that the distribution of equivalised household income
has become more equal. Relative to 1987, the CI of log equivalised house-
hold income is 13% smaller in 2001. Tenure of PHI, however, has evolved in
the opposite direction. Relative to 1987, the distribution of PHI is 11%
more pro-rich.
Doctor visits: As seen in Table 1A, the HI for the probability of visiting a
doctor is positive and significant in 1987, with both income and PHI con-
tributing positively. In 2001, the HI index is not statistically significant, but
this is the result of two antagonistic effects. While in 2001, the contribution
of income is negative (and not significant), the contribution of PHI is still
positive and significant. In Table 1B, we can see that the change in the
contribution of income is driven by a 200% reduction in the size of the
elasticity of the probability of utilisation (and also the decrease in income
inequality). In contrast, as well as becoming more concentrated among the
186 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

rich, tenure of PHI exerts a greater impact on the probability of utilisation.

The relative change in elasticity is about 180%.
Hospitalisations: The case of hospitalisations is similar to that of doctor
visits. There is a reduction in the contribution of income driven by a 28%
reduction in elasticity (plus the reduction in income inequality) but the PHI
elasticity of the probability of utilisation actually increases by 50%. In 2001,
the contribution of PHI is statistically significant, but the lack of significance
of the income contribution renders the HI insignificant.
Emergencies: The HI index is not statistically significant in either 1987 or
2001. But while in 1987 the contributions of income and PHI are both
insignificant, in 2001 the contribution of PHI is positive and significant.
However, as noted above, the exogeneity assumption of PHI in the equation
for emergencies in 2001 is rejected, so these inferences are to be taken with
caution.
In addition to these three services, we have obtained evidence for GP
visits and specialist visits separately for the year 2001 (unfortunately the
data for 1987 do not distinguish between GP visits and specialist visits). The
results are consistent with the evidence obtained by Van Doorslaer et al.
(2004), Rodrı́guez and Stoyanova (2004) and Jones et al. (2007). That is, GP
visits are concentrated among the poor. This is not only due to need being
concentrated among the poor, since the HI index is negative and significant.
That is, the poor and those without PHI have more chances of visiting the
GP than the rich and/or PHI holders with the same level of need. Of course,
this imbalance is compensated by the existence of a good degree of pro-rich
inequity in the probability of visiting a specialist. Indeed, the inequity index
for the probability of visiting a specialist in 2001 is greater than any of the
other HI indices presented in Table 1A. Note that roughly two-fifths of this
index is accounted for by the contribution of PHI.

5. CONCLUSIONS AND WAYS FORWARD

The results presented in the previous section suggest that the Spanish health
system seems to have achieved the goal of ensuring equal utilisation of
doctors, hospitals and emergency services for equal need. In fact, the reason
why the HI indices for the three services are not statistically significant in
2001 is that the contribution of income is negative (doctor visits and emer-
gencies) and/or insignificant (all three services). With the necessary caveats
derived from the fact that this is a pure before–after evaluation exercise, at
least as far as the point estimates are concerned, it seems that the reforms
Public and Private Health Insurance and the Utilisation of Health Care in Spain 187

during the period 1987–2001 have reduced the income elasticity for the
probabilities of utilisation of the three services. Coupled with a reduction
in pure income inequality, this means that income, by 2001, does not lead
to differences in utilisation for the same level of need. This is clearly an
improvement with respect to 1987, a year for which our estimates show a
positive and significant contribution of income to inequity in the utilisation
of doctors.
On closer examination, however, we note that the contribution of PHI to
inequality in utilisation is positive and significant for the three services. The
data reveal that tenure of PHI has become more concentrated among the
rich and, simultaneously, our estimates suggest an increase in the PHI elas-
ticity of the probability of utilisation for the three services. This leads to a
positive and significant contribution of PHI to our measure of inequity in
2001 for the three services. Moreover, if we consider the chances of visiting a
specialist in 2001, the data reveal a substantial degree of inequity with pos-
itive contributions of both income and PHI. These findings are consistent
with previous results in the literature using data for periods prior to 2001
(Urbanos, 2001b; Abásolo et al., 2001; Van Doorslaer et al., 2002, 2004;
Jones et al., 2007), and contribute to confirm the existence of a clear pattern.
An open issue for future research is the exact causal mechanism leading to a
greater responsiveness of utilisation with respect to PHI. One hypothesis
that is worthy of investigation is that in the Spanish health system of 2001,
PHI generates a strong ‘‘access’’ effect (Jones et al., 2007) allowing indi-
viduals to use private services that are perceived as being of superior quality.
This explanation would also be consistent with the reported fact that tenure
of PHI has become more concentrated among the rich.
The implications of these findings for the policy goals stated in the Health
Act of 1986 depend on a value judgement about whether public policy
should be concerned with the inequity effect of PHI. After all, the services
afforded by PHI are privately provided. A crucial point here is that these
services are partially publicly financed through the tax bonuses to PHI.
Should the public purse subsidise better access to some citizens? If so, does it
matter that these citizens tend to be richer than average? Obviously, equity is
not the only relevant issue when assessing the appropriateness of PHI sub-
sidies. Other considerations include the desire to support a private sector
that might introduce competition into the health care market, or the wish to
deviate demand to private outlets in order to decongest the public network.
Concerning the latter, the evidence for the Spanish case (López Nicolás
& Vera Hernández, 2004) suggests that the subsidies are far from self-
financing: their study shows that for each euro given away as a subsidy for
188 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

the purchase of PHI, the public health care network experiences a reduc-
tion in utilisation worth h0.12. Similar evidence is available for the UK
(Emmerson, Frayne, & Goodman, 2001), where tax bonuses were elimi-
nated recently.
While the overall picture obtained in this paper is that the Spanish
National Health Service has advanced towards making utilisation equitable,
further research must find evidence to justify the subsidies for PHI, an
element of the system that this research shows to generate a significant
degree of inequity.

NOTES
1. Mossialos and Thomson (2002) offer a useful taxonomy for the possible roles
of PHI.
2. To emphasise the conceptual similarity between the CI and the Gini index,
think of the familiar plot for the Lorenz curve for income distribution, where the
vertical axis represents the cumulative percent of income and the horizontal axis
represents the cumulative percent of the population ranked by income. If the vertical
axis is replaced by the cumulative percent of health care use then we obtain a
concentration curve of health care use on income. Since health care can be concen-
trated among either the rich or the poor, this curve can lie above or below the
diagonal (in contrast the Lorenz curve always lies below the diagonal). Similar to the
Gini index – defined as twice the area between the diagonal and the Lorenz curve and
taking values between 0 and 1 – the CI is defined as twice the area between the
diagonal concentration curve, so it ranges between 1 and 1, and a value of 0 implies
no income-related inequality in health care use.
3. Additional details on predictive power and goodness-of-fit are available from
the authors on request.

ACKNOWLEDGMENTS

This paper derives from the project ‘‘La dinámica del estado de salud y los
factores socieconómicos a lo largo del ciclo vital. Implicaciones para las
polı́ticas públicas’’, which is supported by the Fundación BBVA. Support
from Ministerio de Educación project SEJ2005-09104-C02-02 is thankfully
acknowledged. We are grateful to Guillem López, Vicente Ortún, David
Casado, Andrew Jones, Xander Koolman, Eddy van Doorslaer, Peter
Lambert and three anonymous referees for useful comments and sugges-
tions. The views expressed in this paper are those of the authors and not
necessarily those of the funders or the authors’ employers.
Public and Private Health Insurance and the Utilisation of Health Care in Spain 189

REFERENCES
Abásolo Alessón, I. (1998). Equidad horizontal en la distribución del gasto público en sanidad
por grupos socioeconómicos en Canarias. Un estudio comparado con el conjunto
español. Hacienda Pública Española, 147, 3–28.
Abásolo, I., Manning, R., & Jones, A. (2001). Equity in utilization of and access to public-
sector GPs in Spain. Applied Economics, 33, 349–364.
Centro de Investigaciones Sociológicas, Encuesta Nacional de Salud 1987.
Centro de Investigaciones Sociológicas, Encuesta Nacional de Salud 2001.
Emmerson, C., Frayne, C., & Goodman, A. (2001), Should private medical insurance be sub-
sidized? Health Care UK. King’s Fund.
European Observatory on Health Care Systems. (2000). Health care systems in transition. Spain.
Gravelle, H. (2003). Measuring income related inequality in health: Standardisation and the
partial concentration index. Health Economics, 12(10), 803–819.
Hausman, J. (1978). Specification tests in econometrics. Econometrica, 46, 1251–1271.
Jones, A., Koolman, X., & Van Doorslaer, E. (2007). The impact of supplementary private
health insurance on the use of specialists in European countries. Annales d’Economie et
de Statistiques, 83.
López Nicolás, A., & Vera Hernández, M. (2004). Are tax subsidies for private medical in-
surance self-financing? Evidence from a microsimulation model for outpatient and in-
patient episodes. Economics and Business Working Paper #632. Pompeu Fabra
University.
Masseria, C., Koolman, K., & Van Doorslaer, E. (2004). Income related inequality in the prob-
ability of a hospital admission in Europe. ECuity III Working Paper #13.
Mossialos, E., & Thomson, S. M. S. (2002). Voluntary health insurance in the European Union:
A critical assessment. International Journal of Health Services, 32(1), 19–88.
OCU. (1997). Seguros de asistencia sanitaria. Dinero y Derechos, 44, 27–32.
Rodrı́guez, M., Calonge, S., & Reñé, J. (1993). Spain. In: F. Rutten, E. Van Doorslaer &
A. Wagstaff (Eds), Equity in the finance and delivery of health care. An international
perspective. New York: Oxford University Press.
Rodrı́guez, M., & Stoyanova, A. (2004). The effect of private insurance access on the choice of
GP/specialist and public/private provider in Spain. Health Economics, 13(7), 689–704.
Schokkaert, E., & Van de Voorde, C. (2004). Risk selection and the specification of the risk
adjustment formula. Journal of Health Economics, 23(6), 1237–1259.
Urbanos, R. (1999). Análisis y evaluación de la equidad horizontal interpersonal en la prestación
pública de servicios sanitarios. Un estudio del caso español para el periodo 1987–1995. PhD
thesis. Universidad Complutense de Madrid.
Urbanos, R. (2001a). Measurement of inequity in the delivery of public health care: Evidence from
Spain (1997). Fedea. Documento de Trabajo 2001–15.
Urbanos, R. (2001b). Explaining inequality in the use of public health care services: Evidence
from Spain. Health Care Management Science, 4, 143–157.
Van Doorslaer, E., & Jones, A. (2003). Inequalities in self-reported health: Validation of a new
approach to measurement. Journal of Health Economics, 22(1), 61–87.
Van Doorslaer, E., & Koolman, X. (2004). Explaining the differences in income-related health
inequalities across European countries. Health Economics, 13(7), 609–628.
Van Doorslaer, E., Koolman, X., & Jones, A. (2004). Explaining income-related inequalities in
doctor utilisation in Europe. Health Economics, 13(7), 629–648.
190 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

Van Doorslaer, E., Koolman, X., & Puffer, F. (2002). Equity in the use of physician visits in
OECD countries: Has equal treatment for equal need been achieved. In: Measuring up:
Improving health systems performance in OECD countries. Paris: OECD.
Van Doorslaer, E., Masseria, C., & Koolman, X. for the OECD Health Equity Group. (2006).
Inequalities in access to medical care by income in developed countries. Canadian Med-
ical Association Journal, 174, 177–183.
Vera-Hernández, M. (1999). Duplicate coverage and the demand for health care: The case of
Catalonia. Health Economics, 8(7), 579–598.
Wagstaff, A., & Van Doorslaer, E. (1996). Measuring and testing for inequity in the delivery of
health care. School of Social Sciences, University of Sussex (mimeo).
Wagstaff, A., Van Doorslaer, E., & Paci, P. (1989). Equity in the finance and delivery of health
care: Some tentative cross-country comparisons. Oxford Review of Economic Policy,
5(1), 89–112.
Wagstaff, A., Van Doorslaer, E., & Watanabe, N. (2003). On decomposing the causes of health
sector inequalities with an application to malnutrition inequalities in Vietnam. Journal of
Econometrics, 112(1), 207–223.
Wooldridge, J. (2003). Introductory econometrics. A modern approach. Mason, Ohio: Thomson,
South-Western.
Public and Private Health Insurance and the Utilisation of Health Care in Spain 191

APPENDIX

Tables A1–A3

Table A1. Descriptive Statistics.

1987 2001

Mean SE N Mean SE N

Visits 0.178 0.382 29,185 0.235 0.424 20,961

Hospital 0.070 0.255 28,849 0.086 0.280 20,952
Emergency visits 0.109 0.311 29,122 0.191 0.393 20,954
GP visits – – 0.159 0.366 20,961
Specialist visits – – 0.076 0.265 20,961
Log income 10.663 0.482 29,218 11.483 0.449 20,978
F20_24 0.054 0.226 29,218 0.044 0.204 20,978
F25_29 0.049 0.215 29,218 0.045 0.207 20,978
F30_34 0.039 0.193 29,218 0.051 0.221 20,978
F35_39 0.037 0.190 29,218 0.045 0.207 20,978
F40_44 0.038 0.190 29,218 0.045 0.207 20,978
F45_49 0.038 0.191 29,218 0.036 0.186 20,978
F50_54 0.048 0.214 29,218 0.038 0.191 20,978
F55_59 0.038 0.191 29,218 0.031 0.173 20,978
F60_64 0.038 0.191 29,218 0.031 0.174 20,978
F65_69 0.039 0.193 29,218 0.041 0.199 20,978
F70_74 0.024 0.154 29,218 0.034 0.180 20,978
F75_79 0.018 0.135 29,218 0.024 0.152 20,978
F80 0.014 0.118 29,218 0.020 0.139 20,978
M16_19 0.049 0.215 29,218 0.036 0.185 20,978
M20_24 0.053 0.224 29,218 0.044 0.205 20,978
M25_29 0.049 0.217 29,218 0.052 0.221 20,978
M30_34 0.039 0.195 29,218 0.047 0.211 20,978
M35_39 0.037 0.190 29,218 0.048 0.213 20,978
M40_44 0.036 0.186 29,218 0.042 0.200 20,978
M45_49 0.035 0.183 29,218 0.034 0.182 20,978
M50_54 0.045 0.207 29,218 0.038 0.190 20,978
M55_59 0.032 0.177 29,218 0.027 0.163 20,978
M60_64 0.032 0.177 29,218 0.030 0.169 20,978
M65_69 0.029 0.169 29,218 0.028 0.165 20,978
M70_74 0.018 0.133 29,218 0.026 0.160 20,978
M75_79 0.012 0.111 29,218 0.020 0.140 20,978
M80 0.011 0.105 29,218 0.014 0.118 20,978
Married 0.620 0.485 29,218 0.580 0.494 20,978
Widowed 0.077 0.266 29,218 0.079 0.269 20,978
Cholesterol 0.061 0.240 29,218 0.109 0.311 20,978
High blood pressure 0.098 0.297 29,218 0.143 0.350 20,978
192 PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

Table A1. (Continued )

1987 2001

Mean SE N Mean SE N

Diabetes 0.040 0.195 29,218 0.055 0.227 20,978

Bronchitis/asthma 0.062 0.241 29,218 0.049 0.215 20,978
Heart 0.044 0.205 29,218 0.051 0.221 20,978
Allergy 0.056 0.230 29,218 0.077 0.266 20,978
Limited by chronic 0.180 0.385 29,218 0.091 0.288 20,978
Limited by pain 0.083 0.275 29,218 0.150 0.357 20,978
SAH good 0.537 0.499 29,218 0.550 0.497 20,978
SAH fair 0.245 0.430 29,218 0.225 0.418 20,978
SAH poor 0.069 0.253 29,218 0.072 0.259 20,978
SAH very poor 0.013 0.114 29,218 0.016 0.127 20,978
Bed 0.043 0.203 29,218 0.072 0.259 20,978
Accident 0.074 0.262 29,218 0.084 0.277 20,978
Private insurance 0.134 0.340 29,218 0.118 0.323 20,978
 Public and Private Health Insurance and the Utilisation of Health Care in Spain 193
Table A2. Linear Probability Model Results for the Probability of Doctor Utilisation in 1987 and 2001.
1987 2001

Total visits Hospital Emergency visits Total visits Hospital Emergency visits GP visits Specialist visits

Log income 0.0078* 0.0034 0.0005 0.0098 0.0027 0.0143 0.0288 0.0189
F20_24 0.0239 0.0288 0.0041 0.0093 0.0106 0.0194 0.0059 0.0152
F25_29 0.0283 0.0467 0.0033 0.0030 0.0523 0.0060 0.0188 0.0158
F30_34 0.0167 0.0030 0.0098 0.0079 0.0561 0.0059 0.0265* 0.0186
F35_39 0.0005 0.0026 0.0236* 0.0229 0.0152 0.0736 0.0220 0.0009
F40_44 0.0204 0.0443 0.0607 0.0091 0.0255 0.0798 0.0359 0.0268
F45_49 0.0195 0.0626 0.0681 0.0260 0.0352 0.1108 0.0050 0.0310
F50_54 0.0003 0.0623 0.0657 0.0284 0.0209 0.0703 0.0030 0.0254*
F55_59 0.0171 0.0711 0.0636 0.0037 0.0490 0.1162 0.0116 0.0078
F60_64 0.0059 0.0688 0.0713 0.0482 0.0239 0.1423 0.0252 0.0231
F65_69 0.0278* 0.0739 0.1048 0.0370* 0.0343 0.1013 0.0231 0.0139
F70_74 0.0630 0.0824 0.1117 0.0718 0.0278* 0.1415 0.0402* 0.0316
F75_79 0.0005 0.0800 0.0981 0.0338 0.0272 0.1630 0.0421* 0.0084
F80 0.0190 0.0793 0.1264 0.0261 0.0514 0.1660 0.0505* 0.0244
M16_19 0.0109 0.0066 0.0026 0.0335* 0.0059 0.0246 0.0388 0.0053
M20_24 0.0209* 0.0077 0.0052 0.0654 0.0077 0.0303 0.0590 0.0064
M25_29 0.0215* 0.0197 0.0146 0.0625 0.0113 0.0325* 0.0493 0.0132
M30_34 0.0372 0.0476 0.0388 0.0428 0.0179* 0.0564 0.0414 0.0014
M35_39 0.0365 0.0522 0.0491 0.0340* 0.0056 0.0690 0.0325 0.0015
M40_44 0.0326 0.0422 0.0558 0.0505 0.0229* 0.0886 0.0474 0.0031
M45_49 0.0340 0.0555 0.0698 0.0403 0.0020 0.0798 0.0427 0.0024
M50_54 0.0413 0.0388 0.0624 0.0413 0.0173 0.0933 0.0337* 0.0076
M55_59 0.0188 0.0365 0.0800 0.0354 0.0191 0.1329 0.0507 0.0154
M60_64 0.0027 0.0635 0.0833 0.0201 0.0132 0.1339 0.0212 0.0011
M65_69 0.0154 0.0456 0.0943 0.0446* 0.0111 0.1267 0.0312 0.0135
M70_74 0.0124 0.0426 0.0820 0.0221 0.0034 0.1246 0.0086 0.0307*
 Table A2. (Continued )

194
1987 2001

Total visits Hospital Emergency visits Total visits Hospital Emergency visits GP visits Specialist visits

M75_79 0.0275 0.0348 0.0751 0.0303 0.0003 0.1145 0.0319 0.0016

M80 0.0027 0.0039 0.0657 0.0147 0.0277 0.1056 0.0116 0.0263
Married 0.0263 0.0520 0.0295 0.0117 0.0198 0.0169 0.0055 0.0062

PILAR GARCÍA GÓMEZ AND ANGEL LÓPEZ NICOLÁS

Widowed 0.0359 0.0362 0.0228 0.0013 0.0019 0.0077 0.0015 0.0028
Cholesterol 0.0347 0.0123* 0.0103 0.0167 0.0127* 0.0100 0.0339 0.0172
High blood pressure 0.0647 0.0106* 0.0006 0.0465 0.0067 0.0074 0.0584 0.0119*
Diabetes 0.0508 0.0126 0.0159 0.0298 0.0185* 0.0007 0.0482 0.0184*
Bronchitis or asthma 0.0383 0.0034 0.0265 0.0260* 0.0165 0.0653 0.0318 0.0059
Heart 0.0458 0.0582 0.0681 0.0267* 0.1011 0.1038 0.0154 0.0422
Allergy 0.0166 0.0044 0.0263 0.0297 0.0293 0.0116 0.0223 0.0074
Limited by chronic 0.0219 0.0687 0.0649 0.0293 0.0558 0.0901 0.0260 0.0033
Limited by pain 0.1849 0.0082 0.0316 0.2564 0.0024 0.1102 0.1877 0.0687
SAH good 0.0242 0.0134 0.0215 0.0505 0.0170 0.0333 0.0295 0.0210
SAH fair 0.1304 0.0440 0.0510 0.1530 0.0926 0.1364 0.0783 0.0747
SAH poor 0.2114 0.1461 0.1273 0.2073 0.2029 0.2102 0.0948 0.1125
SAH very poor 0.1717 0.1786 0.1866 0.0231 0.1032 0.0544* 0.0189 0.0419
Bed 0.2016 0.0385 0.0700 0.1780 0.0603 0.0420 0.1216 0.0564
Accident 0.0663 0.0830 0.2985 0.0580 0.0651 0.4220 0.0082 0.0498
Private insurance 0.0101 0.0160 0.0041 0.0441 0.0328 0.0423 0.0119* 0.0560

Note: Values significantly different from zero (at Po0.05) in bold typeface. *Po0.10.
Public and Private Health Insurance and the Utilisation of Health Care in Spain 195

Table A3. Tests for the Assumption of Exogeneity of Private Health

Insurance in the Utilisation Equations.
1987

Het.-robust F stat. for null hypothesis of F(6.28004) ¼ 112.11

joint no significance of instruments in
reduced form
P-value 0

Total visits Hospital Emergency visits

Het.-robust t-value for null 1.17 0.1 0.71

hypothesis of no significance of
residual from reduced form in
utilisation equation
P-value 0.24 0.92 0.47

2001

Het.-robust F stat. for null hypothesis of F(6.20584) ¼ 9.71

no significance of instruments in
reduced form
P-value 0

Total Hospital Emergency GP visits Specialists

visits visits

Het.-robust t-value for null 1.05 1.71 2.44 0.24 1.91

hypothesis of no significance of
residual from reduced form in
utilisation equation
P-value 0.29 0.08 0.015 0.81 0.056

Note: All test statistics are robust to heteroscedasticity. The set of excluded instruments used for
testing are a set of six occupational dummy variables for the head of the household.
AGING AND INTER-
GENERATIONAL FAIRNESS:
A CANADIAN ANALYSIS

Michael Wolfson and Geoff Rowe

ABSTRACT
Population aging in many countries has become a fundamental concern of
public policy. One reason is fears that increasing numbers of elderly will
place disproportionate burdens on their children in order to fund public
pensions and health-related services. This analysis first discusses basic
principles for assessing this question of intergenerational fairness. It then
applies an empirically-based overlapping cohort dynamic microsimulation
model for a quantitative analysis of the flows of taxes and cash and
in-kind transfers for successive birth cohorts. The simulations cover both
exogenous factors – specifically trends in life expectancy and the strength
of the economy, and policy-related factors – specifically raising the age of
entitlement to public pensions from age 65 to 70, and price versus relative
wage indexing. The analysis concludes, among other points, that inter-
generational differences are significantly smaller than intra-generational
variations, and that the parents of the baby-boom generation are likely to
benefit from the largest lifetime net transfers of any birth cohort from
1890 to 2010.

Equity
Research on Economic Inequality, Volume 15, 197–231
r 2007 Published by Elsevier Ltd.
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15009-2
197
198 MICHAEL WOLFSON AND GEOFF ROWE

To portray America as riven by generational warfare, young against old, is therefore an

exaggeration. Worse, it obscures a deeper divide, of class rather than age. yThe big
problem of the American welfare state is not that the old get too much, but that the rich
do. (Economist, January 11, 1997)

INTRODUCTION

Population aging is increasingly recognized as a worldwide phenomenon. And

the pace of aging over the past few decades, especially in the richest countries,
is unprecedented in human history. While there are many reasons why in-
creased longevity and modest family sizes should be welcomed, population
aging is more often a source of concern. The essential fear is that future
cohorts of elderly, in particular those of the post WW II baby boom, will be
placing an intolerable financial burden on future working age generations.
A claim capturing this concern is that the baby boom generation is the first in
history who cannot expect their children to be better off than they are.
In this paper, we examine these questions for Canada, looking decade by
decade at the financial circumstances of successive birth cohorts born during
the 20th century. More specifically, for each decadal birth cohort, we have
estimated the total amount of income and payroll taxes they pay over their
lifetimes, and the corresponding amounts of income transfers, especially old
age pensions, and transfers in kind they receive in the forms of education
and health care services.
It takes the better part of a lifetime to accrue a pension or to save
adequately for retirement. As a result, a widely accepted criterion for good
public policy in this area is that the rules underpinning public pensions
and other major age-sensitive programs should be reasonably stable and
predictable. Such stability enables individuals to plan better their own private
savings over their life course. On the other hand, this kind of stability in
public sector rules is very difficult to achieve because the future is inherently
unknowable. Still, by indexing and other provisions, governments have de
facto indicated what the responses will be in future to at least some unknown
vicissitudes – for example, in the case of the indexing of benefit levels and
income tax thresholds, to future and as yet unknown rates of inflation.
It would be reassuring for future public pensions, health care, and income
tax policy – the three large government programs most sensitive to popu-
lation aging – to be based on a set of rules or principles (to the extent
possible) that individuals could count on for their long term financial plan-
ning. In turn, this means that these rules or principles would be unlikely to
Aging and Inter-Generational Fairness 199

be amended in future, by whatever democratic coalitions might emerge. In a

phrase, this entails (as a necessary, but not sufficient condition) that the
rules and principles should be designed in such a way that they are currently,
and will continue to be, broadly perceived as inter-generationally fair. (It is
also important that the rules be fair within generations, a point to which we
return later.)
This is a challenging objective for at least two fundamental reasons. First,
there is no widely agreed concept of inter-generational fairness. Second, the
future is inherently unknowable, so that even with an agreed concept of
inter-generational fairness, it is extraordinarily difficult, if not impossible, to
examine such a criterion for all possible eventualities. In other words, it is
impossible to assess the inter-generational fairness of a given system of rules
or principles under all possible future scenarios.
However, it is possible to make a reasoned effort, and to consider a few of
the most important areas of uncertainty. In this spirit, the following analysis
provides a series of birth cohort-specific quantitative reconstructions of the
histories, and projections of the interactions, of successive birth cohorts with
Canada’s major tax and cash and in kind transfer programs. The essential
objective is to provide the information to support judgments as to the inter-
generational fairness of these major public programs.1
In addition to a ‘‘baseline’’ or status quo scenario for the future evolution
of these programs, two further policy-related scenarios will be examined.
Neither represents a specific policy option currently under discussion in
Canada. Rather they represent two stylized and simplified versions of policy
options either under discussion or implemented in other countries. The first
policy alternative essentially raises the age of entitlement to public pensions
from 65 to 70. The second shifts a range of tax and cash transfer program
indexing provisions from the status quo consumer price index (CPI) to the
average wage (AW).
Since a fundamental objective of this analysis is to assess the likely inter-
generational fairness of major age-sensitive public programs to the unknown
vicissitudes of the future, we have also posited several alternative scenarios for
the socio-economic milieu within which future program structures would
apply. Again, we rely on a few highly stylized scenarios. In general, we can
distill two main axes of uncertainty from the literature on public pensions and
inter-generational equity – whither longevity, and whither the economy.2
More concretely, we have therefore constructed four (2
2) ‘‘exogenous’’
socio-economic scenarios within which to embed our examination of the inter-
generational patterns of public policy – high and low projections of future
mortality rate improvements, and high and low future levels of employment.3
200 MICHAEL WOLFSON AND GEOFF ROWE

The vast majority of published literature on inter-generational fairness

and the inter-generational impacts of public programs has been extremely
simplified. In particular, it has typically involved just one representative (or
average) individual (or maybe two, a male and a female) for each birth
cohort (e.g. Kotlikoff, 1992). This is an extremely restrictive assumption,
and one that we will show yields seriously misleading results. This analysis
eschews such representative agents, and instead is fully microanalytic. It is
based on Statistics Canada’s LifePaths microsimulation model (Statistics
Canada). LifePaths is much more than a casual extension of some sort of
spreadsheet analysis. Rather, it is an ‘‘industrial strength’’ policy-oriented
microsimulation model that has been developed over more than a decade in
partnership with a number of central policy ministries of the Canadian
Government.
In the following section, we review some of the main concepts of inter-
generational fairness. Then we briefly outline the LifePaths model. The main
part of the analysis describes the scenarios – both policy and exogenous – in
greater detail and then moves onto the key results.

