Applied Economics, 2014

Vol. 46, No. 5, 573–586, http://dx.doi.org/10.1080/00036846.2013.857003

Going green: does it depend on

education, gender or income?
Dakshina G. De Silvaa and Rachel A. J. Pownallb,c,*
a
Department of Economics, Lancaster University Management School,
Lancaster LA1 4YX, UK
b
Department of Finance, Tilburg University, 5000 LE, Tilburg, The Netherlands
c
Department of Finance, Maastricht Universtiy, 6200 MD, Maastricht, The
Netherlands

Sustainable development entails meeting our present needs without compromis-

ing the ability of future generations to meet their needs. This requires us to treat
economic, social and environmental aspects in an integrated way, but little is
known about the nature of individual preferences towards the trade-offs involved
in this effort. For the first time, we study individual preferences towards the
environment, social well-being and financial well-being by using a survey of over
1400 households in the Netherlands. Using nonparametric, parametric and
matching methods, we find that gender and education are important factors for
sustainability rather than income levels. Moreover, results indicate that educated
females put the greatest value on going green whilst being socially minded.

Keywords: sustainability; financial well-being; heterogenous preferences

JEL Classification: G10; I31

I. Introduction development without paying attention to the environ-

ment, we would degrade the natural capital required to
Financial resources are essential to both ecological and ensure sustained economic well-being. Lastly, if the
social progress. It is a constant challenge for policy- focus is only on economics and environment, neglecting
makers to balance the trade-offs needed to reconcile the social well-being, we would have growing income dis-
three pillars of sustainable development – economic parity and rising unemployment.1 Thus, addressing only
well-being, the environment and social development. It two of the three pillars will result in development that is
has become increasingly obvious that we have to adjust unsustainable. However, in order to achieve this goal, we
our standard of living in order to maintain the environ- need to understand the heterogeneity in preferences
ment at its current level. Whether we are willing to give among groups of individuals towards these trade-offs.
up part of our standard of living in the form of economic, The recent financial crisis has highlighted the need for
environmental or social issues is a highly individualistic improved corporate responsibility and accountability.
choice. As Stevens (2010) points out, if we only stress Coupled with a widening gap between the rich and the
the environmental and social dimensions of sustainable poor, the United States and the United Kingdom find
development in the absence of economics, we neglect the themselves at places 66 and 50, respectively, on the
growth in financial capital. Similarly, if we only build up 2008 Sustainable Society Index. Both the United States
the economic and social pillars of sustainable and the United Kingdom have the fastest growing divides

*Corresponding author. E-mail: [email protected]

