See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/279830115

Unnatural selection

Thesis · August 2014

CITATIONS READS

0 702

1 author:

Hannes Kröger

51 PUBLICATIONS 937 CITATIONS

SEE PROFILE

All content following this page was uploaded by Hannes Kröger on 27 September 2015.

The user has requested enhancement of the downloaded file.

 Unnatural Selection -
The Social Dimension of the Health Selection Hypothesis

Dissertation
zur Erlangung des akademischen Grades
doctor philosophiae
(Dr. phil)

eingereicht an
der Philosophischen Fakultät III
der Humboldt-Universität zu Berlin

von Hannes Kröger

Präsident der Humboldt-Universität zu Berlin

Prof. Dr. Jan-Hendrik Olbertz

Dekanin der Philosophischen Fakultät III

Prof. Dr. Julia von Blumenthal

Gutachter:
1. Prof. Dr. Martin Groß
2. Prof. Dr. Matthias Richter
3. Prof. Dr. Martin Kroh

Tag der Disputation: 20.6.2014

Table of Contents

List of Figures 4

List of Tables 5

1. Introduction 10
1.1. The Challenge of Health Inequalities . . . . . . . . . . . . . . . . . . . . . . 10
1.2. Scientific Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
1.3. Structure of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

2. Theory 17
2.1. Health Selection and Health Inequalities . . . . . . . . . . . . . . . . . . . . . 18
2.1.1. Defining Health Inequalities . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.2. Defining Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.1.3. Social Causation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
2.1.4. Health Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.2. Who Does all the Selection? In Search for Theory in the Literature . . . . . . 23
2.2.1. What Should We Look for? . . . . . . . . . . . . . . . . . . . . . . . 23
2.2.2. Studies with Theoretical Contributions . . . . . . . . . . . . . . . . . 24
2.3. Theory of Health Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
2.4. Health Selection - from Hypothesis to Theory . . . . . . . . . . . . . . . . . . 30
2.4.1. Human Capital Theory . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.4.2. Human Capital Theory and Health . . . . . . . . . . . . . . . . . . . 31
2.4.3. Decreasing Returns to Effort . . . . . . . . . . . . . . . . . . . . . . . 35
2.5. Gender Differences in Health Effects . . . . . . . . . . . . . . . . . . . . . . . 36
2.5.1. Subjective Performance Evaluations . . . . . . . . . . . . . . . . . . . 36
2.5.2. Women’s Disadvantage on the Labor Market . . . . . . . . . . . . . . 39
2.5.3. The Result of Disadvantage and Subjective Evaluation . . . . . . . . . 46
2.5.4. Visibility of Health Problems . . . . . . . . . . . . . . . . . . . . . . . 47
2.6. Health Behavior as an Explanatory Factor . . . . . . . . . . . . . . . . . . . . 49
2.6.1. Gender Differences in Health Behavior . . . . . . . . . . . . . . . . . 51
2.6.2. Consequences of Differences in Health Behavior on Health Effects . . . 51
2.7. Open and Closed Positions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
2.7.1. Open and Closed Positions - Weber’s Theory of Social Closure . . . . . 53
2.7.2. Open and Closed Positions - Sørensen’s theory . . . . . . . . . . . . . 54
2.7.3. Reducing Supply, Increasing Demand - Mechanisms of Social Closure . 56
2.7.3.1. Credentialism and Professions . . . . . . . . . . . . . . . . . 57
2.8. Open and Closed Positions and Health . . . . . . . . . . . . . . . . . . . . . 58
2.8.1. Open and Closed Positions: Incumbents vs. Applicants . . . . . . . . . 59
2.8.1.1. How the Choice of Labor Market Outcome Affects the Role
of Health . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
2.8.1.2. Incumbents vs. Applicants . . . . . . . . . . . . . . . . . . 60

1
 Table of Contents

2.8.1.3. Open and Closed Positions and Discriminated Groups . . . . 62

2.8.2. The Benefits of the Theory of Open and Closed Positions for a Theory
of Health Selection . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
2.9. Summary, Assumptions and Hypotheses . . . . . . . . . . . . . . . . . . . . . 65
2.9.1. Summary and Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . 65
2.9.2. Set of Assumptions . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

3. Review of the Empirical Literature 71

3.1. Overview and Development of a “Good Practice” Approach . . . . . . . . . . 71
3.2. The “Good Practice” Studies in Detail . . . . . . . . . . . . . . . . . . . . . 79
3.2.1. Studies from Medical Sociology and Psychology, Epidemiology, Public
Health, and Sociology of Health . . . . . . . . . . . . . . . . . . . . . 79
3.2.2. Studies from Health Economics . . . . . . . . . . . . . . . . . . . . . 84

4. Methods 88
4.1. Health Selection vs. Social Causation - The Issue of Causality . . . . . . . . . 88
4.1.1. The Counterfactual Model of Causality . . . . . . . . . . . . . . . . . 88
4.1.2. The Counterfactual Argument . . . . . . . . . . . . . . . . . . . . . . 89
4.2. Applying the Counterfactual Model . . . . . . . . . . . . . . . . . . . . . . . 91
4.2.1. Direction of Causality . . . . . . . . . . . . . . . . . . . . . . . . . . 92
4.2.1.1. Closed Positions and Anticipation Effects . . . . . . . . . . . 92
4.2.1.2. Testing Causality in One Model . . . . . . . . . . . . . . . . 93
4.2.2. Spurious Correlation . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2.3. Are the Estimated Effects Causal Effects? . . . . . . . . . . . . . . . . 94
4.3. Measuring Health - Methodological and Theoretical Implications . . . . . . . . 94
4.3.1. Literature on the Measurement of Subjective Health . . . . . . . . . . 95
4.3.2. Measurement of Health in this Study . . . . . . . . . . . . . . . . . . 96
4.4. Confirmatory Factor Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 96
4.4.1. CFA with Categorical Dependent Variables . . . . . . . . . . . . . . . 99
4.4.2. Measurement Invariance . . . . . . . . . . . . . . . . . . . . . . . . . 100
4.4.3. Comparing Fit in CFA Models - Tests and Model Fit Indices . . . . . . 102
4.4.3.1. The Problem of Model Fit Assessment Using CFA . . . . . . 102
4.4.3.2. Fit Statistics Used . . . . . . . . . . . . . . . . . . . . . . . 103
4.5. The Unidirectional Approach: Fixed-Effects Logistic Regression . . . . . . . . 104
4.6. Applicants vs. Incumbents - The Use of Survival Analysis . . . . . . . . . . . 106
4.7. Testing Selection vs. Causation . . . . . . . . . . . . . . . . . . . . . . . . . 108
4.8. Decomposition of Health Inequalities . . . . . . . . . . . . . . . . . . . . . . 110
4.8.1. Indirect Effects Through Time Constant Factors . . . . . . . . . . . . 115
4.9. Data, Variables, and Measurement of Controls . . . . . . . . . . . . . . . . . 121
4.9.1. The Socio-Economic Panel Study - SOEP . . . . . . . . . . . . . . . 121
4.9.2. Control Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
4.10. Measuring Social Closure on the Labor Market . . . . . . . . . . . . . . . . . 124
4.10.1. “Glass-Escalators” and Token Positions . . . . . . . . . . . . . . . . . 124
4.10.2. Professionalism - Knowledge Intensity of Occupations . . . . . . . . . 124

5. Results 126
5.1. Sample Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
5.2. Health and Job Status - Descriptive Evidence . . . . . . . . . . . . . . . . . . 131
5.3. Health as a Latent Variable - Confirmatory Factor Analysis . . . . . . . . . . . 137