JUDGING INTER-GENERATIONAL FAIRNESS AND

SUSTAINABILITY

There is no widely agreed approach to judging whether a society’s tax/

transfer system is inter-generationally fair or sustainable.4 But there is a
considerable literature addressing this question.
One strand builds on conventional economic theory, and dates back at
least implicitly to Ramsay (1928). The basic ‘‘axiom’’ in these inter-temporal
utility function-based analyses (Basu & Mitra, 2005) is that inter-generational
equity is achieved when all generations have identical utility. Unfortunately,
these kinds of formal analyses are highly abstract and embody such simpli-
fying assumptions as to render them of no practical use in the context of
applied analysis.5
Kotlikoff and others have popularized the notion of ‘‘generational
accounting’’ (e.g. Kotlikoff, 1992; Kotlikoff & Burns, 2004). Within this
framework, generational inequity arises when the taxes required of future as
yet unborn generations – to pay off current government debt (including un-
accounted items such as the unfunded longer term liabilities of the U.S. Social
Security pension and Medicare programs) as well as to finance government
services that continue at current levels – exceed the taxes being paid by the
Aging and Inter-Generational Fairness 201

generations alive today. In general, this is a reasonable principle. Note that it

applies to the entire public sector, and not just to the age-sensitive programs
being examined here. However, as just noted, the methods specifically used to
estimate generational accounts (representative agents, constant exponential
growth for all time) are dubious.
There are, in any case, several norms which appear commendable,
generally drawn from public policy documents:
 Inter-generational golden rule (e.g. Canada, 1980, Summary, p. 54) – One
generation, when it becomes old and frail, should not expect to be treated
any better by its children (when they are of working age at that future
time) than it treated its parents’ generation in their old age (when they
themselves were of working age).6
 Sustainability (Canada, 1980, Summary, p. 54; House of Commons, 1983,
p. 15) – The world that parents bequeath to their children should be at
least as good (e.g. economically productive) as the one they in their turn
had inherited.7
 Neutrality – Each generation should pay for its own pensions, i.e. there
should be no inter-generational transfers at all.
 Musgrave (1981) – Per capita transfers to the elderly should be a fixed
proportion of per capita wages less taxes of the working age population.
 House of Commons (1983, p. 17) – Pensioners should expect to share in
real economic growth when it occurs, but correspondingly should not be
completely immune from economic recessions.
 Majority norm – ‘‘If a retirement income system is not, and is seen not to
be, fair in its treatment of successive generations, it will be changed sooner
or later’’ (Canada, Summary, 1980, p. 54). In other words, a tax/transfer
system is sustainable and fair if it is the outcome of a continuing democratic
consensus.
Fig. 1 (Wolfson, Rowe, Gribble, & Lin, 1998) provides a convenient
schema for illustrating these norms. Birth year is shown on the vertical axis,
and calendar time along the horizontal. Each horizontal bar represents one
generation or birth cohort born at time b. In turn, the lifetimes of the bth
birth cohort have been divided into three broad phases: childhood (Cb),
working (Wb), and elderly (Eb). Inter-generational transfers then arise, in
this analysis, only from government tax/transfer activities (both cash and in
kind), and generally speaking involve either Wb-Cb+1 or Wb-Eb1 flows,
as indicated by the short vertical arrows in the diagram.
There are several challenges for assessing inter-generational fairness in
terms of Fig. 1. The first is that the boxes in Fig. 1 (along the vertical axis),
202 MICHAEL WOLFSON AND GEOFF ROWE

Birth
Cohort

1890 C W E
1910 C W E
1930 C W E
1950 C W E
1970 C W E
1990 C W E
Calendar
1890 1990 2090 Year

Fig. 1. Basic Generational Accounting Framework.

in reality, should not all be shown as the same size. Some birth cohorts are
larger than others. Second, the state of the economy varies significantly from
one time period to the next along the horizontal axis. The arrows in Fig. 1
focus on the sequence of contemporaneous (i.e. point-in-time) transfers
occurring across generations. But the diagram has no representation of
savings and investment (or dis-saving and dis-investment), in the sense for
example of the future productivity (or increased environmental degradation)
of the society. How much wealth and income a society is able to generate in
future years is a crucial aspect of assessments of inter-generational fairness,
and is implicit in the first two norms.
Nevertheless, the diagram is still helpful in thinking about the various
norms for inter-generational fairness just outlined. The first norm implies
that the public pensions and health care services expected by the current
working age generation, when it becomes elderly in the future, should not
make any larger claim on resources, relative to the size of the economy, than
the transfers it is financing for the current elderly. In terms of Fig. 1, this
norm implies that the sequence of transfers indicated by the vertical arrows
from Wb to Eb1 should be non-increasing over time (in proportion to the
size of the economy).
The second norm suggests that it is unfair to bequeath to future gener-
ations any kind of substantial liability, such as a large public debt, or a
degraded physical, human, environmental or other kind of capital stock.
This norm is consistent with lifetime consumption or disposable income that
rises from one generation to the next. In other words, each generation of
parents is sacrificing at least somewhat so their children can have a better
life. This norm encompasses much more than income taxes and cash and in
Aging and Inter-Generational Fairness 203

kind transfers made through the public sector. Still, if each generation (e.g. Eb
which had been Wb earlier during its working years) succeeded in leaving its
successors (Wb+1) a wealthier and more productive economy, then it should
be possible for Wb+1 to transfer to Eb an amount that is higher than the
amount Wb had transferred to Eb1 in its turn. In terms of Fig. 1, this means
that transfers from those of working age to elderly should be increasing (or at
least non-decreasing) from one generation to the next. Note that this norm is
inconsistent with the ‘‘anonymity’’ version of inter-generational equity pos-
ited by Basu and Mitra (2005), though it is consistent with the norm implicit
in Kotlikoff-style generational accounting.
The third norm takes a different approach, basically saying that the fairest
system is one in which there are no inter-generational transfers at all. How-
ever, there exists no ‘‘extra-planetary banker’’ who can initially loan funds to
children to fund their consumption and education while they are growing up,
take savings from them when they ‘‘graduate’’ into their working years first to
pay down their ‘‘growing up’’ and educational loans, and then to accumulate
savings for their retirement, and finally gradually disburse their accumulated
savings after they have retired. Instead, the savings and dis-savings of each
generation during its life course inevitably involve de facto contemporaneous
inter-generational transfers (albeit mediated in complex ways both via gov-
ernment taxes and transfers, financial markets, and intra-family transfers).
Still, this norm is (in our view, naively) involved in analyses that compare, for
example, the internal rates of return to different generations for their Social
Security contributions and benefits, and implicitly raises concerns when they
are not all the same (e.g. Beach & Davis, 1998).
This norm could also be taken as the essence of a strong form of the
Kotlikoff-style generational accounting. In these analyses, the key ‘‘empirical’’
result is what the tax rate (presumed constant from today forward) would
have to be for all future unborn generations in order to amortize government
debt (broadly defined). And the presumption, when this tax rate is found to be
much higher than that being paid by currently living generations (as is the case
in the U.S., but not necessarily in Canada; see Oreopoulos & Vaillancourt,
1998), is that taxes on those currently alive should be immediately raised, or
their government benefits cut, such that the two tax rates (those of the living,
and those of the yet unborn) become equal.
The fourth norm (Musgrave, 1981), in a more precise and focused way,
provides a kind of balance point between the first and second norms.
According to this norm, the sequence of transfers Wb to Eb1 are in some
general sense constant, but not constant in simple money terms. Rather,
this norm entails a kind of relative effort and relative benefit constancy. It
204 MICHAEL WOLFSON AND GEOFF ROWE

adjusts automatically in the event of population aging by reducing net

transfers to the elderly; and it adjusts for higher than anticipated per capita
economic growth by raising net transfers.
The fifth norm was developed by a Special Parliamentary Committee
formed in 1983 to review the results of what at the time in Canada was
referred to as the ‘‘great pension reform debate’’. After much (often in
camera) discussion amongst the members of this all party committee, these
Members of Parliament agreed on recommending that public pensions (as
well as income tax thresholds for retirement saving incentives) should be
indexed in such a manner that they are higher when
 real average wage growth is higher,
 labor force participation is higher,
 unemployment is lower, and
 the old-age dependency ratio is lower,
and would be lower in the opposite circumstances.
One metaphor used by committee Members8 to understand this norm in
their own terms was in reference to an extended family in an agrarian
society: When the harvest was good, everyone benefited including the
family’s elderly, even if they were too frail to have contributed much labor.
Correspondingly, when the harvest was poor, everyone was expected to
make do with less. This norm itself is, in fact, a specific and more precise
articulation of the both of the first two norms, and fully in the spirit of the
Musgrave notion.
Interpreting the sixth ‘‘majority’’ norm only in the context of Fig. 1 is
difficult. The main reason is that the population of eligible voters at any
point in time includes not only members of different generations, but also
individuals within a generation who are in widely different circumstances. In
a word, each generation is heterogeneous. It could be, for example, that a
tax/transfer system is (‘‘point in time’’) progressive in a way that lower and
middle income individuals from several adjacent generations (all of whom
are of voting age at that point in time) have more in common than those
with high and low incomes within a given generation. Thus, ‘‘block voting’’
by generation, or generational politics, may not be in many individuals’ self-
interest. As a result, the democratic majority norm need not be consistent
with any of the other norms.9
If we step back from the details, and do not take the numeraire for
measurement too literally, the first five norms, in the end, are quite similar.
And they do, in this sense, align with the Ramsay (1928) implicit assumption
that inter-generational fairness or equity requires some sort of equal
Aging and Inter-Generational Fairness 205

treatment across generations. This, however, is the easy part. The challenge
is how to operationalize ‘‘equal treatment’’ in the real world where, among
other things, the population size of successive birth cohorts varies, along
with rates of economic growth, and patterns of working and saving.
In this regard, the Musgrave norm begins moving toward a more practical
or operational form by taking account of two factors – per capita wages and
tax rates, which in turn will vary with the economic productivity, the
employment to population ratio, and the age structure of the society. The
House of Commons norm depends on essentially identical factors, since ‘‘per
capita wages’’ specified in the Musgrave norm depend primarily on real
average wages, labor force participation and unemployment rates – the
factors explicitly mentioned in the House of Commons norm.
In turn, the implication of the first five norms is that any assessment of inter-
generational fairness requires specific data for each of a sequence of overlap-
ping birth cohorts. In particular, the key data are the incomes net of transfers
received less taxes paid, over complete lifetimes, of successive birth cohorts.
Moreover, in line with Ramsay (1928), ‘‘ywe do not discount later
enjoyments in comparison with earlier ones, a practice which is ethically
indefensible and arises merely from the weakness of the imagination’’. The
quantitative analysis developed in the following sections uses the average
wage as the numeraire – in other words not in nominal dollars, not in
constant dollars (i.e. deflated by the CPI), not in present discounted values
(i.e. deflated by an interest or time-preference rate), but in ‘‘wage-relative’’
dollars (i.e. deflated by the average wage).
The sixth norm reminds us that while these kinds of generational data are
analytically necessary, they are by no means sufficient. We need an anal-
ytical framework that can also unpack and reflect the great heterogeneity of
individuals’ life course experiences within any given generation.
Given the conceptual discussion so far on norms of inter-generational
fairness, it could be argued that two fundamentally distinct notions are still
being confounded – in two words, levels and risks. How would the analysis
unfold if we considered these norms in terms of a pair of questions: (1) if
there were no uncertainty, what should the levels of available resources
(consumption) be for successive generations; and (2) how should the risks
associated with various kinds of uncertainties be shared between genera-
tions?10 For example, if the default assumption is that the inter-generational
‘‘contract’’ implicit in a society’s package of tax and transfer programs is
written in nominal dollars, then the answer to the ‘‘level’’ question would be
clear and easy to understand, but in practice it would be subject to major
risks due to the uncertainties of future inflation rates.
206 MICHAEL WOLFSON AND GEOFF ROWE

In effect, the levels question is the easier one – either a Ramsay approach
with some sort of equality between generations, or a kind of golden rule
approach with increasing levels over time. The risks question is far more
difficult, and what the norms outlined above, in particular the Musgrave
and House of Commons norms, are doing is beginning to specify how levels
of inter-generational transfer should be automatically adjusted in the face of
a range of specific kinds of risks or uncertainties – such as faster or slower
economic growth, more or less rapid population aging, and higher or lower
unemployment rates. In other words, these norms embody answers to both
kinds of questions – what fair levels of inter-generational transfers ought to
be, and what fair methods of sharing risks should be – at least for a few
obvious kinds of risks, or more properly unknowns: future inflation rates,
life expectancy trends, per capita wage growth, etc.
One further and most interesting kind of risk is the marginal utility of
money. The Ramsay (1928) analysis, and most subsequent mainstream eco-
nomic analyses, make explicit reference to this notion by assuming the ex-
istence of well-defined and well-behaved utility functions, and then using
formal mathematics when examining inter-generational equity questions.
However, the debate on the way in which technical progress and ‘‘new goods’’
should be handled in the construction of the Consumer Price Index, i.e. the
way that inflation risk should be removed from inter-generational contracts,
shows the practical impossibility of ever addressing this question.11 The
empirical evidence on subjective well-being, in contrast, suggests that indi-
viduals’ ‘‘utility’’ is more likely driven by a mix of their relative position
within their social group, and person-specific homeostatic set-points. These
latter realities give further support to the use, in this analysis, of the average
wage as the numeraire for assessing inter-generational differences in levels.
One implication of reframing the question of appropriate norms for judg-
ing inter-generational equity into the pair of questions – on levels and on
risks – is that there can never be a completely specified norm. The simple
reason is that the risks are inherently unknowable in full. Perhaps a more
practical approach – call it an ‘‘evolutionary’’ approach to inter-generational
equity – is to plan on the norms having to evolve as new kinds of risks and
uncertainties emerge and become evident. The most obvious case was (not so
much the ‘‘baby boom’’, but rather) the ‘‘baby bust’’ decline in fertility in the
late 1960s. The Social Security and Canada Pension Plan actuaries did not
anticipate this when these public pension plans were first set up, yet now
changes in population age structure are at the heart of much of the discourse
on inter-generational equity.
Aging and Inter-Generational Fairness 207

Finally, the rather general points of the preceding paragraphs serve to em-
phasize the overall weakness of the literature in this area of inter-generational
fairness. The formal theoretical work is far too abstract to be of use; the
sociological and political science writings, while illuminating, offer no practical
guidance; and most of the quantitative empirical work either rests on unre-
alistic simplifying assumptions (e.g. infinite horizon constant exponential
growth in generational accounting), or is too partial (e.g. the internal rate of
return calculations for Social Security in the U.S.).

AN OVERVIEW OF THE LIFEPATHS

MICROSIMULATION MODEL

In order to assess Canada’s tax/transfer system in the light of the norms just
outlined, we draw on Statistics Canada’s LifePaths model. LifePaths is a
computer simulation model designed explicitly to encompass both inter- and
intra-generational analyses simultaneously. Each LifePaths simulation run
generates a representative microcosm of the Canadian population. In other
words, LifePaths is microanalytic. The basic units of observation are indi-
viduals, and the focus is on microlevel dynamics – how individuals move
among various mixtures of socio-economic states over their life courses.
And empirically, LifePaths is metasynthetic – drawing upon multiple data
sets, covering diverse subject matters, and using each in order to assemble
the best possible overall estimate of the information of interest.12
The basic unit of analysis in LifePaths is an individual life history or
stylized biography, as shown in Fig. 2. The ‘‘state space’’ of attributes or
individual characteristics is shown along the vertical axis, with age and
calendar time coincident along the horizontal. The third axis indicates a
representative sample of individuals in the population of interest. These are
not all unrelated individuals; rather, they are juxtaposed to show that family
structure is also included.
Given these microlevel life histories as the basic building blocks, LifePaths
assembles large representative samples of individuals (grouped into nuclear
families) in a sequence of overlapping birth cohorts (Fig. 3). Each ‘‘layer’’ in
the diagram represents one birth cohort, while the sequence of layers
represents successive birth cohorts. A typical population pyramid showing
age structure by sex at a point in time corresponds to a vertical slice through
the overlapping birth cohorts along the line for ‘‘today’’.13
208 MICHAEL WOLFSON AND GEOFF ROWE

population person j+1

sample person j
child 2
child 1
spouse
state Nuptiality
space
Fertility
Education
Labour Market
Disability
Institutionalization
…. etc

time, age

Fig. 2. State Space and Longitudinal Microdata Sample Generated by a LifePaths

Simulation.

Birth
Cohort “today”

Calendar
Year

Heterogeneous
Individuals

Fig. 3. Overlapping Birth Cohorts with Heterogeneous Members.

LifePaths essentially creates a large sample of representative individual life

histories, where the individuals have been born throughout the 20th century
in accord with historical population data. The historical reconstruction and
subsequent projection processes proceed by data synthesis using longitudinal
microsimulation: each individual’s life history is synthesized, starting at birth
and then recursively generating the suite of events and characteristics shown
along the vertical axis of Fig. 2 over time until death. Then another family of
Aging and Inter-Generational Fairness 209

individuals is synthetically generated, and again, and again, until a very large
sample (e.g. 1,000,000s) is generated.
The result is our ‘‘fitted’’ population microcosm (for years prior to
‘‘today’’), plus microlevel extrapolations of each life history beyond ‘‘today’’
(if still alive) over coming decades. The result is a very large longitudinal
sample of synthesized individuals that – when appropriately cross-tabulated
or otherwise examined – reproduces a diversity of observed data, such as
population characteristics from censuses, mortality and fertility rates dating
back to about 1900, age- and sex-specific employment/population ratios
since the 1970s, and aggregate wages and income taxes back to the 1920s.
Underlying any LifePaths simulation is a detailed set of empirically based
state transition dynamics. As a result, dynamics are represented by transition
probabilities (more precisely, by transition probability functions of a range
of time-varying covariates/co-evolving characteristics).
For example, the nuptiality transitions explicitly modeled are shown in
Fig. 4. The different states are given by the boxes, while the arrows indicate
the possible transitions. For each arrow, there is an empirically estimated

Single Common law union

Married Widowed Widowed CLU

Separated CLU Separated

Divorced CLU Divorced

Married (2nd) Widowed (2nd)

Separated (2nd)

Divorced (2nd)

Fig. 4. Nuptiality States and Transitions.

210 MICHAEL WOLFSON AND GEOFF ROWE

transition probability function of time-varying covariates. The transition

probability functions have been estimated initially from survey data and then
(where possible) adjusted so that LifePaths as a whole reproduces the dis-
tribution of families by marital status observed in Canada’s 1996 population
census.
Some added detail on LifePaths is given in the appendix, and further
information is available at www.statcan.ca/english/spsd/LifePaths.htm.

EXOGENOUS SCENARIOS

Given this background discussion, we turn to the heart of this analysis.

First, in order to assess the impact of exogenous changes on the projected
inter-generational profiles of major government programs on Canadian
birth cohorts, a series of four scenarios characterizing the socio-economic
milieu have been constructed – high and low mortality, and high and low
economic circumstances.

Mortality

As noted above, the mortality process is simulated in terms of hazard rates.

These hazards, or transition probability functions, are differentiated by
birth year, age, sex, as well as institutional status. Up to 2004, observed data
are used.
Under the high mortality/lower life expectancy scenario, Statistics
Canada’s ‘‘high mortality’’ scenario in the official demographic projections
(Statistics Canada, 2005) is used up to 2051, the end of that projection
period. For those members of birth cohorts who survive beyond 2051 in
these simulations, mortality rates remain constant at their 2051 levels.
For the lower mortality/higher life expectancy scenario, the low age-sex
specific mortality hazards from the demographic projections were used
from 2005 to 2051. For years after 2051, mortality hazard rates (differen-
tiated by age and sex) were assumed to continue improving at the constant
rate embodied in the demographic projections for the interval from 2050
to 2051.14
These two scenarios result in considerable divergence in life expectancy, as
shown in Table 1. Life expectancies for each sex differ by about three years
for the cohort born in the 1990s, and by almost five years for those born in
the 2002–2011 interval.15
Aging and Inter-Generational Fairness 211

Table 1. High and Low Life Expectancy Scenarios.

Simulated Life Expectancies

Cohort Scenario Females Males

Born in 1992–2001
Low life expectancy 87.1 82.2
High life expectancy 89.9 85.2
Difference 2.9 3.0

Born after 2001

Low life expectancy 87.6 83.1
High life expectancy 92.2 88.0
Difference 4.6 4.9

Strength of the Economy

The other major axis of exogenous uncertainty that will be considered is the
strength of the economy. This can be conceptualized in a variety of ways, but
for this analysis we focus on employment. A ‘‘strong’’ economy is defined as
one where employment levels are higher; while a weak economy is one where
employment is lower.
In order to implement these alternative employment scenarios, advantage
was taken of the fact that the employment dynamics module in LifePaths is
based on a set of mutually interacting transitions, each one corresponding to
an arrow in Fig. 5, where these rates are in turn functions of calendar year
dummy variables for each of the years from 1976 to 2004, as well as a range
of other factors (age, sex, educational attainment, duration in employment
state, province of residence, presence of a spouse, and spouse’s employment
status). The year dummy variables appear both on their own, and as inter-
action terms respectively with age (linear and quadratic terms, age 65, age
65+), presence of pre-school children, low education level (oHigh School),
and presence of pre-school children
low education level.
Given this detailed and richly specified structure for the employment
dynamics module, the ‘‘high employment’’ scenario is based simply on the
assumption that the most recent business cycle peak year, 2004, will apply for
the years 2005 and on when evaluating the set of employment state transition
probability functions as a ‘‘fixed effect’’ for all future periods.
The ‘‘low employment’’ scenario makes an analogous assumption, but in
this case uses the year of the most recent business cycle trough, 1993, as a
fixed effect for all future periods after 2011. Actual transition hazards for
212 MICHAEL WOLFSON AND GEOFF ROWE

Paid
employee

Self Not
employed employed

Fig. 5. Employment States and Transition Possibilities.

$ 20,000

$17,500

$15,000

Low Life Expectancy Low Employment

$12,500 High Life Expectancy Low Employment
Low Life Expectancy High Employment
High Life Expectancy High Employment
$10,000
1970 1980 1990 2000 2010 2020 2030 2040 2050

Fig. 6. Simulated Earnings (2001$) per Capita for Mortality and Employment
Scenarios.

1994–2004 are still used. From the 2004 business cycle peak, a smooth tran-
sition has been assumed for the dummy variables to the 1993 business cycle
trough over the interval 2005–2011. After that, the dummies remain constant
at their 1993 levels.
Fig. 6 shows the effects of these two employment scenarios on per capita
earnings. For most of the projection period, the divergence in employment
Aging and Inter-Generational Fairness 213

transition dynamics, between those of the 2004 peak and the 1993 trough
business cycle years, amounts to over $2,000 (in 2001 dollars) or over 10%
of per capita earnings.
Fig. 6 also shows the effects on earnings of the two life expectancy
scenarios. Since most of the variation in mortality rates arises after prime
working ages, and the differences do not affect the total population that
greatly (even though they change life expectancy by as much as almost five
years), there is no appreciable difference in per capita earnings as a result of
differences in life expectancy.

PROVISION FOR RETIREMENT IN THE CANADIAN

SYSTEM

In addition to incorporating a wide variety of socio-economic characteris-

tics, this LifePaths analysis must also realistically model Canada’s major tax
and transfer programs. In this section, we first give a brief overview of
Canada’s current public pension system, and then describe the taxes and in
kind transfers that have also been explicitly modeled.
Canada’s public pension system is often described as having three tiers.
The first tier is a pair of cash transfers to the elderly (generally age 65+)
based only on their current income, and financed out of general taxation.
One is a taxable ‘‘demogrant’’ called the Old Age Security (OAS) pension. It
started in 1952 paying monthly benefits to those over age 70. This entitle-
ment age was subsequently lowered to 65, and at July 2006 rates, annual
benefits are $5,904 per year (subject to a sufficient period of prior residency).
The other basic cash transfer is an income-tested benefit, the Guaranteed
Income Supplement (GIS) program, starting in 1967, and an extension, the
Spouse’s Allowance (SPA) program, starting in 1976. They provide non-
taxable monthly benefits to Canadians age 65 and over (or age 60–64 with a
spouse age 65 or over in the case of SPA). Together, these major programs
provide basic income guarantees of up to $14,139 and $18,884 for Canada’s
senior individuals and couples in 2006. As a result of a number of ad hoc
increases over the years, their combined benefit levels have become such that
very few of Canada’s elderly have incomes below the ‘‘low income line’’.
The next tier is the Canada and Quebec Pension Plans (C/QPP), an
earnings-related public pension plan that pays out a retirement pension
essentially equal to 25% of average (updated) pre-retirement earnings.
(While there are two plans, one for Quebec and one for the rest of Canada,
214 MICHAEL WOLFSON AND GEOFF ROWE

they have virtually identical contribution and benefit provisions, and are
completely integrated from the viewpoint of individuals moving between
them.) The maximum pension in 2006, $10,135, is based on 25% of the
average over the past five years of the year’s maximum pensionable earnings
(YMPE). This in turn is equal to the average annual wage in Canada, and
was $40,500 in 2004.16
The plans are financed by a payroll tax. Until recently, the rate was set so
as to assure a reserve equal to the payout of about two years of benefits.
Recently, the payroll tax rate has been increased so that accumulated funds
are projected to rise to about 4.4 times benefits in 2010, and about 6.3 times
in 2050 (OSFI 2005b). Still, overall, the plans remain funded essentially on a
pay-as-you-go basis.
The third tier of Canada’s public pension system is a set of tax incentives
for private saving for retirement, either via individual accounts called
Registered Retirement Savings Plans (RRSPs) or employer-sponsored plans
(Registered Pension Plans or RPPs). The tax expenditure (foregone income
tax revenue) in respect to these provisions amounted to about $19 billion
in 2005 (Canada, 2006), while the total cost of OAS/GIS was about
$30 billion in the same year (OSFI, 2005a), and C/QPP also paid out about
$30 billion (for retirement and survivor pensions, based on OSFI 2005b).
Thus, income tax incentives are a significant component of the public system,
and they are used disproportionately by those in upper income brackets.
Beyond the public system, and the significant volume of private saving
accumulated under registered plans for retirement purposes, home owner-
ship is a significant form of de facto saving for retirement. However, about
half of all Canadians enter retirement without owning a house, and with
relatively little in the way of accumulated savings of any form. They are
therefore highly dependent on the public pension system.17
Beyond public pensions and other provisions for retirement, this analysis
also takes explicit account of personal income taxes (both federal and pro-
vincial), payroll taxes, (un)employment insurance transfers, and in kind edu-
cation and health care benefits. In the latter two cases, benefits are imputed
based on highly simplified formulae. Health care in kind service benefits are
assumed to vary only by age and sex, while education in kind service benefits
vary only by the type of educational institution attended (Cameron &
Wolfson, 1994). These unit costs are projected simply in line with the growth
in average wages. It is, of course, well known that non-demographic factors
are typically far more important in determining the trends in these unit costs
(Evans, McGrail, Morgan, Barer, & Hertzman, 2001). However, considera-
tion of these factors is beyond the scope of this analysis.18
Aging and Inter-Generational Fairness 215

POLICY SCENARIOS

As noted earlier, the core of this analysis is the simulation of the impacts of
Canada’s major age-sensitive tax/transfer programs on inter-generational
fairness according to several basic norms. The four scenarios to test the
robustness of the policy scenarios to unknown future uncertainties (high and
low life expectancy and high and low employment) were outlined above.
From the policy perspective, we consider two stylized alternative policy
scenarios in addition to the status quo scenario, as follows.

Extended Work

One approach to the aging of the population is to redefine retirement,

essentially by raising the age of entitlement to public pensions. At the time
von Bismark first set an age for pensions in 1889 at age 70 (it was subse-
quently lowered to age 65 in 1916, see SSA http://www.ssa.gov/history/
ottob.html) hardly anyone even survived to that age. In 1983, the U.S.
amended the law so that the age of entitlement for full Social Security re-
tirement pensions would begin rising gradually to age 67.19 More recently,
Mankiw (2006), for example, has called for further change in this direction.
In Canada, the Lazar Report (Canada, 1980) considered whether the age of
entitlement to public pensions (both C/QPP and OAS) should be raised
gradually,20 but concluded that the uncertainties were such that instead,
some sort of trigger criteria for considering this kind of change, with a
10-year lead time, would be more appropriate. The criteria could include
dependency ratios, taxpayer burdens, or labor force participation rates
(Canada, 1980, p. 328). More recently, the Swedes reformed their public
pension system in a way that implicitly indexes pension benefits to life ex-
pectancy – by requiring that the benefit about to come into pay at the point
of formal retirement be based on an actuarial annuity calculation, in turn
based on whatever mortality rates are then currently projected (Flood, 2003).
In order to reflect this broad class of possible options for responding to
concerns about the costs of public pensions, we have defined an ‘‘extended
work’’ scenario. This is implemented in the model by ‘‘delayed aging’’
of persons for purposes of employment transitions and for public pension
(C/QPP, OAS, and GIS/SA) eligibility and take-up. This delayed aging
occurs not only in future, but also in the past. It is as if the policy of
increasing the age of entitlement to public pensions in Canada had started in
1976, and proceeded at a very gradual rate until 2005 when everyone’s
entitlement would be at age 70.
216 MICHAEL WOLFSON AND GEOFF ROWE

More specifically, delayed aging is implemented by creating an artificial

alternative age variable to be used by all the appropriate simulation modules,
including both those governing behavioral dynamics and those determining
program eligibility and participation. This alternative age variable kicks in at
age 55 and remains fixed at age 55 for as long as five years. The onset age of 55
was chosen because it is an age where employment rates are high. In effect, by
starting the ‘‘delayed aging’’ process at age 55, the decline in age-specific
employment to population ratios observed after age 55 could be reasonably
attenuated and thereby generate a more realistic scenario in line with the likely
behavioral impacts of increasingly delayed entitlement to public pensions.
The general intention is to implement delayed aging gradually: first,
delaying the aging of the cohort that turned 55 in 1976 by two months; then,
delaying the aging of the cohort that turned 55 in 1977 by four months, and
so on – gradually implementing the full five-year delay over a period of 30
years (1976–2005). However, many of the relevant modules in the LifePaths
model are not sensitive to fractional ages. As a result, the gradual 30-year
phase-in was implemented by: first, delaying the aging of only a (randomly
chosen) 1/6th of the cohort that turned 55 in 1976 by one year; then, delaying
the aging of only a (randomly chosen) 2/6th of the cohort that turned 55 in
1977 by one year; y; delaying the aging of all of the cohort that turned 55 in
1981 by one year; delaying the aging of all of the cohort that turned 55 in 1982
by one year and further delaying the aging of a (randomly chosen) 1/6th of
that cohort by an additional one year; y and so on. All members of cohorts
that turn 55 on or after 2005 experience delayed aging of a full five years.21
In terms of the norms for inter-generational equity outlined earlier, this
policy scenario is a rather arbitrary and approximate response to population
aging. It does not offer an explicit adjustment mechanism, as in the Swedish
reforms, that automates and thereby clarifies for successive cohorts the
‘‘inter-generational contract’’ with regard to retirement pensions.

Relative Indexing

Another broad set of approaches to concerns about growing public pension

costs in the face of population aging is to change the way the dollar value of
pensions is updated from one year to the next, i.e. the indexing provisions. For
example, both the Musgrave (1981) and the House of Commons (1983) norms
for inter-generational fairness described above reflect forms of indexing.
Canada has a somewhat complex and unusual system of indexing
provisions. For the major earnings-related public pension, C/QPP, benefits
Aging and Inter-Generational Fairness 217

are implicitly indexed to AW during working years. Then, after retirement

when the pension comes into pay, it is indexed by the CPI. However, in
Canada the earnings-related pension constitutes less than half of all the
publicly provided old age cash benefits. The other major programs, OAS
and GIS/SA, are indexed to the CPI. Moreover, the OAS is now subject to a
degree of income testing, and the threshold where the income testing begins
is itself indexed to the CPI. As a result, any real average wage growth results
in these public pensions declining relative to the size of the economy.
In contrast, the U.S. public pension system, for example, which is dom-
inated by Social Security, is much closer to being fully wage indexed (even
though, as with Canada’s C/QPP, pensions in pay are indexed to the CPI).
Similarly, in many European countries, retirement pensions were substan-
tially wage indexed, though over the last decade a number have moved away
from AW indexing and closer to CPI indexing.
On the one hand, CPI rather than AW indexing means that in future,
pension costs will be lower. On the other hand, CPI compared to wage
indexing means that for pensioners, their pension incomes will be lower, and
they will be more likely to fall below a low income line.
The importance of indexing provisions has been well known to the eco-
nomically informed for decades. For example, as early as the 1980s, officials
in the IMF (Heller, Hemming, & Kohnert, 1986) projected that Canada was
relatively unique in not facing a pension affordability problem – essentially
because of the CPI indexing of the OAS and GIS. However, there have been
virtually no analyses showing the counterpart implication of falling relative
individual income levels among the future elderly (with the exceptions of
Murphy & Wolfson, 1991; Wolfson & Murphy, 1997).
Given the importance of indexing scenarios from these earlier analyses, the
second stylized and illustrative policy scenario explores wage indexing as an
alternative to the current price indexing of major government programs –
including not only public pensions but also the income tax system and its
associated set of refundable income tax credits which de facto are very much
like cash transfers. This scenario, in principle, also moves the system in a
direction that is much closer to most of the norms for inter-generational
equity outlined above. The shift to wage indexing is assumed to occur in 2001.