1
See Stevens (2010) for an insightful overview.

© 2013 Taylor & Francis 573

574 D. G. De Silva and R. A. J. Pownall
between the rich and the poor in the OECD area. gender effects risk-taking could be due to a number of
Currently, the focus has been tilted heavily towards eco- reasons. Loewenstein et al. (2001) focus on the risk-as-
nomic growth at the cost of both the environment and feelings hypothesis to explain differences in risk-taking due
social welfare. to differences in emotional reactions to risky situations.
As far as we know, this article provides the first sys- Findings from the psychology literature indicate that
tematic analysis in measuring the extent to which indivi- women experience emotions more strongly than men so
duals rate these trade-offs between the three pillars of that the risk-as-feelings hypothesis can explain why
sustainable development. We concentrate on individuals women are more ‘affected’ by risk than men and are sub-
to measure their sustainable values, focusing on the trade- sequently more risk-averse. Kuhnen and Knutson (2011)
off in terms of social welfare, the environment and finan- try to determine whether emotions influence behaviour by
cial well-being. Little is known about how heterogeneity changing risk preferences, or if the belief formation is
affects these values. Hence, we explicitly ask a represen- updated or if both influences are together at play. We are
tative sample from the Dutch population – more than 1400 able to control for a number of these issues by including a
individuals – about their preference to live in a society number of important variables in our sample, such as
which strives towards greater financial well-being or a gender, education and income. However, how much of a
society which strives towards reducing carbon emissions. role does gender play in establishing social and environ-
We ask a similar question regarding social welfare and the mental values within the sustainability debate is open to
environment. question. Furthermore, does age, education, work status
The literature on environmental sustainability is not and if one is a homeowner or not affect these decisions?
new. Eichholtz et al. (2010) examine the relationship In this study, we are able to shed some light towards
between commercial real estate prices and the energy understanding how the heterogeneity in society affects
star ratings of these buildings. Their results indicate that how individuals value the trade-offs between people,
‘green’ buildings’ rental rates are about 3% higher. planet and profit. Evaluation of these trade-offs helps
Waddock and Graves (1997) mention that ‘corporate policy-makers to take these preferences into account
social responsibility (CSR)’ has become a social norm when structuring policy towards green social economic
when firms consider their production inputs. According growth. To our knowledge, this is the first article to
to Social Investment Forum (2010), evaluation of CSR has analyse the effects of gender and education on individual
become an investment decision for some. Firms believe preferences on environmental sustainability and financial
that better CSR policies could outperform other strategies. welfare and/or social welfare using field survey data. Our
Turban and Greening (1997) and Fombrun and Shanley results, using parametric, nonparametric and matching
(1990) note that good CSR policies could improve a methods, show that values towards people, planet and
company’s reputation. Baron (2001) and Lyon and profit are indeed strongly heterogenous. We find signifi-
Maxwell (2011) show that better CSR policies could cant evidence that education and gender play an impor-
lead to less scrutiny from governments or other concerned tant role in determining the trade-offs between these
organizations. domains. Interestingly, educated females put the greatest
It has generally been observed that gender has as an value on going green. We also find that age, being a
effect on value-based decisions (see, for example, Becker homeowner, and work status are vital factors. Less
(1974), Andreoni and Vesterlund (2001), Charness and important factors are income levels, whether the main
Rabin (2002) and Falk and Fischbacher (2006)). Croson residence is in a city or the number of children. Economic
and Gneezy (2009) suggest that women are more sensitive policy that takes account of this heterogeneity will be
to social cues than men. The literature also shows that there more effective in achieving the goal of sustained eco-
are significant gender-related differences in making finan- nomic development. Further implications of the results
cial decisions. It is generally observed from a large litera- are directed towards decisions for sustainable businesses.
ture on experimental economics that women are more If individuals have different values towards people, pla-
averse to risk than men. Watson and McNaughton (2007) net and profits, then investors might be found to value
link gender to risk-taking and find that women tend to be companies differently, which would affect corporate
more risk-averse than men and this affects their choice of financial decision-making. Furthermore, consumers
retirement investments. Lusardi and Mitchell (2008) also would value products differently, which would affect
find that gender-related differences can be seen with respect the demand for sustainable products and would influence
to financial literacy. Women are also more financially illit- pricing decisions.
erate when it comes to making financial decisions, which is The article is structured as follows. In Section II, we
a concern for old age and retirement provisions. Croson and presents the methodology behind the survey data. In
Gneezy (2009) provide a broad survey of the experimental Sections III and IV, we give a description of the data
evidence on gender differences. Studies also find that single and describe the empirical analysis, respectively.
men are the most willing to take risk. The reasons why Nonparametric results using Racine and Li (2004)
Going green 575
method are presented in section ‘Nonparametric results households when they compare the DNB Household
using Racine and Li method’. In sections ‘Reduced-form Survey results to national accounts data and microdata
estimation’ and ‘Matching estimation’, we provide on household wealth published by Statistics Netherlands.
further empirical results using reduced form and match- Although no household survey can ever be entirely free of
ing estimation procedures. Finally, in Section V, we con- potential biases caused by nonresponse, their findings
clude by providing a discussion of the results and the suggest that this problem is limited in the DNB
implications for sustainable policy and decision-making Household Survey.
within the context of policy towards sustained economic For the present study, the question used to evaluate
development. households’ overall attitude to sustainability was mea-
sured by asking the following question: ‘To what extent
would you be willing to reduce your standard of living in
II. Survey Data Methodology order to maintain the environment at the same level for the
next generation?’ Using a scale from 1 to 10, with 1
We are interested in examining individuals’ social prefer- indicating ‘not at all willing’, the households rated their
ences. We focus, in particular, on how social preferences willingness. We also look at the three direct trade-off’s
differ across genders, by educational attainment and level between people, planet and profits by asking participants
of income using survey data rather than experimental to indicate on a scale from 1 to 10: ‘I prefer to live in a
evidence. Our survey is designed to observe the trade- society that strives for financial well-being than striving to
offs between the three pillars of sustainable development. reduce carbon emissions’ (the range = 1 (financial well-
In order to understand these trade-off’s, we designed a being) to 10 (reduce carbon emissions)). Similarly, we ask
survey to ask individuals to rate each of the trade-offs participants two further trade-off questions: ‘I prefer to
between the three pillars. The individuals’ choice variable live in a society that strives for social well-being than
is the extent to which they rate the importance of living in striving to reduce carbon emissions’ (the range = 1 (social
a society which focuses on striving for financial well- well-being) to 10 (reduce carbon emissions)) and ‘I prefer
being than striving to reduce carbon emissions or for to live in a society that strives for financial well-being than
greater social welfare. We structured the questions in this striving for greater social welfare’ (the range = 1 (financial
format so that we could measure the utility derived from well-being) to 10 (social welfare)). In the next section, we
living in a society which strives for a monetary pay-off or provide summary statistics on these trade-off questions by
spending now on future environmental conservation (or gender, education attainment and homeownership.
social benefits).
We collect data on individuals’ attitudes towards sus-
tainability by focusing on three main questions to assess III. Data Description
the trade-off between people, planet and profits. During
August 2010, we sent out a separate survey to a subsection Table 1 presents summary statistics. In general, both gen-
of the sample population used for the Dutch National ders and those with or without a college degree are con-
Bank (DNB) Household Survey. This enabled us to ask a cerned about greenhouse gases and social welfare and are
number of direct questions to the households about their willing to give up their current standard of living or
preferences towards the environment, social well-being financial well-being to preserve the environment for the
and financial well-being. We have a sample of 1433 future and for better social welfare. However, we are
households from The Netherlands. The survey is adminis- interested in examining differences in responses due to
tered and conducted by CentERdata at Tilburg University. gender and for individuals with or without a college
The purpose of this survey is to collect household-level degree.
data to study the economic and psychological determi- In the sample, we have 789 males and 644 females. Out
nants of households’ decision-making behaviour. of this, 319 males and 243 females have college degrees.
Household members who are at least 16 years old are These unconditional results also indicate that females are
interviewed. CentERdata includes individual information more willing to reduce their current standard of living in
about subjects’ gender, educational attainment, homeow- order to maintain the environment for future generations
nership, annual household and personal income, marital compared to males. Similarly, they prefer reducing green-
status, number of children, regional location, living area house gases than social welfare and prefer social welfare
building environment and host of other individual char- to financial well-being compared to men. However, there
acteristics which provide us with a rich database with are no observable differences between males and females
which to analyse attitudes towards sustainability in con- for preferences in reducing greenhouse gases at the
junction with heterogeneous economic variables. A good expense of social welfare. When considering differences
introduction to this data is given in Alessie et al. (2004). due to educational attainment, these simple summary sta-
They find that the data is a representative panel of Dutch tistics indicate that individuals with at least a college
576 D. G. De Silva and R. A. J. Pownall
Table 1. Summary statistics by gender and education

Male Female

Variable College No college College No college

Number of observations 319 470 243 401

Willing to reduce standard of living to maintain the environment 6.047 (2.508) 5.770 (2.250) 6.630 (1.990) 6.020 (2.102)
Prefer reducing greenhouse gases than social welfare 5.088 (2.332) 5.291 (2.222) 5.169 (2.144) 5.087 (1.968)
Prefer social welfare than financial well-being 7.445 (2.117) 7.548 (1.849) 8.008 (1.737) 7.269 (1.862)

Note: Standard deviations are in parentheses.