2
 Table of Contents

5.3.1. Establishing Measurement Invariance . . . . . . . . . . . . . . . . . . 137

5.3.2. External Reliability - Latent Health as a Predictor of Mortality . . . . . 143
5.4. Basic Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144
5.4.1. Other Model Specifications . . . . . . . . . . . . . . . . . . . . . . . 146
5.4.2. Taking a Closer Look at Closure in the Private Sector: Glass Escalator
and Professionalism . . . . . . . . . . . . . . . . . . . . . . . . . . . 148
5.5. Effects of Long Term Health Conditions . . . . . . . . . . . . . . . . . . . . . 152
5.6. Health Effects and the Status of Applicants vs. Incumbents . . . . . . . . . . 153
5.7. Testing Health Selection vs. Social Causation . . . . . . . . . . . . . . . . . . 156
5.8. Comparing the Health Effect to Other Predictors . . . . . . . . . . . . . . . . 164
5.9. Decomposing Overall Health Inequalities . . . . . . . . . . . . . . . . . . . . 167
5.10. Summary of Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.10.1. Does Health Selection Contribute to Job Status Related Health Inequal-
ities? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 173
5.10.2. Does Social Context Matter? . . . . . . . . . . . . . . . . . . . . . . 174
5.10.3. What is More Important: Health Selection, Social Causation or Third
Factors? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 175
5.10.4. Are There Gender Differences in Health Effects? . . . . . . . . . . . . 176

6. Conclusion 177
6.1. Overview . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
6.2. Key Results and Unresolved Questions . . . . . . . . . . . . . . . . . . . . . 178
6.3. Caveats and Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180
6.4. Generalizing the Theory of Health Selection . . . . . . . . . . . . . . . . . . . 181
6.5. Further Research . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182

Bibliography 185

A. Appendix i
A.1. Results Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . i
A.2. Strategies of Closure on the Labor Market . . . . . . . . . . . . . . . . . . . xxiii
A.2.1. Protection Against Dismissal . . . . . . . . . . . . . . . . . . . . . . xxiii
A.2.2. Unionization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xxiv
A.2.3. Licensing and Registration . . . . . . . . . . . . . . . . . . . . . . . . xxiv
A.2.4. Monopoly and Ownership . . . . . . . . . . . . . . . . . . . . . . . . xxv
A.3. Open and Closed Positions - Segmented Labor Market Theory . . . . . . . . . xxvi
A.4. Review of the Literature on Reporting Heterogeneity in Self-Rated Health . . . xxix

3
List of Figures
4.1. Health as a Latent Variable . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.2. A Cross-Lagged Model with Fixed-Effects . . . . . . . . . . . . . . . . . . . . 110
4.3. Indirect Effect of Health Through Time Constant Factor u . . . . . . . . . . . 118

5.1. Health Inequality due to Job Status - Subjective Health . . . . . . . . . . . . 132

5.2. Health Inequality due to Job Status - Sickness Absence . . . . . . . . . . . . 134
5.3. Size of Health Inequalities - Job Status Compared to Education . . . . . . . . 136
5.4. Confirmatory Factor Analysis - Gender . . . . . . . . . . . . . . . . . . . . . 139
5.5. Confirmatory Factor Analysis - Time . . . . . . . . . . . . . . . . . . . . . . 141
5.6. Confirmatory Factor Analysis - Job Status . . . . . . . . . . . . . . . . . . . 142
5.7. Change in Health as a Predictor of Change in Job Status . . . . . . . . . . . 145
5.8. Change in Sickness Absence as a Predictor of Change in Job Status . . . . . . 146
5.9. Different Model Specifications of Results - Health . . . . . . . . . . . . . . . 147
5.10. Different Model Specifications of Results - Sickness Absence . . . . . . . . . . 147
5.11. Health Effect Depending on Knowledge Intensity of Occupation . . . . . . . . 149
5.12. Effect of log. Days of Sickness Absence Depending on Knowledge Intensity of
Occupation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 149
5.13. Health Effect Depending on Male or Female Dominated Occupations . . . . . 151
5.14. Effect of log. Days of Sickness Absence Depending on Male or Female Domi-
nated Occupations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
5.15. Change in Long Term Bad Health as a Predictor of Change in Job Status . . . 153
5.16. Health Effect on Status Change Depending on the Position of Applicant or
Incumbent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 155
5.17. Effect of Sickness Absence on Status Change Depending on the Position of
Applicant or Incumbent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
5.18. Health Effects - Allowing for Reversed Causality in a Cross-Lagged Model with
Fixed-Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157
5.19. Effects of High Status Jobs on Health - Allowing for Reversed Causality in a
Cross-Lagged Model with Fixed-Effects . . . . . . . . . . . . . . . . . . . . . 158
5.20. Effects of High Status Jobs on Health in Cross-Section - Allowing for Reversed
Causality in a Cross-Lagged Model . . . . . . . . . . . . . . . . . . . . . . . 161
5.21. log. Days of Sickness Absence Effects - Allowing for Reversed Causality in a
Cross-Lagged Model with Fixed-Effects . . . . . . . . . . . . . . . . . . . . . 162
5.22. Effects of High Status Jobs on log. Days of Sickness Absence - Allowing for
Reversed Causality in a Cross-Lagged Model with Fixed-Effects . . . . . . . . 163
5.23. Effects of High Status Jobs on log. Days of Sickness Absence in Cross-Section
- Allowing for Reversed Causality in a Cross-Lagged Model . . . . . . . . . . . 163
5.24. Women in the Private Sector: Proportion of Health Inequalities due to Factors
which are... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168

4
List of Tables
2.1. Expected Effect of Health Depending on the Type of Labor Market Position . . 63
2.2. Change of the Degree of Closure of a Labor Market Position Depending on
Discrimination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

3.1. Overview of Studies Involving Health Selection . . . . . . . . . . . . . . . . . 73

4.1. Meaning of Different Value Ranges of Decomposition - in Percent . . . . . . . 115

4.2. Description of Control Variables . . . . . . . . . . . . . . . . . . . . . . . . . 123
4.3. Proportion of Men and Women According to Hultin’s Classification of Occupa-
tions - in Percent . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

5.1. Sample and Sample Restrictions for Women . . . . . . . . . . . . . . . . . . 127

5.2. Sample and Sample Restrictions for Men . . . . . . . . . . . . . . . . . . . . 128
5.3. Sample Statistics for Women . . . . . . . . . . . . . . . . . . . . . . . . . . 129
5.4. Sample Statistics for Men . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
5.5. Latent Health Factor Score as Predictor of Mortality . . . . . . . . . . . . . . 143
5.6. Variable Groups for Assessing Mediated Effects . . . . . . . . . . . . . . . . . 159
5.7. Relative Size of the Health Effect Compared to Other Predictors of Job Status
- in Percentage Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166
5.8. Relative Size of the Effect of Sickness Absence Compared to Other Predictors
of Job Status - in Percentage Points . . . . . . . . . . . . . . . . . . . . . . 166
5.9. Decomposition of Health Inequalities - Subjective Health . . . . . . . . . . . . 170
5.10. Decomposition of Health Inequalities - log. Days of Sickness Absence . . . . . 172
5.11. Health Selection in Open and Closed Positions . . . . . . . . . . . . . . . . . 175

A.1. The Effects of Subjective Health on Job Status . . . . . . . . . . . . . . . . . i