SIMULATION RESULTS

In this section we present the main results from a series of LifePaths sim-
ulations, based on the exogenous and policy scenarios just described. In all
218 MICHAEL WOLFSON AND GEOFF ROWE

cases, the focus is on the net balance between income and payroll taxes paid,
and cash transfers plus in kind health and education benefits received. These
are lifetime net balances, summed over representative samples of individuals
in each decadal birth cohort.
The sums are net present discounted values, where the discount rate is the
same as the growth rate of average wages. This growth rate, in turn, is
assumed at 1% per annum – slightly lower than the 1.1% and 1.2%
assumed by the Chief Actuary in his previous two actuarial reports on the
CPP (OSFI, 2005b).22
Graph 1 shows these lifetime net present values (NPVs) (2001 $000s) on
the vertical axis, for each decadal birth cohort along the horizontal axis. The
different curves correspond to four different points in the NPV distributions
for each birth cohort – the first quartile (Q1), the median, the mean, and the
third quartile (Q3). In this case, we have shown the ‘‘base’’ policy scenario,
and the low life expectancy and low employment exogenous scenarios.23
To begin with the middle of the distribution of lifetime NPVs, the median
curve peaks with the 1920s and 1930s birth cohorts at over $150,000. These
are the birth cohorts who experienced the first benefits from the fully phased

400.0

300.0

200.0

100.0

0.0
1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1990s 2000s
1980s
(100.0)
Q1
Median
(200.0) Mean
Q3
(300.0)

(400.0)

Graph 1. Net Present Values of Lifetime [TransfersTaxes]; Scenario ¼ le emp

base.
Aging and Inter-Generational Fairness 219

in C/QPP (introduced in 1966 and fully phased in by 1976), after having

made minimal payroll tax ‘‘contributions’’, as well as being the first
generation to benefit from improvements to the OAS and the introduction
of the GIS. But after this point, median lifetime NPVs tail off, and become
slightly negative for the cohort being born in the current decade.
From this perspective, the status quo system (under the specific ‘‘low–low’’
exogenous life expectancy and employment scenario) is not balancing the
first two fairness norms outlined above – successive birth cohorts after the
1920s are receiving declining net transfers.
But this graph also indicates dramatic differences across individuals
within each birth cohort. The mean NPV is generally lower than the median.
This reflects a negative skewness in the NPV distribution, and in turn the
fact that taxes are unbounded above, while cash and in kind transfers are
bounded below at zero.
More importantly, the curves in Graph 1 for the first and third quartiles
show a very wide dispersion in NPVs, on the order of $400,000. This is an
extraordinarily clear support of the old adage, ‘‘beware of the mean’’. The
usual analyses of inter-generational fairness (e.g. Kotlikoff, 1992), which
are based on representative agents, completely ignore these tremendous
variations within generations.
Graph 2 reinforces this point by showing two measures of dispersion, the
inter-quartile range for the same low–low scenario as in Graph 1, and
the standard deviation of NPVs, for each birth cohort. Additionally, for the
standard deviations, four different curves are plotted, one for each of the
exogenous scenarios.24
Two key messages arise. The first is that both indicators of dispersion give
similar results – there is a great deal of heterogeneity in individual circum-
stances within each birth cohort. Indeed, that variation within birth cohorts
is far larger than that between birth cohorts (e.g. the mean or median)
shown in Graph 1.
Second, the extent of this variation is essentially unaffected by which of
the exogenous scenarios is chosen. Whether life expectancy is high or low,
and whether employment is high or low, has almost no effect on the dis-
persion of lifetime NPVs.
A major growth of income taxes occurred in the 1940s war years and
subsequently with the growth of the ‘‘welfare state’’. Similarly, government
expenditures on health, education, and old age pensions and benefits became
significant only in the 1960s. As a result, the early cohorts starting with those
born in the 1890s had little opportunity over their lifetimes to be ‘‘exposed’’
to large government tax/transfer programs. In turn, the opportunity for
220 MICHAEL WOLFSON AND GEOFF ROWE

500

450

400

350

300

250

q3 - q1
200
le emp base
le EMP base
150
LE emp base
LE EMP base
100

50

0
1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s

Graph 2. Inter-Quartile Range of NPV of Tax – Tran and Standard Deviation of

NPVs by Exogenous Scenario, Base Policy Scenario.

dispersion in the resulting lifetime net benefits was relatively small for the
1890s cohort, grew significantly up to the 1930s birth cohort, and has con-
tinued to grow for later cohorts, but much more slowly, as the welfare state
programs to which these cohorts have been and are projected to be exposed
are largely mature.
While Graph 2 shows little effect from the exogenous life expectancy and
employment scenarios, Graph 3 shows that sex does make a big difference.
In this case, we are looking at the same scenario set as in Graph 1 – the
status quo policy scenario, and the low–low exogenous scenario, though for
clarity the mean NPV curves have been dropped. Graph 3 therefore shows
six curves: the three quartiles for males and for females.
The dashed lines for females are almost everywhere above the solid lines
for males. The median NPVs for females in the 1920s and 1930s birth
cohorts are on the order of $300,000 higher than those of their male
counterparts, though this declines to about $200,000 for the current decadal
birth cohort.
Aging and Inter-Generational Fairness 221

500

400

300

200

100

0
1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s
-100

-200 m-q1
m-med
-300 m-q3
f-q1
f-med
-400 f-q3

-500

Graph 3. Males and Females by Quartile of NPV; Scenario ¼ le emp base.

First quartile males (those with high incomes so therefore paying more
income tax and receiving less transfers) end up with NPVs of $300,000 or
lower for the 1970s and successive birth cohorts, while the corresponding
birth cohorts of females have NPVs about $200,000 or higher. These massive
transfers from men to women are intuitively plausible when one considers
that with women’s greater life expectancy and generally higher morbidity,
they consume more in kind health care services and receive more survivor
and old age demogrant (OAS) benefits; and with women’s generally lower
incomes, they pay less in taxes and receive more in income-tested benefits.
Even though the exogenous scenarios have virtually no effect on the dis-
persion in lifetime NPVs (Graph 2 above), they do have an impact on the
typical net present value of taxes minus transfer for successive birth cohorts.
Graph 4 shows the impacts on the mean (solid lines) and the median (dashed
lines) NPVs of the high and low life expectancy scenarios (‘‘LE’’ and ‘‘le’’,
respectively) and high and low employment scenarios (‘‘EMP’’ and ‘‘emp’’,
respectively) for the status quo (‘‘base’’) policy scenario (for both sexes
combined).
The main result here is that the strength of the economy has a much larger
impact than the pace of improvement in life expectancy. Not surprisingly,
higher life expectancy increases the NPVs of transfers minus taxes, as indi-
viduals living longer have more years of entitlement to public pension benefits
and use more health care services, but are not paying correspondingly more
222 MICHAEL WOLFSON AND GEOFF ROWE

200

150

100

50

0
1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s
(50)

le emp base
(100) le EMP base
LE emp base
(150) LE EMP base
le emp base
le EMP base
(200)
LE emp base
LE EMP base
(250)

(300)

Graph 4. Medians and Means by Exogenous Scenario.

income and payroll taxes over their longer lifetimes. On the other hand, a
stronger economy reduces NPVs as income taxes in particular are higher,
while the demogrant (OAS) and income tested (GIS) portions of the public
pension system are relatively unaffected.
In quantitative terms, for the baby boom and subsequent birth cohorts, a
change in life expectancy of as much as five years has impacts on lifetime
NPVs that are only about one fifth as large as an improvement in employ-
ment with the effect of raising per capita wages by about 10%.
Finally, Graph 5 shows how the two stylized policy alternatives compare.
Since the low and high employment scenarios had a much larger impact
than the high and low life expectancy scenarios, we focus on only the two
exogenous employment scenarios (both assuming low life expectancy
change). The light lines show the status quo (‘‘base’’) scenarios; the dashed
lines show the scenario where individuals work longer and the age of
entitlement to pensions rises gradually from 65 to 70 over the period
1976–2005; and the heavy lines show the scenario where indexing has been
shifted from a price index basis to an index of average wages, while leaving
the age of entitlement at 65.
The most dramatic result here is the relatively weak impacts of the work-
ing longer/delayed retirement scenarios compared to the wage indexing
Aging and Inter-Generational Fairness 223

200

150

100

50

0
1890s 1900s 1910s 1920s 1930s 1940s 1950s 1960s 1970s 1980s 1990s 2000s
(50)

(100) le emp base

le emp WORK
le emp INDEX
(150) le EMP base
le EMP WORK
(200) le EMP INDEX

(250)

Graph 5. Lifetime NPVs for Three Policy and Two Employment Scenarios.

scenarios. The shift to delayed retirement (from light solid to dashed lines) is
much smaller than the shift from price to wage indexing (from light to heavy
solid lines) – on the order of a reduction of $20,000 compared to an increases
well over $100,000 in lifetime NPVs of transfers minus taxes. Given the
increasing policy discourse in many quarters on the importance of delaying
retirement, these results are a sobering indication that such a change may not
be as ‘‘helpful’’ as many expect.
Rather, and as shown earlier, an improvement in the economy (from
‘‘emp’’ to ‘‘EMP’’) is quantitatively much more important, and tends to
shift all the curves up on the order of $100,000. Thus, a stronger economy
(of the order posited in the exogenous scenarios simulated here, in turn
based on the actual low and high points in the Canadian economy since the
1990s) turns out roughly to offset the shift from price to wage indexing.
Finally, in terms of the basic focus of this analysis, on inter-generational
fairness, the shift from price to wage indexing has the effect of leveling out
the curves of NPVs across generations. (Recall that these NPVs are dis-
counted, based on the growth rate of average wages.) In terms of the norms
of inter-generational fairness outlined above, these wage indexed scenarios
appear most in accord.
224 MICHAEL WOLFSON AND GEOFF ROWE

CONCLUDING COMMENTS

Population aging is more often a source of concern for public policy than
cause for celebration. One reason is the expectation that future cohorts of the
elderly, particularly the post WW II baby boom generation, will place intol-
erable burdens on future working age generations in their retirement years in
order to finance their public pensions and insured health care services. How-
ever, this analysis suggests the opposite. Under current program rules, and a
range of scenarios for future economic growth and longevity, birth cohorts
after those born in the 1920s and 1930s will experience successively smaller
lifetime net transfers (both cash, and in kind for health and education).
The main factor underlying this (perhaps) unexpected result is the index
broadly used to update cash transfer benefit levels and income tax thresholds
and other related parameters. This index is the CPI. However, if history is
any guide, nominal wages and the economy more generally will most likely
grow faster than inflation. In other words, we can likely expect real per capita
economic growth. In such scenarios, CPI-indexed benefits will gradually
shrink relative to the average incomes of those of working age, and
CPI-indexed tax bracket thresholds will result in taxpayers gradually finding
themselves in ever higher tax brackets (‘‘bracket creep’’), hence paying a
larger proportion of their incomes in income tax.
While there is great concern about the effects of increasing longevity on
pension costs and hence on inter-generational fairness, our simulations
suggest that for quite a wide range of life expectancy scenarios, this has a
much smaller impact than the strength of the economy – judged by the range
of employment over the most recent business cycle – specifically the 1993
trough and the 2004 peak.
Also, notwithstanding the focus of this analysis on widely expressed
concerns regarding inter-generational fairness, our results show that differ-
ences within generations are far larger than those between generations.
Women’s net lifetime transfers minus taxes are hundreds of thousands of
dollars greater than those for men, while the differences between the poor
and the rich within any given generation are larger still.
Finally, one of the most widely discussed responses to population aging,
in the context of public pensions, is raising the age of entitlement and
otherwise encouraging delayed retirement. One of the least discussed issues,
on the other hand, is the nature of the indexing of pensions as well as income
taxes. That these government programs should be indexed to the CPI is
largely taken for granted. However, our analysis suggests that the impact of
Aging and Inter-Generational Fairness 225

the indexing provisions is far larger than delaying the age of entitlement
from age 65 to 70.
Continuing with CPI indexing results in a continuing fall from one gene-
ration to the next in the net lifetime value of transfers minus taxes. Moving to
wage indexing results in a leveling off of these net values – a situation much
more in accord with the norms for inter-generational fairness found in both
the economics literature and in major Canadian policy documents. Still,
under the scenarios examined here, cohorts born after the 1930s never receive
the net transfers minus taxes that the 1920s and 1930s birth cohorts do.

NOTES
1. Of course, inter-generational transfers occur in many ways, from unrequited
payments among family members of different generations, to broader investments
(or dis-investments) in the productive capacity of the economy and the quality of the
environment. The latter are beyond the scope of this analysis, even though they may
enter into policy debates about the inter-generational fairness of the taxes and
transfers being examined here.
2. A third major axis, which is beyond the scope of this analysis, is the future
health or disability status of the population, over and above life expectancy. This has
been treated in Wolfson and Rowe (2004).
3. Employment is only one of several ways to measure the strength of the econ-
omy. Another would be per capita economic growth rates, in turn usually linked to
productivity growth. Similarly, there is considerable interest in the role of immigra-
tion among demographic factors. Alternative scenarios of this sort are easily feasible
with LifePaths, the analytical tool being used. The focus here on life expectancy and
employment is simply for the convenience of a manageable range of scenarios for this
initial exploratory analysis.
4. A widely agreed norm for intra-generational fairness is progressivity, i.e. that
the tax/transfer system is generally redistributive from those with higher to those
with lower (more often contemporaneous than lifetime) income.
5. For example, simplifying assumptions include no heterogeneity within gener-
ations, generations not overlapping each other in time, and the existence of well-
defined smooth precise utility functions in the first place. Indeed, the thrust of the
Basu and Mitra analysis is to show formally the impossibility of a social welfare
function that at one and the same time obeys the axiom of anonymity, which they
equate with inter-generational equity, and even the weakest of Pareto principles
which would allow future generations to be better off than their parents – a prop-
osition which seems intuitively obvious.
6. ‘‘(t)hose now working could build up a moral claim on future pension enti-
tlements by making transfers to the current elderly of at least the same magnitude as
they would expect to receive when their time came. This would set in motion a kind
of intergenerational golden rule’’ (House of Commons, 1983).
226 MICHAEL WOLFSON AND GEOFF ROWE

7. This is essentially identical to the concept of ‘‘sustainable development’’ artic-

ulated by the Brundtland Report (Bruntland et al., 1987, p. 27), defined as devel-
opment that ‘‘meets the needs of the present without compromising the ability of
future generations to meet their own needs’’.
8. Personal communication with Wolfson.
9. Indeed, Wolfson et al. (1998) show such an example. Also, as one referee has
commented, majority support for a rule or norm is not necessarily an assurance that
it is fair.
10. This way of framing the issue is more in line with current discussions of the so-
called ‘‘crisis’’ in private defined benefit pension plans, though in that case, there is
the added factor of the differential risks borne by the employer as plan sponsor, and
workers as plan beneficiaries.
11. The ‘‘Boskin Committee’’ for the U.S. Senate (Advisory Committee, 1996)
highlighted the importance of this question in the context of the U.S. Social Security
system. Wolfson (1999), however, has shown the practical impossibility of ever con-
structing a CPI that will suffice for inter-generational comparisons given that the
time spans involved are such that major ‘‘new goods problems’’ are inevitable.
12. The term ‘‘metasynthesis’’ is used in contrast to the epidemiological term
‘‘meta-analysis’’, which refers to the combination of results from a number of data
sets, all of which pertain to the same question. Metasynthesis, in contrast, refers to
the combination of results from data sets covering diverse subject matters. The
estimation of GDP and the range of tables comprising the System of National
Accounts (SNA) is an excellent example of metasynthesis – drawing together data
from a wide variety of sources in order to produce the best possible estimate of a
given set of concepts. The main difference is that the concepts in the SNA are
aggregate, while those which are the object of LifePaths are micro and distributional
– a representative sample of individual life course trajectories.
13. Although the diagram implies that time is discrete, LifePaths represents and
models all events in continuous time.
14. Since the mortality rates underlying the demographic projections are based on
a smooth mathematical function, using the rate of change over the last single year is
reasonable and convenient for extrapolating the underlying longer term trend.
15. The Chief Actuary’s projections in his most recent report (OSFI, 2005b) are
80.7 and 84.1 for males and females, respectively. Note that Chief Actuary’s are
period life expectancies in 2050, rather than cohort life expectancies used here, in
turn based on mortality projections extending to the end of the 21st century.
16. The C/QPP also provide pre-and post-retirement survivor pensions, orphan
and disability pensions, and a lump sum death benefit. However, in this analysis we
consider only the retirement pension and the post-retirement survivor pension. These
comprise about three-quarters of the total benefits provided by the plans.
17. Note that while private tax-assisted employer-sponsored pension plan saving
(RPPs) is explicitly modeled in LifePaths, private tax-assisted retirement savings via
RRSPs and home ownership are not.
18. Future versions of this analysis could build on the EUs projection approaches
which explicitly model improvements in health status and hence declining age-
specific health care costs, for example. See DG ECFIN (2005).
Aging and Inter-Generational Fairness 227

19. This is based on a phase in according to an individual’s year of birth – age 65 if

born in 1937 or earlier, then rising in two-month steps per year of birth to age 67 if
born in 1960 or later.
20. The report strongly recommended against any lowering of this age.
21. It should be noted that this scenario increases income and payroll tax rev-
enues, as a result of greater employment in the 55–70 age range, and reduces payouts
of public pensions in the 65–70 age range. Consequently, the government’s fiscal
balance improves. Notwithstanding, no adjustments are made to income or payroll
taxes or to any other aspect of cash transfers. While this is likely an unrealistic
scenario, it is simpler and aids interpretation of results.
22. The results are sensitive to the choice of discount rate. For this kind of anal-
ysis, from a social and inter-generational perspective, as contrasted for example to an
individual perspective on a short or medium term investment choice, it is arguable
that the growth rate in per capita wages is the most appropriate discount rate. And as
noted earlier, this kind of indexing is closest to enabling assessment of the ‘‘equal
utility’’ type of inter-generational equity norm.
23. Even though the LifePaths simulations cover everyone in Canada, these
graphs cover only persons who survived to at least age 15 and were born in
a province west of Newfoundland. The reason, simply, is that immigrants receive
no transfers and pay no taxes until they arrive in Canada, and Newfoundlanders
receive no transfers and pay no taxes until after Confederation 1949. These two
cases create anomalous spikes in the frequency distribution of transfers minus taxes
at zero.
24. Note that this dispersion does not include the further effects of differential
mortality by income. This is well known, for example Wolfson, Rowe, Gentleman,
and Tomiak (1993). More recently, the Chief Actuary (OSFI, 2006) has estimated
that life expectancy at age 65 in 2001 varies by 4.5 years for males and 3.6 years for
females between those with low incomes (roughly under $10,000) and those with high
incomes (roughly over $50,000).

REFERENCES
Advisory Committee to Study the Consumer Price Index. (1996). (‘‘Boskin Committee’’:
M. J. Boskin, E. R. Duhlberger, R. J. Gordon, Z. Grilliches, & D. Jorgenson), Toward
a more accurate measure of the cost of living. Final Report to the Senate Finance
Committee, Washington D.C., December 4.
Basu, K., & Mitra, T. (2005). Possibility theorems for aggregating infinite utility streams eq-
uitably. CAE Working Paper #05-05; http://www.arts.cornell.edu/econ/cae/05-05.pdf
(referenced Nov 12/06).
Beach, W. W., & Davis, G. E. (1998). Social security’s rate of return from the Heritage Foundation,
http://www.heritage.org/Research/SocialSecurity/CDA98-01.cfm (referenced Nov 12/06).
Bruntland, G. et al. (1987). Our common future, http://www.are.admin.ch/imperia/md/content/
are/nachhaltigeentwicklung/brundtland_bericht.pdf?PHPSESSID=e6d4ef5966d56;fa8532c
fc13cf38e2bb (referenced Nov 12/06); also Oxford University Press.
228 MICHAEL WOLFSON AND GEOFF ROWE

Cameron, G., & Wolfson, M. C. (1994). Missing transfers: Adjusting household incomes for
non-cash benefits. IARIW 23rd General Conference, St Andrews.
Canada. (1980). The retirement income system in Canada: Problems and alternative policies for
reform (Lazar Report), Ottawa.
Canada. (2006). Tax expenditures per http://www.fin.gc.ca/taxexp/2005/taxexp05_2e.html#
Table1, referenced July 27, 2006.
DG ECFIN. (2005). Annex, the 2005 EPC projections of age-related expenditure (2004–2050)
for the EU-25 member states: Underlying assumptions and projection methodologies.
Report prepared by the Economic Policy Committee and the European Commission (DG
ECFIN), Special Report No 4/2005, accessed at http://ec.europa.eu/economy_finance/
publications/european_economy/2005/eespecialreport0405_en.htm, July 27, 2006.
Evans, R. G., McGrail, K., Morgan, S., Barer, M. L., & Hertzman, C. (2001). Apocalypse no:
Population aging and the future of the health care system. Canadian Journal on Aging,
20(Suppl. 1), 160–191.
Flood, L. (2003). Can we afford the future? An Evaluation of the New Swedish Pension System.
http://www.sesim.org/Documents/Flood_Natsem_2003.pdf referenced Nov 12/06.
Heller, P. S., Hemming, R., & Kohnert, P. W. (1986). Aging and social expenditures in the
major industrial countries, 1980–2025. Occasional paper No. 47, International Monetary
Fund, Washington D.C., September 1986.
House of Commons. (1983). Report of the parliamentary committee on pension reform (Frith
Committee), Ottawa.
Kotlikoff, L. J. (1992). Generational accounting: Knowing who pays, and what, for what we spend.
NY: Free Press.
Kotlikoff, L. J., & Burns, S. (2004). The coming generational storm: What you need to know
about America’s economic future. Cambridge, MA: MIT Press.
Mankiw, N. G. (2006). Mr Paulson’s Challenge. Wall Street Journal, May 31, A12.
Murphy, B., & Wolfson, M. C. (1991). When the baby boom grows old: Impacts on
Canada’s public sector. Statistical Journal of the United Nations Economic Commission
for Europe, 8(1).
Musgrave, R. A. (1981). A reappraisal of financing social security. In: F. Skidmore (Ed.), Social
security financing. Cambridge, MA: MIT Press.
Oreopoulos, P., & Vaillancourt, F. (1998). Applying generational accounting to Canada:
Findings and fallacies. In: M. Corak (Ed.), Government Finances and Generational
Equity, Statistics Canada Cat. No. 68-513-XPB, Ottawa.
OSFI. (2005a). Actuarial Report (7th) on the Old Age Security Program as at 31 December,
2003, http://www.osfi-bsif.gc.ca/app/DocRepository/1/eng/reports/oca/oas7_e.pdf,
referenced July 27, 2006.
OSFI. (2005b). Actuarial Report (21st) on the Canada Pension Plan as at 31 December, 2003,
http://www.osfi-bsif.gc.ca/app/DocRepository/1/eng/oca/reports/21/CPP2104_e.pdf,
referenced July 27,2006.
OSFI. (2006). Old age security program mortality experience, Actuarial Study No. 5, http://
www.osfi-bsif.gc.ca/app/DocRepository/1/eng/oca/studies/Mortality_Exp_No5_e.pdf,
referenced July 27, 2006.
Ramsay, F. P. (1928). A mathematical theory of savings. Economic Journal, 38, 543–559.
Statistics Canada. (2005). Population projections for Canada, provinces and territories,
Catalogue no. 91-520-XIE.
Statistics Canada – LifePaths: www.statcan.ca/english/spsd/LifePaths.htm
Aging and Inter-Generational Fairness 229

Wolfson, M. C. (1996). LifePaths: A new framework for socio-economic accounts. IARIW 24th
General Conference, Lillehammer, Norway, August.
Wolfson, M. C. (1999). New goods and the measurement of real economic growth. Canadian
Journal of Economics, 32(2), 447–470.
Wolfson, M. C., Rowe, G., Gribble, S., & Lin, X. (1998). Historical generational accounting
with heterogeneous populations. In: M. Corak (Ed.), Government Finances and Gene-
rational Equity, Statistics Canada Cat. No. 68-513-XPB, Ottawa.
Wolfson, M. C., & Murphy, B. (1997). Aging and Canada’s public sector: Retrospect and
prospect. In: K. Banting & R. Boadway (Eds), Reform of retirement income policy:
International and Canadian perspectives. Kingston: School of Policy Studies, Queens
University.
Wolfson, M. C., & Rowe, G. (2004). Disability and informal support: Prospects for Canada. In:
S. B. Cohen & J. M. Lepkowski (Eds), Eighth Conference on Health Survey Research
Methods, National Center for Health Statistics, Hyattsville, MD. (http://www.cdc.gov/
nchs/data/misc/proceedings_hsrm2004.pdf referenced March 5, 2006).
Wolfson, M. C., Rowe, G., Gentleman, J., & Tomiak, M. (1993). Career earnings and death: A
longitudinal analysis of older Canadian men. Journal of Gerontology: Social Sciences,
48(4), S167–S179.

APPENDIX: DATA AND METHODS

This analysis draws on extensions to the LifePaths (Wolfson, 1996) family of

models being developed at Statistics Canada. These are dynamic Monte
Carlo microsimulation models which generate representative population
cohorts. The cohorts are built up as longitudinal samples of millions of syn-
thetic but highly realistic individual biographies or life paths – particularly
in respect to their educational participation and attainment, employment,
earnings, fertility, nuptiality, government taxes and transfers, and mortality
trajectories over their lifetimes – hence their LifePaths.
The analysis starts with the cohort born in the 1890s, and extends for two
centuries, to the ultimate demise of the children being born in the 2002–2011
decade. A major effort has been made to ground the analysis using quan-
titative data. However, the combination of an absence of detailed historical
data, with the need to make long-run projections, means that relatively
stylized representations of the main socio-demographic processes and
components of Canada’s tax/transfer system have had to be used.
LifePaths is in a constant state of development and refinement. The most
recent version was employed in the analysis reported here. However, a rea-
sonably up-to-date description of most of the components dealing with
demography, education, employment, and earnings can be found on the
Statistics Canada website at www.statcan.ca/english/spsd/LifePaths.htm.
230 MICHAEL WOLFSON AND GEOFF ROWE

Components of the model employed for this study that are not described
on the website include Income Taxes and Cash Transfers, Other Sources of
Income, and In Kind Transfers. The following provides a brief description
of these:

Income Taxes and Cash Transfers

Federal income taxes have been implemented explicitly using historical tax
regulations. Structural changes through time and legislated changes have
been implemented, as have surtaxes and surtax reductions from the inception
of the income tax in Canada in 1917 to present. Income taxes are calculated
at year end using the simulated detail on each individual’s income: including
income from working (both employment and self-employment), pension in-
come (CPP/QPP Retirement Benefits, CPP/QPP Survivors Benefits, CPP/
QPP Death Benefits, RPP Benefits, OAS Benefits, and Unemployment/
Employment Insurance (UI/EI) Benefits). Net Income is determined after
repayment of social benefits (Family Allowances, OAS Benefits, UI/EI
Benefits). Deductions and exemptions (recently converted to non-refundable
income tax credits) accounted for include basic personal amount, age
amount, pension income amount, married and equivalent to married
amount, dependent amount, education amount, CPP/QPP contributions,
and UI/EI Premiums. Refundable income tax credits that have been taken
into account explicitly include: Child Tax Credit, Federal Sales Tax Credit,
and Goods and Services Tax Credit. Major sources of provincial transfer
income have been included: Provincial Family Allowances, Quebec Newborn
Allowance, and Quebec Child Supplements. However, provincial income
taxes have been modeled simply as a weighted proportion of basic federal
income taxes. The main programs that are currently not implemented are
Provincial GIS top-ups and Provincial Child Tax Benefit Programs. Current
CPI or CPI – 3% partial indexing is assumed to continue into the future,
under one of the scenarios to be considered. This is a critical assumption, as
shown in Wolfson and Murphy (1997).

Other Sources of Income

Selected special sources of income are imputed at year end just before the
year’s income tax calculation takes place. These include components of
income that are otherwise difficult to model: provincial Social Assistance
(‘‘welfare’’), workers compensation, veterans’ benefits, investment and
Aging and Inter-Generational Fairness 231

dividend income, and alimony. The imputation equations were estimated

from census microdata, and take account of sex, age, immigration status,
student and employment status, education level, marital status and number
of children at home, as well as weeks worked and earnings in the previous 12
months. The imputation was carried out for three separate source groups:
Other Transfer Income, Investment Income, and Other Miscellaneous
Income. In each case, imputation was carried out in two steps: First, it was
determined whether the imputed value was to be non-zero using a logistic
regression equation. Then, if it was to be non-zero, a random value was
imputed from an appropriate distribution using three quartile regression
equations to reflect the location, dispersion and asymmetry of the empirical
distribution. The last step in the imputation process involved rescaling the
imputed values to values more appropriate to the simulated calendar year of
imputation. For that purpose, Other Transfer Income was rescaled by the
consumer price index, Investment Income was rescaled by the bank rate and
Other Miscellaneous Income was rescaled by the average industrial wage.

In Kind Transfers

The major in kind government transfers are health care and education. These
are modeled based on unit costs by age, and sex in the case of health care, and
unit costs based on the kind of educational institution attended (elementary-
secondary, community college, university; Cameron & Wolfson, 1994).
CHANGING POVERTY OR
CHANGING POVERTY AVERSION?

Daniel L. Millimet, Daniel Slottje and

Peter J. Lambert

ABSTRACT

Supposing that decisionmakers in any country and at any point in time

tolerate a certain fixed level of perceived poverty, differences in poverty
aversion are called for to explain observed international and intertempo-
ral variations in poverty statistics. Under the Natural Rate of Subjective
Poverty hypothesis advanced in this paper, variations in the degree of
poverty aversion are estimable and can be explained by political and
socioeconomic factors. The methodology is applied to US data from 1975
to 1998 and across nations using cross-section data from the mid-1990s.
Factors such as the political affiliation of government officials, public
expenditure, per capita income, and economic growth account for much of
the variation in poverty aversion implied by our hypothesis. The rela-
tionship between inequality aversion and poverty aversion is also
explored, with the aid of a parallel ‘‘natural rate’’ hypothesis for
inequality (Lambert et al., 2003). Our findings provide a new framework
in which to interpret observed correlations between poverty, inequality,
and social welfare.