Table 2. Summary statistics

Variable Mean (SD) Variable Mean (SD)

Main cities (Amsterdam, Rotterdam and the Hague) .170 (.376) Number of children .606 (.999)
North .121 (.327) Net house hold income (per month) 2771.33 (1397.63)
East .214 (.410) Net personal income (per month) 1622.49 (1057.67)
West .278 (.448) Paid worker .442 (.497)
South .214 (.410) Family company .006 (.075)
Densely built up .149 (.357) Freelance .041 (.199)
Built up .259 (.438) Unemployed and looking for work .017 (.128)
Town .225 (.418) Unemployed and looking for first job .004 (.059)
Less built up .197 (.398) Student .026 (.159)
No built up .166 (.372) State benefits .001 (.026)
Age 55.050 (15.207) House worker .105 (.307)
No high school .320 (.467) Retired .293 (.455)
High school education .288 (.453) Unable to work .045 (.208)
College and above .392 (.488) Volunteer worker .017 (.128)
Homeowner .759 (.428) Other worker .003 (.059)

education are willing to reduce their current standard of IV. Empirical Analysis
living in order to maintain the environment for future
generations and prefer social welfare to financial well- Nonparametric results using Racine and Li method
being compared to individuals without a college
We are interested to know as to how individual character-
education. These differences are even larger for college-
istics affect social preferences and our findings so far. For
educated females compared to any other group.
example, how age, gender and income levels influence an
Overall summary statistics for individual characteristics
individuals’ trade-off between people, profits and the pla-
are presented in Table 2. It shows that 17% of the sample
net. Following List (2006), we analyse the trade-off S as a
population live in three main cities in the Netherlands
function of X, which controls for individual characteristics.
(Amsterdam, Rotterdam and Haag). Roughly equal pro-
First, we draw conditional densities using the nonpara-
portions of the sample are distributed over the four quad-
metric regression technique proposed by Racine and Li
rants of the Netherlands when splitting the sample into
(2004). The main advantage of this data-driven estimation
North, East, West and South regions. The average age of
technique is that it provides a modelling framework for the
the sample is 55 years and the average household income
relation among variables, applying a kernel method of
is almost 2800, while the net personal income is just over
density estimation to discrete variables that admit no nat-
€1600 monthly. Note that 78% of the survey subjects ural ordering, such as education attainment or mortgage
identify themselves as homeowners. Results also show status, which are used frequently in this study. Note that
that about 30% of the sample population is retired and this technique has been shown to have higher predictive
only 2.6% are students. While Table 1 suggests that edu- power than other conventional approaches in the presence
cation and gender are important factors for ‘going green’, of categorical variables. We consider the following
we also need to be cautious in interpreting these results, empirical model:
since there are no controls yet for individual characteris-
tics. Therefore, our next section presents some in-depth Si ¼ hðXi Þ þ ni (1)
analysis to describe more fully the differences that educa-
tion, gender and other individual characteristics have on where si is the dependent variable. We have four depen-
‘going green’. dent variables (survey questions regarding households’
Going green 577
overall attitude towards sustainability): (1) willingness to distribution functions. We can reject the null hypothesis
reduce standard of living to maintain the environment for of equality between the distributions at the 95% confi-
future generations, (2) preference for reducing greenhouse dence interval for all responses other than those who
gases rather than financial well-being, (3) preference for ‘prefer reducing greenhouse gases than social welfare’.
reducing greenhouse gases rather than social welfare and Figure 2 shows the conditional density graphs by educa-
(4) preference for social welfare rather than financial well- tional achievement. K–S tests show similar patterns as in
being. Here hðÞ has an unknown functional form, and Xi gender for college graduates. We also conduct a two-
represents a set of continuous and discrete regressors. We sample t-test on the differences of predicted conditional
define Xi ¼ Xic ; Xid with Xic representing the subset of densities. These results are presented in Table 3. We also
continuous variables and Xid the discrete variables. First, estimate the conditional densities by gender for those with
we construct the conditional densities by gender. In our college degrees. In this case, we see that educated females
case, the continuous variables are log age, log household tend to care more about the green economy compared to
income and number of children. Our discrete variables are their educated male counterparts. We have omitted the
gender (if gender is female then female = 1, otherwise = 0), graphs but differences of these predicted conditional den-
mortgage status (if holds a mortgage then mortgage = 1, sities are presented in Table 3.3
otherwise = 0), education attainment (if college educated From these conditional density figures and results in
then college = 1, otherwise = 0) and residential location (if Table 3, we see an interesting pattern. We can test if the
in three main cities = 1, otherwise = 0.) The optimal properties of completeness and transitivity hold. Females
smoothing parameters for hðÞ were chosen using the and college graduates indicate that they ‘prefer to reduce
‘leave-one-out cross-validation’ mechanism when esti- current standard of living in order to maintain the environ-
mating the fitted values.2 Figure 1 shows the conditional ment for future generations’ compared to males and indi-
density or graphs of the predicted responses by survey viduals without college degrees. Females and college
question. These figures suggest that the male conditional graduates also ‘prefer reducing greenhouse gases than
response distributions first order stochastically dominate financial well-being’ compared to males and people with-
the female distributions for all responses except the pre- out a college degree. They also indicate that they are
ference to reduce greenhouse gases over improving social indifferent between ‘reducing greenhouse gases at the
welfare. These results provide initial supporting evidence expense of social welfare’. This pattern indicates that
consistent with our summary statistics. We perform a these two groups are capable of expressing preferences
Kolmogorov–Smirnov (K–S) test for equality of the (or indifference) between all possible bundles and thus the
.3
.3
.2

.2
Density
Density
.1

.1
0

0 2 4 6 8 10 0 2 4 6 8 10
Willing to reduce standard of living to maintain the environment Prefer reducing greenhouse gases than financial well-being
Male Female Male Female
.3

.3
.2

.2
Density

Density
.1

.1
0

0 2 4 6 8 10 0 2 4 6 8 10
Prefer social welfare than reducing carbon emissions Prefer social welfare than financial well-being
Male Female Male Female

Fig. 1. Conditional densities by gender

2
Bandwidths of variables were chosen using standard Silverman’s rule of thumb and biweight kernels when estimating results.
3
Graphs can be provided upon request.
578 D. G. De Silva and R. A. J. Pownall

.3

.3
.2

.2
Density

Density
.1

.1
0

0
0 2 4 6 8 10 0 2 4 6 8 10
Willing to reduce standard of living to maintain the environment Prefer reducing greenhouse gases than financial well-being
No college College No college College
.3

.3
.2

.2
Density
Density
.1

.1
0
0

0 2 4 6 8 10 0 2 4 6 8 10
Prefer social welfare than reducing carbon emission Prefer social welfare than financial well-being
No college College No college College

Fig. 2. Conditional densities by educational attainment

Table 3. Conditional distributions’ differences by gender and education

Full sample College sample

Variable Female – Male College – No college Female – Male

Willing to reduce standard of living to maintain the environment .375** (.086) .394** (.088) .583** (.030)
Prefer reducing greenhouse gases than financial well-being .434** (.083) .381** (.085) .808** (.113)
Prefer reducing greenhouse gases than social welfare −.079 (.080) −.056 (.082) .081 (.109)
Prefer social welfare than financial well-being .391** (.079) .572** (.081) .568** (.112)

Notes: Standard deviations are in parentheses.