A.2. The Effects of log. Days of Sickness Absence on Job Status . . . . . . . . . . ii
A.3. The Effects of Subjective Health on Job Status - Mediated by Knowledge Intensity iii
A.4. The Effects of log. Days of Sickness Absence on Job Status - Mediated by
Knowledge Intensity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
A.5. The Effects of Subjective Health on Job Status in Different Occupations - Women v
A.6. The Effects of Subjective Health on Job Status in Different Occupations - Men vi
A.7. The Effects of log. Days of Sickness Absence on Job Status in Different
Occupations - Women . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
A.8. The Effects of log. Days of Sickness Absence on Job Status in Different
Occupations - Men . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii
A.9. The Effects of Long-Term Poor Health on Job Status . . . . . . . . . . . . . ix
A.10.The Effects of Subjective Health on Job Status - Applicants . . . . . . . . . . x
A.11.The Effects of Subjective Health on Job Status - Incumbents . . . . . . . . . xi
A.12.The Effects of log. Days of Sickness Absence on Job Status - Applicants . . . xii
A.13.The Effects of log. Days of Sickness Absence on Job Status - Incumbents . . . xiii
A.14.Cross-Lagged Panel Model - Cross-Sectional . . . . . . . . . . . . . . . . . . xiv

5
 List of Tables

A.15.Cross-Lagged Panel Model with Fixed-Effects . . . . . . . . . . . . . . . . . . xvii

A.16.Cross-Lagged Panel Model for log. Days of Sickness Absence - Cross-Sectional xix
A.17.Cross-Lagged Panel Model with Fixed-Effects for log. Days of Sickness Absence xxi

6
Abstract
In this PhD-thesis the social context of health selection processes on the German labor market
are investigated theoretically and empirically. Based on human capital theory a number of
hypotheses about the causal effect of subjective health and sickness absence on job status
are derived. The theory is modified to allow the effect of health to vary with the degree of
disadvantage a person faces and the degree of social closure of the job. In concrete terms, the
moderating effect of gender and public versus private sector are investigated, as well as the
occupational gender composition. The empirical analyses are based on the Socio-economic
panel study (SOEP) using different methods to estimate causal effects of subjective health on
subsequent job status. A decomposition of overall health inequalities into effects attributable to
time-constant, time-varying confounders and into health selection processes is presented. The
results show that health selection is present for women in the private sector, but not for men
nor in the public sector. Sickness absence shows the strongest effects for men in the private
sector, but not for women nor in the public sector. For the chosen setting, health selection
processes are strongest in open positions and for groups that are disadvantaged.

7
Zusammenfassung
Die vorliegende Dissertation untersucht theoretisch und empirisch gesundheitliche Selektion-
sprozesse auf dem deutschen Arbeitsmarkt und wie diese durch soziale Kontextfaktoren bee-
influsst werden. Aufbauend auf der Humankaiptaltheorie werden eine reihe von Hypothesen
aufgestellt über den kausalen Effekt der subjektiven Gesundheit und der krankheitsbedingten
Fehltage auf den Jobstatus. Die Humankapitaltheorie wird so erweitert, dass der Effekt
der Gesundheit mit dem Grad der Benachteiligung einer Person und dem Grad der sozialen
Schließung des Jobs variieren kann. Konkret werden die moderierenden Einflüsse des Geschlechts
und des öffentlichen versus privaten Sektors sowie der Geschlechteranteile im Beruf unter-
sucht. Die empirische Analysen werden auf Basis des Sozio-oekonomischen Panels (SOEP)
durchgeführt. Verschiedene Methoden werden angewandt, um kausale Effekte der subjek-
tiven Gesundheit auf den Jobstatus zu schätzen. Weiterhin wird eine Dekomposition der
gesamten gesundheitlichen Ungleichheit in zeit-konstante und zeit-variante Faktoren und auf
gesundheitlicher Selektion basierende Prozesse durchgeführt. Die Ergebnisse zeigen, dass
gesundheitliche Selektion für Frauen im privaten Sektor gefunden werden kann, nicht aber
im öffentlichen Sektor und auch nicht für Männer. Für krankheitsbedingte Fehltage kann ein
kausaler Effekt für Männer im privaten Sektor gefunden werden, nicht aber im öffentlichen
Sektor oder für Frauen. Unter den gewählten Bedingungen der Studie, sind gesundheitliche
Selektionsprozesse daher eher in offenen Positionen und für Gruppen, die einer Benachteiligung
gegenüber stehen zu finden.

8
Acknowledgements
During my work on this thesis, I have received much support, both academically and personally.
Therefore, I would like to take the opportunity to thank my supervisor Martin Groß for
counseling me through my studies and my PhD-thesis. I also want to thank Matthias Richter,
and especially Martin Kroh for excellent feedback and helpful guidance. From my graduate
school I would like to express my gratitude to Frederike Esche, Timo Lochocki, Jasmin Rocha,
Nicolas Grießhaber, Matthias Orlowski, Jana Rieckmann and Christoph Raiser, all SESS.PhD-
candidates, and the whole 2010-generation. Further, I am grateful to support from Bettina
Sonnenberg and Johannes Langer. My greatest thanks I owe to Lea Kröger. Of course, none
of the aforementioned should be implicated in any mistakes, which might remain in the thesis.

9
1. Introduction

1.1. The Challenge of Health Inequalities

“Health” that is in several languages the short form of making a toast . In Italian it is “Alla
Salute”, in French it is “Au Santé”, in Russian “sa sdorowje”, in Greek “jamas”. Wishing for
health seems to be an important part of cultures. But it is not only when we look at traditional
toasts that we get the impression that health is of central importance to modern societies. In
the scientific world we see such a trend as well.
Recently, traditional measures of welfare like the GDP are gradually replaced and complemented
by more comprehensive measures of welfare. All of these indicators include at least one form of
measuring public health in a society.
The so called Stieglitz-Sen-Fitoussi commission1 proposed that one essential dimension of
well-being is health. Similarly, the enquete commission of the German Bundestag proposed
health as one important dimension of measuring prosperity. They suggest that life expectancy
and especially healthy years of life are an adequate and important measure for the prosperity
of a country2 .
Health disparities are a form of inequality which faces one of the lowest acceptance rates in
society (Blacksher, Rigby & Espey 2010). Most governments have therefore, in one form or
another, pledged to reduce health inequalities. This includes traditionally social-democratic
welfare states like Sweden (Socialstyrelsen 2010), conservative welfare states likes Germany
(BMAS 2013) and liberal welfare states like the US (HHS 2011) and the UK (Smith 2000).
The EU also has a strategy for reducing health inequalities (EPEH 2007). It is fascinating that
although there seems to be universal agreement to abolish health inequalities, they still persist
for decades throughout continents and throughout different welfare regimes. This puzzle has
kept scientists busy for some time now, resulting in several theories which try to explain this
phenomenon. So far, no general conclusion has been reached (Mackenbach 2012).
The goals of various OECD-countries therefore imply, that health is supposed to be universally
enjoyed regardless of social position. Abolishing health inequalities would thus mean that social
position does not correspond with health status. This normative approach to public health
can be supplemented by a more instrumental understanding of the role health and health

1
The “Report by the Commission on the Measurement of Economic Performance and Social Progress”
(Stiglitz, Sen & Fitoussi 2009).
2
For the report see Bundestag (2011).

10
 1. Introduction

inequalities play. We can arrive at this understanding if we ask the following questions. Is
health only an end in itself? Something that is valued for its own sake? Or is it also a means
to achieve other things in life? Do individuals use their health strategically to get into certain
social positions? Does such a strategic use lead to selection of the healthy from the unhealthy
in favorable positions? Are health inequalities affected by such selection processes? If such a
selection process existed, this would mean that we would observe differences in health status
between social positions, because those in good health get into good positions while those
in poor health only get into low positions. And if health inequalities are generated in such a
fashion, does this explain all health inequalities, or do inequalities remain after we take selective
processes into account? Would such a health selective mechanism convert understanding of
health inequalities? Would this lead us to the conclusion that these inequalities are natural
inequalities a society has to live with? Can health selection reduce health inequalities to the
realm of biological processes?
These are the questions, which inspired this thesis. The role of health selection in the generation
of health inequalities is investigated theoretically and exemplified with an analysis of health
inequalities and selective processes between jobs of high and normal status.