Equity
Research on Economic Inequality, Volume 15, 233–268
Copyright r 2007 by Elsevier Ltd.
All rights of reproduction in any form reserved
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15010-9
233
234 DANIEL L. MILLIMET ET AL.

Many individuals and organizations both in Canada and abroad understandably want to
know how many people and families live in ‘poverty,’ and how these levels change.
Reflecting this need, different groups have at different times developed various measures
which purported to divide the population into those who were poor and those who were
not. In spite of these efforts, there is still no internationally-accepted definition of
povertyy This is not surprising, perhaps, given the absence of an international con-
sensus on what poverty is and how it should be measuredy The underlying difficulty is
due to the fact that poverty is intrinsically a question of social consensus, at a given point
in time and in the context of a given countryy It is through the political process
that democratic societies achieve social consensus in domains that are intrinsically
judgmental.
Ivan P. Fellegi, Chief Statistician of Canada, September 19971

1. INTRODUCTION

Poverty is an illusive term, one which economists and policymakers have

long struggled to define and measure, as evidenced by the quote above.
Similar sentiments have been voiced by policymakers in the UK and else-
where. Nonetheless, the media in many countries regularly publish or
broadcast news of changes in the poverty rate (however defined). Implicit in
such (usually stark) reports is the presumption that the level of poverty is an
exogenous event, occurring within the confines of the existing social and
institutional framework. That is, it is presumed that policymakers are
merely bystanders, compelled into action only in the wake of a pessimistic
report. Our primary hypothesis is that such a view is incorrect. Rather, as
stated in the quote above, poverty is determined through social consensus.
Specifically, we maintain that decisionmakers in any country and at any
point in time tolerate a certain fixed level of poverty. However, the extent of
poverty inherent in the current income distribution is entirely subjective.
Thus, while the subjective level of poverty – the level of poverty perceived by
policymakers – is, we posit, constant over time and place, objective or
absolute measures of poverty may change depending on the preferences
dictated by social consensus (or some unilateral decisionmaker in a non-
democracy). Under such a view of the world, preferences for poverty (i.e.,
poverty aversion) are time- and country-specific. Moreover, to maintain a
constant level of subjective poverty, periods of greater aversion to poverty
should be characterized by lower objective poverty (i.e., the type reported in
the media). Conversely, periods of lower poverty aversion should be char-
acterized by greater objective poverty. As a result, policymakers are not
Changing Poverty or Changing Poverty Aversion? 235

reacting to changing objective poverty levels; rather, such levels are being
explicitly determined by policymakers according to social mandates.
Such a claim is of course hardly new.2 The innovation proffered in this
paper is that the degree of poverty aversion is quantifiable and the deter-
minants of intertemporal and international variations in poverty aversion
are estimable. Our ability to identify the socially determined level of poverty
aversion stems from the assumption that subjective poverty is constant
across time and space. We will refer to this assumption as the Natural Rate
of Subjective Poverty (NRSP) hypothesis from now on. According to the
NRSP hypothesis, subjective poverty is held constant, whilst poverty
aversion is time- and country-specific. Objective poverty accordingly varies
from time to time and from place to place in such a way as to maintain this
state of affairs as an ‘‘equilibrium.’’
We should note from the outset that our characterization of some poverty
measures as ‘‘objective’’ in what follows is necessarily simplistic. All meas-
ures of poverty – even those such as the headcount ratio which are
purported to be absolute measures – entail implicit assumptions regarding
preferences surrounding the issue of poverty. Given the manner in which
such indices are treated by some, however, it is not unreasonable to
characterize them as objective.
Two well-known poverty indices which we characterize as objective for
present purposes are the headcount ratio (proportion of the population who
are poor) and the normalized poverty deficit (aggregate income shortfall of
poor persons or households expressed per capita of the overall population
and normalized by the poverty line). Both of these poverty statistics are
widely quoted in applied work; the headcount ratio is particularly popular
with the media. The headcount ratio is insensitive to the extent of shortfall
of incomes from the poverty line; the normalized poverty deficit is insen-
sitive to the distribution of income among the poor.
Several parametric families of poverty indices have followed upon Sen’s
(1976) axiomatic treatment of the poverty measurement issue, in which the
parameter purports to capture aversion to poverty (see Zheng’s (1997)
comprehensive survey article). Any one of these is a potential candidate for
our subjective measure. Sen himself argued that the incidence, intensity, and
inequality of poverty all matter and he isolated a particular poverty index
which is a mix of the headcount ratio, income gap ratio and Gini coefficient
of income among the poor. Kakwani (1980) generalized Sen’s index by
introducing a ‘‘sensitivity parameter,’’ increases in which make the index
more sensitive to transfers of income among those with large poverty gaps
and also more sensitive to small income changes at the bottom end of the
236 DANIEL L. MILLIMET ET AL.

distribution. Clark et al. (1981) proposed two parametric families, in one of

which the parameter measures aversion to inequality in poverty gaps and in
the other, to inequality in basic (censored) incomes; this latter parameter has
been called ‘‘aversion to inequality in poverty’’ by Foster and Sen (1997,
p. 178); see also Chakravarty (1983a, 1983b) on this. The so-called FGT
family of Foster, Greer, and Thorbecke (1984, p. 761) has a parameter
which is described by the authors as an indicator of ‘‘aversion to poverty,’’
but it has a strange property: the more averse to poverty is the observer, the
lower the poverty value assigned to any fixed income distribution. Only in
Zheng (2000b) is the concept of poverty aversion placed on rigorous footing.
Zheng (2000b, p. 121) identifies a family of constant poverty aversion
indices with formal properties: ‘‘The notion of ‘poverty aversion’ that
researchers have in mind means a lot more than just ‘disliking poverty’.’’
The Zheng index is our favored ‘‘subjective’’ measure of poverty.3
The paper unfolds as follows. Section 2 gives a brief overview of the
poverty measures used in the analysis and describes the NRSP hypothesis
and estimation strategy. In Section 3, we analyze the intertemporal variation
in poverty aversion in the US over the last quarter of the 20th century under
the natural rate hypothesis. Section 4 similarly examines the empirical
heterogeneity observed in the level of poverty aversion across countries in
the mid-1990s under the natural rate hypothesis. Section 5 considers
relationships with other recent literature and provides concluding remarks
which point to opportunities for future research.

2. MODEL

2.1. Preliminaries

For the case of a discrete income distribution F ¼ {y1, y2, y, yN}, Zheng
(2000a, 2000b) discusses the following constant distribution-sensitivity
(CDS) measure of poverty:
1 X gðz
yi Þ
q
PgF ¼ ½e
1 g40 (1)
N i¼1
where z is the poverty line, q (qoN) is the number of individuals with
income below the poverty line, y is income, and g is the measure of
distribution-sensitivity. Distribution-sensitivity measures the decrease in
poverty as a result of a progressive income transfer. Zheng (2000a) shows
that the distribution-sensitivity in Eq. (1) is constant and given by g. The
Changing Poverty or Changing Poverty Aversion? 237

author further argues that distribution-sensitivity is an appropriate formal

definition of poverty aversion (see also Zheng (2000b)).
We can re-write Eq. (1) in terms of the ratio of each individual’s income to
the poverty line as
1 X gzð1
si Þ
q
PgF ¼ ½e
1 (2)
N i¼1

where si ¼ yi/z. If the data (discussed below) is partitioned into K groups,

and if si is assumed constant within each group, Eq. (2) can be re-written as
K

q 1X K X X nk gzð1
sk Þ
PgF ¼
þ egzð1
si Þ ¼
H F þ e (3)
N N k¼1 i2k k¼1
N

where HF is the headcount index, nk is the number of individuals in group k,

and sk is the ratio of income to the poverty line for those in group k.
The measure PgF differs from the headcount index (HF) and other
summary statistics-based poverty indices in its explicitly ethical foundation.
It embodies the poverty aversion parameter of a decisionmaker; thus, it is
referred to as subjective in contrast to, for example, the headcount index.
Note that
@PgF
40 8F (4)
@g
Thus, a more poverty-averse social decisionmaker will perceive greater
poverty in any given distribution. Although the headcount index entails
ethical judgments as well, this index is regarded as authoritative by many;
we call it ‘‘objective’’ for the purposes of the present study as noted above.
For robustness, we also utilize the normalized poverty deficit as an
objective poverty measure. For the same income distribution, F, this
measure is given by
1 X
z
yi 
q
NPDF ¼ (5)
N i¼1 z

for discrete data and

X
nk  X
nk 
NPDF ¼ ð1
sk Þ ¼ H F
sk (6)
k
N k
N

for data partitioned into k groups. The NPD, which is equivalent to the
Foster–Greer–Thorbecke (FGT) measure with a ¼ 1, entails ethical judg-
ments just as the headcount index does, but it is not distribution-sensitive
and we characterize it as objective for present purposes as well.
238 DANIEL L. MILLIMET ET AL.

According to the PgF measure, two entirely different income distributions

F1 and F2 could be attributed the same level of subjective poverty by two
different decisionmakers:
g g
PF11 ¼ PF22 g1 ag2 (7)
The relationships in Eqs. (4) and (7) are at the crux of the NRSP
hypothesis.

2.2. The Natural Rate of Subjective Poverty

To analyze differences in poverty a version either within a particular country

over time or across countries, let Fi be the income distribution function for
observation i (i may index either years or countries). Given the natural rate
of subjective poverty, p, we first identify the poverty aversion parameter, gi,
such that the subjective poverty in distribution Fi equals p:
g
PF11 ¼ p 8i (8)

We then analyze the determinants of the poverty aversion parameters, gi,

identifying the empirical factors associated with the changing degree of
aversion. If x is a vector of possible explanatory variables and xj the jth
component, we wish to estimate a function c(x) such that
gi  cðxi Þ (9)

where xi indicates the values of x for observation i; or, at least, to sign the
partial derivatives qc/qxj. The x’s we examine later include variables re-
flecting the political and economic climate of a particular country or period,
such as income, population, education, unemployment, female empower-
ment, and corruption.
With the explanatory variables in x and function c(x) determined, new
political or social conditions that alter one of the explanatory variables, say
xij ; will cause the level of subjective poverty to diverge from the historical
natural rate of poverty p. For example, suppose qc/qxj>0 and that xij
increases to xij þ Dxij : Then gi increases to gi+Dgi, where Dgi ¼
ð@c=@xj ÞDxij 40: This causes the level of subjective poverty to increase since
the decisionmaker is now more averse to poverty:
" g
! #
gi þDgi @PFi i
PF i ¼pþ Dgi 4p (10)
@gi
Changing Poverty or Changing Poverty Aversion? 239

Given the new degree of poverty aversion gi+Dgi for observation i; can
the natural rate of poverty p be restored? Yes, if redistributive policies (e.g.,
directly through the tax system or indirectly through increased transfer
payments, government sponsored training programs, raising the minimum
wage, etc.) are undertaken that improve the incomes of those below the
poverty line and consequently alter the income distribution, from Fi to Gi
say, where:
g þDgi g g þDgi
PGi i ¼ PFi i ¼ poPFi i (11)

The condition in Eq. (11) requires Gi to contain objectively less poverty

than Fi (so that a decisionmaker with poverty aversion gi+Dgi is less con-
cerned about poverty in Gi than in Fi). Taking the headcount index as our
measure of objective poverty, then Gi should have a lower headcount index
than Fi (i.e., H Gi oH F i ).
If this is the mechanism whereby the natural rate p is restored, then
according to the NRSP, our measures of objective poverty should be
inversely related to the same set of explanatory variables, x. For example,
we should be able to express the headcount ratio for observation i as:

H F i  oðxi Þ (12)
and if qc/qxj>0 (as assumed previously), then qo/qxjo0. This constitutes
the testable implication of the NRSP.4

3. POVERTY AVERSION IN THE UNITED STATES

3.1. Data

To examine the intertemporal variation in poverty and poverty aversion in

the US, we use Census Bureau statistics on the extent of poverty (and
poverty thresholds) over the period 1975–1998.5 Table A1 in the appendix
contains the annual cumulative percentage of the total population and sub-
population of individuals over age 65 with income below a certain fraction
of the poverty line. Combining this data with the Zheng index for parti-
tioned data in Eq. (3) (and ignoring inequality within quantiles6), we find the
value of g that achieves a given level of subjective poverty, p, by conducting
a grid search over potential values. To do this, we compute the value of
Eq. (3) for g between 1.0  10
06 and 0.15 with a step size of 1.0  10
06.
240 DANIEL L. MILLIMET ET AL.

We perform this exercise for several possible values of p since we do not

presume to know the ‘‘true’’ value of the natural rate p. Tables A2 and A3 in
the appendix report the values of g which yield a Zheng index value of 0.02,
0.04, 0.06, 0.08, and 0.10 for the total population and the sub-population
aged 65 and over, respectively.7,8 Finally, to ensure a consistent measure of
objective poverty, we use the headcount index reported in the same Census
Bureau table, as well as the NPD measure estimated using Eq. (6). Table 1
contains the relevant summary statistics and sources for the potential
determinants of poverty aversion we examine below.

3.2. Results

3.2.1. Preliminaries
To analyze those factors associated with the level of poverty aversion in the
US, we first must ensure our analysis is robust to the choice of p. Thus, we
use several values for p, ranging from 0.02 to 0.10. Table 2 presents the
correlation matrix between the time-specific poverty aversion parameters for
each value of p, g0.02 refers to the value of g such that PgF ¼ 0:02; etc. For all
values of p, the correlations are extremely close to unity, implying that over
a wide range of p, the arbitrary choice of p does not appear to be prob-
lematic. Also included in Table 2 are the correlations between the time-
specific aversion parameters and our two measures of objective poverty. In
all cases, the correlations between the g’s and the objective poverty measures
are negative and close to unity in absolute value. Consequently, periods of
higher objective poverty are characterized by less poverty averse decision-
makers in the US, as the NRSP theory predicts.
To further illustrate these points, Fig. 1 plots g0.04, g0.06, g0.08, and g0.10
versus our two measures of objective poverty (the headcount ratio and the
normalized poverty deficit) for both the entire population as well as the over
65 sub-population.9 The relationships are downward-sloping (although not
monotonically), and within each panel the four cases are nearly parallel to
one another. Fig. 2 plots the level of aversion over time. The top two panels
plot g0.04, g0.06, g0.08, and g0.10 for the two samples, marking times of
presidential changes. As in Fig. 1, the four cases yield similar insights into
relative changes in the degree of poverty aversion over this period. The
bottom panels re-plot g0.02 and g0.10 against time, super-imposing one atop
the other (i.e., with the axes re-scaled) to illustrate how each measure yields
virtually identical inferences concerning changes in relative poverty aver-
sion. As a result, not knowing the actual NRSP, p, does not impede our
 Changing Poverty or Changing Poverty Aversion?
Table 1. Summary Statistics: US Sample.
Variable Years Mean Std. Dev. Source

President (1 ¼ Democrat) 1975–1998 0.46 0.50 http://www.policsci.com/almanac/history/polidivs.htm

Senate (# Democrats) 1975–1998 52.79 5.99 http://www.policsci.com/almanac/history/polidivs.htm
Representatives (# Democrats) 1975–1998 256.50 27.32 http://www.policsci.com/almanac/history/polidivs.htm
Annual growth rate (GDP) 1975–1998 0.03 0.02 http://www.bea.doc.gov
Median household income (1000s 97 US$) 1975–1998 41.86 16.00 http://www.census.gov
Total public education expenditures (% GDP) 1975–1998 0.14 0.08 http://www.bea.doc.gov
Unemployment rate 1975–1997 0.07 0.01 http://stats.bls.gov
% HS+, males 1975–1998 0.74 0.06 http://nces.ed.gov
% HS+, females 1975–1998 0.74 0.07 http://nces.ed.gov
% 4 years college+, males 1975–1998 0.23 0.03 http://nces.ed.gov
% 4 years college+, females 1975–1998 0.17 0.03 http://nces.ed.gov
Union share 1975–1998 0.16 0.06 http://www.demographia.com
Minimum wage (96 US$) 1975–1997 3.30 0.44 http://epinet.org/datazone/minimumwage.html

241
242 DANIEL L. MILLIMET ET AL.

Table 2. Poverty Aversion (Whole Population): Correlation Matrix,

1975–1998.
g0.02 g0.04 g0.06 g0.08 g0.10 Headcount

g0.02 1.0000
g0.04 0.9026 1.0000
g0.06 0.9234 0.9705 1.0000
g0.08 0.9284 0.9715 0.9781 1.0000
g0.10 0.9430 0.9712 0.9790 0.9837 1.0000
Headcount –0.8792 –0.9375 –0.9212 –0.9310 –0.9419 1.0000
NPD –0.9406 –0.9781 –0.9783 –0.9836 –0.9887 0.9690

ability to make inferences regarding intertemporal variation in poverty

aversion.
The decline in overall poverty aversion (depicted in the top-left panel of
Fig. 2), along with the negative correlation between poverty aversion and
objective poverty (listed in Tables 2 and 3), can be interpreted in the context
of the recent rise in (objective) poverty in the US.10 Possible explanations
have ranged from structural changes in the labor market (Cutler & Katz,
1991), negative inner city neighborhood effects (Cutler & Glaeser, 1997),
and an increase in female labor force participation (Topel, 1994) and female-
headed households (Blank & Hanratty, 1992), to name but a few.11 How-
ever, the NRSP hypothesis maintains that these are at best only indirect
explanations for the rise in poverty. To understand the rise in objective
poverty, one must analyze the determinants of poverty aversion. The NRSP
hypothesis asserts that it is poverty aversion among policymakers that leads
to actions (or inactions) that determine the level of objective poverty. For
example, Blank and Hanratty (1992) argue that the lack of generous transfer
payments (e.g., welfare) is responsible for the rise in poverty. This is entirely
consistent with the NRSP hypothesis at work.
The NRSP hypothesis may also shed light on another puzzle: the inability
of some state governments to effectively reach poor residents. Ravallion
(1999b, p. 373) states that, ‘‘As a rule, governments of poor states (provinces
or countries) do not seem to be very good at targeting public spending to
their poor. Yet anti-poverty programs often target poor states, in the hope
of reaching poor people.’’ However, if poverty aversion differs across states
or, more to the point, if state governments have different aversion levels
than the federal government, and subjective poverty is equalized across
states, then states with low levels of poverty aversion should have high
objective poverty according to the NRSP. Moreover, state governments
 Changing Poverty or Changing Poverty Aversion?
gamma_0.04 gamma_0.04 gamma_0.06
gamma_0.06 gamma_0.08 gamma_0.10
0.3 gamma_0.08 gamma_0.10 0.4

Poverty Aversion

Poverty Aversion
0.25
0.3

0.2

0.2
0.15

0.1 0.1
0.11 0.12 0.13 0.14 0.15 0.1 0.12 0.14 0.16
Headcount Index Headcount Index
Whole Population Population Aged 65+

gamma_0.04 gamma_0.06 gamma_0.04 gamma_0.06

gamma_0.08 gamma_0.10 gamma_0.08 gamma_0.10
0.3 0.4
Poverty Aversion

Poverty Aversion
0.25
0.3

0.2

0.2
0.15

0.1 0.1
0.04 0.05 0.06 0.07 0.03 0.035 0.04 0.045
Normalized Poverty Deficit Normalized Poverty Deficit
Whole Population Population Aged 65+

243
Fig. 1. Poverty Aversion and Objective Poverty.
 244
gamma_0.04 gamma_0.06 gamma_0.04 gamma_0.06
gamma_0.08 gamma_0.10 gamma_0.08 gamma_0.10
0.3 0.4

Poverty Aversion

Poverty Aversion
0.25
0.3

0.2

0.2
0.15

0.1 0.1
1975 1980 1985 1990 1998 1975 1980 1985 1990 1998
Year Year
Whole Population Population Aged 65+

gamma_0.02 gamma_0.10 gamma_0.02 gamma_0.10

0.08 0.28 0.11 0.38

DANIEL L. MILLIMET ET AL.

0.26 0.36

gamma_0.10
gamma_0.10
gamma_0.02
gamma_0.02

0.07 0.1

0.24 0.34

0.06 0.09
0.22 0.32

0.05 0.2 0.08 0.3

1975 1980 1985 1990 1998 1975 1980 1985 1990 1998
Year Year
Whole Population Population Aged 65+

Fig. 2. Poverty Aversion in the US: 1975–1998.

Changing Poverty or Changing Poverty Aversion? 245

Table 3. Poverty Aversion (Population Aged 65+): Correlation Matrix,

1975–1998.
g0.02 g0.04 g0.06 g0.08 g0.10 Headcount

g0.02 1.0000
g0.04 0.8668 1.0000
g0.06 0.9105 0.9558 1.0000
g0.08 0.9097 0.9600 0.9715 1.0000
g0.10 0.9162 0.9515 0.9715 0.9839 1.0000
Headcount
0.8842
0.7960
0.8135
0.7635
0.7711 1.0000
NPD
0.9544
0.9559
0.9719
0.9592
0.9651 0.8982

with low poverty aversion will not find it ‘‘optimal’’ to direct federal aid to
combat poverty since this would only push the states out of equilibrium.
Finally, we note that overall aversion to poverty is negatively correlated
with aversion to poverty in the sub-population aged 65 and over. The pair-
wise correlation between the overall g’s and the sub-population g’s (for a
fixed value of p) range from
0.11 (for p ¼ 0.04) to
0.35 (p ¼ 0.02). In-
terestingly, then, it appears as if policymakers do not focus on poverty as a
whole, but rather focus on poverty within subgroups of the population at
any given time.12

3.2.2. Correlates of US Poverty Aversion

We now shift our focus to exploring the empirical factors associated with the
observed variation in poverty aversion.13 To proceed, Figs. 3 and 4 present
time plots of various US attributes and social policy choices and poverty
aversion. Several interesting associations emerge. First, the number of Con-
gressional representatives and senators affiliated with the democratic party
is positively correlated with overall poverty aversion. The correlation co-
efficients are 0.59 and 0.63, respectively. However, a democratic Congress is
negatively correlated with poverty aversion for the over 65 sub-population
(correlation coefficients of
0.36 and
0.25, respectively). Second, perhaps
not surprisingly, union share is positively correlated with overall poverty
aversion (0.81), but negatively correlated with over 65 poverty aversion
(
0.47). Third, while we find little correlation between lagged growth in
per capita income and overall poverty aversion (
0.09 and
0.04 using a
three-year and five-year moving average, respectively), we do find a large
positive correlation between lagged growth and aversion to poverty in
the over 65 sub-population (correlations of 0.34 and 0.42 using the same
moving averages). Finally, in terms of policy outcomes, we find significant
 246
DANIEL L. MILLIMET ET AL.
Fig. 3. Poverty Aversion (Whole Population) and Select US Attributes: 1975–1998.
 gamma_0.10 # democrats gamma_0.10 # democrats gamma_0.10 Union Share

Changing Poverty or Changing Poverty Aversion?

0.38 0.38 300 0.38 30

Union Share (% non-ag.,

House (# of democrats)
Senate (# of democrats)
60

private employment)
Poverty Aversion
Poverty Aversion

Poverty Aversion
0.36 0.36 0.36 25
55
0.34 0.34 250 0.34 20

0.32 50 0.32 15
0.32

0.3 45 0.3 200 0.3 10

1975 1980 1985 1990 1998 1975 1980 1985 1990 1998 1975 1980 1985 1990 1998
Year Year Year
Panel A Panel B Panel C
gamma_0.10 median family gamma_0.10 3-year Moving gamma_0.10 5-year Moving
income, 97$ Average Average

GDP Growth (Per Capita)

Median Household Income
46000

GDP Growth (Per Capita)

0.38 0.38 0.04 0.38 0.04

Poverty Aversion
Poverty Aversion

Poverty Aversion
0.36 44000 0.36 0.36 0.03
0.02
0.34 42000 0.34 0.34 0.02
0
0.32 40000 0.32 0.32 0.01

0.3 38000 0.3 -0.02 0.3 0

1975 1980 1985 1990 1998 1975 1980 1985 1990 1998 1975 1980 1985 1990 1998
Year Year Year
Panel D Panel E Panel F
gamma_0.10 unemployment rate gamma_0.10 Minimum Wage gamma_0.10 Expenditure
0.38 10 0.38 0.38 40

Expenditure (% of GDP)
4

Government Education
Unemployment Rate
Poverty Aversion

Poverty Aversion
Poverty Aversion

Minimum Wage
0.36 0.36 0.36
8 3.5 30
0.34 0.34 0.34
6 3 20
0.32 0.32 0.32
10
0.3 4 0.3 2.5 0.3
1975 1980 1985 1990 1998 1975 1980 1985 1990 1998 1975 1980 1985 1990 1998
Year Year Year
Panel G Panel H Panel I

247
Fig. 4. Poverty Aversion (Population Aged 65+) and Select US Attributes: 1975–1998.
248 DANIEL L. MILLIMET ET AL.

associations between poverty aversion and the minimum wage (correlation

coefficient of 0.77 for overall aversion;
0.57 for over 65 aversion) and
government expenditure on education as a fraction of GDP (0.85 for overall
aversion;
0.42 for over 65 aversion).
To further test the NRSP hypothesis and better analyze the determinants
of poverty aversion in the US, Tables 4 (whole population) and 5 (over 65
sub-population) present the results of some simple AR(1) regressions using
ln(g0.10), ln(H), the natural logarithm of the headcount ratio, and ln(NPD),
the natural logarithm of the NPD, as the dependent variables.14 The most
important result to examine – which is at the crux of the NRSP hypothesis –
is whether changes in various attributes that alter the poverty aversion of
decisionmakers also result in changes in the level of objective poverty, such
that the natural rate of poverty is maintained. Thus, factors associated with
greater aversion should also be associated with lower levels of objective
poverty. Examining Tables 4 and 5 verifies that this is in fact so. In almost
every case, the coefficients are identical in terms of statistical significance,
but of the opposite sign, in the model using g0.10 as the dependent variable
and the models using our objective poverty measures.

Table 4. Determinants of Poverty Aversion (Whole Population) and

Objective Poverty: Semi-log AR(1) Specificationa.
Independent Variable Dependent Variable

g0.10 Headcount NPD

President (1 ¼ Democrat) 0.05 [p ¼ 0.00]
0.05 [p ¼ 0.00]
0.07 [p ¼ 0.00]

ln (Senate (# Democrats))
0.07 [p ¼ 0.48]
0.06 [p ¼ 0.47]
0.08 [p ¼ 0.57]
ln (House of Reps. (# Democrats)) 0.18 [p ¼ 0.07] 0.04 [p ¼ 0.56] 0.11 [p ¼ 0.39]
ln (Median household income)
0.27 [p ¼ 0.24] 0.02 [p ¼ 0.95] 0.16 [p ¼ 0.58]
Unemployment rate
0.06 [p ¼ 0.00] 0.04 [p ¼ 0.00] 0.06 [p ¼ 0.00]
Growth rate
0.01 [p ¼ 0.02] 0.00 [p ¼ 0.43] 0.00 [p ¼ 0.13]
Female LFPR
0.03 [p ¼ 0.05]
0.00 [p ¼ 0.78] 0.02 [p ¼ 0.28]
% HS+, males
0.18 [p ¼ 0.00] 0.16 [p ¼ 0.00] 0.23 [p ¼ 0.00]
% HS+, females 0.12 [p ¼ 0.00]
0.10 [p ¼ 0.00]
0.16 [p ¼ 0.00]
% 4 years college+, males 0.16 [p ¼ 0.00]
0.11 [p ¼ 0.01]
0.17 [p ¼ 0.00]
% 4 years college+, females
0.01 [p ¼ 0.16] 0.02 [p ¼ 0.06] 0.03 [p ¼ 0.02]

r
0.86
0.85
0.91

Observations 22
a
p-values associated with the test H0: b ¼ 0 in brackets. Refer to the text and/or appendix for
variable definitions.
Changing Poverty or Changing Poverty Aversion? 249

Table 5. Determinants of Poverty Aversion (Population Aged 65+)

and Objective Poverty: Semi-log AR(1) Specificationa.
Independent Variable Dependent Variable

g0.10 Headcount NPD

President (1 ¼ Democrat)
0.01 [p ¼ 0.73]
0.02 [p ¼ 0.58] 0.00 [p ¼ 0.97]

ln (Senate (# Democrats)) 0.16 [p ¼ 0.36] 0.21 [p ¼ 0.30] 0.08 [p ¼ 0.72]
ln (House of Reps. (# Democrats))
0.32 [p ¼ 0.06] 0.25 [p ¼ 0.22] 0.25 [p ¼ 0.24]
ln (Median household income)
0.79 [p ¼ 0.18]
0.06 [p ¼ 0.90] 0.12 [p ¼ 0.87]
Unemployment rate
0.01 [p ¼ 0.34] 0.01 [p ¼ 0.51] 0.01 [p ¼ 0.46]
Growth rate 0.02 [p ¼ 0.00]
0.02 [p ¼ 0.00]
0.03 [p ¼ 0.00]
Female LFPR 0.01 [p ¼ 0.68]
0.01 [p ¼ 0.79] 0.01 [p ¼ 0.73]
% HS+, males
0.32 [p ¼ 0.00] 0.29 [p ¼ 0.01] 0.39 [p ¼ 0.00]
% HS+, females 0.25 [p ¼ 0.00]
0.28 [p ¼ 0.00]
0.34 [p ¼ 0.00]
% 4 years college+, males 0.22 [p ¼ 0.01]
0.11 [p ¼ 0.31]
0.27 [p ¼ 0.02]
% 4 years college+, females
0.07 [p ¼ 0.05] 0.09 [p ¼ 0.02] 0.11 [p ¼ 0.02]

r
0.86
0.83
0.86

Observations 22
a
See Table 4.