** denotes statistical significance at the 5% level.

property of completeness holds.4 Next, if the property of the same level for the next generation varies by gender. We
transitivity holds for females and college graduates, then also look at this in the context of the preference to live in a
they should prefer ‘social welfare than financial well- society that strives to reduce greenhouse gases rather than
being’. Our Table 3 results indicate that, in fact, this is striving for financial well-being, the preference to live in a
the case for females and college graduates compared to society that prefers reducing greenhouse gases rather than
males and individuals without college degrees. Similar improving social welfare and the preference to live in a
conclusions by gender are also observed for the sample society that strives for social welfare rather than striving
of individuals with only college degrees. for financial well-being and again see how this varies due
to gender differences. We consider the following simple
regression model:
Reduced-form estimation
0 0
In order to deepen our understanding of the patterns of Si ¼ Di δ þ Ci β þ εi (2)
survey question outcomes due to gender and education, we
present a set of reduced-form regressions that show how an where si represent the dependent variables as noted before.
individual’s willingness to reduce his or her own current The independent variables include D’s that control for
standard of living in order to maintain the environment at gender and education and C’s that control for individual
4
It is worth noting that those who ‘prefer to reduce current standard of living in order to maintain the environment for future generations’
and those who ‘prefer reducing greenhouse gases than financial well-being’ are similar in context. In this case, one should expect to see
that a person who indicates that they are ‘willing to reduce current standard of living in order to maintain the environment for future
generations’ will indicate that they ‘prefer reducing greenhouse gases than financial well-being’.
Going green 579
Table 4. Regression results for ‘willing to reduce standard of living to maintain the environment’

Variable (1) (2) (3) (4) (5) (6)

Female .487** (.119) .479** (.122) .477** (.122) .478** (.122) .572** (.127)
.158 (.215)
College and above .480** (.144) .493** (.144) .485** (.145) .464** (.145) .212 (.208)
College and above  female .610** (.286)
High school education .184 (.147) .186 (.148) .176 (.149) .178 (.148) .057 (.208)
High school education  female .247 (.296)
Mortgage owner .366** (.152) .385** (.147) .378** (.148) .444** (.148) .449** (.149) .390** (.148)
Log of age 1.072** (.196) 1.070** (.199) 1.060** (.199) 1.118** (.200) .980** (.277) 1.118*** (.201)
Number of children .111* (.066) .111* (.066) .110 (.067) .120* (.067) .113 (.068) .106 (.066)
Log of net house hold income .058 (.136)
Log of net personal income −.002 (.027) −.002 (.027) −.002 (.027) −.014 (.027)
Main cities (Amsterdam, .128 (.175) .130 (.175) .138 (.176)
Rotterdam and the Hague)
Regional effects Yes Yes
Urban density effects Yes
Employment status effects Yes
Number of observations 1433 1433 1433 1433 1433 1433
Adj. R2 .034 .033 .032 .036 .025 .034

Notes: Robust standard errors are in parentheses.

*** denotes statistical significance at the 1% level, ** denotes statistical significance at the 5% level and *denotes statistical significance
at the 10% level.

characteristics. In D, females and college graduates take the Table 5 presents the results on individuals’ preferences
value of 1. The individual characteristics are age, number of to reducing greenhouse gases compared to being finan-
children, homeownership, household or net personal cially well off. This survey question is very similar to the
income, residential location (by three main cities or question posed on the ‘willingness to reduce standard of
regions), residential location by built environment (urban living to maintain the environment for future generations’
density or congestion measure) and employment status. and can be used as a robustness check for Table 4 results.
Tables 4 through 7 present these OLS regression results. Again, the results indicate that females and individuals
Table 4 presents the results for ‘willing to reduce stan- with at least a college education prefer reducing green-
dard of living to maintain the environment for future gen- house gases to being financially well off as compared to
erations’. Our main interest is in examining whether there men. This result holds for all specifications. As in our
are any systematic differences due to gender or education conditional density estimation, we see that homeowners
attainment in one’s response. The results indicate that prefer financial well-being instead of reducing green
females and college graduates are more willing to reduce house gases compared to nonmortgage holders.
their current standard of living to maintain the environment Considering other variables, we observe that age has a
for future generations as compared to men. This result positive effect on the dependent variable. However, num-
holds for all specifications. We also observe that age, num- ber of children, income, residential location, urban density
ber of children and homeownership have a positive and and employment status do not influence the wiliness to
significant effect on the willingness to reduce the standard reduce greenhouse gases rather than being financially
of living to maintain the environment for future genera- well off. These general results also hold for all the specifi-
tions. These general results also hold for almost all speci- cations. Again, in accordance with Table 3 results, Table 5
fications. Income, residential location, urban density and results indicate that females with college degrees are more
employment status do not tend to influence the willingness willing to reduce greenhouse gases compared to being
to reduce standard of living to maintain the environment for financially well off compared to any other group.
future generations. It is also interesting to examine whether Table 6 presents the results on preference to reduce green-
education attainment affects gender differently regarding house gases compared to improving social welfare. Here, we
the willingness to live in a more green economy. In column observe that there are no statistical differences in the
6, we therefore present results where the gender dummy responses between either males and females, or for indivi-
has been interacted with dummy variables for education duals with and without a college degree and also not for
attainment. Our results provide strong support that females homeowners and nonhomeowners. This result holds for all
with college degrees are more willing to reduce their cur- specifications. Finally, in Table 7, we present the results on
rent standard of living to maintain the environment for preference for improving social welfare compared to being
future generations compared to any other group. This is in financially well off. From the early results, we can see that, in
accordance with our Table 3 results. general, more females than males and more college graduates
580 D. G. De Silva and R. A. J. Pownall
Table 5. Regression results for ‘prefer reducing greenhouse gases than financial well-being’