Health inequalities present a fundamental challenge to societies that are built on equality
and guaranteed human rights. Is it justified that those who are poor die several years earlier
than those who are well off? I will not answer this question, because a normative discussion
of such matters is not conducive for the purpose of the study. However, I do strongly believe
that a discussion of such questions is important. The answer to the question whether these
inequalities are justified or not, could be contingent on how these inequalities actually come
into existence. My thesis makes a small contribution to the discussion about mechanisms of
production and reproduction of health inequalities.
Health inequalities have corollaries for the live of individuals beyond the sphere of physical
health. We will see in my theoretical discussion, that impaired health will reduce chances in
the competition for important resources like educational credentials or jobs. Health is not only
a fundamental indicator for social inequalities, but it is also an unequally distributed resource
which leads to varying chances in different social spheres. This makes health inequalities a form
of structural inequality. The ramifications of health inequalities go further than those of other
forms of inequality, which are often juxtaposed to health inequalities, like social inequalities in
happiness, or life satisfaction.
Following the logic of Bourdieu (2012) health as a resource or capital (Grossman 1972) certainly
has influence on many parts of an individual’s life. One important aspect of today’s life is
particularly subject to the consequences of an individual’s health. This is the labor market
and all the rewards an individual can gain there. For example, if wages are paid according to
productivity, and sick workers are less productive, they will have smaller paychecks than healthy
workers. This is the health selection hypothesis for the labor market in a nutshell.

11
 1. Introduction

For high status jobs it can be argued that you need to show excellent performance to beat
your competitors for the position. If your health is impaired, you face significantly worse
odds in such a competition. From a population perspective, part of the health inequalities
between incumbents of high status jobs and incumbents of regular jobs might stem from
the fact that these highly desirable social positions go to those persons who are healthy and
not to those who are sick. This is an example of the health selection hypothesis and the
proposed mechanisms applied to a certain aspect of the labor market. While the argument
seems convincing, in a wide range of literature of public health research it has received scant
to no attention. If health selection was discussed it was deemed negligible (for some of the
studies which shed a critical light on health selection, see e.g. Blane 1985, Bartley 1988, Smith,
Bartley & Blane 1990, Blane, Smith & Bartley 1993, Bartley & Plewis 1997, Blane, Harding &
Rosato 1999, Chandola, Bartley, Sacker, Jenkinson & Marmot 2003, Claussen, Smits, Naess &
Davey Smith 2005, McMunn, Bartley, Hardy & Kuh 2006, Bartley, Ferrie & Montgomery 2006).
During this thesis it will become clear that the neglect of health selection in the analysis of
health inequalities is an aberration. I assert that we will find instances where persons are
selected according to their health and we will find cases where health plays only a minor role in
determining societal outcomes. What we lack is a guiding theory which tells us where to look
for health selection. I try to make a first step to close this gap. I exemplify the potential of
my approach with the analysis of the influence of health on job status on the German labor
market. The results show that under some circumstances health is a valuable asset. It is a
characteristic on which workers are sometimes, but not always, selected.

The thesis is guided by the following research questions.

A) Through what mechanisms does health selection influence health inequalities in society?
Are these mechanisms natural or social processes? My argument is that theoretical work
so far has approached the health selection hypothesis only in a superficial way. A more
detailed and explicit theory, which accounts for variation in health selection processes due
to social context can be a remedy to this situation. I use the word “theory” for simplicity’s
sake. In this context it means that I develop a set of assumptions. From these assumptions
I derive expectations about the social world, which coalesce with most of the predictions
commonly subsumed under the health selection hypothesis. I do not claim to develop a
grand theory. If anything, it is rather a middle-range theory (Merton 1968).

B) I follow up with several empirical question that put the theory to the test. These are:
1) Does subjective health influence job status?
2) Does this selective process vary with social context?
3) How do gender and competition moderate the health selection process?
The generalized hypotheses I develop in my theoretical part are:

12
 1. Introduction

1) Better health increases the chance of attaining or keeping a high status job.
2) Health influences job status stronger for women than for men.
3) The higher the competition on the labor market, the stronger the impact of health is
on job status.

1.2. Scientific Contributions

While investigating a central aspect of inequality in contemporary societies, this thesis makes
several important scientific contributions to the existing literature.
First of all, it provides the first comprehensive approach to a theory of contextualized health
selection. Such a theory allows to bridge the theoretical gap between advocates and opponents
of the health selection thesis. This includes a bridge between the research fields of public health
and health economics. Health economists often use the health selection hypothesis implicitly,
without calling it health selection. Studies from the field of public health on the other hand
often dismiss health selection outright.
Beyond the bridging function, my theory of health selection highlights the societal relevance
of health selection. It describes how it influences the social position of the individual and
how in turn health selection is influenced by social context. Further, the theory allows the
identification of key conditions that are necessary for health selection or which increase or
decrease the likelihood of health selection processes. In addition, the theory facilitates the
derivation of empirically testable hypotheses about the existence or absence of health selection
processes on the labor market and beyond.
Second, there are several methodological innovations in my thesis: The measurement of
subjective health as a latent variable based on self-reported health indicators has not been
done so far. This measurement procedure is a very close approximation of the methodological
theory of what construct should actually be measured and is therefore a judicious alternative
to existing measures. A latent variable approach has the innate feature of offering empirical
criteria, which allow me to decide whether subjective health can be plausibly compared between
social groups in my data set.
I also use a fixed-effects approach with a cross-lagged panel model, combining two strengths of
methods used in public health and health economics. This combination allows me to control
for time constant unobserved factors and to test both paths of causality in one model: Health
influencing job status (health selection) versus job status influencing health (social causation).
Further, I discuss a common approach to the decomposition of health inequalities and prove
that it is inadequate for its intention. An improved method based on path analysis is presented
and used in this thesis. Path analysis is also used to develop a method for decomposing health
inequalities into time constant and time varying parts including measures of uncertainty for
this decomposition. To the best of my knowledge, this constitutes a methodological approach,

13
 1. Introduction

which has not been developed or used so far.

Third, I contribute to the empirical literature on health selection. For Germany there are only
a few studies which deal with this problem. In the analyses I focus on job status, a labor
market outcome, which has only rarely been studied with regard to health selection. In addition,
I conduct systematic sub-analyses according to the degree of social closure of positions on
the labor market highlighting the importance of social context. And I make a systematic
differentiation between health problems that are directly visible to employers and those who
are not and compare respective results.
Developing a first theoretical approach to health selection, using and developing interdisciplinary,
advanced methods, and providing empirical evidence on health selection in understudied labor
market contexts, my thesis thus makes important contributions to the recent scientific literature
on health inequalities and labor market research.