According to the specific point estimates, several interesting findings

emerge. First, overall poverty aversion and overall objective poverty are
significantly associated with the political affiliation of the president,
although independent of the political affiliation of Congress at standard
significance levels. As one might expect, a Democrat-controlled White
House is associated with periods of greater (lower) poverty aversion
(objective poverty). Second, during periods of greater economic mobility,
overall poverty aversion (objective poverty) is lower (higher); whilst poverty
aversion (objective poverty) among those over 65 is higher (lower). The fact
that overall aversion to poverty falls during periods of economic growth is
consistent with the positions in Ravallion and Lokshin (2000), Bénabou and
Ok (2001), and Hirschman and Rothschild (1973). These studies argue
that preferences for redistribution depend not just on the median voter’s
current position in the income distribution, but also his/her expectations of
future income mobility.15 However, the positive effect of economic growth
on aversion to poverty among those over 65 is possibly due to the same
mentality that causes charitable giving to increase with income. Via an
increase in poverty aversion, the benefits of economic growth are passed
along to those no longer in the workforce.
Third, we find a significant, negative association between the unemploy-
ment rate and overall poverty aversion. Thus, during times of low
250 DANIEL L. MILLIMET ET AL.

unemployment, poverty aversion intensifies. Again, this is consistent with

the observed rise in charitable giving as incomes increase. It is also con-
sonant with the findings in Blank and Blinder (1986), which report a similar
increase in (objective) poverty as unemployment worsens. Finally, we find
significant effects of education levels. An increase in the share of males
(females) with at least a high school diploma is associated with a decline
(increase) in poverty aversion for both samples; conversely for objective
poverty. College completion rates for males, on the other hand, have a
positive effect on poverty aversion in both samples and there is a negative
effect of female college completion rates on poverty aversion for those over
65. The fact that an increase in males with only a high school degree lowers
poverty aversion, but increases in college-educated men increase poverty
aversion is not surprising. Moreover, the fact that an increase in the share of
(at least) high school educated females increase poverty aversion is also not
surprising given the literature on women being more risk averse than men
(see, e.g., Jianakoplos and Bernasek (1998), as well as the findings in
Ravallion and Lokshin (2000) that women tend to favor redistribution).
However, the fact an increase in college-educated women is associated with
diminished aversion to poverty amongst the sub-population aged 65 and
over is perhaps unexpected.

4. INTERNATIONAL DIFFERENCES IN POVERTY

AVERSION

If one accepts the NRSP hypothesis, it may apply to not only intertemporal
comparisons within a given country, but also to cross-sectional comparisons
across countries. In its widest sense, subjective poverty may be constant
across both time and countries.

4.1. Data

To test the NRSP hypothesis and analyze heterogeneity in terms of poverty

aversion across countries, we use data from the World Development Report
2000/2001 (WDR) and 1999 World Development Indicators published by the
World Bank. Table A4 in the appendix contains the percentage of individ-
uals living on less than $1/day and $2/day for 51 countries.16 The countries
are primarily low income and/or market transition economies.17 We assume
all countries have an identical poverty line of $2/day. According to the
Changing Poverty or Changing Poverty Aversion? 251

Table 6. Summary Statistics: International Sample.

Variable Mean Std. Dev. Source

Annual population growth, 1.81 0.94 http://www.undp.org/hdro

1995–2015
Enrollment rate (combined 1st, 56.42 18.50 http://www.undp.org/hdro
2nd, and 3rd level), 1995
Adult literacy rate, 1995 70.18 25.46 http://www.undp.org/hdro
Real GDP (per capita), 1995 3062 2147 http://www.undp.org/hdro
% Females in gov’t, ministerial 6.19 4.87 http://www.undp.org/hdro
level, 1995
% Females in gov’t, sub- 8.28 7.09 http://www.undp.org/hdro
ministerial level, 1995
Corruption index (CPI), 1998 3.29 1.26 http://www.gwdg.de/uwvw

WDR (2000), 2.8 billion of the world’s 6 billion individuals live on less than
$2/day; 1.2 billion on less than $1/day. In addition, there is considerable
variation across countries not only in the percentage of individuals surviving
on incomes below these thresholds at any given time, but also in the trend
for these figures. For example, the number of individuals living on less than
$1/day fell by 130 million in East Asia over the period 1987–1998. Over this
same time span, the number in Sub-Saharan Africa rose by over 50 million
(see also Ravallion, 1994b).
As in the previous section, we combine the data listed in Table A4 with
the Zheng index for partitioned data in Eq. (3) and find the value of g that
achieves a given level of subjective poverty, p, by conducting a grid search
over potential values. As before, we perform this exercise for p ¼ 0.02, 0.04,
0.06, 0.08, and 0.10. Table A5 in the appendix reports the values of g.18,19
Table 6 contains the relevant summary statistics and sources for the
variables we analyze as possible determinants of country-specific poverty
aversion.

4.2. Results

4.2.1. Preliminaries
As in the prior section, we must verify our analysis is robust to the choice of
p. Table 7 is analogous to Table 2 and presents the correlation matrix
between the country-specific poverty aversion parameters for each value of
p. As in Table 2, the g’s are nearly perfectly correlated. Moreover, the
correlations between the aversion parameters and our two measures of
252 DANIEL L. MILLIMET ET AL.

Table 7. International Poverty Aversion: Correlation Matrix.

g0.02 g0.04 g0.06 g0.08 g0.10 Headcount

g0.02 1.0000
g0.04 0.9997 1.0000
g0.06 0.9991 0.9998 1.0000
g0.08 0.9982 0.9993 0.9999 1.0000
g0.10 0.9972 0.9986 0.9995 0.9999 1.0000
Headcount –0.8810 –0.8865 –0.8911 –0.8949 –0.8981 1.0000
NPD –0.8152 –0.8237 –0.8310 –0.8374 –0.8430 0.9519

objective poverty are close to minus one. Thus, as with the time series data
from the US, countries with higher objective poverty are characterized by
less poverty averse social policymakers, as the NRSP theory predicts.

4.2.2. Correlates of Country-Specific Poverty Aversion

Table 8 presents the estimates from two different cross-sectional regression
models where the dependent variables used within each model are the nat-
ural logarithms of g0.10, the headcount index, and the NPD.20 The variables
we examine as potential determinants of the country-specific level of poverty
aversion are: annual population growth, literacy and school enrollment
rates, real per capita GDP, the proportion of females at various levels of the
government, and corruption.21 Due to data availability, Model I (omitting
corruption) has 50 observations, while Model II (including corruption) has
35 observations. Note that, as in the previous section, in almost every case
the coefficients are of the opposite sign and have the same level of statistical
significance in the models using g0.10 as the dependent variable and the
models using our objective poverty measures.
Examining the results from both models, we find significant, negative
(positive) effects of population growth on poverty aversion (objective pov-
erty) and significant, positive (negative) effects of per capita GDP and cor-
ruption on poverty aversion (objective poverty). The lack of concern for
hardship in countries with high population growth may reflect the percep-
tion of policymakers that such hardship is self-induced through excess fer-
tility rates (e.g., Lanjouw & Ravallion, 1995). In any event, the negative
consequences of high population growth are disconcerting given the World
Bank’s projection that world population will increase by two billion over the
next 25 years, with over 97% occurring in developing countries (WDR,
2000).
 Changing Poverty or Changing Poverty Aversion?
Table 8. Determinants of International Poverty Aversion and Objective Povertya.
Independent Dependent Variable
Variable
Model I Model II

g0.10 Headcount NPD g0.10 Headcount NPD

Population
0.26 0.24 0.31
0.29 0.24 0.33

growth (
2.50) (2.51) (2.59) (
2.40) (2.15) (2.45)
Literacy rate
4.39  10–04
1.01  10–03 5.46  10–04 2.81  10–03
0.01
3.54  10–03
(
0.08) (
0.20) (0.09) (0.42) (
0.92) (
0.46)
Enrollment rate 2.44  10–03
5.50  10–04
2.58  10–03
0.01 4.73  10–03 0.01
(0.30) (
0.08) (
0.28) (0.32) (0.48) (0.60)
Real GDP (per 0.32
0.25
0.35 0.32
0.23
0.34
capita) (2.15) (
1.87) (
2.08) (1.81) (
1.36) (
1.70)
% females in
0.01 0.01 0.01
0.01 0.01 0.01
gov’t, (
0.62) (0.47) (0.59) (
0.71) (0.44) (0.66)
ministerial
level
% females in
2.26  10
03
0.01 1.34  10–03
0.01 1.01  10–03 0.01
gov’t, sub- (
0.211) (
0.64) (0.11) (
0.62) (0.09) (0.58)
ministerial
level
Corruption 0.13
0.13
0.15
(1.93) (
2.02) (
1.98)
2 0.51 0.50 0.51 0.51 0.51 0.51
R̄
Observations 50 35
a
Estimation is by OLS. T-statistics in parentheses.

253
254 DANIEL L. MILLIMET ET AL.

The positive association between per capita income and poverty aversion
is not surprising. The fact that social decisionmakers become more con-
cerned over the fate of the least fortunate as per capita incomes rise is
consonant with the findings in the previous section with respect to the neg-
ative effect of unemployment on poverty aversion in the US. Finally, we find
an unexpected positive association between corruption and poverty aver-
sion. One possible explanation for this finding is reverse causation. If pov-
erty averse governments implement measures aimed at combatting poverty,
these programs may create new opportunities for rents to be extracted by
corrupt officials.22

5. CONCLUDING REMARKS

This study began by noting that defining and measuring poverty is perhaps a
fraught exercise. Sawhill (1988, p. 1082) concludes that ‘‘it should be clear
by now that poverty is in the eye of the beholder and reflects off the green
eyeshades of the statistician.’’ However, rather than being discouraged, we
embrace this fact. We do this by fully admitting that the amount of poverty
inherent in a particular income distribution is subjective. For a given dis-
tribution, policymakers averse to poverty will ‘‘see’’ greater poverty than
those less averse. In addition, if we assume that all societies tolerate a con-
stant level of subjective poverty, then the amount of objective poverty will
move inversely with the level of poverty aversion mandated by social con-
sensus. We refer to this assumption as the NRSP hypothesis.
The hypothesis offers more than an intellectual exercise. Applying the
NRSP hypothesis to Zheng’s (2000a, 2000b) measure of subjective poverty,
we are able to quantify the level of poverty aversion maintained by pol-
icymakers in a given economy. We then verify – using both time-series data
from the US and cross-sectional data from 51 countries – that poverty
aversion and objective poverty (measured by the headcount ratio and the
normalized poverty deficit) have a strong, inverse relationship. Thus, both
data are consistent with the NRSP hypothesis. Finally, we examine the
determinants of poverty aversion. In the US, we find significant effects of
various economic and political variables such as the political affiliation of
the president and Congress, economic growth, unemployment, and educa-
tion levels. In our cross-section of countries, we find significant effects of
population growth, per capita income, and corruption.
The study we have undertaken here complements the analysis of Lambert,
Millimet, and Slottje (2003) and Millimet, Slottje, and Lambert (2000), in
Changing Poverty or Changing Poverty Aversion? 255

which a parallel hypothesis for inequality is articulated and explored.

According to the Natural Rate of Subjective Inequality (NRSI) hypothesis
posited therein, social planners maintain a fixed and constant value of sub-
jective inequality at all times and in all places, this being measured à la
Atkinson (1970) as the percentage of total income that could be given up
without welfare loss, the rest being distributed equally. To explain this
hypothesis, inequality aversion must be time- and location-specific, with
adjustment processes from changes in political and socioeconomic explana-
tory variables via inequality aversion to the income distribution explaining
observed differences in objective inequality (measured by the Gini coefficient
in Lambert et al. (2003) and Millimet et al. (2000)). All of this research begs
an intriguing question: do social decisionmakers obeying our NRSP hypoth-
esis also obey the NRSI hypothesis? In other words, can policymakers
simultaneously arrange their countries’ affairs to maintain both subjective
inequality and poverty at their respective natural rates?23 An affirmative
answer would provide a new perspective on the old and vexed question of
links between inequality, welfare, and poverty.
Perhaps not surprisingly, we find a strong positive correlation in the US
between our measure of overall poverty aversion and the level of inequality
aversion reported in Millimet et al. (2000) (correlation coefficient 0.75). We
also find a weak positive correlation between the level of country-specific
poverty aversion and the level of inequality aversion reported in Lambert
et al. (2003) (correlation coefficient 0.15). On the other hand, the correlation
between aversion to poverty among those over age 65 in the US and in-
equality aversion in the US is –0.50. Fig. 5 plots inequality aversion and
poverty aversion over time for each of the two US samples.24
The correlation of 0.75 in the US between poverty and inequality aversion
offers insight into recent findings regarding the linkages between inequality
and poverty. Gottschalk and Danziger (1984), among others, have attrib-
uted the increase of late in (objective) poverty in the US to burgeoning
inequality. Sawhill (1988, p. 1092) states that ‘‘this has merely substituted
one puzzle for another’’ since we are unsure why inequality has risen.
However, combining the NRSP and NRSI hypotheses provides a coherent
explanation for both phenomena. Social attitudes in the US, responding
perhaps to the increase in high school educated males and college educated
females, have become less averse to both inequality and poverty. Conse-
quently, objective measures of each (e.g., the headcount ratio and the Gini
index) have both risen.
Other researchers have also been concerned with the link between pov-
erty, inequality, and well-being. Atkinson (1987) discusses the relationship
256 DANIEL L. MILLIMET ET AL.

gamma_0.10 inequality aversion

0.28 0.55

0.26 0.5
gamma_0.10

e_0.10
0.24 0.45

0.22 0.4

0.2 0.35
1975 1980 1985 1990 1998
Year
Entire Population

gamma_0.10 inequality aversion

0.38 0.55

0.36 0.5
gamma_0.10

e_0.10

0.34 0.45

0.32 0.4

0.3 0.35
1975 1980 1985 1990 1998
Year
Population Aged 65+

Fig. 5. Poverty and Inequality Aversion in the US: 1975–1998.

Changing Poverty or Changing Poverty Aversion? 257

between the size of the pie, the cost of inequality and the ‘‘cost of poverty.’’
Massey (1996), in his presidential address before the Population Association
of America, emphasized the dual rise in poverty and inequality, as well as
the increase in the spatial concentration of the impoverished. Kakwani
(1999) argues that inequality and poverty indices can be derived from com-
mon underpinnings in terms of deprivation; but see also Ravallion (1999a).
There is also a growing literature on the linkages between inequality and
poverty trends in transition economies (see, e.g., Milanovic (1998), World
Bank (2000), and, most recently, Ivaschenko (2002)). Our framework offers
a fresh perspective, suggesting the examination of the dual effects of the
various political and socioeconomic changes that typically characterize such
transitions on poverty and inequality aversions.
Moreover, Ravallion (1994a) characterizes social welfare functions as in-
clusive measures of well-being (i.e., including the whole population) and
poverty indices as exclusive measures (giving zero weight to the incomes of
the non-poor). The author investigates empirically the correlations between
measures of poverty and social welfare in developing countries, finding these
to be very high, giving reasons and identifying factors which in practice tend
to blur the theoretical differences between exclusive (poverty) and inclusive
(social welfare) measures. Our twin NRSI and NRSP hypotheses provide an
alternative explanation. One should expect to simultaneously observe high
poverty and high social welfare given that aversion to inequality and pov-
erty are highly correlated (these aversions both being determined by the
same set of economic and political variables).
On the other hand, the strong, negative correlation between aversion to
poverty for the over 65 sub-population and inequality aversion – as well as the
negative correlation between overall poverty aversion and poverty aversion for
those over 65 reported in Section 3 – suggests that policymakers in the US
tend to focus on the economic status of either younger or older cohorts at a
particular time. The nature of politics may hinder the government’s ability to
simultaneously address the needs of both groups. Clearly, more research is
warranted, not only in terms of the relationship between poverty aversion
across population sub-groups, but also addressing simultaneous changes in the
level of and aversion to poverty and inequality within a coherent framework.

NOTES
1. Re-printed from http://www.statcan.ca/english/concepts/poverty/pauv.htm
2. Sawhill (1988, p. 1073) states that despite the growth in poverty-related re-
search, ‘‘discussions of poverty have become more, not less, ideological.’’ Moreover,
258 DANIEL L. MILLIMET ET AL.

anecdotal evidence of the subjectivity involved in poverty measurements abounds.

For example, assessments of the US’s War on Poverty are divided along partisan
lines. Conservatives claim that either the war has been won or that the program
actually increased poverty; liberals that greater action is required (Sawhill, 1988). In
addition, in the inequality literature, Epstein and Spiegel (2001, p. 472) discuss the
implications of deviations from an ‘‘acceptable level of inequality’’ or ‘‘natural in-
equality.’’ Lambert et al. (2003) propose a natural rate of subjective inequality that is
constant across countries.
3. Each of the headcount ratio, normalized poverty deficit and Watts index, which
we have characterized as objective, involves an implicit choice of parameter in one or
other of the families referred to here. Thus, the headcount ratio and normalized
poverty deficit belong to the FGT family, whilst the Watts index is an increasing
transformation of an index in Clark et al.’s second family. See Zheng (1997).
4. One should resist the temptation to conclude that the negative relationship
predicted by the NRSP hypothesis is mechanical in nature. If there were an exact
relationship between Zheng’s poverty index, the poverty aversion parameter, g, and
either of our ‘‘objective’’ measures of poverty, Q, so that PgF ¼ gðg; QF Þ; say, with
qg/qg>0 and qg/qQF>0, which held across all income distributions, then certainly
any xj which had a positive effect on poverty aversion would necessarily have a
negative effect on objective poverty, since we hold g(g,QF) constant and equal to p.
This would indeed void the test of NRSP. However, the headcount and NPD are
entirely insensitive to the distribution of income among the poor, whereas Zheng’s
index most certainly is not; thus, there could be no such relationship.
5. The validity of the poverty lines used by the US is beyond the scope of this
paper. Future research may examine the possible simultaneous determination of
poverty aversion and poverty lines by decisionmakers.
6. Ignoring within-quantile inequality is unfortunate, but necessary, given the
data available. This restriction may be addressed in future research, utilizing large
time-series of cross-section data from the UK or the CPS in the US. For now, we
note that we have tested the robustness of the results to different assumed equal
share values within quantiles. The results are unaltered.
7. The corresponding values of PgF are not presented in the table, but in all cases
we are able to find a value of g which brings PgF to within 72.8  10
04 of the desired
value.
8. As suggested by an anonymous referee, we also obtained values of g for values
of the Zheng index of 0.15, 0.20, 0.25, and 0.30. While these are available upon
request, we note that corresponding values are strongly correlated with the values for
g obtained using a value of the Zheng index of 0.10. Specifically, using the total
population (sub-population of individuals over age 65), the correlation is above 0.99
(0.95) in all cases.
9. Note, for the reader’s ease, we multiply g, the poverty aversion parameter, by
1,000 in the plots.
10. The headcount index has risen from 11.4% in 1978 to 15.1% in 1993 and the
Watts measure (NPD) rose by 57% (52%) over the same period (Table A1). See also
Levernier, Partridge, and Rickman (2000), Blank and Hanratty (1992) and Sawhill
(1988).
11. Sawhill (1988) provides a nice survey.
Changing Poverty or Changing Poverty Aversion? 259

12. Ideally one would want to compute a measure of poverty aversion for those
under age 65 and those over age 65. However, US poverty data from the census is
only defined (by age) amongst the total population and the over 65 population.
While one could use data on the total population by age category to estimate poverty
rates for those under age 65, there is no corresponding poverty threshold (to our
knowledge) defined for this sub-group. Nonetheless, we can reasonably infer that if
overall poverty aversion is negatively correlated with poverty aversion for the over 65
sub-population, then the correlation between poverty aversion amongst only those
under 65 and those over 65 is even more negatively correlated.
13. We use the word ‘‘determinants’’ loosely as we do no attempt to deal with
issues of endogeneity.
14. For robustness, we also estimated the models using ln(g0.30) as the dependent
variable. The pattern of signs on the coefficients is unchanged, and the majority of
the statistically significant results remain. The results are available upon request.
15. Alternatively, Gottschalk and Danziger (1984) and others have argued that
greater economic growth has led to a widening of the income distribution and con-
sequently higher objective poverty (see Sawhill (1988) for a review). We return to this
point in Section 5.
16. Again, we abstract from the choice of these ‘poverty lines’ and take them as
given. See footnote 5.
17. We do not attempt to limit our analysis to countries deemed democratic.
While the process by which the social consensus is mapped into the level of poverty
aversion by policymakers may be less obvious in non-democracies, such avenues do
exist. For example, ‘‘fair wage’’ models and models based on social custom argue
that societal standards and perceptions of fairness may affect the income distribution
through employment decisions, work effort conditional on employment, propensity
for collective action on the part of workers, wage-setting practices of employers, as
well as other channels (see, e.g., Agell & Lundborg, 1992; Ackerlof & Yellen, 1990;
Blinder & Choi, 1990; Naylor, 1989; Ackerlof, 1980). Moreover, it is not trivial to
distinguish between democracies and non-democracies. Finally, since (as we shall
see) the data based on the full sample support our natural rate hypothesis, if indeed
the relevant avenues are missing in non-democracies, then the results would be even
stronger upon restricting our attention to the sub-sample of democratic regimes.
18. The corresponding values of PgF are not presented, but in all cases we are able
to find a value of g which brings PgF to exactly the desired value.
19. Again, we also obtained values of g for values of the Zheng index of 0.15, 0.20,
0.25, and 0.30. While these are available upon request, we note that corresponding
values are strongly correlated with the values for g obtained using a value of the
Zheng index of 0.10. Specifically, the correlation is above 0.89 in all cases.
20. For robustness, we also estimated the models using ln(g0.30) as the dependent
variable. The pattern of signs on the coefficients is unchanged, and the majority of
the statistically significant results remain. The results are available upon request.
21. The corruption measure is obtained from Transparency International. Ac-
cording to their website: ‘‘The CPI [Corruption Perception Index] is a means of
enhancing understanding of levels of corruption from one country to another. It does
not attempt to assess the degree of corruption practiced by nationals outside their
own countries. In an area as complex and controversial as corruption, no single
260 DANIEL L. MILLIMET ET AL.

source, or polling method, has yet been developed that combines a perfect sampling
frame, large enough country coverage, and a fully convincing methodology to pro-
duce comparative assessments. This is why the CPI has adopted the approach of a
composite index. It is a ‘poll of polls’. It consists of credible surveys using different
sampling frames and varying methodologies and is the most statistically robust
means of measuring perceptions of corruption. The 1998 CPI includes data from the
Economist Intelligence Unit (Country Risk Service and Country Forecasts), Gallup
International (50th Anniversary Survey), the Institute for Management Development
(World Competitiveness Yearbook), the Political & Economic Risk Consultancy
(Asian Intelligence Issue), the Political Risk Services (International Country Risk
Guide), World Development Report (Private Sector Survey), and the World Eco-
nomic Forum (Global Competitiveness Report). The index ranges from 0 (least
corrupt) to 10 (most corrupt).’’
22. For example, De Soto (1989) and Carino (1986) examine the effect of various
government interventions on corruption. See also Acemoglu and Verdier (2000).
23. Relatedly, one may ask why policymakers would choose to do so. Such a
question ventures into political economy dimensions. For example, voters may eval-
uate policymakers by several factors, among which is included a constraint on the
perceived level of inequality and poverty. Exploration of such behavior is an inter-
esting issue for future research.
24. Inequality aversion is given by e0.10, the value of e necessary for the Atkinson
index of inequality to be equal to 0.10 (Millimet et al., 2000).

ACKNOWLEDGMENTS

The authors wish to thank John Bishop, an anonymous referee, and Buhong
Zheng for insightful comments, as well as conference participants at the 6th
International Meeting of the Society for Social Choice and Welfare, Cal
Tech University, July 2002, and at the John Formby retirement conference,
Alabama, May 2003, and seminar participants at the University of Oregon
and University of British Columbia.

REFERENCES
Acemoglu, D., & Verdier, T. (2000). The choice between market failures and corruption.
American Economic Review, 90, 194–211.
Ackerlof, G. A. (1980). A theory of social custom, of which unemployment may be one con-
sequence. Quarterly Journal of Economics, 94, 749–775.
Ackerlof, G. A., & Yellen, J. L. (1990). The fair wage-effort hypothesis and unemployment.
Quarterly Journal of Economics, 105, 255–283.
Agell, J., & Lundborg, P. (1992). Fair wages, involuntary unemployment and tax policies in the
simple general equilibrium model. Journal of Public Economics, 47, 299–320.
Changing Poverty or Changing Poverty Aversion? 261

Atkinson, A. B. (1970). On the measurement of inequality. Journal of Economic Theory, 2,

244–263.
Atkinson, A. B. (1987). On the measurement of poverty. Econometrica, 55, 749–764.
Bénabou, R., & Ok, E. A. (2001). Social mobility and the demand for redistribution. The
POUM Hypothesis. Quarterly Journal of Economics, 116, 447–487.
Blank, R.M., & Blinder, A. (1986). Macroeconomics, income distribution, and poverty.
In: S. Danzinger & S. Weinberg (Eds), Fighting poverty: What works and what doesn’t
(pp. 180–208). Cambridge: Harvard University Press.
Blank, R. M., & Hanratty, M. (1992). Down and out in North America: Recent trends in
poverty rates in the United States and Canada. Quarterly Journal of Economics, 107,
232–254.
Blinder, A. S., & Choi, D. H. (1990). A shred of evidence on theories of wage stickiness.
Quarterly Journal of Economics, 105, 1003–1015.
Carino, L. V. (1986). Bureaucratic corruption in Asia: Causes, consequences and controls.
Manila: JMC Press.
Chakravarty, S. R. (1983a). Ethically flexible measures of poverty. Canadian Journal of
Economics, 16, 74–85.
Chakravarty, S. R. (1983b). A new index of poverty. Mathematical Social Science, 6, 307–313.
Clark, S., Hemming, R., & Ulph, D. (1981). On indices for the measurement of poverty.
Economic Journal, 91, 515–526.
Cutler, D. M., & Glaeser, E. L. (1997). Are ghettos good or bad? Quarterly Journal of
Economics, 112, 827–872.
Cutler, D. M., & Katz, L. F. (1991). Macroeconomic performance and the disadvantaged.
Brookings Papers on Economic Activity, 2, 1–74.
De Soto, H. (1989). The other path: The invisible revolution in the Third World. New York:
Harper.
Epstein, G. S., & Spiegel, U. (2001). Natural inequality, production, and economic growth.
Labour Economics, 8, 463–473.
Foster, J., Greer, J., & Thorbecke, E. (1984). A class of decomposable poverty measures.
Econometrica, 52, 761–766.
Foster, J. E., & Sen, A. (1997). On economic inequality after a quarter century. In: A. Sen (Ed.),
On economic inequality (pp. 107–219). Oxford: Clarendon Press (expanded edition).
Gottschalk, P., & Danziger, S. (1984). Macroeconomic conditions, income transfers, and the
trend in poverty. In: D. L. Bawden (Ed.), The social contract revisited (pp. 185–215).
Washington, DC: Urban Institute Press.
Hirschman, A. O., & Rothschild, M. (1973). The changing tolerance for income inequality in
the course of economic development. Quarterly Journal of Economics, 87, 544–566.
Ivaschenko, O. (2002). Growth and inequality: Evidence from transitional economies, CESifo
Working Paper No. 746, Munich.
Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse? Economic Inquiry, 36,
620–630.
Kakwani, N. C. (1980). On a class of poverty measures. Econometrica, 48, 437–446.
Kakwani, N. C. (1999). Inequality, welfare and poverty: Three interrelated phenomena. In:
J. Silber (Ed.), Handbook of income inequality measurement (pp. 599–628). Norwell, MA:
Kluwer Academic Publishing.
Lambert, P. J., Millimet, D. L., & Slottje, D. J. (2003). Inequality aversion and the natural rate
of subjective inequality. Journal of Public Economics, 87, 1061–1090.
262 DANIEL L. MILLIMET ET AL.

Lanjouw, P., & Ravallion, M. (1995). Poverty and household size. Economic Journal, 105,
1415–1434.
Levernier, W., Partridge, M. D., & Rickman, D. S. (2000). The causes of regional variations in
U.S. poverty: A cross-county analysis. Journal of Regional Science, 40, 473–497.
Massey, D. S. (1996). The age of extremes: Concentrated affluence and poverty in the twenty-
first century. Demography, 33, 395–412.
Milanovic, B. (1998). Income, inequality and poverty during the transition from planned to market
economy. Washington, DC: The World Bank.
Millimet, D. L., Slottje, D., & Lambert, P. J. (2000). Inequality aversion, income inequality, and
social policy in the US: 1947–1998. Mimeo, Southern Methodist University.
Naylor, R. (1989). Strikes, free riders, and social custom. Quarterly Journal of Economics, 104,
771–785.
Ravallion, M. (1994a). Measuring social welfare with and without poverty lines. American
Economic Review (AEA Papers and Proceedings), 84, 359–364.
Ravallion, M. (1994b). Is poverty increasing in the developing world? Review of Income and
Wealth, 40, 359–376.
Ravallion, M. (1999a). Comment on Kakwani’s inequality, welfare and poverty: Three inter-
related phenomena. In: J. Silber (Ed.), Handbook of income inequality measurement
(pp. 629–634). Norwell, MA: Kluwer Academic Publishing.
Ravallion, M. (1999b). Are poorer states worse at targeting their poor? Economics Letters, 65,
373–377.
Ravallion, M., & Lokshin, M. (2000). Who wants to redistribute? The Tunnel Effect in 1990s
Russia. Journal of Public Economics, 76, 87–104.
Sawhill, I. V. (1988). Poverty in the U.S.: Why is it so persistent? Journal of Economic Lit-
erature, 26, 1073–1119.
Sen, A. (1976). Poverty: An ordinal approach to measurement. Econometrica, 44, 219–231
Reprinted as Chapter 17 in Sen (1982).
Topel, R. H. (1994). Regional labor markets and the determination of wage inequality.
American Economic Review, 84, 17–22.
World Bank. (2000). Making transition work for everyone: Poverty and inequality in Europe and
Central Asia. Washington, DC.
World Development Report 2000/2001: Attacking Poverty (2000). World Bank and Oxford
University Press.
Zheng, B. (1997). Aggregate poverty measures. Journal of Economic Surveys, 11, 123–162.
Zheng, B. (2000a). Poverty orderings. Journal of Economic Surveys, 14, 427–466.
Zheng, B. (2000b). Minimum distribution-sensitivity, poverty aversion and poverty orderings.
Journal of Economic Theory, 95, 116–137.
Changing Poverty or Changing Poverty Aversion? 263

APPENDIX

Table A1. Poverty Rates in the US, 1975–1998.

Year Whole Population Population Aged 65+

% with % with % with % with % with % with

income income income income income income
below below below below below below
50% of 75% of poverty 50% of 75% of poverty
poverty poverty line poverty poverty line
line line line line

1998 0.051 0.085 0.127 0.023 0.047 0.105

1997 0.054 0.090 0.133 0.022 0.046 0.105
1996 0.054 0.093 0.137 0.021 0.049 0.108
1995 0.053 0.093 0.138 0.019 0.047 0.105
1994 0.059 0.101 0.145 0.025 0.056 0.117
1993 0.062 0.105 0.151 0.024 0.056 0.122
1992 0.061 0.102 0.148 0.023 0.056 0.129
1991 0.056 0.097 0.142 0.022 0.052 0.124
1990 0.052 0.091 0.135 0.021 0.052 0.122
1989 0.049 0.084 0.128 0.020 0.046 0.114
1988 0.052 0.088 0.130 0.019 0.047 0.120
1987 0.052 0.090 0.134 0.019 0.053 0.125
1986 0.053 0.094 0.140 0.021 0.047 0.124
1985 0.052 0.094 0.136 0.020 0.050 0.126
1984 0.055 0.097 0.144 0.017 0.046 0.124
1983 0.059 0.102 0.152 0.022 0.053 0.138
1982 0.056 0.101 0.150 0.025 0.060 0.146
1981 0.049 0.091 0.140 0.020 0.063 0.153
1980 0.044 0.083 0.130 0.021 0.063 0.157
1979 0.038 0.073 0.117 0.024 0.066 0.152
1978 0.036 0.069 0.114 0.017 0.053 0.140
1977 0.035 0.070 0.116 0.017 0.054 0.141
1976 0.033 0.070 0.118 0.019 0.058 0.150
1975 0.037 0.073 0.123 0.020 0.057 0.153
264 DANIEL L. MILLIMET ET AL.