Variable (1) (2) (3) (4) (5) (6)

Female .491** (.113) .540** (.117) .540** (.118) .547** (.117) .607** (.121)
.072 (.215)
College and above .589** (.144) .520** (.143) .515** (.144) .501** (.144) .101 (.200)
College and above  female .928*** (.277)
High school education .205 (.141) .182 (.141) .175 (.142) .174 (.142) .020 (.197)
High school education  female .316 (.281)
Mortgage owner −.269* (.140) −.328** (.131) −.330** (.131) −.264* (.132) −.242* (.131) −.313** (.130)
Log of age 1.125** (.186) 1.107** (.188) 1.099** (.188) 1.145** (.189) 1.182** (.265) 1.187*** (.188)
Number of children .063 (.064) .071 (.064) .076 (.065) .074 (.065) .064 (.065) .070 (.064)
Log of net house hold income −.185 (.154)
Log of net personal income .026 (.028) .027 (.028) .027 (.029) .010 (.028)
Main cities (Amsterdam, Rotterdam and the Hague)
−.114 (.158) −.122 (.159) −.126 (.160)
Regional effects Yes Yes
Urban density effects Yes
Employment status effects Yes
Number of observations 1433 1433 1433 1433 1433 1433
Adj. R2 .043 .042 .041 .044 .034 .048

Notes: Robust standard errors are in parentheses.

*** denotes statistical significance at the 1% level, ** denotes statistical significance at the 5% level and * denotes statistical significance
at the 10% level.

Table 6. Regression results for ‘prefer reducing greenhouse gases than social welfare’

Variable (1) (2) (3) (4) (5) (6)

Female −.037 (.115) .002 (.121) −.002 (.121) −.002 (.120) −.091 (.227)
.044 (.123)
College and above −.056 (.146) −.092 (.150) −.103 (.150) −.111 (.151) −.234 (.211)
College and above  female .305 (.294)
High school education .013 (.142) −.014 (.144) −.019 (.145) −.024 (.145) .028 (.207)
High school education  female −.121 (.287)
Mortgage owner −.172 (.143) −.152 (.136) −.145 (.136) −.152 (.138) −.210 (.136) −.144 (.136)
Log of age .778** (.195) .730** (.196) .734** (.196) .746** (.196) .948** (.274) .766*** (.198)
Number of children −.008 (.064) .003 (.065) −.001 (.065) .007 (.065) .007 (.066) −.000 (.065)
Log of net house hold income .041 (.130)
Log of net personal income .035 (.026) .033 (.026) .034 (.026) .028 (.027)
Main cities (Amsterdam, −.013 (.153) −.013 (.153) −.016 (.154)
Rotterdam and the Hague)
Regional effects Yes Yes
Urban density effects Yes
Employment status effects Yes
Number of observations 1433 1433 1433 1433 1433 1433
Adj. R2 .010 .011 .010 .014 .009 .010

Notes: Robust standard errors are in parentheses.

*** denotes statistical significance at the 1% level and ** denotes statistical significance at the 5% level.

than noncollege graduates are willing to reduce their standard When examining the differences in responses, one con-
of living to maintain the environment for future generations, cern that we have is that we may not be making as tight a
prefer reducing greenhouse gases to being financially well comparison as possible between males and females and
off and are similar to males in their responses regarding college graduates and noncollege graduates. Therefore, in
preference for reducing greenhouse gases over improving the next section, we explore this by using matching tech-
social welfare. In this case, one should expect women and niques in order to compare more similar individuals.
college graduates to respond favourably, compared to men
and noncollege graduates, to improve social welfare over
Matching estimation
being financially well off. Our results support this conjecture
and hold for all specifications including the specification in We use matching techniques to evaluate whether the treat-
column 6 where we have interacted education attainment ment group (females or college graduates) compared to
dummies with a female dummy. the control group (males or people who did not complete
Going green 581
Table 7. Regression results for ‘prefer social welfare than financial well-being’

Variable (1) (2) (3) (4) (5) (6)

Female .479** (.104) .514** (.108) .515** (.108) .522** (.108) .561** (.113)
.212 (.202)
College and above .838** (.132) .786** (.132) .779** (.132) .775** (.132) .535** (.189)
College and above  female .543** (.258)
High school education .386** (.131) .371** (.132) .360** (.133) .369** (.132) .224 (.190)
High school education  female .290 (.262)
Mortgage owner −.123 (.132) −.176 (.127) −.172 (.127) −.120 (.130) −.077 (.127) −.160 (.127)
Log of age 1.231** (.186) 1.224** (.188) 1.210** (.189) 1.254** (.189) 1.389** (.255) 1.261*** (.190)
Number of children .071 (.062) .075 (.063) .082 (.063) .076 (.064) .052 (.064) .077 (.063)
Log of net house hold income −.162 (.136)
Log of net personal income .017 (.026) .017 (.026) .018 (.026) .007 (.026)
Main cities (Amsterdam, Rotterdam and Haag)
−.086 (.140) −.092 (.140) −.079 (.143)
Regional effects Yes Yes
Urban density effects Yes
Employment status effects Yes
Number of observations 1433 1433 1433 1433 1433 1433
Adj. R2 .063 .062 .061 .064 .037 .063

Notes: Robust standard errors are in parentheses.