1.3. Structure of the Thesis

The thesis is structured as follows.
The second chapter develops a theory of health selection. At the beginning, health inequali-
ties are defined and their link to health selection is discussed. A review of the theoretical work
in the scientific literature follows. The literature is summarized and deficiencies are highlighted.
In a next step, I present and de-construct the often dismissive argument that health selection
cannot contribute anything substantial in explaining health inequalities. My argument is that
health selection is not a form of natural but unnatural selection in the sense that it is social in
nature and its mechanisms are susceptible to change through social context.
I make an attempt at building a theory of health selection. Using a modified version of human
capital theory, I can show how health might become a selective factor with regard to labor
market success. The main actors in a health selection process are the employer and the
employee. Effort links health to labor market rewards. Furthermore, I introduce a distinction
between actual and perceived effort and argue that women face a disadvantage with regard to
perceived effort. This leads to a differentiation of health selection theory for men and women,
stating that health should be more important for women than for men with regard to their
labor market success. Alternatively, I suggest that men do not adjust their effort at work to
the same degree as women do. This leads to the hypothesis that health is a selective factor for
women, but not for men.
In addition to the gender dimension of the theory, I look at the context which facilitates or
hinders health selection. The main argument is derived from the implicit assumption of a
competitive market situation, which lies at the heart of human capital theory. This leads to
the hypothesis that health selection only occurs if there is competition for jobs.
While a competitive situation is one of the main assumptions of human capital theory, socio-
logical labor market research has argued for decades that there are substantial parts of the

14
 1. Introduction

labor market that do not follow market mechanisms and where job competition is low. The
most refined theory which allows me to make such distinctions is the theory of open and closed
positions. Drawing on the theory of open and closed positions, I propose that in labor market
positions with a high degree of closure health is not a selective factor. In positions with a low
degree of closure the health selection mechanisms work as described in my modified human
capital model. The integration of sociological labor market theories corroborates the impact of
social context on health selection mechanisms.
In the end of the theory chapter, I summarize my theoretical argument, list the set of assump-
tions on which it rests, and propose a series of hypotheses to test the theory.

The third chapter gives an overview of the studies which already dealt with the health
selection hypothesis. I group the studies according to various characteristics and develop
criteria for “good practice” in empirical studies. I present the studies, which fulfill the “good
practice” criteria in detail to show differences and similarities in their study design compared to
my study.

Chapter four explains my methodological approach to the identification of health selection

effects on the German labor market. Health is defined in the first part of the chapter. I propose
a latent variable approach which allows to address the question of comparability of subjective
health items between men and women on an empirical basis. The question of causality is
addressed in a separate section. I explain in what sense my thesis can provide evidence for
causal relationships. Subsequently, the statistical approaches for this thesis are presented. This
includes a brief discussion of survival analysis, fixed-effects logistic regression, and cross-lagged
panel models with fixed-effects, as used in my thesis. I also describe an improved approach
for decomposing health inequalities into different factors, including time constant and time
varying factors. Finally, I present the data set and discuss the empirical measures of the various
constructs from the theory.

In the fifth chapter, I discuss the results of my analysis. I begin with basic descriptives of
health inequalities. All analyses are done separately for men and women, and separately for
public and private sector, yielding four basic groups of analysis. The estimation of the effect
of health on job status is based on fixed-effects logistic regression. Afterwards, the effect is
allowed to vary according to the degree of occupational closure.
In the next step I use survival analysis to differentiate whether a person wants to acquire a
high status job or whether the person already occupies such a position. This exemplifies how
the individual position interacts with health selective mechanisms.
I use cross-lagged panel fixed-effects models to test for reversed causality and to assess both
health selection versus social causation in one model. The model controls for time constant
unobserved factors making a causal interpretation of the results more plausible. The last part of

15
 1. Introduction

the analysis decomposes the observed health inequalities into social causation, health selection,
and third factors. It also compares the strength of the health effect to other covariates. This
allows an assessment of the importance of health selection in generating health inequalities.
The last part of the chapter summarizes the results with a focus on the main research questions
and hypotheses from the theory.

The sixth chapter concludes. A focus will be put on a generalization of the health selection
model to non-labor market situations. In addition to suggestions for further research I call for
a serious theoretical and empirical treatment of health selection as one relevant factor which
contributes to the explanation of health inequalities.

16
2. Theory
As explained in the introduction my thesis has more than one point of focus. The most abstract
level will deal with the question whether health selection processes should be viewed as just
a nuisance in the analysis of health inequalities or if they should play a separate role as one
mechanism among others in explaining health inequalities.
Then I will address the question under which circumstances we should look for health selection
processes. The area which I will apply my theoretical discussion to, will be the labor market in
Germany, and high status jobs in particular. On this concrete third level I will investigate to
what degree health can have different effects on labor market rewards for men and women and
under different degrees of competitiveness of the labor market. This is where the empirical
analysis comes into play. Based on the example of the German labor market the conditions
for health selection will be assessed both theoretically and empirically. The fact that context
matters for health selection processes underlines that it is not a negligible or natural process.
It is unnatural selection in the sense of being directly influenced and shaped by social actors
and structures. Its relevance is to be determined empirically. Therefore it should find a place
in every discussion of health inequalities.

To be more specific my theoretical part will deal with the following points:

1. Defining conditions under which health selection mechanisms can be expected to con-
tribute to health inequalities using a classical human capital approach to explain health
selection.

2. Explaining gender differences in health effects on job status through two complementary
approaches:
a) women’s structural disadvantages on the labor market, which focuses on the role of
the employer.
b) gender differences in preferences for health and career, which focuses on the role of
the employee.

3. Using the theory of open and closed positions to explain the absence of health effects in
some parts of the labor market.

4. Modifying the theory of open and closed positions to explain different health effects for
incumbents and applicants.

17
 2. Theory

5. Deriving hypotheses from the set of theoretical assumptions that can be empirically
tested.

The chapter is organized as follows. First, I will define health and health inequalities and its
three major explanations: social causation, health selection, and spurious correlation (section
2.3). I review the literature on health selection with regard to theoretical contributions showing
that most of the literature lacks an explicit theory of health selection. I then argue that health
selection has to move from a hypothesis to a theory to fully appreciate its contribution and
assess its limitations in explaining health inequalities.
Second, I use an economic model of effort and labor market rewards - as proposed by Gary
Becker (1985) - and introduce health as an exogenous factor influencing effort (section 2.4).
I develop the concept of perceived instead of actual effort as a determinant of labor market
rewards and argue that employers’ perception of effort is biased against women. Several
complementary theories justifying the assumption of this bias are proposed (sections 2.5.1 &
2.5.2). In conclusion this leads to the hypothesis that - with regard to labor market rewards
- health is more important for women than for men. As a complementary explanation I use
gender differences in employees’ preferences for health and career which also leads to the
conclusion that health has a stronger effect for women than for men (section 2.6.1).
Third, I will point out that there might be differences in the effect of health problems with
regard to the visibility to the employer. I discuss how this matters for health selection and how
it can be tested empirically (section 2.5.4). Fourth, I draw on the notion of open and closed
positions (section 2.7). Following the argument that performance and rewards are not related
to each other in closed positions I conclude that health effects on labor market rewards should
only be found in open positions (section 2.8).
However, fifth, I argue that a modification of the theory of open and closed positions is
necessary. Health effects in open and closed positions are different for incumbents than for
applicants (section 2.8.1.2). The role of disadvantaged groups in modifying the degree of
closure of a position is discussed with implications for gender differences.
At the end of the chapter I sum up the argument, list explicitly all assumptions which my
theory relies on, and derive a number of hypotheses which I will test in my empirical part.

2.1. Health Selection and Health Inequalities

When discussing health selection and its contribution to health inequalities it is helpful to
first define health inequalities. I will discuss the concept of health inequalities and how the
contribution of health selection to health inequalities is seen in the literature. Then, I will make
the case for a better standing of health selection within the scientific debate of health inequalities.