Table A2. Poverty Aversion, US (Whole Population).

Year g0.02 g0.04 g0.06 g0.08 g0.10

1998 0.00006 0.00012 0.00016 0.00020 0.000230

1997 0.00006 0.00011 0.00015 0.00019 0.000230
1996 0.00006 0.00011 0.00015 0.00019 0.000220
1995 0.00006 0.00011 0.00015 0.00019 0.000220
1994 0.00005 0.00010 0.00014 0.00018 0.000210
1993 0.00005 0.00010 0.00014 0.00017 0.000200
1992 0.00005 0.00010 0.00014 0.00018 0.000210
1991 0.00006 0.00010 0.00015 0.00018 0.000220
1990 0.00006 0.00011 0.00015 0.00019 0.000230
1989 0.00006 0.00012 0.00016 0.00020 0.000240
1988 0.00006 0.00011 0.00016 0.00020 0.000230
1987 0.00006 0.00011 0.00016 0.00019 0.000230
1986 0.00006 0.00011 0.00015 0.00019 0.000220
1985 0.00006 0.00011 0.00015 0.00019 0.000230
1984 0.00006 0.00011 0.00015 0.00018 0.000220
1983 0.00005 0.00010 0.00014 0.00018 0.000210
1982 0.00006 0.00010 0.00014 0.00018 0.000210
1981 0.00006 0.00011 0.00016 0.00020 0.000230
1980 0.00007 0.00012 0.00017 0.00021 0.000250
1979 0.00007 0.00014 0.00019 0.00023 0.000270
1978 0.00008 0.00014 0.00019 0.00024 0.000280
1977 0.00008 0.00014 0.00019 0.00024 0.000280
1976 0.00008 0.00014 0.00020 0.00024 0.000280
1975 0.00007 0.00014 0.00019 0.00023 0.000270

Table A3. Poverty Aversion, US (Population Aged 65+).

Year g0.02 g0.04 g0.06 g0.08 g0.10

1998 0.00011 0.00019 0.00026 0.00032 0.00037

1997 0.00011 0.00020 0.00026 0.00032 0.00037
1996 0.00011 0.00019 0.00026 0.00032 0.00037
1995 0.00011 0.00020 0.00027 0.00033 0.00038
1994 0.00010 0.00017 0.00024 0.00029 0.00034
1993 0.00010 0.00017 0.00024 0.00029 0.00034
1992 0.00010 0.00017 0.00024 0.00029 0.00034
Changing Poverty or Changing Poverty Aversion? 265

Table A3. (Continued )

Year g0.02 g0.04 g0.06 g0.08 g0.10

1991 0.00010 0.00018 0.00025 0.00030 0.00035

1990 0.00010 0.00018 0.00025 0.00031 0.00036
1989 0.00011 0.00020 0.00027 0.00033 0.00038
1988 0.00011 0.00019 0.00026 0.00032 0.00038
1987 0.00010 0.00019 0.00025 0.00031 0.00036
1986 0.00010 0.00019 0.00026 0.00031 0.00036
1985 0.00010 0.00019 0.00025 0.00031 0.00036
1984 0.00011 0.00020 0.00027 0.00033 0.00038
1983 0.00010 0.00017 0.00024 0.00030 0.00034
1982 0.00009 0.00016 0.00022 0.00027 0.00032
1981 0.00009 0.00016 0.00023 0.00028 0.00033
1980 0.00009 0.00016 0.00022 0.00028 0.00032
1979 0.00008 0.00016 0.00022 0.00027 0.00031
1978 0.00010 0.00018 0.00025 0.00031 0.00036
1977 0.00010 0.00018 0.00025 0.00031 0.00036
1976 0.00009 0.00017 0.00024 0.00029 0.00034
1975 0.00009 0.00017 0.00023 0.00029 0.00034

Table A4. International Poverty Rates.

Country Year % With Income % With Income

Below $1/Day Below $2/Day

Bangladesh 1996 0.291 0.778

Bolivia 1990 0.113 0.386
Brazil 1997 0.051 0.174
Burkina Faso 1994 0.612 0.858
Chile 1994 0.042 0.203
China 1998 0.185 0.537
Colombia 1996 0.110 0.287
Costa Rica 1996 0.096 0.263
Cote d’lvoire 1995 0.123 0.494
Dominican Republic 1996 0.032 0.160
Ecuador 1995 0.202 0.523
Egypt, Arab Rep. 1991 0.031 0.527
El Salvador 1996 0.253 0.519
Estonia 1995 0.049 0.177
Ethiopia 1995 0.313 0.764
266 DANIEL L. MILLIMET ET AL.

Table A4. (Continued )

Country Year % With Income % With Income
Below $1/Day Below $2/Day

Guatemala 1989 0.398 0.643

Honduras 1996 0.405 0.688
India 1997 0.442 0.862
Indonesia 1999 0.152 0.661
Jamaica 1996 0.032 0.252
Kazakhstan 1996 0.015 0.153
Kenya 1994 0.265 0.623
Lesotho 1993 0.431 0.657
Madagascar 1993 0.602 0.888
Mali 1994 0.728 0.906
Mauritania 1995 0.038 0.221
Mexico 1995 0.179 0.425
Moldova 1992 0.073 0.319
Mongolia 1995 0.139 0.500
Nepal 1995 0.377 0.825
Niger 1995 0.614 0.853
Nigeria 1997 0.702 0.908
Pakistan 1996 0.310 0.847
Panama 1997 0.103 0.251
Paraguay 1995 0.194 0.385
Peru 1996 0.155 0.414
Poland 1993 0.054 0.105
Romania 1994 0.028 0.275
Russian Federation 1998 0.071 0.251
Rwanda 1983–5 0.357 0.846
Senegal 1995 0.263 0.678
Sierra Leone 1989 0.570 0.745
South Africa 1993 0.115 0.358
Sri Lanka 1995 0.066 0.454
Tanzania 1993 0.199 0.597
Turkmenistan 1993 0.209 0.590
Uganda 1992 0.367 0.772
Venezuela 1993 0.147 0.364
Yemen, Rep. 1998 0.051 0.355
Zambia 1996 0.726 0.917
Zimbabwe 1990–1 0.360 0.642
Changing Poverty or Changing Poverty Aversion? 267

Table A5. International Poverty Aversion.

Country g0.02 g0.04 g0.06 g0.08 g0.10

Bangladesh 0.02893 0.05692 0.08405 0.11034 0.13585

Bolivia 0.06320 0.12239 0.17802 0.23044 0.27998
Brazil 0.13477 0.25231 0.35621 0.44911 0.53297
Burkina Faso 0.01896 0.03744 0.05545 0.07303 0.09019
Chile 0.13089 0.24701 0.35104 0.44503 0.53058
China 0.04305 0.08412 0.12336 0.16092 0.19693
Colombia 0.07550 0.14488 0.20900 0.26855 0.32409
Costa Rica 0.08378 0.16019 0.23032 0.29508 0.35517
Cote d’lvoire 0.05264 0.10262 0.15016 0.19547 0.23873
Dominican Republic 0.16504 0.30732 0.43179 0.54204 0.64072
Ecuador 0.04211 0.08225 0.12059 0.15726 0.19241
Egypt, Arab Rep. 0.06643 0.13004 0.19104 0.24958 0.30584
El Salvador 0.03811 0.07448 0.10927 0.14260 0.17458
Estonia 0.13540 0.25368 0.35838 0.45207 0.53669
Ethiopia 0.02830 0.05568 0.08221 0.10791 0.13285
Guatemala 0.02729 0.05363 0.07907 0.10367 0.12749
Honduras 0.02624 0.05161 0.07616 0.09994 0.12299
India 0.02258 0.04454 0.06591 0.08670 0.10697
Indonesia 0.04063 0.07971 0.11732 0.15357 0.18854
Jamaica 0.12049 0.23002 0.33018 0.42224 0.50727
Kazakhstan 0.20223 0.37675 0.52936 0.66432 0.78484
Kenya 0.03399 0.06666 0.09809 0.12837 0.15757
Lesotho 0.02587 0.05087 0.07505 0.09846 0.12114
Madagascar 0.01887 0.03728 0.05523 0.07274 0.08985
Mali 0.01673 0.03308 0.04905 0.06466 0.07994
Mauritania 0.12717 0.24109 0.34397 0.43753 0.52315
Mexico 0.04959 0.09641 0.14072 0.18277 0.22276
Moldova 0.08259 0.15894 0.22981 0.29587 0.35766
Mongolia 0.05009 0.09767 0.14297 0.18618 0.22745
Nepal 0.02495 0.04916 0.07268 0.09553 0.11775
Niger 0.01897 0.03745 0.05547 0.07305 0.09021
Nigeria 0.01709 0.03378 0.05009 0.06603 0.08161
Pakistan 0.02685 0.05291 0.07821 0.10279 0.12669
Panama 0.08329 0.15904 0.22843 0.29239 0.35165
Paraguay 0.05013 0.09728 0.14177 0.18386 0.22379
Peru 0.05357 0.10402 0.15166 0.19678 0.23961
268 DANIEL L. MILLIMET ET AL.

Table A5. (Continued )

Country g0.02 g0.04 g0.06 g0.08 g0.10

Poland 0.16841 0.30672 0.42376 0.52502 0.61415

Romania 0.11561 0.22169 0.31950 0.41005 0.49421
Russian Federation 0.09670 0.18436 0.26440 0.33792 0.40583
Rwanda 0.02526 0.04978 0.07360 0.09676 0.11928
Senegal 0.03260 0.06401 0.09430 0.12356 0.15183
Sierra Leone 0.02091 0.04122 0.06097 0.08017 0.09888
South Africa 0.06562 0.12681 0.18409 0.23788 0.28855
Sri Lanka 0.06637 0.12918 0.18875 0.24534 0.29921
Tanzania 0.03933 0.07702 0.11319 0.14794 0.18137
Turkmenistan 0.03882 0.07601 0.11169 0.14596 0.17893
Uganda 0.02613 0.05145 0.07600 0.09981 0.12294
Venezuela 0.05872 0.11362 0.16514 0.21365 0.25946
Yemen, Rep. 0.08446 0.16323 0.23694 0.30611 0.37121
Zambia 0.01669 0.03298 0.04891 0.06449 0.07973
Zimbabwe 0.02882 0.05661 0.08342 0.10933 0.13439
A GENDER-FOCUSED
MACRO-MICRO ANALYSIS OF
THE POVERTY IMPACTS OF
TRADE LIBERALIZATION IN
SOUTH AFRICA$

John Cockburn, Ismael Fofana, Bernard Decaluwe,

Ramos Mabugu and Margaret Chitiga

ABSTRACT
Despite the general presumption in favor of trade liberalization, the
question of how to implement it in a way to ensure equitable income
distribution and sustainable poverty alleviation in developing countries is
at the core of the current trade debate. We build a macroeconomic
framework that integrates both market and non-market activities, while
distinguishing male and female workers throughout, in order to evaluate
impacts of tariffs elimination on men and women in South Africa. Our

$
This work emanates from the medium-term sub-programme (2001–2005) being implemented
by the African Centre for Gender and Development (ACGD) of the United Nations Economic
Commission for Africa (ECA). This work also benefited from funding from the Poverty
and Economic Policy (PEP) Research Network, financed by the International Development
Research Centre (IDRC).

Equity
Research on Economic Inequality, Volume 15, 269–305
r 2007 Published by Elsevier Ltd.
ISSN: 1049-2585/doi:10.1016/S1049-2585(07)15011-0
269
270 JOHN COCKBURN ET AL.

study reveals a strong gender bias against women with a decrease in their
labor market participation, while men participate more in the market
economy. This strong result is due to the fact that female workers
are concentrated in contracting sectors that were initially among the
protected sectors and that benefit little from the fall in input prices. In
contrast, male workers are more concentrated in the expanding export-
intensive sectors. Female labor market participation drops particularly
for Black African women, as they are more concentrated in contracting
sectors. As male labor market participation and real wages increase more
than for their female counterparts, their income share increases within the
household. Women continue to suffer nonetheless from a heavy time use
burden given their increased domestic work with trade liberalization.

1. INTRODUCTION

Over the past decade, developing countries especially in Sub-Saharan Africa

have committed themselves to meeting targets in various international and
bilateral agreements including the millennium development goals (MDGs).
Many countries have implemented policy reforms aimed at achieving these
goals. These policy reforms, including trade liberalization, will have signifi-
cant repercussions on the economy of these countries and on income dis-
tribution and poverty reduction. Although the principal argument in favor
of openness to trade has been the benefit brought to all nations, its partisans
recognize that it creates winners and losers in all countries. The problem
of the distribution of gains from trade in the developing countries, where
national income is already unequally distributed, is at the core of the current
trade debate.
Gender poverty and inequality has become an important issue in devel-
oping countries. Many studies show that, compared to men, women are
more vulnerable to chronic poverty because of gender inequalities in the
distribution of income, access to productive inputs such as credit, asset
management and labor market conditions. Several other studies have
focused on a variety of gender issues, including the impact of trade liber-
alization on gender inequalities. Most of these studies1 note a significant
increase in female labor market participation during the last decade,
corresponding to the period of liberalization in the majority of the develo-
ping countries. The feminization of work in export sectors was found to
be stronger in the industrial sector and semi-industrialized economies,
than in agricultural sectors economies. In semi-industrialized countries,
Gender-Focused Macro-Micro Analysis 271

some reservations have been expressed about the welfare impacts, mostly
concerning the conditions under which female work grows. In agricultural
economies, many studies2 mention that market opportunities benefit men
more than women, because of the difficulties for women to access produc-
tive assets (loans, land, new technologies, knowledge, etc.).
Recently, gendered macroeconomic models3 have been used to analyze
the economy-wide impact of trade liberalization on male and female welfare.
These studies conclude that trade liberalization expands female work and
income more than their male counterparts in the economies analyzed.
Although, the expansion of female market work is seen as enhancing their
negotiating power within the household, it could constitute for them a bur-
den if there is not an equivalent reduction in their domestic work. Its
perverse effects on female leisure and domestic work leave some skeptical of
its benefits for women, children and other household members.
Fiscal policy is a key policy instrument in influencing men and women’s
welfare and their prospects for economic empowerment. It can contribute to
narrowing or widening gender gaps in time use, incomes, health, education,
nutrition, etc. There is also increasing concern about how gender inequality
can constrain the outcomes of macroeconomic policy. Recent studies4 show
that economic reforms with decreased incentives can reduce women’s output
or restrict access to education, and thus hinder women’s ability to develop
their human resources. They also observe that ignoring household non-
market work may affect macroeconomic outcomes by constraining labor
mobility and the supply response, as well as affecting the demand for close
market substitutes to home produced commodities.5 Therefore, interactions
between male and female work on the one hand, and market and non-market
activities on the other hand, may play key roles in policy impact analysis.
Therefore, the analysis of the impact of trade reform policy on income
distribution and poverty reduction in South Africa, and its differentiated
impacts on women and men is of crucial importance.
Conventional economics and most economic statistics ignore the enor-
mous volume of unpaid work and the undeniably valuable output of services
by the household or ‘‘care’’ economy. Households devote a large proportion
of their time to produce ‘‘home’’ commodities, which can neither be pur-
chased nor sold on the market and which, therefore, are consumed entirely
by the household themselves. Although, many of these commodities have
their equivalents in the market economy, economics is generally blind to the
unpaid work and production of women (and men) within households.
Advances in economic theory have stressed that important produc-
tive activities occur within the household, and that more attention should be
272 JOHN COCKBURN ET AL.

devoted to distinguishing domestic work activities from leisure activities,

as it is not likely that home production and leisure activities will be affected
in the same way by changes in technology, wage rates or socioeconomic
variables.
Although these gender-related development issues have prompted serious
debate, the absence of appropriate gender-focused macroeconomic analyti-
cal tools has prevented quantitative analysis. A related constraint is the
inadequate data and statistical indicators for effective gender-sensitive
policy-making, monitoring and evaluation.6
We address this deficiency by developing a gender-focused Computable
General Equilibrium (CGE) for the South African economy that distin-
guishes male and female workers throughout, as well as breaking down
market and non-market activities. This approach allows us to evaluate im-
pacts of trade liberalization on male and female time allocation between
work (at the market and at home) and leisure, as well as on household
welfare. CGE models are powerful tools to capture, in a general equili-
brium framework, all direct and indirect effects of macroeconomic shocks
(wherever the shock occurs in the economy) on sectoral production and
factor demands.

2. OVERVIEW OF SOUTH AFRICAN

CHARACTERISTICS AND POLICY ON TRADE,
POVERTY AND GENDER ISSUES

2.1. International Trade

South Africa was reintegrated into the world economy following a credible
transition to democracy symbolized by the elections in 1994. The new South
African government immediately adopted the Reconstruction and Develop-
ment Programme (RDP), which set the broad framework of the new gov-
ernment’s economic and social policy. This was followed in 1996 by the
launching of the Growth, Employment and Redistribution (GEAR) pro-
gramme, which defined policy instruments and objectives for the five years
until 2001. The pace of trade liberalization quickened after South Africa
became a signatory to the Marrakech Agreement. Initial progress in
rationalizing the very complex tariff regime and lowering the overall level of
nominal and effective protection was relatively fast. Between 1990 and 1999,
the number of tariff lines was reduced from 12,500 in 200 tariff bands to
Gender-Focused Macro-Micro Analysis 273

Table 1. Reform of South African Tariffs.

All Rates All Rates All Rates Positive Rates
(1990) (1996) (1999) (1999)

Number of lines 12500 8250 7743 2463

Number of bands 200 49 47 45
Minimum rate (%) 0 0 0 1
Maximum rate (%) 1389 61 55 55
Unweighted mean rate (%) 27.5 9.5 7.1 16.5
Standard deviation (%) n.a. n.a. 10.0 8.6
Coefficient of variation (%) 159.8 134.0 140.3 52.2

Source: Lewis (2001).

7,743 in only 47 tariff bands. In fact, if the numerous cases of zero tariffs are
ignored, the number of tariff lines had been reduced to fewer than 2,500 by
1999. At the same time, the maximum existing tariff has been reduced from
almost 1,400 to 55 percent and the average economy-wide tariff fell from
28 to 7.1 percent (Table 1).
The aggregate response of trade to liberalization has been quite dramatic.
The average annual growth rate of the trade ratio, as measured by the sum
of export and import values to GDP (in current prices), was 5.5 percent
between 1993 and 1996, 0.8 percent between 1997 and 1999 and 9.8 percent
between 2000 and 2002 (Davies & van Seventer, 2003). Closer inspection
shows that the trade ratio started to grow in 1992, perhaps reflecting the
post-apartheid reintegration. The slowdown in 1997–1999 was probably
related to the Asian crisis, but may also reflect the ending of the initial
impetus provided by the ending of apartheid. The acceleration after 1999
likely reflects both world recovery and domestic liberalization policies
starting to make an impact.
A broad look at the performance of exports in the apartheid, transition
and liberalized periods suggests that the reforms may have stimulated export
growth. During the pre-democracy period (1981–1990), the value of exports
declined by an average of 2.6 percent. Between 1991 and 1996, export
growth averaged 6.4 percent per year. During the liberalized period
(1997–2002), exports grew by only 3.7 percent per annum. However, this
latter figure masks a sharp downturn in the mining sector in recent years.
As to the composition of exports, the share of manufacturing products
increased from 41.2 percent in 1991–1996 to 53.3 percent in 1997–2001,
while the share of mining diminished from 42.3 percent in 1991–1996 to 26.9
percent in 1997–2001. Gold, the main South African export, still accounted
274 JOHN COCKBURN ET AL.

for more than 20 percent of total exports in 1997–2001, down from more
than 35 percent in 1991–1996 (TIPS, 2002). The performance of imports in
the transition and liberalized periods suggests that the reforms may have
also stimulated import growth, albeit marginally. TIPS (2002) shows that
imports grew by an average of 11.7 percent in the transition period between
1991 and 1996, but grew by only 0.1 percent between 1997 and 2001. The
relatively low growth in the latter years is puzzling given that tariffs were
falling in this period, although GDP growth was muted. Imports, just as
exports, are dominated by manufactured and mining products, accounting
for over 86 percent of imports over the two periods.
In terms of trade competitiveness, South Africa has on average experi-
enced a 50 percent decline in terms of trade over the two periods, with the
exclusion of gold (Ndlela & Nkala, 2003). Although the terms of trade
inclusive of gold have increased by about 20 percent during the same period,
the overall decline in the terms of trade reflects a critical weakness in the
structure of the country’s trade composition. Like many developing coun-
tries, a large proportion of exports consist of unprocessed raw materials
with, in the case of South Africa, the mining industry contributing the
greatest proportion to the country’s total exports. The proportion of man-
ufactured goods in exports has however experienced a significant rise, with a
higher proportion of raw materials being processed before export. Major
export commodities are gold, diamonds, platinum, wool, sugar, manganese
and chrome ores, asbestos, atomic energy materials and base minerals such
as coal, antimony, copper and iron ore. Exports of chemicals, metal pro-
ducts, machinery, transport equipment and manufactured goods have in-
creased, particularly to Africa, in recent years.

2.2. Income Distribution and Poverty

South Africa has one of the worst income distributions in the world. The
Gini coefficient, which measures the degree of income inequality, was 0.56 in
1995 and 0.57 in 2000, implying that the income distribution has also been
getting worse across the country. In addition, the Gini coefficient not
only takes high values for the country as a whole, but also for individual
population categories, suggesting a high degree of inequality within each
major ethnic group. The Gini coefficient is higher for Africans than for
Whites. More significantly, according to ILO (1999), income distribu-
tion within population groups worsened between 1990 and 1995 while
inequality decreased between groups. This may reflect the fact that the end
Gender-Focused Macro-Micro Analysis 275

of apartheid-based discrimination has created new employment opportuni-

ties for highly skilled Africans. Income is also unevenly distributed among
rural and urban areas.
There is general agreement that poverty has grown worse since liberali-
zation. The World Bank (1999) notes that extreme poverty is concentrated
mainly in rural areas, where over 75 percent of the households cannot meet
their minimum food requirements. In urban areas poverty is much less
acute, with only about 10 percent of the households below the poverty line.
The same study argues that poverty has a strong gender dimension, showing
that female-headed households have a 50 percent higher poverty rate than
male-headed households. In addition, unemployment figures have shown
that females suffer more from unemployment than males. Although no
concrete evidence is available, there is a general consensus that this pattern
has persisted over time. The UNDP (2000) gives the rate of poverty as
45 percent. This is despite the fact that South Africa is classified as an upper
middle-income country. Poverty differs greatly by region and by race, with
the majority of the poor being Black Africans. According to Klasen and
Woolard (1998), there is a strong correlation between unemployment and
poverty. They estimate that the unemployment rate among the 20 percent
poorest households is 53 percent compared to 4 percent in the case of the
richest 20 percent households. The problem is not restricted to persons with
low levels of formal education. Although education reduces the likelihood of
unemployment, rates are extremely high amongst African women, irrespec-
tive of whether or not they have completed secondary education. It should
be noted, however, that significant differences exist in the quality of edu-
cation that has been provided to different population groups. This is a
legacy of deliberate discrimination under the apartheid system. Only tertiary
education seems to substantially reduce the risk of unemployment for both
men and women.

2.3. Gender in the South African Economy

Male and female time allocation is presented in Fig. 1. At the national level,
men are more active in the labor market than women, contributing roughly
60 percent of total market labor. Women, meanwhile, perform more than
75 percent of domestic unpaid work.
Looking at employment in formal versus informal sectors, we note that,
in broad terms, women are concentrated in informal services and trade. In
1995, 33 percent of economically active African women were own account
276 JOHN COCKBURN ET AL.

100%

80%

60% Female

40% Male

20%

0%
Market work Domestic work Leisure

Fig. 1. Gender Time Allocation in 2000. Source: Statistics South Africa (2001b).

workers compared to 6 percent of men. Women are generally concentrated

within the low profit activities, generally earn less and tend to have smaller
activities in the informal sector (Baden, Hasim, & Meintjes, 1998). Although
in general more women than men work in the informal sector, the sectoral
distribution of men and women in the informal sector as well as the
occupations are very similar to the formal sector. Valodia (2000) gives more
corroborating evidence validating this.
A broad look at the participation of women in formal employment in the
apartheid, transition and liberalized periods suggests that female partici-
pation rates have been on a general increase. Standing, Sender, and Weeks
(1996) show that this trend started in the apartheid era, with female labor
force participation increasing from 23 percent to 41 percent between 1960
and 1991. According to the World Bank (2002), female labor force as
a percentage of total labor force has moved from 37 percent in 1990 to 38
percent in 2000. According to Casale and Posel (2002), the post-apartheid
period from 1995 to 1999 witnessed a continued feminization of the labor
force in South Africa. In 1995, 38 percent of all females between the ages of
15 and 65 were either working or actively looking for work. By 1999, this
had increased to 47 percent.
A gender segregated labor market in South Africa may be explained by
discrimination against women in education and training during the apart-
heid period. However, post-apartheid primary and secondary enrolment and
literacy rates suggest a dramatic improvement in these indicators for women
(World Bank, 2004). Men and women tend to work in different sectors.
Some sectors are male-intensive (i.e. mining, food, beverage and tobacco,
heavy manufacturing and construction), while others are female-intensive
(i.e. textile, private services). Women are engaged primarily in tertiary
Gender-Focused Macro-Micro Analysis 277

activities, while men are spread throughout primary and secondary sectors.
This trend between men and women has not changed much, as seen in the
2001 census data represented in Fig. 2.
As shown in Table 2, unemployment rates for women are much higher
than for men. This is true for all population groups, especially in urban
areas where unemployment was estimated at over 28 percent for women,
compared to 24.1 percent for men. Higher female unemployment may be
explained, inter alia, by lower education and literacy rates. In situations of
declining demand during the liberalized and deflationary period, women
were pushed into the informal sector.
Overall, the post-apartheid period and increased globalization have been
associated with higher female participation rates in the labor force. How-
ever, we notice that the feminization of the labor force is accompanied by an
increase in female unemployment. Where employment has grown, this seems
to have been mostly in self-employment in the informal sector. Accordingly,
given that there has been no appreciable increase in the demand for female
labor in the formal sector, these findings may reflect an increasing number of
women who are ‘‘pushed’’ into the labor market (Casale & Posel, 2002).
Data on earnings distribution by gender in the apartheid period are sparse
but are generally thought to confirm discrimination along racial lines.
Fallon and Lucas (1998) find that, while education and experience are
important determinants of earnings, other factors such as discrimination by
race and gender and barriers to mobility (i.e. geographic location and for-
mal/informal economic activity) are associated with larger differentials than
usually found in studies for other countries. In a recent study, Rospabe
(2002) shows that Black Africans earn the lowest incomes, followed by
Coloureds and Indians, while Whites have the highest earnings.
According to the 1997 October Household Survey, African workers
earned, on average, 63 percent less than White workers. As later surveys
have shown, this racial wage gap, though still significant, is smaller in com-
munity, social and personal services than in other sectors. The gap has
tended to narrow in the long run, though data should be interpreted with
caution due to methodological changes (see Hofmeyr, 1993; Crankshaw,
1997). According to ILO (1999), during the last few years the gap seems to
have remained unchanged. The continuing increase of the relative wages of
African workers in community services may be due to policy changes in the
public sector after the end of apartheid.
The same narrowing in the globalized era has not been perceived in the
gender wage gaps from available anecdotal evidence. There is a growing
literature showing gender differences are also seen in terms of earnings.
 Ag

278
ric

0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%

100%
ul

Fig. 2.
Fo ture
Land transport re
st
Fi ry
Water transport Pe M shin
tro ine g
Air transport le
um coa
Transport sppt & l
M Min ga
in e s
e
Post, telecomm m Go
O eta ld
Financial th l
Fo M er m ore
Insurance, Pensin Te od ine ini s
xt ; B S ng
ile ev er
Auxiliary finance , c , T vic
lo o e
th ba
,l c
Real estate ea o
Fo the
Rent machinery M ot r
nf w
ct ea
r, r
Computers F
M uel
Research, devlpt M ine ;
n r
M fr. al

Census.
Other business M n. e me
nf le ta
Male

r. c l

Male
Administration, defence M ele tric
nf ct al
r. ro
Education M tran nic;
nf sp
Health, social work El r. f or
ec ur t
tri nit
community; social, personal ci ur
ty e
Female

Female
;g
Membership organisations a
C W s;
on a
s te
Sports, culture, recreational W tru r
/s ct
al io
Other service R e, t n
Sa eta rad
Exterritorial orgs H le il tr e
ot
el veh ade
foreign governments , r ic
es le
ta s;
Private households ur
an
t
Undetermined
JOHN COCKBURN ET AL.

Employment by Sector and Gender. Source: Calculation from the 2001

Gender-Focused Macro-Micro Analysis 279

Table 2. Unemployment (Official Definition) by Race and Gender

(2001).
Population Group and Area Unemployed Rate (%)

Male Female

All population groups 24.8 28.0

Urban 24.1 28.6
Non-urban 26.4 26.9
African 30.0 32.3
Urban 31.0 35.7
Non-urban 28.6 27.9
Colored 21.1 22.8
Urban 23.9 23.7
Non-urban 8.8 17.1
Indian/Asian 13.9 23.0
Urban 13.5 22.9
Non-urban 31.4 30.1
White 5.6 7.8
Urban 5.7 8.0
Non-urban 4.9 5.4

Source: Statistics South Africa (2001a).

According to Fig. 3, there are substantial monthly earning differentials in

favor of men.
Women are estimated to earn 65–95 percent less than men in formal
sector employment (Valodia, 1996). Using the October household survey for
1999, Rospabe (2002) shows that the cumulative earnings distribution of
female workers is first order dominated by the distribution of income for
male workers. Further, it was found that women’s earnings were more
severely affected in manufacturing than in other sectors, with women’s
earnings equal to 73 percent of men’s in metropolitan areas. This is mainly
because of the lower positions of women (Pillay, 1993 cited in Budlender,
1995). Budlender (1997) also found that, even for those with the same
qualification, there was substantial discrimination in earnings between men
and women. Rospabe (2002) estimated the average earnings gap between
males and females to be about 20 percent. These differences are attributed
to differences in productivity between the two genders and to labor market
discrimination against women (Rospabe, 2002).
280 JOHN COCKBURN ET AL.