*** denotes statistical significance at the 1% level and ** denotes statistical significance at the 5% level.

college) are different in their willingness to reduce their where X is defined as observable individual characteris-
current standard of living in order to maintain the environ- tics. To derive the above equation, three assumptions
ment at the same level for the next generation. In a similar need to be satisfied6: balancing of observable variables,
manner, we look at the other trade-off for each treatment unconfoundedness and the common support condition.
group compared to the control group. Rosenbaum and When the balancing property is met, observations with
Rubin (1983) proposed propensity score matching as a the same propensity score have the same distribution of
method to evaluate a ‘treatment effect’.5 They defined the observable individual characteristics independent of gen-
propensity score as the conditional probability of der (or college graduates) status. Unconfoundedness
receiving a treatment given observable (pre-treatment) assumes that, conditioning on observed individual char-
characteristics. When taken in our context, the basic idea acteristics, gender (or college graduates) assignment is
of the matching method is to compare, for example, the independent of the survey question outcome. Our situa-
outcomes of females and males who have similar distribu- tion is a typical matching case where the gender assign-
tions conditioning on the observable individual character- ment is random. This may not be the case for the
istics. Let D ¼ 1 when we observe females (or in the attainment of college education. In order to control for
analogous case, considering college graduates) and this nonrandom assignment, we match college graduates
D ¼ 0 when we observe males (or people who did not and people who did not complete college using a well-
complete college). The variables Q0 and Q0  Q1 are defined set of individual characteristics. Therefore, in all
survey question outcomes for males and females, cases, we ensure that the balancing property is satisfied,
respectively, and we are interested in the difference while estimating the propensity score. Finally, the com-
between – Q1 . The average effect of treatment on the mon support condition assumes that for each female or
treated (AETT) can be written as follows: treated unit, there are male or control units with similar
observable characteristics. When these assumptions are
met, the observed outcome of males can be used to
τ ¼ E fQ1i  Q0i jDi ¼ 1g
estimate the counterfactual outcome of males (or people
¼ E fEfQ1i  Q0i jDi ¼ 1; pðXl Þgg who did not complete college) in the case of being
¼ E fEfQ1i jDi ¼ 1; pðXi Þg  EfQ0i jDi ¼ 0; pðXi ÞgjDi g females (or in the analogous case of being a college
(3) graduate).

5
Imbens (2000) and Lechner (2001) extend the method of propensity score matching to multiple mutually exclusive programmes.
Froelich (2004) discusses different impact evaluation methods, including those based on the conditional independent assumption in a
similar context. Also, see Lechner (2002a, b).
6
For a formal proof, see Rosenbaum and Rubin (1983) and Imbens (2000).
582 D. G. De Silva and R. A. J. Pownall
Next, we used probit models to estimate propensity in OLS regressions. In Table 10, we report these results.
scores that are employed in matching techniques. The We report only the specification similar to the one used in
results of the probit models are reported in Table A1. The Racine and Li (2004) and column 1 in the OLS Tables.
results in column 1 indicate that, in the sample, females are Other results can be provided upon request, and they are
younger and less educated than males. In column 2, we all qualitatively similar to the ones reported in Table 10.
control for employment types. Employment types indicate We are interested in examining whether there is a differ-
that women in the sample have a higher probability to hold ence between the estimated coefficient of females’ (and
paid work, run family-own businesses, be at home (house- college graduates) across the 0.25, 0.50 and 0.75 quan-
wives), be unable to work or be involved in volunteer work tiles. The difference across the three quantiles tested from
compared to men. These results are not shown in column 2 the models are statistically insignificant. However, we
of Table A1 in order to save space and can be provided upon observe that the effect of mortgage ownership does not
request. Therefore, it is even more important to match hold for all quantiles for the question regarding those
individuals with similar characteristics by gender when individuals who ‘prefer reducing greenhouse gases than
making inferences regarding their willingness to reduce financial well-being’, and it is significant only at the lower
their current standard of living in order to maintain the quantiles. Results also indicate that mortgage status does
environment at the same level for the next generation (or not matter when considering the questions ‘prefer redu-
other survey questions). In columns 3 and 4, we report the cing greenhouse gases than social welfare’ and ‘prefer
probit results for college graduates. Note that the coefficient social welfare than financial well-being’. We also estimate
for ‘female’ is statistically insignificant, meaning that, in a quantile model for college sample. Results are qualita-
this sample, equal proportions of both males and females tively similar to those we observe in OLS and matching
have attained college degrees. We also want to examine the models. We do not report these results in order to save
differences between males and females in the college- space and again can be provided upon request.
educated sample. Therefore, we use our college only sam-
ple and estimate propensity scores to be used in the match-
ing technique. These probit results are reported in column 5 V. Discussion and Conclusions
of Table A1.
In Table 8 (a–d), we report the effect of gender on the This article is the first to provide evidence of the value
four survey questions. Full sample results are presented in structure that individuals apply to the three pillars
columns 1 and 2, and the college only sample results are required for sustainable development, namely, economic,
provided in column 3. Here, we use two matching techni- social and environmental concerns. Results from the
ques: (1) radius matching and (2) kernel density match- analysis provide strong evidence that the willingness of
ing.7 In radius matching, we specify the radius to be 0.005. households to reduce their current standard of living in
The counterfactual results indicate that men would have order to maintain the environment at its current levels
reacted very differently to these survey questions if they depends on heterogenous factors. We use a unique survey
were females. This result holds for the college only sample that asks households directly about their attitudes to
too. Table 9 presents the AETT effects by college attain- sustainable values. We are able to derive the willingness
ment. The results indicate that noncollege graduates of households to give up their current financial well-
would have responded very differently to these questions being in order to reduce carbon emissions. We find that
if they were college graduates. The differences are very females play a significant and positive role in establishing
similar in magnitude and statistical significance to those positive values towards both people and planet, in terms
shown in conditional distribution differences in Table 3 of social welfare and reducing carbon emissions, com-
and the OLS regression Tables 4–7 are consistent for all pared to men. Similar observations can be made when
the dependent variables. comparing college graduates to individuals without a
The OLS and matching results provide the mean differ- college degree. Those with a university education place
ences between males and females (educational attain- more value on reducing carbon emissions and on social
ment). Next, as a robustness check, we test if these welfare. This effect is even larger for college-educated
differences hold across the distribution as well. females compared to any other group. We also observe
Therefore, we use the quantile regression technique intro- that this effect trickles down to those with a high school
duced by Koenker and Bassett (1982). We restrict estima- education when faced with the preference to live in an
tion to three quantiles – 0.25, 0.50 and 0.75 – and estimate economy that strives towards social welfare or one that
the models. These models are similar to the ones we used strives towards financial well-being. Age is also an

7
A detailed discussion about how to implement these matching techniques is given by Becker and Ichino (2002).
 Going green

Table 8. Matching results by gender

Full sample College sample

(1) (2) (3)