18
 2. Theory

2.1.1. Defining Health Inequalities

The term health inequalities is usually used if social inequalities are related to health or illness
(Richter & Hurrelmann 2009, 11). This means that not all differences in health are called
inequalities, but only if they relate to socio-economic inequalities and are therefore in a sense
avoidable (Bommier & Stecklov 2002, 502).
Generally speaking, health inequalities can be explained by three major mechanisms. First,
there is social causation. This view claims that social circumstances influence health. Second,
there is health selection, which claims that health influences the social position of an individual.
Third, there are common background factors that might lead to spurious correlation. In this
case it is assumed that some (possibly unknown or unmeasured) factors influence both the
social position and health (Goldman 2001, 10068), but that they are in fact not causally related.

2.1.2. Defining Health

Before I go into detail on theoretical models explaining health inequalities I want to state the
definition of health which underlies all my theoretical arguments. My definition of health needs
to satisfy three conditions.

1. It needs to make a link to performance on the labor market plausible.

2. It must be measured in a large scale longitudinal survey that also has detailed information
about the household context and job characteristics

3. It must at least partly be comparable to definitions of health in previous research to allow

an evaluation of the results within the wider context of the literature.

s
The well-known World Health Organization definition (WHO 1946) of health is

“Health is a state of complete physical, mental and social well-being and not merely
the absence of disease or infirmity.”

I will stick to this definition which is sometimes labeled the social model of health in contrast
to the medical model of health (Townsend & Davidson 1982, 42). As this definition has a
holistic approach to health, I choose to let every person evaluate for themselves how good
or bad their health is. This has the disadvantage that sometimes the same health conditions
might lead to different responses on a subjective health item. However, the advantage is that
this approach covers all aspects of health. Most of the theory developed in this chapter is not
specific to certain diseases or limitations. Theories specializing on certain health conditions are
very useful, but are beyond the more general approach, which I adopt. For now the purpose of
the thesis is best served by this broad definition.

19
 2. Theory

2.1.3. Social Causation

I now turn to the two hypotheses which postulate a causal link between social conditions and
health. I start with social causation. This section will offer only a very brief introduction into
the logic of social causation in general. Some theoretical ideas on how job status in particular
and health might be linked by the logic of social causation are stated as well. I do not claim
that the discussion presented here is exhaustive. Far from it, it is a reminder of what is the
dominant way of explaining health inequalities in the literature. The main focus of the thesis
is on a discussion of the health selection hypothesis within social context. It would go way
beyond the scope of the thesis to give a full account of health selection and social causation
for the matter at hand. The relatively short section on social causation does not mean however
that it is less important or theoretically less developed than the health selection theory. The
opposite is true. I focus on health selection, because theoretical and empirical work on this
hypothesis lags behind in comparison to the social causation approach. The relative neglect of
social causation is the price I have to pay for the extended discussion of health selection. It
should be noted, however, that although the empirical analyses also focus on health selection
they include tests of both causal pathways.

Usually health inequalities are explained by disadvantages certain groups suffer in relation
to other groups. The theory states that these disadvantages (directly or indirectly) impair
the health of members of that group. Therefore general health status is worse in the disad-
vantaged group. The mechanisms might vary by context and application, the logic is always
the same. Ever since the Black Report (Townsend & Davidson 1982) this logic has domi-
nated health inequalities research, at least in the field of medical sociology and epidemiology
(Smith 1999, Goldman 2001). I will refer to this argument in general as social causation.

The mechanisms of how social group membership can influence health are very diverse. I
mention the most important ones (compare e.g. Richter & Hurrelmann 2009).
First, there is material deprivation. Some persons cannot get enough (healthy) nutrition, clean
water or have no access to health care and medical treatment, because they cannot afford
these things. This will have negative health consequences and create health inequalities in
comparison to persons who can afford it. This kind of health inequality is mostly associated
with disparities in household income (Marmot 2005, 1101).
Second, there are environmental factors (Borg & Kristensen 2000). Air pollution, environmen-
tal hazards, occupational injuries, physical or psychological strain at work or at home might
severely endanger a person’s health. Some economic theorists argue that such dangers should
be compensated by increased wages or reduced rents (Cousineau, Lacroix & Girard 1992, Leeth
& Ruser 2003, Viscusi & Moore 1987). Nevertheless, there will be health disparities between
those exposed to such risks and those who are not. An important factor is the interaction with

20
 2. Theory

health knowledge, because the risk of exposure might not always be adequately assessed by
the individual. As these risks are not randomly distributed within society, some groups are
more exposed to such risks than others (Hoven & Siegrist 2013, Wahrendorf, Blane, Bartley,
Dragano & Siegrist 2013).
Third, health behavior strongly influences health and can create health inequalities. Some
social groups smoke more often, visit general practioners less often, eat rather unhealthy, etc.
Preventive health measures can lead to lower prevalences of (serious) illnesses over the life
course. Healthy diet can increase life expectancy or reduce chance of getting diseases like
diabetes or strokes. Exercise and physical activity also contribute to health inequalities between
different social groups. These difference in health behavior are again strongly associated with
income and education (Cockerham 2005, Kenkel 1991, Tunistra, Groothoff, van den Heuvel &
Post 1998, Wickrama, Conger, Wallace & Elder 1999, Richter, Moor & Lenthe 2012).
Fourth, there are social-psychological mechanisms like relative deprivation, inequality, or strain-
ing life events which can have a negative impact on (mental) health. The argument is that
mental but also physical health can be influenced by psychological processes. Feelings of inferi-
ority, lack of control, loss or isolation play an important role in explaining prevalence of various
mental illnesses or coping behavior with existing other diseases. The psycho-social strain depends
strongly on the social position, social networks, work, and family of the individual, generating
structural, not random differences in health outcomes (Siegrist & Marmot 2004, Schneider-
man, Ironson & Siegel 2005, Schwartz, Friedman, Tucker, Tomlinson-Keasey, Wingard &
Criqui 1995, Levenstein, Smith & Kaplan 2001, Lynch, Kaplan & Salonen 1997, Lynch, Smith,
Kaplan & House 2000).

For the example of job status and health there are also some arguments that can claim a
causal connection in the sense of social causation. Health inequalities between high and low
job status can be linked to these approaches in several ways. There might be indirect links,
or spurious correlations, because persons in high status jobs are more often well educated,
and might share certain personal characteristics which are beneficial to occupational success
and health. In addition, there is a resource advantage of incumbents of high status jobs with
regard to e.g. wages which can lead to health inequalities. Non-pecuniary rewards might also
be relatively higher, and working conditions could be less noisy, dirty, or physical straining as
soon as supervision or highly specialized tasks are assumed in high status.
There are actually only a few empirical studies on this exact issue. However, a phenomenon
related to job status is promotion. For promotions, Anderson & Marmot (2012) demonstrate
that a promotion among English white collar workers is followed by a substantial reduction of
risk of heart disease. Chandola et al. (2003) can show that an increase in employment grade
among UK white collar workers reduces risk of negative health conditions. This shows that job
status can clearly be interpreted as a social factor which influences health through possibly
material, but mostly environmental-psychological factors. Although social causation is not

21
 2. Theory

the main focus of this thesis, these results and ideas clearly require that social causation is
accounted for in any analyses of health inequalities. Therefore, the empirical part will allow the
existing health inequalities to be explained by the classical social causation approach as well.
This allows contrasting the social causation approach with the reexamination of the health
selection hypothesis.

2.1.4. Health Selection

The second way to explain health inequalities I call health selection1 (see e.g. Warren 2009).
Persons in good health can replace persons with less good health (and related characteristics)
in their social positions. This allows for social upward mobility of healthy persons and social
downward mobility of persons with impaired health. Aggregated this can lead to a higher level
of persons with impaired health in groups with lower social status (Bartley & Plewis 1997,
376). The process of health selection is iterative and cumulative over the life course, although
there are certainly critical junctures in life that are of particular importance. Health selection
can also work indirectly through risky or unhealthy attitudes, behavior and lifestyles (Virtanen,
Vahtera, Kivimäki, Pentti & Ferrie 2002, 697). In this thesis I consider only direct selection.