Men Women
14000

12000

10000

8000

6000

4000

2000

0
Formal White Formal Informal Informal Domestic Domestic Agricultural Agricultural
urban urban non-urban (urban) (non-urban) (formal) (informal)
(African) (African) (African) (African) (African) (African) (African)

Fig. 3. Mean Monthly Income by Gender (1999). Source: Statistics South Africa
(1999).

Market work Domestic work Personal care Extra-leisure time

100%

80%

60%

40%

20%

0%
Male Female Male Female Male Female Male Female Male Female

African Coloured Indian White All

Fig. 4. Household and Gender Time Allocation.

The 2000 South African survey of time use shows that men have more
market labor and leisure time. Women do more of the work of rearing and
caring for children, caring for other household members, cooking and
cleaning (Fig. 4).

3. BUILDING A GENDER-FOCUSED INTEGRATED

MACRO-MICRO MODEL

Our gender-focused integrated micro-macro model is constructed in three

steps. First, we prepare an accounting framework that brings together
Gender-Focused Macro-Micro Analysis 281

market and non-market activities using macro- and microeconomic datasets

for South Africa. Then, we incorporate into a standard CGE model labor
market segmentation between male and female workers. These are con-
sidered as different factors of production in the same way workers are
differentiated according to skill or geographical location in other contexts.
Finally, we introduce non-market activities and leisure time into the model
with the recognition that women are more likely to perform house-
hold work, while men are more active in the labor market and have more
leisure time.

3.1. Building a Gender-Focused Social Accounting Matrix with

Household-Level Data

A CGE model is generally built on the basis of a social accounting matrix

(SAM). A gender-focused SAM further distinguishes labor factors by gen-
der. Integrating real households from a representative survey of the popu-
lation requires vectors of household income and expenditures, as well as
data on their market and non-market time allocation. In this process, we
use both macro- and microlevel datasets, which we reconcile in a single
framework.
Fig. 5 illustrates the procedure used in building a gender-focused SAM
with household-level data on income, expenditures and the allocation of
time to various activities. We first bring together the Supply and Use Tables
(SUT) and the integrated economic accounts (IEA), both for year 2000, in a
single framework: a standard SAM (step one). Then, household-level data
on income and expenditure as well as male and female market work, are
computed from household surveys, i.e. the Income and Expenditure Survey
(IES) and the September Labor Force Survey7 (LFS), once again both for
year 2000 (step two). We then reconcile these household-level data and the
standard SAM to generate a gender-disaggregate SAM (step three). Fourth,
the time use survey (TUS) for year 2000 is combined with household income
and expenditure data, and individual market work data, to impute time
spent by individuals in non-market activities, i.e. domestic work, leisure and
personal care activities (step four). Fifth, household non-market work,
leisure and personal care time are incorporated into the gender-disaggregate
SAM to generate a gender-focused SAM (step five). The concept of a
National Satellite Account of Household Production (NSAHP) is used
to incorporate non-market activities (household production of services
and leisure activities) into the standard SAM as recommended by the 1993
282 JOHN COCKBURN ET AL.

Supply and Use Integrated Economic Income and Expenditure Labor Force
Tables for 2000 Accounts for 2000 Survey for 2000 Survey for 2000

Step 1
Step 2

Standard Social Accounting

Matrix for 2000
Gender Disaggregated Data on
Step 3 Household Income +
Expenditure + Market
Time Use
Gender-Disaggregate Social Step 4 Survey for
Accounting Matrix for 2000 2000

NSAHP=Non-SNA Prod +
Step 5 SNA Prod + Leisure =GHP

Step 6
Extended Gender-Aware Gender-Aware CGE Micro
Social Accounting Matrix Simulation Model
with NSAHP for 2000 Integrated approach

Fig. 5. From NSAHP to a Gender-Focused CGE Microsimulation Model.

System of National Accounts of the United Nations. The sixth and final
stage is to distinguish each individual household within the SAM in order to
obtain a gender-focused SAM with real households.
In this procedure, time spent by individuals on non-market work and
leisure activities is converted into monetary value by assigning a price. The
opportunity cost approach is used to impute a unitary value to the time
spent by individuals on various non-market activities. This price is approxi-
mated to the ‘‘expected’’ wage rate that an individual would have received if
he or she had sold his/her time (or labor services) to the market rather than
performing non-market activities. The expected wage rate is predicted for
each individual in the household based on individual characteristics (age,
gender, etc.).
The estimated value of non-market labor is used as an indicator of the
value of household production. Therefore, home-produced goods ‘‘directly’’
require neither capital nor inputs by assumption. Substitution and com-
plementarity between durable and non-durable goods in the home produc-
tion of services are ‘‘indirectly’’ integrated in the consumption decisions
of households.
Gender-Focused Macro-Micro Analysis 283

3.2. The Gender-Focused Integrated Macro-Micro Model

This study uses a CGE model based on the neoclassical-structuralist speci-

fication presented in Decaluwe, Martens, and Savard (2001). The model
seeks to explain production, consumption and prices in an economy in
which consumers and producers respond to relative prices based on wel-
fare and profit maximizing consumption and production behavior, and
markets simultaneously adjust relative prices in order to clear markets.
Though, most of the equations have strong microeconomic foundations,
the conceptualization of the economy also allows a strict macroeconomic
analysis such that the behavior of agents is consistent with macroeconomic
constraints.
The model incorporates additional features of particular interest for
developing countries. The model explicitly treats trade and transportation
margins for commodities that enter the market sphere. A constant trade and
transportation margins coefficient is added to each transaction, included in
the price, and the corresponding revenues generated are a source of demand
for the trade and transportation sector.
Labor markets have been treated to reflect empirical evidence in deve-
loping countries and South African specificities. Initially, there are eight
categories of workers distinguished by their residential area (urban and
rural), age (child and adult) and skill categories in the case of adult workers
(high, medium and low). The model explicitly treats unemployment as a
consequence of labor market imperfections in South Africa.
Most standard CGE models make the implicit assumption that male and
female workers are perfect substitutes in market production and thus do not
distinguish them. However, many studies underline the fact that there is
segmentation in the labor market between men and women, and different
levels of market work flexibility according to the domestic tasks they per-
form. Also, it is observed that male and female workers tend to concentrate
in different sectors and occupations, which further undermines the hypothe-
sis of perfect substitutability. Finally, it is widely recognized that there is
often a gender bias against women in the labor market in term of wage
earnings and job opportunities. Indeed, the 2002 report on men and women
in South African shows that the unemployment rate is higher for women
than for men within each population group, and in both urban and rural
areas. Formal sector work is far more common for men than for women.
Employed women tend to cluster into a small number of industries
compared to men; and women are significantly more likely than men to be
employed in clerical jobs while men are primarily employed as operators.
284 JOHN COCKBURN ET AL.

Mean hourly earnings are higher for men than women across all population
groups. Therefore, the first step of the modeling exercise will consist in the
segmentation of the labor market into male and female workers to highlight
the gender bias observed in the South African economy.
Therefore, male labor (LDmal fem
i ) and female labor (LDi ) are imperfect
substitutes in the aggregate sector i labor demand (LDi ). The conditional
demand of male and female labor depends on initial sectoral shares, wage
rates and their degree of substitution in sectoral production.
LDi ¼ f ðLDmal fem
i ; LDi Þ

The gross earnings of male and female workers are equal to the volume of
labor services demanded by productive sectors valued at market wage rates.
Household labor income (Yh) consists of male and female wage incomes.
Assuming labor market imperfections and the presence of unemployment,
only a proportion of the total hours supplied by households to the labor
market is hired. LSh is household h labor supply, w and u the rate of wage
and the rate of unemployment, respectively.
Y h ¼ f ðLSmal fem
h ; LSh ; w
mal
; wfem ; umal ; ufem Þ
We introduce non-market activities into the model with the recognition
that women are more likely to perform household work while men are
more active in the labor market and have more leisure time. Furthermore,
modeling non-market activities alongside market activities makes it possible
to assess (i) the importance of household production of services which are
intensive in female work. These services, which are not sold in the market
and therefore entirely consumed by the household, enter in competition with
their market substitutes,8 (ii) the impact of constraints faced by women at
the household level (because of their involvement in family tasks) that may
negatively affect their labor market participation and the performance of the
overall economy, (iii) the impacts on female leisure time. Increased female
participation in the labor force will not necessarily improve their welfare if
they still perform most of the domestic work and must therefore reduce
their leisure time, and (iv) the impact on child education at the household
level. If children, especially girls, are required to assume the household tasks
of female adults who have entered the labor market, their education and
leisure time could be negatively affected.
Men and women substitute the time devoted to leisure and to the pro-
duction of home goods, which are imperfect substitutes for market goods.9
Male (LZmal fem
h ) and female domestic work (LZh ) are imperfect substitutes in
home good production (Zh), which, by assumption, does not require either
Gender-Focused Macro-Micro Analysis 285

intermediate goods or capital.10

Z h ¼ f ðLZmal fem
h ; LZh Þ

The relative demand for male and female labor in home production de-
pends on their relative share (amal
h and afem
h ) in home production, male and
mal
female expected wage rates (w and wfem ) and the degree of trade-off
between men and women in home production represented by the elasticity
of substitution11 (eh).

LZmal
h
¼ f ðamal fem
h ; ah ; w
mal
; wfem ;
h Þ
LZfem
h

The value of home-produced goods is equal to the value of the labor

devoted to their production, where non-market labor is valued at its
opportunity cost as measured by the expected market wage rates.
An extended linear expenditure system is specified to derive household
demand for home goods subject to full income; where ‘malh and ‘fem
h represent
z m
male and female leisure time, respectively; C h and C h;i are home produced
goods and market goods consumptions, respectively.

U h ¼ f ðC m z mal fem
h;i ; C h ; ‘h ; ‘ h Þ

Men and women allocate their total available time in two steps. First, the
total exogenous time (hours) available for market and non-market activities
are allocated to domestic activities (according to home goods production
requirement and the degree of substitutability among men and women in
home production), to leisure activities (the demand for male and female
leisure is derived from the utility function) and to market activities as resi-
duals. Second, the hours of labor supplied to the market is allocated be-
tween work and unemployment.12

4. SIMULATIONS AND RESULTS

Our first simulation involves the elimination of all import tariffs where
government revenue is held constant through the introduction of an en-
dogenous adjustment in indirect taxes. Trade liberalization emerges as one
of the key policy issues of the GEAR agenda discussed in previous sections.
Although other trade barriers still exist, tariffs constitute the principal
286 JOHN COCKBURN ET AL.

protectionist measure in South Africa. The removal of tariffs in South

Africa modifies the entire price structure and, consequently, factor returns.
The impact on households depends on their factor endowments and their
consumption patterns. Trade liberalization also has differential impacts on
men and women depending on the sectors in which they are intensively
employed and the household to which they belong.

4.1. Trade and Output Effects

The initial impact of the removal of all tariffs is a fall in the domestic price
of imports that is particularly strong in the highly protected sectors. Local
consumers react to the fall in import prices by increasing their imports by
sector roughly in the same proportion as the fall in sectoral tariff rates.
Given a fixed current account balance, the increase in imports leads to a
4.9 percent exchange rate depreciation, which partially offsets the fall in
import prices in these sectors and leads to an increase in the domestic price
of imports in some other sectors (see Appendix 1). Of course, this import
surge comes at the expense of domestic competitors, who experience a
decline in the volume and price of their sales on the local market. Given
the imperfect substitution between local and imported goods (CES), as
well as the relatively small initial import intensities (imports/domestic
consumption), the changes here are proportionally much smaller than
the variations in import volumes. Nonetheless, the sectors with the most
substantial reductions in local sales are the highly protected sectors.
The exchange rate depreciation also results in a 1.8 percent increase in
export volume. Exports increase most in the export-intensive sectors, i.e. the
sectors with the highest initial export intensity ratios (Exports/Output).
These sectors are identified in italics in Appendix 1. Variations in exports
and domestic sales determine changes in total output. Given their loss in
domestic sales, the highly protected sectors also experience the strongest
output declines.
Output prices are averages of export prices, which are assumed fixed, and
domestic prices, weighted by the share of sales on each of these markets.
Thus, it is unsurprising that they fall strongly in most of the highly protected
sectors, while they increase in sectors with initially low tariff rates. Note that
these are after-tax output price variations and thus include a 13.4 percent
increase in the indirect tax rate required to balance the government’s budget,
whereas the import and domestic sales price variations are shown net of
indirect taxes.
Gender-Focused Macro-Micro Analysis 287

4.2. Factor Effects

We now examine how the trade and output effects above influence factor
prices and unemployment rates, crucial components of the ultimate welfare
and poverty effects.
To understand these results, note that factor prices are driven by value-
added prices as the source of their remuneration. While changes in value-
added prices generally reflect output price variations, their evolution is more
positive (less negative) when input costs rise less (fall more) than output
prices. Thus value-added prices generally fall most among the highly pro-
tected sectors whereas they increase among sectors with low initial tariffs.
However, export-oriented industries are among the sectors with the strongest
increases in value-added prices. Indeed, beyond the modest increase in their
output prices, these sectors benefit more from falling input costs, given the
high share of their inputs that come from the initially highly protected sectors.
The contrary is true for the high value-added (low input) agricultural sector,
in which value-added prices fall. The rest of this table shows the shares of
total income that each factor derives from each of the sectors (Appendix 2).
The relationship between the wage rate and the unemployment rate is
represented by a downward-sloping relation (the ‘‘wage curve’’). As a result,
rising (falling) wage rates are associated with falling (rising) unemployment
rates. Essentially, rising demand for certain types of labor translates into
increased wages and employment, whereas falling demand has the opposite
effect. Consequently, we focus our analysis here on the wage impacts with the
understanding that the unemployment effects are generally the mirror image.
To explain wage impacts, we refer throughout to the value-added price
variations and factor intensities in Appendix 2. Whereas public sector
employment and wage rates are assumed fixed, private sector workers are
assumed to be mobile between sectors with wage rates that equalize across
all private sectors. We note substantial differences in private sector wage
and unemployment rate changes according to the gender, skills and location
of workers (Table 3).
Male wage rates generally evolve more favorably than female wage rates,
with the exception of high-skilled urban and low-skilled rural workers.
Female workers are penalized by their greater participation in garments, as
well as health and social work, for which value-added prices fall (Appendix
2). In contrast, male workers benefit from their strong participation
in mining activities, which offsets their dependency on agricultural wages,
especially in rural areas. Among child workers, in both rural and urban
South Africa, the decline in wage rates for girls can be traced primarily
 288
Table 3. Wage and Unemployment Rate Variations.
Urban Urban All Rural Rural All All

High skill Medium skill Low skill Child High skill Medium skill Low skill Child

Change in wage rates

Males
0.01 0.05 0.02
0.09 0.02
0.01 0.02
0.12
0.24
0.01 0.02
Females 0.01 0.01
0.01
0.09 0.01
0.04
0.06
0.09
0.64
0.06 0.00
All
0.01 0.04 0.01
0.09 0.02
0.02 0.00
0.11
0.31
0.03 0.01
Change in unemployment rates
Males 0.14
0.53
0.24 0.83
0.31 0.13
0.09 1.17 1.85 0.09
0.27

JOHN COCKBURN ET AL.

Females
0.10
0.10 0.07 0.37
0.08 0.40 0.58 0.88 6.59 0.56
0.03
All 0.05
0.38
0.10 0.52
0.23 0.25 0.04 1.05 2.77 0.23
0.18
Gender-Focused Macro-Micro Analysis 289

to their greater participation in agriculture, whereas boys diversify into a

number of services sectors where prices generally rise modestly.
Capital is assumed to be sector-specific because of the short-term horizon
of our analysis. As a result, variations in the rates of returns to capital
closely follow changes in the value-added prices of their respective sectors.
These rates fall most in the highly protected sectors and increase in the
expanding export-oriented sectors (Appendix 2).

4.3. Time Allocation Analysis

We have already noted that men are more active in the labor market than
women, whereas women are more heavily involved in domestic work. We
also noted that men and women tend to work in different sectors. Most
sectors are male intensive with the notable exceptions of textiles and
garments and a number of service sectors.
Households respond to the changes in real wage rates for male and female
workers by changing their allocation of time between market and non-
market activities. Labor market participation decisions depend on labor and
non-labor income effects, which are taken at the household level. As a
consequence, higher real wage rates will not necessarily induce an increase in
the labor market participation of workers as labor income from other
members, non-labor income and non-market activities may also play
important roles.
Women, especially in urban areas, increase their market participation
while male market participation stagnates (Table 4). Male and female
market participation in rural areas fall as their real wage rates decrease.
Female market participation increases within the Black population cate-
gory, because of their low endowment of high- and medium-skilled workers
that win from tariff elimination in South Africa, and decrease within other
household categories.
Men and women work more at home, as they substitute market goods
with rising prices by home produced goods. Although females already do
more domestic work than men, they continue to carry out most of the
domestic tasks, especially in urban areas. Their market and non-market
work increase is roughly double that of men, at the expense of their pure
leisure time. As a large proportion of their time spent outside market work is
devoted to leisure activities rather than domestic work, men perform even
less domestic work with trade liberalization, especially in urban areas and
within female-headed household categories.
 290
Table 4. Change in Hours Worked (Percent Variation).
Market Work Domestic Work

Urban Urban Urban Urban All Rural Rural Rural Rural All All Urban Urban Urban Urban All Rural Rural Rural Rural All All
high medium low child urban high medium low child rural high medium low child urban high medium low child rural
skill skill skill skill skill skill skill skill skill skill skill skill

South Africa
Male
0.01
0.01 0.01 0.01 0.00 0.00
0.04
0.04
0.55
0.02 0.00
0.01 0.01
0.09
0.03
0.04 0.03 0.01 0.89 0.87 0.56 0.07
Female 0.00
0.02 0.07 0.03 0.04
0.03
0.04
0.54 0.00
0.03 0.03 0.08 0.00
0.01 0.05 0.03 0.23
0.05 0.47 1.11 0.54 0.14

Head of household
Male head
Male
0.01
0.01 0.01
0.01 0.00
0.01
0.06
8.04 0.00
0.02 0.00
0.01 0.01
0.06 0.00
0.02 0.04 0.03 1.02 0.90 0.61 0.09
Female 0.00
0.01 0.07
0.16 0.04
0.02
0.02
0.07 0.00
0.02 0.03 0.08
0.02
0.05
0.01
0.02 0.21
0.10 0.52 1.01 0.47 0.08

Female head
Male
0.01 0.00 0.09 0.04 0.03 0.04
0.03 0.00
2.85
0.03 0.02 0.04 0.01
0.17
0.08
0.10
0.06 0.00 0.69 0.83 0.47 0.04
Female
0.01
0.02 0.04 0.09 0.04
0.11
0.05
4.66 0.00
0.07 0.02 0.09 0.01 0.05 0.10 0.08 0.30
0.02 0.42 1.19 0.60 0.20

Population group
Black
Male
0.01 0.00 0.00 0.01 0.00 0.00
0.04
0.04 0.00
0.02 0.00 0.03
0.02
0.10
0.04
0.05 0.04 0.01 0.90 0.86 0.53 0.08
Female 0.00
0.01 0.09 0.03 0.06
0.03
0.04
0.79 0.00
0.03 0.04 0.06
0.02
0.01 0.05 0.03 0.25
0.05 0.47 1.11 0.55 0.14

JOHN COCKBURN ET AL.

Colored
Male
0.01 0.00 0.02
0.02 0.00 0.00
0.04 0.00
5.76
0.04 0.00 0.04
0.02
0.08
0.02
0.04
0.01 0.01 0.91 0.87 0.57 0.10
Female 0.00 0.00 0.04
0.13 0.02
0.08
0.01
0.14 0.00
0.06
0.02 0.00
0.11 0.03 0.06 0.05 0.15
0.04 0.35 1.01 0.26 0.13

Asian
Male 0.00
0.02 0.01
0.01 0.00 0.02 0.00 0.00 0.00 0.00 0.00 0.08 0.02
0.07
0.03
0.03 0.00
0.05 1.03 0.84 0.68 0.09
Female 0.00 0.00 0.00 0.00 0.00
0.02 0.00 0.00 0.00
0.02
0.02 0.00 0.00 0.00
0.02
0.02 0.10 0.00 0.00 1.15 0.54 0.07

White
Male 0.00
0.01 0.02 0.02 0.00 0.00
0.08 0.00 0.00
0.02 0.00
0.03 0.01
0.09
0.02
0.03 0.04 0.11 0.86 0.94 0.68 0.04
Female 0.00
0.13 0.00 0.01 0.00 0.03 0.03 0.00 0.00 0.03 0.00 0.11 0.10 0.05 0.04 0.05 0.23
0.24 0.45 1.20 0.73 0.10

Unspecified
Male
0.04
0.62
0.44 0.00
0.13 0.03
0.10 0.00 0.00
0.02
0.13 0.04 0.44
0.06
0.04
0.01 0.02
0.01 0.51 0.87 0.42 0.04
Female 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Gender-Focused Macro-Micro Analysis 291

4.4. Income, Consumption and Welfare Effects

With the CGE microsimulation approach, we obtain income variations

for each individual household, which we can then group in a variety of ways.
We present results grouped according to residential area, population group
and the gender of the household head. We first note that wages are the
principal source of income in South Africa, followed by various forms of
transfers (from other households, government, etc.), although there are
substantial differences between household groups.
Overall, incomes do not change significantly (see Table 5). Incomes evolve
more favorably in urban than rural areas, as a consequence of rising wage
rates and market participation, and the capital remuneration effects. Male-
headed households benefit from trade reform policy, whereas female-headed
households suffer from a significant drop in the returns to their capital. All
population groups except ‘‘White’’, which shows a stagnation in income,
benefit from free trade in terms of income. The ‘‘Asian’’ population group
is the big winner, as they benefit from a significant increase of the returns to
their capital.
In addition to its income effects, trade liberalization influences household
welfare by changing consumer prices. While pre-tax prices fall both for
imports and, in the face of increased import competition, domestic goods,
consumer prices increase by 0.9 percent due to exchange rate depreciation
and the increase in the sales tax required to offset lost tariff revenues. The
required increase in the initial indirect tax rate (surtax) is quite small (13.4
percent) due to the relative small share of tariffs in government revenue
and the average sale tax of 4.5 percent. However, consumption prices for
agriculture goods increase less than services and manufacturing goods, as
they present the smallest average tax rate.
Each household is affected differently by consumer price reductions
according to its consumption patterns. In this respect, we note that con-
sumer prices increase for all household, consequently, tariff elimination on
imports leads to a welfare loss in South Africa.

4.5. Poverty and Inequality Analysis

The advantage of the integrated CGE microsimulation approach is its

capacity to capture the heterogeneity of household income sources and
consumption patterns in order to perform the between- and within-group
distribution, poverty and inequality analysis. The model is used to generate
 292
Table 5. Household Income and Expenditure Effects (in Percent).
South Africa Residential Area Head of Household Population Group

Urban Rural Male head Female head Black Colored Asian White Unspecified

All incomes 0.04 0.09
0.27 0.06
0.04 0.04 0.05 0.16 0.00 0.46
Income taxes
0.21
0.21
0.16
0.23 0.00 0.04 0.08 0.05
0.29 0.12
Transfers out 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Savings 0.22 0.44
1.22 0.40
0.10 0.07 0.02 0.23 2.67 1.68
Consumption 0.05 0.09
0.13 0.07
0.02 0.04 0.07 0.12 0.05 0.05
Consumer price index 0.92 0.89 1.05 0.90 0.99 0.86 0.79 1.19 0.99 0.72
EV/initial income
0.23
0.21
0.35
0.24
0.15
0.17
0.11
0.17
0.31
0.33

JOHN COCKBURN ET AL.

Gender-Focused Macro-Micro Analysis 293

the post-simulation data. Then these data and the base year data drawn
from the income and expenditure survey are used to compute and compare
standard consumption-based poverty and inequality indicators.
Foster–Greene–Thorbecke (FGT) poverty indicators (i.e. headcount in-
dex, poverty gap and squared poverty gap) and the Theil inequality index
are adopted. We define the poverty line as 3,864 South African rands per
year in 2000 prices, a lower bound poverty line suggested by Hoogeveen
and Özler (2004). Post-liberalization consumption data are deflated by the
Laspeyres economy-wide consumer price index to account for the change in
the general price level.
Results presented in Table 6 suggest that the impacts of complete tariff
removal on poverty are small. Poverty and inequality increase slightly.
Poverty indicators increase more in rural areas than in urban areas. This is
confirmed by the increasing inequality in rural areas. Poverty increase more
among female-headed, colored and Black households, whereas they increase
slightly or remain stable for male-headed, Asian, White and unspecified
households.
Our analysis until now has been at the household level. However, given
our preoccupation with the gender impacts of trade liberalization, we
exploit the microsimulation aspect of our analysis, including detailed in-
formation on all individuals in the sample households, to analyze poverty

Table 6. Poverty and Inequality Indexes (in Percent).

Initial Values Variation

P0 P1 P2 Theil index P0 P1 P2 Theil index

South Africa 53.0 25.3 15.0 1.6 0.29 0.26 0.20 0.06
Residential area
Urban 42.4 18.4 10.2 1.6 0.23 0.21 0.14
0.01
Rural 68.3 35.4 22.1 1.0 0.37 0.34 0.27 0.07

Head of household
Male 43.6 19.5 11.1 1.6 0.19 0.22 0.15 0.03
Female 65.8 33.4 20.5 0.8 0.43 0.32 0.26 0.03

Population group
Black household 61.0 29.5 17.6 1.1 0.31 0.30 0.23 0.07
Colored household 36.2 14.7 7.8 0.8 0.45 0.19 0.12 0.01
Asian household 6.4 2.3 0.8 0.3 0.00 0.04 0.03 0.01
White household 0.1 0.0 0.0 1.0 0.00 0.00 0.00 0.06
Unspecified household 11.4 3.1 0.8 1.7 0.00 0.08 0.04
0.06

Notes: P0 ¼ poverty headcount; P1 ¼ poverty gap; P2 ¼ poverty severity.

294 JOHN COCKBURN ET AL.

and inequality impacts separately for men, women and children. Note that
we do not attempt to integrate issues of intra-household allocation and
simply assume that income and consumption are shared evenly. Thus, men,
women and children are considered to be poor if they belong to a poor
household, i.e. a household for which consumption expenditure per capita is
less than the poverty line. It can be shown that these results are robust for a
wide range of poverty lines.
Table 7 indicates that, in South Africa, 63 percent of children and 51
percent of women are poor (live in poor households), as compared to only
44 percent of men. This hierarchy is reproduced for all household categories,
with the exception of female-headed households, in which the incidence

Table 7. Poverty Indexes by Gender and Age.

Men Women Children

P0 P1 P2 P0 P1 P2 P0 P1 P2

Base year values (%)

South Africa 43.8 19.9 11.5 50.8 23.9 14.0 62.7 31.2 19.0
Urban area 35.1 14.9 8.1 41.6 18.1 10.0 51.7 22.8 12.8
Rural area 61.5 30.2 18.5 65.9 33.4 20.5 73.7 39.7 25.4
Male headed 36.6 15.6 8.7 41.9 18.7 10.6 53.6 24.9 14.6
Female headed 66.0 33.2 20.3 59.4 28.9 17.3 72.6 38.1 23.9
Black 51.8 23.8 13.9 60.1 28.5 16.8 68.9 34.7 21.3
Colored 30.8 11.9 6.1 34.6 14.3 7.7 43.0 17.6 9.3
Asian 5.5 2.0 0.8 2.9 1.0 0.3 12.3 4.3 1.6
White 0.0 0.0 0.0 0.2 0.1 0.1 0.0 0.0 0.0
Unspecified 0.0 0.0 0.0 7.9 2.1 0.6 21.8 5.8 1.6
Variations after simulation (%)
South Africa 0.22 0.22 0.16 0.31 0.26 0.19 0.32 0.30 0.23
Urban area 0.26 0.17 0.11 0.26 0.21 0.15 0.15 0.25 0.17
Rural area 0.13 0.32 0.24 0.39 0.34 0.27 0.50 0.35 0.29
Male headed 0.19 0.19 0.13 0.23 0.21 0.15 0.14 0.26 0.19
Female headed 0.30 0.31 0.25 0.38 0.31 0.23 0.53 0.35 0.28
Black 0.23 0.26 0.19 0.35 0.31 0.23 0.33 0.33 0.26
Colored 0.42 0.17 0.10 0.37 0.18 0.12 0.55 0.22 0.15
Asian 0.00 0.03 0.02 0.00 0.02 0.01 0.00 0.07 0.05
White 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00
Unspecified 0.00 0.00 0.00 0.00 0.05 0.03 0.00 0.15 0.08

Notes: P0 ¼ headcount index; P1 ¼ poverty gap; P2 ¼ squared poverty gap (poverty severity
index).
Gender-Focused Macro-Micro Analysis 295

of poverty is less among women than among men. Note that this does
not reflect intra-household allocation, as this is ignored. Instead, it indicates
that there tends to be a higher ratio of men to women in poor female-headed
households than in non-poor female-headed households. The results of
our trade liberalization scenario in the bottom half of Table 7 indicate
that poverty increases slightly more among women and children than among
their male counterparts. In particular, the elimination of import tariffs
is likely to increase more among women and children living in poverty
than men. This gender and age bias in the poverty results is particularly
strong for individuals in rural areas, female-headed and Black house-
holds. It can be shown also that those results are robust for a wide range of
poverty lines.

5. CONCLUSIONS

South Africa is in the midst of an ambitious trade liberalization program,

notably in the context of various regional and international trade agree-
ments. These policies are likely to have wide-ranging effects on the South
African economy, in particular its international trade, production, govern-
ment revenues, factor markets, household incomes and consumer price
structure. To analyze the poverty impacts on South African men, women
and children, we construct a CGE microsimulation model including 4,000
actual households from a nationally representative household survey and
featuring the explicit modeling of male and female market and domestic
work activities and leisure time.
South Africa has a very discriminatory tariff structure with rates varying
from 0 to 112.8 percent. The high protection sectors are predominantly
comprised of light manufacturing activities such as garments, beverages and
tobacco, structural metal, electrical equipment and household appliances.
These sectors are found to suffer from a contraction in output and value-
added prices subsequent to trade liberalization. In contrast, export-oriented
sectors such as mining, transport and communication equipment, machinery
and medical instruments expand as a result of the import-driven exchange
rate depreciation and the fall in input costs.
As male workers tend to be more heavily involved in export-oriented
sectors, whereas women work more in the highly protected light manufac-
turing activities and services, male wage rates rise with respect to female
296 JOHN COCKBURN ET AL.

wage rates. An interesting contrast emerges between urban and rural work-
ers. While urban wage rates tend to perform better than rural wage rates,
results vary substantially between skill categories. Medium- and urban low-
skilled workers, especially men, are the big winners from our trade experi-
ment. In contrast, high- and rural low-skilled, and child workers fare worse
as a result of their dependency on wages from the agricultural sector in rural
areas, and from social services in urban areas. Consequently, wage effects
are in favor of the more skilled workers in rural areas and in favor of the less
skilled urban workers. As capital is assumed to be immobile in the short run,
rates of return closely mirror the evolutions in sectoral value-added prices.
The upshot of all these changes is an increase in the incomes of urban and
male-headed households relative to their rural and female-headed counter-
parts. These results compound with a greater reduction in consumer prices
among urban and male-headed households to generate for them a smaller
increase in poverty and a reduction in inequality.