Treatment Control Treatment Control Treatment Control

Matching estimator n n AETT n n AETT n n AETT

a. Effect of gender on ‘willing to reduce standard of living to maintain the environment’

Radius matching (r ¼ :005) 638 774 .399** (.120) 617 777 .418** (.121) 212 296 .701** (.235)
Kernel matching 644 766 .373** (.128) 644 788 .388** (.109) 243 311 .593** (.171)
b. Effect of gender on ‘prefer reducing greenhouse gases than financial well-being’
Radius matching (r ¼ :005) 638 774 .402** (.126) 617 777 .443** (.123) 212 296 .741** (.244)
Kernel matching 644 766 .416** (.112) 644 788 .457** (.107) 243 311 .821** (.158)
c. Effect of gender on ‘prefer reducing greenhouse gases than social welfare’
Radius matching
(r ¼ :005) 638 774 −.144 (.114) 617 777 −.038 (.133) 212 296 .168 (.186)
Kernel matching 644 766 −.094 (.119) 644 788 −.061 (.117) 243 311 .087 (.188)
d. Effect of gender on ‘prefer social welfare than financial well-being’
Radius matching (r ¼ :005) 638 774 .381** (.102) 617 777 .350** (.119) 212 296 .635** (.202)
Kernel matching 644 766 .376** (.118) 644 788 .407** (.088) 243 311 .560** (.159)

Notes: ** Denotes statistical significance at the 5% level.

Bootstrapped standard errors are in parentheses.
583
584 D. G. De Silva and R. A. J. Pownall
Table 9. Matching results by educational attainment

(1) (2)

Treatment Control Treatment Control

Matching estimator n n AETT n n AETT

a. Effect of college on ‘willing to reduce standard of living to maintain the environment’

Radius matching (r ¼ :005) 542 446 .467** (.163) 541 829 .316** (.139)
Kernel matching 562 459 .437** (.147) 562 860 .411** (.113)
b. Effect of college on ‘prefer reducing greenhouse gases than financial well-being’
Radius matching (r ¼ :005) 542 446 .622** (.186) 541 829 .289** (.121)
Kernel matching 562 459 .439** (.162) 562 860 .367** (.125)
c. Effect of college on ‘prefer reducing greenhouse gases than social welfare’
Radius matching (r ¼ :005) 542 446 −.009 (.177) 541 829 −.204 (.132)
Kernel matching 562 459 −.093 (.150) 562 860 −.101 (.126)
d. Effect of college on ‘prefer social welfare than financial well-being’
Radius matching (r ¼ :005) 542 446 .840** (.155) 541 829 .456** (.121)
Kernel matching 562 459 .691** (.137) 562 860 .563** (.110)

Notes: ** Denotes statistical significance at the 5% level.

Bootstrapped standard errors are in parentheses.

Table 10. Quantile regression results

Quantile

Variable .25 .50 .75

a. Results for ‘willing to reduce standard of living to maintain the environment’

Female ðβ1 Þ .473* (.126) .303** (.138) .257** (.128)
College and above ðβ2 Þ .416** (.150) .754** (.166) .485** (.147)
Number of Observations 1433 1433
Pseudo R2 .004 .023 .033
b. Results for ‘prefer reducing greenhouse gases than financial well-being’
Female ðβ1 Þ .486* (.141) .450** (.143) .476** (.131)
College and above ðβ2 Þ .908** (.170) .754** (.172) .447** (.155)
Number of Observations 1433 1433 1433
Pseudo R2 .028 .023 .017
c. Results for ‘prefer reducing greenhouse gases than social welfare’
Female ðβ1 Þ .226 (.238) −.080 (.116) −.432 (.263)
College and above ðβ2 Þ −.051 (.285) −.002 (.139) −.087 (.177)
Number of Obseravtions 1433 1433 1433
Pseudo R2 .002 .019 .007
d. Results for ‘prefer social welfare than financial well-being’
Female ð1 Þ .457** (.115) .395** (.132) .489** (.183)
College and above ðβ2 Þ 1.145** (.137) .886** (.159) .851** (.219)
Number of Observations 1433 1433 1433
Pseudo R2 .025 .054 .031

Notes: ** Denotes statistical significance at the 5% level and * denotes statistical significance at the 10% level.
Bootstrapped standard errors are in parentheses.
We test H0 : β25
1 ¼ β 1 ¼ β 1 and H0 : β1 ¼ β1 ¼ β1 for all Tables 10a–d. Our test results fail to reject H0 for all tests. These results
50 75 25 50 75

can be provided upon request.