In spite of the fact that there is a good theoretical foundation to support the health
selection thesis, different health researchers deem the actual contribution of health selec-
tion processes to overall health inequalities within societies as rather small (e.g Richter &
Hurrelmann 2009, Townsend & Davidson 1982, Warren 2009). Actually a lot of introductory
or overview readings on health inequalities hardly mention health selection. If they mention
health selection it is usually stated that research has shown that it is negligible at least in
comparison to social causation (Smith 2003, Graham 2009, Dowler & Spencer 2007). The
editors of the collection of Richter & Hurrelmann (2009) state with due scientific caution that
they do not want to claim that health selection does not exist, but that their current volume is
focused on social causation. From their perspective this is legitimate, because research has
shown that health selection can make only a minor contribution to the explanation of health
inequalities. Two articles in the collection discuss the health selection thesis, but still rate it
as clearly inferior in explanatory power than the social causation approach (Jungbauer-Gans,
Gross, Richter & Hurrelmann 2009, Dragano & Siegrist 2009).
In the collection of Bauer, Bittlingmayer & Richter (2008) one article explicitly deals with
the health selection thesis for the case of inequality in health between married and unmarried
persons. Unger (2008) claims in the article that for Germany the health gap between married
and unmarried persons can almost entirely be explained by better prior health status, strongly
supporting the health selection hypothesis.

1
In other contexts this is referred to as drift hypothesis (Goldberg & Morrison 1963) or natural and
social selection (Townsend & Davidson 1982). The same concept stands behind these different terms.

22
 2. Theory

The collection of Wendt & Wolf (2006) does not discuss the health selection approach as one
of the possible approaches in the sociology of health. Blane, Smith & Bartley (1993) point to
the idea that indirect selection, which means that health and social class are affected at the
same time by third group variables such as education, is more promising in explaining health
inequalities. It is interesting that they still refer to it as indirect selection and not simply as
spurious correlation. A lot of studies claiming that health selection has been shown to be
of lesser importance interstingly point to studies by British scholars around the trio of Mel
Bartley, David Blane, and Davey Smith (for some of their work which sheds a critical light
on health selection, see e.g. Blane 1985, Bartley 1988, Smith, Bartley & Blane 1990, Blane,
Smith & Bartley 1993, Bartley & Plewis 1997, Blane, Harding & Rosato 1999, Chandola
et al. 2003, Claussen et al. 2005, McMunn et al. 2006, Bartley, Ferrie & Montgomery 2006).
While the quality of their work is beyond question, and their scientific renown well deserved it
seems that a more differentiated view and approach to the subject is possible.
More than twenty years ago, West (1991) already criticized this stark opposition to the health
selection thesis and claimed that both theoretical and methodological inaccuracy and misun-
derstandings of the selection hypothesis might have biased such results.

Another issue might be that health sociologists, epidemiologists, medical sociologists, and
scholars of public health are concerned that the attention should be focused on processes of
social causation, because these processes imply greater challenges for social and health policy.
This approach seems to be implicitly taken by a lot of scholars as a citation from a study for
the UK EU-presidency shows:

”The unspoken assumption in debates about the role of selection versus causation
often was, that social selection is less of a problem for public policy than
social causation. This assumption was incorrect, however, because limiting
the social consequences of health problems is one of the classical objectives
of social security and public health policies in many European countries.”
(Mackenbach 2006, 31)

The author of the study still maintains, however, that health selection contributes far less to
the explanation of health inequalities than social causation.

2.2. Who Does all the Selection? In Search for Theory in

the Literature

2.2.1. What Should We Look for?

So far we have seen that health selection as a mechanism for generating health inequalities is
doubted to carry much importance. In this part, I will review the theoretical work which has

23
 2. Theory

been done so far on the health selection hypothesis. There has been no systematic attempt to
address the issue of moving from a hypothesis to a theory. Unsurprisingly, we will see that the
theoretical work is very fragmented. Some studies contribute some substantial insights that will
be picked up in the theory development in this thesis. Other studies make minor theoretical
points which are also interesting. Most studies however, have a complete lack of theory. They
usually just state two things:

1. There is a health selection hypothesis that proposes reversed causality for the common
explanation of health inequalities.

2. Health inequalities arise, because individuals in poor health are selected into unfavorable
social positions.

They make no mention of who selects the selected and why: It is also unclear according
to what criteria they are chosen and under which circumstances this happens. The reason is
surely, that those studies do not aim at theory building. They want to make a contribution to
the empirical literature on health inequalities. As empirical analyses have trumped theoretical
discussion so far, I think that a more systematic approach to theory building in this thesis is
warranted.
This section is not meant to systematically build a theory on already developed theoretical
approaches in the literature, because the theoretical ideas are too fragmented for such an
approach. While reading the studies I was looking for an answer to the following questions:

• Which actors are involved in the selection process? Especially, who selects and why?

• Under what circumstances or in what context does health selection occur?

• Is health selection a social process?

• What are the mechanisms linking health status to unfavorable social position?

2.2.2. Studies with Theoretical Contributions

McLeod & Pavalko (2008, 77) define health selection, as it is commonly used in the literature:

“The most common conceptualization of selection effects is as the influence of

physical and mental health on the statuses and attainments of individuals, or what
we refer to as reverse causation.”

Some studies just very briefly make statements about the relation between social position or
labor market outcomes and health. Examples are: Haan & Myck (2009, 1116) simply state
that poor health reduces productivity. Haas (2006, 340) hypothesizes that a drift downward in
SES through poor health works via lower labor market participation and reduced wages. Cai &

24
 2. Theory

Kalb (2006, 242) state that poor health reduces productivity and employment chances given
the same wage.

Naming actors involved in the process is key to understanding the mechanisms behind health
selection and attempting to formulate a theory of health selection. This approach allows
to model individuals with capacity for self management within certain constraints, and to
make predictions about their behavior and outcomes given certain circumstances. In a slightly
different context, Pavalko, Gong & Long (2007, 354) write:

“If we view individuals as actively managing and negotiating role demands within
structural constraints, these selection processes become critical for understanding
the relationships among work, family, and health.”

Two studies which deal with health selection and permanent versus temporary jobs shed a
little light on theoretical mechanisms lying behind the health selection hypotheses.
Wagenaar, Kompier, Houtman, van den Bossche & Taris (2012, 1192) specify which actors
are involved in the selection process. When making decisions about downsizing employers
dismiss those with most health problems. In addition, they state that healthy individuals are
more often hired into permanent employment, enjoying all its benefits (e.g. job security, higher
wages). So, one relevant actor is the employer.
A differentiation within the labor market between core and periphery is made in the study of
Virtanen, Janlert & Hammarström (2012). They state that finding a permanent job is hindered
by poor health. Poor health also increases the chances of being dismissed during periods of
downsizing. Most importantly, the authors acknowledge that legislation, rules and labor market
structure play a mediating role in the effect of health on permanent job attainment. The
mechanisms of health selection will vary with context.
This a generalization of the argument made decades ago by Perrott & Sydenstricker (1935).
They argue that increased competition is associated with a reduced likelihood to be successful
on the labor market when health condition is poor. For unemployment the argument is picked
up by Bartley (1988) suggesting that health selection into unemployment is more important in
times of high overall unemployment. This important insight will be picked up later where it is
framed within a broader sociological theory (see section 2.7).
Paul & Moser (2009, 268) point out that mental health is likely to influence the job search
in two ways. First, psychological problems might lead to less effort and efficiency in the job
search process. Second, distress caused by poor state of mental health on part of the job seeker
might negatively influence the employer’s decision to hire the applicant.
McLeod & Pavalko (2008, 80) agree with other studies that job search is harder if health is
impaired. Deterioration of health when already on the job might lead to reduced productivity,
implicitly leading to dismissal, or reduction in compensation or status.