NOTES
1. Elson and Pearson (1981), Standing (1989), Cagatay and Ozler (1995), Joekes
(1995, 1999) and Ozler (2000, 2001).
2. Gladwin (1991) and Fontana, Joekes, and Masika (1998).
3. Fontana and Wood (2000), Fontana (2001, 2002) for Bangladesh and Zambia;
Fofana, Cockburn, and Decaluwe (2003, 2005) for Nepal; and Siddiqui (2004) for
Pakistan.
4. Haddad, Brown, Richter, and Smith (1995), Cagatay, Elson and Grown (1995)
and Palmer (1994).
5. Elson (1995), Sinha (1999, 2000) and Fofana et al. (2003).
6. Latigo and Ironmonger (2004).
7. The year 2000 Income and Expenditure Survey (IES) is based on the same
sample of households as the September 2000 Labor Force Survey (LFS: 2).
8. We observe some complementarity between market and non-market productive
activities.
9. Gronau (1977) and Solberg and Wong (1992) assume that home goods are
perfect substitutes for market goods. However, in other versions of Gronau’s model,
these goods are imperfect substitutes. For other assumptions to simplify the mode-
ling aspects, see Fofana et al. (2003).
10. Domestic paid labor, capital goods and intermediate goods are included in the
household utility function and indirectly substitute to domestic unpaid labor which is
referred here as home goods.
11. We assume that there are very limited substitution possibilities between men
and women in the production of home goods, reflected by a low elasticity of sub-
stitution (0.5) between male and female domestic work.
12. For details on technical aspect, refer to Fofana et al. (2003).
Gender-Focused Macro-Micro Analysis 297

ACKNOWLEDGMENTS

The contribution of both Dr. Alfred Latigo and Mr. Omar Abdourahaman
from the ACGD/Economic Commission for Africa in providing comments
for this study is greatly appreciated. However, the opinion expressed in this
paper is the only responsibility of the authors without engaging the ACGD
or the Economic Commission of Africa.

REFERENCES
Baden, S., Hasim, S., & Meintjes, S. (1998). Country gender profile: South Africa. Report
prepared for the Swedish International Development Office, Pretoria, Republic of South
Africa, Bridge Report no. 45.
Budlender, D. (1995). Women and the labor market paper for ILO DRAFT mimeo.
Budlender, D. (Ed.). (1997). Women and men in South Africa. Statistical booklet prepared for
the Central Statistical Services (mimeo).
Cagatay, N., Elson, D., & Grown, C. (1995). Introduction. World Development, 23(11),
1827–1836.
Cagatay, N., & Ozler, S. (1995). Feminization of the labor force: The effects of long-term
economic development and structural adjustment. World Development, 23(8), 1883–1894.
Casale, D., & Posel, D. (2002). The continued feminization of the labor force in South Africa:
An analysis of recent data and trends. The South African Journal of Economics. [Die
Suid-Afrikaanse Tydskrif vir Ekonomie], 70, 1 March-Maart 2002.
Crankshaw, O. (1997). Race, class and the changing division of labor under apartheid. London:
Routledge.
Davies, R., & van Seventer, D. (2003). A gap and macro decomposition analysis for South Africa,
1993–2002. Development Policy Research Unit, School of Economics, University of
Cape Town.
Decaluwe, B., Martens, A., & Savard, L. (2001). La politique économique du développement et les
modèles d’équilibre général calculable. Presses de l’Université de Montréal, Montreal,
Canada.
Elson, D. (1995). Gender awareness in modeling structural adjustment. World Development,
23(11), 1851–1868.
Elson, D., & Pearson, R. (1981). Nimble fingers make cheap workers. Feminist Review, 7,
87–107.
Fallon, P., & Lucas, R. (1998). South African labor markets. Adjustment and inequalities.
Discussion Paper no. 12. World Bank, Southern Africa Department, Washington.
Fofana, I., Cockburn, J., & Decaluwe, B. (2003). Modeling male and female work in a com-
putable general equilibrium. CIRPÉE, Université Laval, Québec, Canada (mimeo).
Fofana, I., Cockburn, J., & Decaluwe, B. (2005). Developing country superwomen: Impact of
trade liberalization on female market and domestic work. Working Paper 05-19. CIRPEE,
University Laval.
Fontana, M. (2001). Modeling the effects of trade on women: A closer look at Bangladesh.
Working Paper 139. Institute of Development Studies, Brighton.
298 JOHN COCKBURN ET AL.

Fontana, M. (2002). Modeling the effects of trade on women: The case of Zambia. Working
Paper 155. Institute of Development Studies, Brighton.
Fontana, M., Joekes, S., & Masika, R. (1998). Global trade expansion and liberalization: Gender
issue and impacts. Bridge Report no. 42. IDS, Brighton, UK.
Fontana, M., & Wood, A. (2000). Modeling the effects of trade on women at work and at
home. World Development, 28(7), 1173–1190.
Gladwin, C. (1991). Structural adjustment and the African women farmers. Economic perform-
ance in Latin America and the Caribbean. New York: Oxford.
Gronau, R. (1977). Leisure, home production, and work – The theory of the allocation of time
revised. Journal of Political Economy, 85(6), 1099–1124.
Haddad, L., Brown, L. R., Richter, A., & Smith, L. (1995). The gender dimensions of economic
adjustment policies: Potential interactions and evidence to date. World Development,
23(6), 881–896.
Hofmeyr, J. F. (1993). African wage movements in the 1980s. South African Journal of
Economics, 61(4), 266–280.
Hoogeveen, J. G., & Özler, B. (2004). Not separate, not equal poverty and inequality in post-
apartheid South Africa. World Bank, 1818H Street NW, Washington, DC 20433, USA.
ILO. (1999). ILO report on social impact of globalization. Geneva: ILO.
Joekes, S. (1995). Trade-related employment for women in industry and services in developing
countries. UNRISD Occasional Paper, Geneva.
Joekes, S. (1999). A gender-analytical perspective on trade and sustainable development.
In: UNCTAD, Trade sustainable development and gender. New York and Geneva:
UNCTAD.
Klasen, S., & Woolard, J. (1998). Levels, trends and consistency of employment and unemploy-
ment figures in South Africa. Working Paper, unpublished. University of Munich/
University of Port Elizabeth.
Latigo, A., & Ironmonger, D. (2004). The missing link in growth and sustainable development:
Closing the gender gap. An Issues Paper presented at the ADB/ECA Symposium of
African Ministers of Finance, Planning and Economic Development, 24 May 2004,
Kampala, Uganda.
Lewis, J. (2001). Reform and opportunity: The changing role and patterns of trade in South Africa
and SADC. Africa Region Working Paper Series no. 14. The World Bank, Washington,
DC.
Ndlela, T., & Nkala, P. (2003). Determinants of exports from SADC and the role of market
access. Paper presented at the Trade and Industrial Policy Strategies (TIPS), 2003
Annual Forum at Glenburn Lodge, Muldersdrift.
Ozler, S. (2000). Export orientation and female share of employment: The evidence from
Turkey. World Development, 28(7), 1239–1248.
Ozler, S. (2001). Export led industrialization and gender differences in job creation and destruc-
tion: Micro evidence from the Turkish manufacturing sector. Unpublished paper,
Department of Economics, UCLA.
Palmer, I. (1994). Social and gender issues in macro-economic policy advice. Social Policy Series
no. 13. GTZ, Eschborn.
Rospabe, S. (2002). An evaluation of gender discrimination in employment, occupation attainment
and wage in South Africa in the late 1990s. Development Policy Research Institute,
University of Cape Town, mimeo.
Gender-Focused Macro-Micro Analysis 299

Siddiqui, R. (2004). Modelling gender dimensions of the impact of economic reforms on time
allocation among market work, households work and leisure. Research Report no. 185.
Pakistan Institute of Development Economics, Islamabad, Pakistan.
Sinha, A. (1999). Gender awareness in macroeconomic analysis. An Indian experiment presented
at the conference on Feminist Economics at Calton University, Ottawa, Canada.
Sinha, A. (2000). Gender in macroeconomic framework, a CGE model analysis. Presented at the
2nd Annual Gender Planning Network Meeting, Kathmandu.
Solberg, J. E., & Wong, D. C. (1992). Family time use: Leisure, home production, market work,
and work related travel. The Journal of Human Resources, 27(3), 485–510.
Standing, G. (1989). Global feminization through flexible labor. World Development, 17(7),
1077–1095.
Standing, G., Sender, J., & Weeks, J. (1996). Restructuring the labor market: The South African
challenge. An ILO country review. Geneva: ILO.
Statistics South Africa. (1999). October household survey South Africa.
Statistics South Africa. (2001a). Labor force survey. February 2001 Statistical release P0210.
Statistics South Africa. (2001b). A survey of time use: How South African women and men spend
their time. Pretoria, South Africa.
TIPS. (2002). A review of the changing composition of the South African economy. Report
compiled by Trade and Industrial Policy Strategies.
United Nations Development Programme. (UNDP). (2000). Human development report 1997.
New York, USA.
Valodia, I. (1996). Work, Chapter 3. In: D. Budlender, (Ed.), The women’s budget (pp. 53–96).
Cape Town: IDASA.
Valodia, I. (2000). Economic policy and women’s informal and flexible work in South Africa.
Paper presented at the Tips 2000 Annual Forum.
World Bank. (1999). South Africa country assistance strategy: Building a knowledge partnership.
The World Bank Group, Africa Region, Washington.
World Bank. (2002). Gender stats – Database for gender statistics. http://devdata.worldbank.
org/genderstats
World Bank. (2004). World development. http://devdata.worldbank.org
 300
APPENDIX 1. TRADE AND OUTPUT EFFECTS
Sectors Tariff Sectoral Shares IPR EIR Volume Changes Price Changes

VA Imp Exp Imp Exp Output Dom Imp Dom Output

AGRICULTURE 1.1 3.3 1.6 2.6 7.0 12.4
5.9 2.9
0.5
1.0 3.7 0.3 0.6
Gold and uranium 0.0 2.1 0.0 9.9 0.0 85.3
5.8 0.0 0.0 0.0 4.9 0.8 0.8
Other mining 0.0 4.7 13.7 24.8 73.0 87.5
0.4 0.0 0.4 2.7 4.9 2.8 1.3
MININGa 0.0 6.8 13.7 34.7 65.8 86.8
0.4 0.0 0.3 1.9 4.9 2.2 1.2
Meat and vegetables 4.0 0.5 1.5 1.8 11.9 16.4
1.6 2.6
0.8
1.1 0.9 0.6 0.8
Dairy 7.6 0.2 0.2 0.2 6.1 6.8 3.6 2.7
1.0
1.1
2.5 0.6 0.6
Grain milling 3.2 0.4 0.6 0.3 9.1 4.5
2.8 2.6
1.1
1.2 1.6 0.5 0.6
Other food 5.5 0.8 0.5 1.2 6.4 16.8
0.3 2.7
1.0
1.5
0.6 0.2 0.6
Beverages and tobacco 112.8 1.4 0.4 1.6 3.4 14.1 147.7 5.3
4.9
5.9
50.7
6.0
5.4
Textiles 6.7 0.2 1.0 0.4 28.7 17.4 0.4 2.1
1.9
2.5
1.7 0.2 0.7
Garments 17.2 0.7 1.5 1.1 18.0 18.5 14.8 2.0
2.7
2.9
10.5 0.1 0.3
Leather 1.9 0.1 0.4 0.5 40.2 47.7
2.3 3.4 1.9 0.7 3.0 0.9 1.6
Footwear 11.7 0.1 0.7 0.1 34.8 8.4 7.3 2.5
2.1
2.3
6.1 0.0 0.1

JOHN COCKBURN ET AL.

Wood 0.9 0.5 0.6 0.8 13.2 19.3
4.1 3.7 1.4 0.9 4.0 0.5 0.9
Paper 7.7 0.9 1.3 2.4 12.6 24.3 3.8 3.2 0.2
0.6
2.6 0.2 0.8
Printing 1.9 0.7 1.1 0.4 17.0 7.4
2.8 3.5 0.8 0.6 2.9 0.6 0.7
Petroleum 0.7 1.5 1.1 3.4 7.5 24.3
5.2 2.1
0.5
1.0 4.2 1.3 1.6
Basic chemicals 1.2 1.0 5.2 3.7 39.2 34.6
1.0 2.3 1.4 0.9 3.7 2.4 2.7
Other chemicals 1.4 1.2 5.4 1.8 28.8 13.8
2.6 2.4 0.3 0.0 3.5 1.6 1.8
Rubber products 6.4 0.2 0.9 0.4 32.4 22.0 2.4 2.3
0.6
1.2
1.4 1.0 1.4
Plastic products 1.4 0.6 1.0 0.3 18.9 7.9
2.6 2.7 0.5 0.3 3.4 1.4 1.5
Glass products 1.5 0.1 0.3 0.2 24.8 16.3
2.8 3.0 0.8 0.5 3.4 1.1 1.3
Non-metallic mineral 1.5 0.6 1.1 0.4 17.8 7.7
1.7 2.4 0.5 0.4 3.3 1.9 2.0
 Gender-Focused Macro-Micro Analysis
Iron and steel 0.5 2.2 3.1 10.5 17.2 46.1
1.2 2.1 1.5 1.1 4.4 2.8 3.1
Structural metal 20.7 0.4 0.1 0.4 2.5 11.7 25.2 2.2
0.4
0.8
13.1 1.4 1.7
Other fabricated metal 2.2 0.8 1.8 0.7 20.3 10.0
0.9 2.6 0.6 0.4 2.6 1.7 1.8
General purpose machinery 2.2 0.4 4.3 2.6 72.2 71.4
0.3 4.0 3.3 2.1 2.6 1.0 1.8
Special purpose machinery 2.7 0.4 5.8 1.4 60.7 32.9
0.4 3.0 1.4 0.8 2.2 1.4 1.8
Household appliances 10.6 0.1 0.5 0.1 35.8 14.1 5.5 1.8
2.7
3.1
5.1 0.4 0.6
Electric motors 7.2 0.1 0.6 0.2 42.9 25.4 2.0 2.5
0.7
1.7
2.1 0.3 1.0
Electricity distribution 0.7 0.1 0.9 0.2 56.9 20.3
1.4 3.2 2.2 1.9 4.2 1.9 2.1
Insulated wire and cable 14.9 0.1 0.2 0.1 11.9 6.4 12.8 1.9
2.2
2.4
8.7 0.6 0.8
Accumulators 1.3 0.1 0.2 0.1 27.0 12.8
2.6 2.8 0.7 0.4 3.5 1.4 1.6
Lighting equipment 1.5 0.0 0.3 0.0 39.8 11.5
1.9 3.2 1.2 1.0 3.3 1.3 1.4
Other electrical equipment 10.5 0.2 0.7 0.4 28.9 23.6 6.0 2.4
1.6
2.2
5.0 0.2 0.6
Communication equipment 2.6 0.2 6.1 0.7 77.8 33.7
0.8 3.4 1.7 1.0 2.3 1.1 1.5
Medical instruments 0.5 0.1 3.0 0.4 83.5 57.9
0.8 3.9 3.4 3.2 4.4 1.7 1.8
Motor vehicles 5.7 0.9 4.9 3.8 23.4 21.2 2.1 1.6
0.9
1.5
0.8 1.6 2.1
Motor vehicle parts 0.0 0.5 8.5 1.2 64.6 23.3
1.7 2.9 2.1 2.0 4.9 2.3 2.4
Other transport equipment 0.0 0.1 3.6 1.1 85.1 69.0
0.3 3.4 3.3 3.2 4.9 2.5 2.6
Furniture 7.0 0.3 0.3 1.1 11.3 39.9 3.0 2.6
0.7
1.0
2.0 0.7 0.9
Other manufacturing 2.4 0.2 5.4 1.8 75.1 68.5
0.7 2.2 1.0 0.0 2.5 2.0 2.6
INDUSTRYa 3.3 18.8 74.9 47.7 29.4 24.1 0.6 2.6
0.1
0.7 1.5 0.8 1.3
Electricity and gas 0.0 2.4 0.0 0.4 0.0 3.6 0.0 2.9
0.1
0.3 0.0 0.8 0.8
Water 0.0 0.4 0.0 0.0 0.2 0.4
6.9 3.0
0.6
0.6 4.9 0.4 0.4
Building 0.0 1.6 0.1 0.0 0.5 0.1
5.0 2.6 0.0 0.0 4.9 1.4 1.4
Construction 0.0 1.3 0.2 0.0 1.9 0.2
5.4 2.6
0.2
0.2 4.9 1.3 1.3
Trade services 0.0 11.1 0.1 0.2 0.1 0.3
5.7 3.2 0.2 0.2 4.9 0.8 0.8
Hotel and restaurant 0.0 2.0 2.0 2.6 20.6 27.4
5.6 3.0 0.9 0.1 4.9 0.9 1.4
Transport services 0.0 6.2 2.8 5.2 7.6 14.5
4.7 2.7 0.6 0.2 4.9 1.4 1.7
Post and telecommunications 0.0 3.9 1.6 1.1 6.6 5.1
5.4 3.0 0.3 0.2 4.9 1.0 1.1
Financial services 0.0 10.0 0.8 2.9 1.4 5.5
7.2 3.4
0.2
0.4 4.9 0.1 0.2

301
 302
APPENDIX 1. (Continued )

Sectors Tariff Sectoral Shares IPR EIR Volume Changes Price Changes

VA Imp Exp Imp Exp Output Dom Imp Dom Output

Real estate 0.0 6.0 0.2 0.4 0.6 1.5
5.6 3.0 0.1 0.0 4.9 0.9 0.9
Other business 0.0 2.9 1.0 0.6 5.5 3.4
5.5 3.2 0.4 0.3 4.9 0.8 0.9
General government 0.0 16.5 0.0 0.5 0.0 0.7 0.0 3.3 0.0 0.0 0.0 0.5 0.5
Health and social work 0.0 1.9 0.1 0.3 0.6 2.6
6.7 2.7
0.7
0.8 4.9 0.7 0.8
Other services 0.0 4.8 0.9 0.8 4.0 4.0
6.7 3.2
0.2
0.3 4.9 0.4 0.5
SERVICEa 0.0 71.1 9.8 15.0 3.1 4.0
5.6 3.0 0.1
0.1 4.9 0.7 0.8
ALLa 2.5 100.0 100.0 100.0 16.1 15.8
0.3 1.8 0.0
0.3 2.3 0.7 1.0

Notes: VA ¼ value added; Imp ¼ imports; Exp ¼ exports; Dom ¼ local sales of domestic output; IPR ¼ import penetration ratio;
EIR ¼ export intensity ratio.
a
Average variation for volumes – Laspeyres index for prices.

JOHN COCKBURN ET AL.

 Gender-Focused Macro-Micro Analysis
APPENDIX 2. FACTOR EFFECTS
Sectors VA Price Share in Male Wages Share in Female Wages Capital

Urban Rural Urban Rural Share Returns

Hi Med Lo Kid Hi Med Lo Kid Hi Med Lo Kid Hi Med Lo Kid

AGRICULTURE
1.0 0.1 1.1 4.6 50.6 14.2 13.4 33.3 75.8 0.4 0.5 2.4 88.2 1.7 8.3 20.1 100.0 4.8
1.5
Gold and uranium 0.0 1.8 5.8 5.4 0.0 0.0 0.8 0.2 0.0 0.2 0.2 0.1 0.0 0.0 0.0 0.0 0.0 1.4 0.0
Other mining 0.7 2.2 5.1 2.3 0.0 2.9 11.6 7.3 0.0 0.3 0.5 0.3 0.0 0.0 2.7 0.1 0.0 6.5 1.2
MINING 0.5 4.0 11.0 7.7 0.0 2.9 12.3 7.5 0.0 0.5 0.7 0.4 0.0 0.0 2.7 0.1 0.0 7.9 1.0
Meat and vegetables
0.5 0.4 0.5 1.3 0.0 0.6 0.4 1.0 0.0 0.2 0.5 2.1 0.0 1.3 0.9 0.7 0.0 0.5
1.3
Dairy
0.5 0.3 0.5 0.5 0.0 0.7 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.2 0.0 0.2
1.5
Grain milling
1.2 0.5 0.2 0.5 0.0 0.4 0.3 0.3 0.0 0.2 0.1 0.1 0.0 0.0 0.6 0.1 0.0 0.4
2.3
Other food
0.7 0.4 0.9 2.8 0.0 0.3 1.5 1.4 0.0 0.1 0.9 1.4 0.0 0.0 1.2 2.3 0.0 0.7
1.6
Beverages and tobacco
11.4 1.7 0.4 1.0 0.0 0.1 0.4 0.4 0.0 0.3 0.3 0.1 0.7 0.0 0.8 0.2 0.0 2.2
15.8
Textiles
0.6 0.1 0.3 0.7 0.0 0.0 0.3 0.0 0.0 0.1 0.9 0.2 0.0 0.0 0.7 0.0 0.0 0.1
2.4
Garments
0.5 0.5 0.6 1.4 0.0 0.1 0.3 0.3 0.0 1.2 3.9 2.4 0.0 0.7 7.2 0.9 0.0 0.2
3.1
Leather 1.3 0.1 0.1 0.0 0.0 0.0 0.1 0.2 0.0 0.1 0.1 0.2 0.0 0.0 0.0 0.1 0.0 0.1 3.3
Footwear
2.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.3 0.3 0.0 0.0 0.7 0.1 0.0 0.1
4.1
Wood 0.5 0.2 1.0 1.0 0.0 0.3 1.8 1.9 0.0 0.2 0.4 0.2 0.0 0.0 1.2 1.0 0.0 0.3 1.9
Paper 0.3 0.5 1.0 0.6 0.0 0.0 0.4 0.4 0.0 1.0 0.6 0.9 0.0 0.0 0.4 1.6 0.0 1.1 0.5
Printing 0.3 1.1 0.9 1.2 7.2 0.0 0.2 0.2 0.0 1.7 1.1 0.5 0.0 1.0 0.4 0.0 0.0 0.4 1.0
Petroleum
2.9 0.8 0.5 0.1 0.0 0.0 0.1 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.7
3.4
Basic chemicals 2.2 1.2 0.6 0.2 0.0 1.9 0.4 0.0 0.0 0.3 0.3 0.1 0.0 0.0 0.0 0.0 0.0 1.3 3.6
Other chemicals 0.1 3.0 0.4 1.4 0.0 0.0 0.6 0.5 0.0 1.9 0.4 2.8 0.0 0.0 0.7 0.2 0.0 0.9 0.4
Rubber products
0.2 0.1 0.6 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.1 0.0 0.1
0.8
Plastic products 0.1 0.7 0.9 1.5 0.0 0.2 0.0 0.1 0.0 0.7 1.2 3.6 0.0 0.0 0.2 1.0 0.0 0.1 0.5
Glass products 0.4 0.1 0.3 0.4 0.0 0.0 0.2 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.1 0.0 0.0 0.1 1.2
Non-metallic mineral 0.9 0.5 0.5 0.3 0.0 0.1 0.8 0.6 0.0 0.1 0.1 0.3 0.0 0.0 0.8 0.2 0.0 0.9 1.5

303
Iron and steel 2.9 1.0 1.9 3.5 0.0 0.8 1.4 0.3 0.0 1.2 1.1 0.1 0.0 0.0 0.2 0.0 0.0 3.2 4.5
 304
APPENDIX 2. (Continued )
Sectors VA Price Share in Male Wages Share in Female Wages Capital

Urban Rural Urban Rural Share Returns

Hi Med Lo Kid Hi Med Lo Kid Hi Med Lo Kid Hi Med Lo Kid

Structural metal
0.1 0.3 1.0 0.5 0.0 0.7 0.6 0.0 0.0 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.2
0.5
Other fabricated metal 0.4 1.2 1.0 1.4 0.0 0.0 0.3 0.1 0.0 1.3 0.1 0.0 0.0 0.2 0.6 0.1 0.0 0.6 1.0
General purpose machinery 0.6 0.6 0.5 1.2 0.0 0.0 0.0 0.2 0.0 0.2 1.2 0.0 0.0 0.0 0.0 0.3 0.0 0.1 3.9
Special purpose machinery 0.4 1.0 0.7 0.9 0.0 0.0 0.8 0.0 0.0 0.1 0.2 0.0 0.0 0.0 0.0 0.2 0.0 0.2 1.7
Household appliances
0.8 0.1 0.1 0.1 0.0 0.2 0.0 0.0 0.0 0.2 0.0 0.2 0.0 0.0 0.0 0.1 0.0 0.0
3.5
Electric motors
0.2 0.2 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.3 0.0 0.0 0.0 0.0 0.0 0.0 0.0
1.0
Electricity distribution 2.6 0.0 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.1 0.1 0.0 0.0 0.0 0.0 0.0 0.1 4.8
Insulated wire and cable
3.0 0.1 0.2 0.0 0.0 0.0 0.0 0.6 0.0 0.0 0.1 0.1 0.0 0.0 0.3 0.0 0.0 0.2
5.1
Accumulators 0.7 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 1.1 0.0 0.0 0.0 0.0 0.0 0.1 1.4
Lighting equipment 1.1 0.0 0.0 0.0 0.0 0.0 0.5 0.0 0.0 0.0 0.1 0.1 0.0 0.0 0.0 0.0 0.0 0.0 2.4
Other electrical equipment
1.3 0.4 0.1 0.0 13.8 0.0 0.0 0.0 0.0 0.0 0.2 0.1 0.0 0.0 0.4 0.1 0.0 0.2
2.9
Communication equipment 0.7 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 2.0 0.1 0.1 0.0 0.0 0.0 0.0 0.0 0.1 2.4
Medical instruments 1.3 0.1 0.0 0.1 0.0 0.0 0.0 0.0 0.0 0.0 0.4 0.1 0.0 0.0 0.0 0.0 0.0 0.0 4.7

JOHN COCKBURN ET AL.

Motor vehicles
0.8 1.5 1.0 0.3 0.0 1.2 0.3 0.2 0.0 0.1 0.3 0.6 0.0 0.0 0.7 0.1 0.0 1.0
1.7
Motor vehicle parts 1.1 0.4 1.0 1.5 0.0 0.3 0.8 0.0 0.0 0.2 0.4 0.3 0.0 0.0 0.0 0.0 0.0 0.4 3.3
Other transport equipment 0.3 0.5 0.3 0.2 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 0.0 3.6
Furniture
0.2 0.4 0.4 0.7 0.0 0.0 0.7 0.2 0.0 0.0 0.3 0.0 0.0 0.0 0.1 0.4 0.0 0.2
0.9
Other manufacturing 0.8 0.1 0.2 0.4 0.0 0.7 0.1 0.1 0.0 0.2 0.4 0.4 0.0 0.0 0.2 0.0 0.0 0.2 1.8
INDUSTRY
0.6 20.2 19.0 25.8 21.0 8.4 13.9 9.2 0.0 13.9 16.7 18.8 0.7 3.2 18.6 9.9 0.0 19.4
2.5
Electricity and gas
0.3 2.3 1.8 1.0 0.0 0.0 0.7 1.4 0.0 0.4 0.9 0.7 0.0 0.0 0.3 0.0 0.0 3.5
0.4
Water
1.2 0.4 0.2 0.0 0.0 0.5 0.8 1.0 0.0 0.0 0.1 0.0 0.0 0.0 1.2 0.3 0.0 0.6
1.8
Building 0.0 0.6 4.5 0.8 11.4 0.2 3.2 0.9 0.0 0.0 0.5 0.2 0.0 0.0 0.5 0.0 0.0 1.3 0.1
Construction
0.1 0.6 1.5 10.9 5.9 0.6 2.8 9.5 0.0 0.1 0.3 0.2 0.0 0.0 0.6 0.5 0.0 1.2
0.3
Trade services 0.1 8.1 12.2 16.3 2.7 9.6 13.2 7.5 24.2 9.5 16.8 19.2 2.2 6.2 24.4 15.8 0.0 10.2 0.3
Hotel and restaurant 2.1 0.5 0.9 0.5 0.0 0.6 1.8 0.7 0.0 1.0 2.7 1.1 0.0 0.2 5.1 1.9 0.0 3.1 3.0
 Gender-Focused Macro-Micro Analysis
Transport services 0.7 4.6 7.4 6.0 0.0 11.6 6.0 4.5 0.0 4.3 1.8 0.6 0.0 3.3 0.3 0.3 0.0 7.5 1.2
Post and telecommunications 0.5 5.0 1.5 0.9 0.0 0.3 0.2 1.2 0.0 2.1 4.8 0.3 0.0 0.1 3.2 0.9 0.0 5.2 0.8
Financial services
0.3 12.7 4.0 1.0 0.0 9.0 0.5 0.1 0.0 8.7 9.3 1.1 0.0 3.5 1.2 0.0 0.0 13.5
0.5
Real estate 0.8 1.7 0.1 0.5 0.0 0.3 0.0 0.0 0.0 2.0 0.5 1.0 0.0 0.0 0.0 0.0 0.0 12.4 0.8
Other business 0.1 7.7 2.4 1.7 0.0 0.3 1.5 0.8 0.0 6.8 3.8 4.3 0.1 1.8 1.8 2.2 0.0 1.1 0.5
General government 0.0 23.5 31.1 18.9 0.0 19.6 27.7 19.9 0.0 20.5 35.3 13.2 0.0 8.3 26.2 20.7 0.0 4.2 0.0
Health and social work
0.6 1.4 0.3 1.1 0.0 3.3 0.4 0.8 0.0 9.0 2.0 2.8 2.1 13.3 2.3 3.0 0.0 1.9
1.3
Other services
0.1 6.4 1.1 2.3 8.3 18.8 1.4 1.9 0.0 20.9 3.2 33.8 6.8 58.3 3.5 24.4 0.0 2.3
0.2
SERVICES 0.2 75.6 69.0 61.9 28.4 74.6 60.3 50.0 24.2 85.2 82.1 78.4 11.2 95.1 70.5 69.9 0.0 67.9 0.4
ALLa 0 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 100 0

Notes: VA ¼ value added; Hi, Med, Lo, Kid ¼ high, medium, low skilled and child workers.
a
Laspeyres price index.

305