important factor in determining the willingness to reduce Of further interest is the finding that when households
carbon emissions for financial well-being. Interestingly, are also homeowners, they value financial welfare higher
income does not drive any of our results. In all cases, than do those renting their homes. When the household
income is not statistically significant in determining the has a large amount of debt outstanding on their residential
willingness to make any of the trade-offs. Again, our property, they appear to be more concerned with the over-
results are robust across all income levels. all financial well-being of society as a whole. When the
Going green 585
question is phrased in terms of living standards, home- Andreoni, J. and Vesterlund, L. (2001) Which is the fair sex?
owners are willing to reduce their standard of living to Gender differences in Altruism, Quarterly Journal of
Economics, 116, 293–312.
maintain the environment but not when phrased in terms
Baron, D. P. (2001) Private politics, corporate social responsi-
of their financial well-being. This raises some interesting bility, and integrated strategy, Journal of Economics and
concerns regarding the impact of housing wealth on sus- Management Strategy, 10, 7–45.
tainable values. Homeowners who have a large mortgage Becker, G. (1974) A theory of social interactions, Journal of
on their home are significantly concerned with this finan- Political Economy, 82, 1063–93.
Becker, S. O. and Inchino, A. (2002) Estimation of average
cial outlay. They are willing to reduce their standard of
treatment effects based on propensity scores, The Stata
living to maintain the environment, but favour striving Journal, 2002, 358–77.
for financial well-being over reducing greenhouse gases. Charness, G. and Rabin, M. (2002) Understanding social prefer-
We had hoped that the size of the household’s home ences with simple tests, Quarterly Journal of Economics,
equity would play a significant role in addressing these 117, 817–69.
Croson, R. and Gneezy, U. (2009) Gender differences in prefer-
values. The size of the mortgages to the value of the
ences, Journal of Economic Literature, XLVII, 448–74.
house is consistently large over the sample, about 95% Eichholtz, P., Kok, N. and Quigley, J. M. (2010) Doing well by
or more. The system of tax incentives in the Netherlands doing good? Green office buildings, American Economic
aids in maintaining high debt-to-equity ratios on personal Review, 100, 2492–509.
property. We would like to investigate this in greater Falk, A. and Fischbacher, U. (2006) A theory of reciprocity,
Games and Economic Behavior, 54, 293–315.
detail, with more specific information on loan-to-value
Fombrun, C. J. and Shanley, M. (1990) What’s in a name?
ratios, since the current debt-to-equity rates are main- Reputation building and corporate strategy, Academy of
tained at high levels throughout the term of the loan in Management Journal, 33, 233–58.
the Netherlands. The issue of housing wealth having an Froelich, M. (2004) Programme evaluation with multiple treat-
influential role on sustainable values is worthy of greater ments, Journal of Economic Surveys, 18, 181–224.
Imbens, G. W. (2000) The role of propensity score in estimating
investigation. The results have also been taken during the
dose – response functions, Biometrika, 87, 706–10.
current downturn in the real estate market, so those with Koenker, R. and Bassett, G. (1982) Robust tests for heterosce-
mortgages are likely to be much more concerned about dasticity based on regression quantiles, Econometrica, 50,
the level of their wealth, and when asked if they are 43–61.
willing to reduce their current standard of living in Kuhnen, C. and Knutson, B. (2011) The influence of affect
on beliefs, preferences, and financial decisions,
order to maintain the environment, they state that they
Journal of FInancial and Quantitative Analysis, 46,
are in favour of cutting their current standard of living; 605–26.
homeowners generally prefer to strive for financial Lechner, M. (2001) Identification and estimation of causal
values over the planet. The home may be regarded as effects of multiple treatments under the conditional inde-
the homeowner’s ‘financial castle’ that he is not willing pendence assumption, in Econometric Evaluation of
Labour Market Policies, Pfeiffer, F. and Lechner, M.
to give up.
(Eds), Physica-Verlag, Heidelberg, pp. 43–58.
An understanding of how heterogeneity plays an impor- Lechner, M. (2002a) Programme heterogeneity and propensity
tant role in assessing sustainable values will contribute score matching: an application to the evaluation of active
towards making economic policy more effective in labor market policies, The Review of Economics and
achieving the goal of sustainable development. Our results Statistics, 84, 205–20.
Lechner, M. (2002b) Some practical issues in the evaluation of
provide a first crucial step towards understanding how we
heterogeneous labour market programmes by matching
can treat economic, social and environmental aspects in an methods, Journal of the Royal Statistical Society: Series a
integrated way in order to best meet our current needs (Statistics in Society), 165, 59–82.
without compromising the ability of future generations to List, J. (2006) Field Experiments: a bridge between lab and
meet their needs. naturally occurring data, Advances in Economic Analysis
and Policy, 6, Article 8.
Loewenstein, G. F., Weber, E., Hsee, C. K. et al. (2001) Risk as
feelings, Pscyhological Bulletin, 127, 267–86.
Acknowledgements Lusardi, A. and Mitchell, O. (2008) Planning and financial
We would like to thank Ronald Peeters, Reyer Gerlach, literacy: how do women fare?, American Economic
Andres Vargas, Ian Walker, Aart de Zeeuw and the Royal Review, 98, 413–17.
Lyon, T. P. and Maxwell, J. W. (2011) Greenwash: corporate
Economic Society Annual Conference (2012) participants environmental disclosure under threat of audit. Journal of
for helpful comments. Economics and Management Strategy, 20, 3–41.
Racine, J. and Li, Q. (2004) Nonparametric estimation of regres-
sion functions with both categorical and continuous data,
References Journal of Econometrics, 119, 99–130.
Alessie, R., Hochguertel, S. and van Soest, A. (2004) Ownership Rosenbaum, P. R. and Rubin, D. B. (1983) The central role of the
of stocks and mutual funds: a panel data analysis, Review of propensity score in observational studies for causal effects,
Economics and Statistics, 86, 783–96. Biometrika, 70, 41–55.
586 D. G. De Silva and R. A. J. Pownall
Social Investment Forum. 2010. 2010 Report on Socially prospective employees, Academy of Management Journal,
Responsible Investing Trends in the United States, Social 40, 658–72.
Investment Forum Foundation, Washington, DC. Available Waddock, S. A. and Graves, S. B. (1997) The corporate social
at http://www.greenmoneyjournal.com/summer-2011/the- performance-financial performance link, Strategic
2010-report-on-socially-responsible-investing-trends-in-the- Management Journal, 18, 303–19.
united-states/ (accessed 22 October 2013). Watson, J. and McNaughton, M. (2007) Gender
Stevens, C. (2010) Are women the key to sustainable develop- differences in risk aversion and expected retire-
ment? Sustainable Development Insights, 3, 1–8. ment benefits, Financial Analysts Journal, 63,
Turban, D. B. and Greening, D. W. (1997) Corporate social 52–62.
performance and organizational attractiveness to

Appendix

Table A1. Probit results

Full sample College sample

Female College graduate Female

Variable (1) (2) (3) (4) (5)

Log of age −.213** (.050) −.293** (.074) −.077 (.047) −.112 (.071) −.465** (.082)
Female −.007 (.027) .037 (.029)
Number of children −.005 (.016) −.043** (.017) −.052** (.016) −.036** (.016) .000 (.025)
Mortgage owner −.047 (.034) −.063* (.035) −.073 (.063)
High school education −.109** (.034)
College and above −.055* (.033)
Log of net house hold income −.100** (.031) .218** (.045) −.125** (.054)
Main cities (Amsterdam, −.072** (.036) −.062*(.038) .019 (.035) −.007 (.035) −.041 (.059)
Rotterdam and the Hague)
Employment status effects Yes Yes
Number of observations 1433 1433 1433 1433 562
Wald χ 2 47.66 166.24 31.29 92.41 47.41
Pseudo R2 .026 .142 .042 .053 .070

Notes: Robust standard errors are in parentheses.

** denotes statistical significance at the 5% level and * denotes statistical significance at the 10% level.