25
 2. Theory

Mastekaasa (1996, 191) uses an economic approach to explain health selection. He presents
the most systematic discussion of health selection mechanisms. My theoretical argument will
resemble his in several regards, as we will see in the next chapter. Mastekaasa (1996, 191)
names the employer as the main actor, making decisions about dismissing employees. These
decisions are made within certain legal and social constraints.

”However, both legal and social factors limit employers’ freedom of action. [...]
[...],only two actors are generally involved, i.e. a job applicant and an employer.”
(Mastekaasa 1996, 192)

The second important actor is therefore identified as the employee. The author assumes a
rational profit maximizing behavior of employers resulting in a lay off of the least productive
workers (Mastekaasa 1996, 191). He states that both sickness absence and reduced effort-
per-hour are results of impaired mental and physical health. Therefore health becomes an
important factor in the decision which employees to keep and which to dismiss. If precedence
is given to seniority over productivity when choosing which employees to keep, mental and
physical health should play only a minor role for the employer. One example the author give
when this might be the case is if unions have a stronger influence on the process of downsizing
(Mastekaasa 1996, 192). Overall the information available about the employees’ productivity
are key for health selection processes.

“Generally, the union policy is to give as much weight to the seniority criterion
as possible, and to resist attempts by the employer to use productivity criteria. To
the extent that such union policies prevail, mental or physical health will have no
impact on the probability of job loss.” (Mastekaasa 1996, 192)

Less healthy employees will have a harder time convincing employers that they are productive.
In addition, they will most often have more days of sickness absence, which are objectively
measurable and visible to the employer (Mastekaasa 1996, 192).

Cardano, Costa & Demaria (2004, 1564) argue strongly that health selection processes
are embedded in social context and are of social not biological nature. Selection processes
are closely related to certain forms of discrimination or social closure. The conscious and
unconscious decisions of real actors under certain constraints should be taken into account. I
interpret this as a call for a rigorous sociological analysis of health selection. It is worth citing
a longer passage from their study. Together with the theoretical work of Mastekaasa (1996),
their theoretical idea of a truly unnatural selection process is the most important building block
in the struggle to fully integrate health selection and social causation as social explanations of
health inequalities for future research:

“The selection hypothesis has usually been interpreted - not always without
justification - as embodying an ideological attempt to deny the ethical and political

26
 2. Theory

importance of social inequalities in health [...]. For if it is individuals’ health which

determines their social position, differences in the health status of the various
classes cannot be seen as unjust, or rather cannot be seen as socially unjust, for
the root cause is a cynical “lottery of nature”, which gives some people sound
and others poor health. This has led to the idea that the hypothesis which sees
flows of social mobility as being affected by health is a form of “social Darwinism”
(West, 1991), which portrays inequalities in health as part of a biological selection
of the fittest for the posts of higher social responsibility.” (Cardano, Costa &
Demaria 2004, 1564)

The most important and most cited discussion of the health selection theory comes from
West (1991). He analyzes the discussion on health selection in public health, epidemiology, and
medical sociology since the publication of the “Black-Report” (Townsend & Davidson 1982).
He states that health selection is portrayed as an asocial theory which “[...] appears merely
to vindicate capitalist values” (West 1991, 373). He points out that health selection could
be seen as a form of discrimination. In his view defining the actual mechanisms and actors
involved in such a process of health related selection appears to be a difficult, but important
task. In his words health selection can be conceptualized:

“not as inevitable outcomes of the properties of individuals, but as the result of

the focused attention of social agencies and institutions engaged in the business of
controlling entry into and movement within occupational opportunity structures.”
(West 1991, 374).

West (1991, 374) also speaks against the attribute of natural with respect to health selection.
He proposes health-related mobility as a more useful term. West (1991, 379) identifies one
important argument of those scholars trying to dismiss health selection in the analysis of health
inequalities. This is the argument that health inequalities between classes are by far too great
that health selection could play a major role in explaining them.
In my view, making a general statement that health selection cannot explain the great class
differences encountered in health and mortality makes as much sense as stating that social-
psychological factors on their own (as part of the social causation approach) are not enough
to explain health inequalities completely and should therefore not be considered. Or stating
that health behavioral factors might not be enough. Of course, no one approach is enough
to explain health inequalities completely. It is a very complex construct influenced by many
factors. Therefore, a comprehensive approach including all feasible explanation should be
used. And of course the context needs to be considered. It is very unlikely that all forms
of health inequalities related to any kind of measure of social inequality for all social groups
is determined by the same causal factors. The literature has shown abundantly in the last
decades that under different circumstances for different groups of individuals different parts
of the social causation approach have varying degrees of explanatory power. The same will

27
 2. Theory

be true for health selection. In some circumstances it will not play a role, in other cases it
might be a major factor, or one factor among many. And what if it can “only” explain 10%
of health inequalities in some cases. Is that not worth mentioning as well? It should also
be noted that health selection might decrease health inequalities in some cases. Thus, not
taking health selection into account might lead to an understatement of health inequalities
generated by processes of social causation. So we can see that this argument is rightly refuted
by West (1991). His exact words after reporting the argument from the literature are: “[...],
as if this was enough in itself to win the argument” (West 1991, 379). This marks the point
that the critics of health selection tried to win an argument where there should be nothing to win.

West (1991, 380) fiercely argues in favor of contextualizing health selection and seeing it
as a social process, same as Cardano, Costa & Demaria (2004) do in their study. He states
that social agents, bound by social constraints and norms, interpret health related attributes
of other individuals and give them a meaning that can lead to a selection of the ill from the
healthy. Health selection does not adhere to a simple cause-effect logic, but is necessarily
mediated by social processes. If values or norms were to change and different meaning would
be attributed to health related characteristics, health selection might work differently, or might
not be an issue at all.

“In a fundamental sense, health selection–direct and indirect–does not occur in

a social vacuum; it is the outcome of an interaction between more or less valued
attributes of individuals and the opportunity structures and the institutions and
social agencies which control access to and processes within them” (West 1991,
380)

In my view it seems a little paradox that critics of the health selection hypothesis would argue
it is Social Darwinist, asocial or perpetuating capitalist values. At the same time, they do
not seem to realize that health selection only becomes natural if one accepts these ways of
thinking, of making arguments, and of legitimizing inequalities as the natural order of human
life. Only those who accept those capitalist values in an ahistoric fashion as natural laws could
see health selection as natural selection. These values and norms are subject to change and
so are the results of health related selection processes. This makes health inequalities caused
by health selection a social phenomenon changeable and accessible to changes in culture or
policies. For empirical research this informs us that there will be differences in the degree of
health selection between and within societies (West 1991, 383).

Summing up, we can say that only very few studies name the main actors of health selection,
which are employers and employees. Selection is sometimes done by the employer and sometimes
as self-selection by the employee. Only one study makes this and the selection criteria in the
process explicit. A few studies give examples of how context might influence selection, but a

28
 2. Theory

systematic account is not given.

This underlines the importance of my intent to go into the direction of a systematic theory,
which allows to make predictions about health selection under different circumstances and for
different groups of actors. This task will be addressed in the next sections.

2.3. Theory of Health Selection

What theoretical foundation is there to support the health selection hypothesis? So far there
is actually very little theory explaining the mechanisms behind health selection. The study
with the best theoretical foundation and a very important exception to that rule is the study
by Mastekaasa (1996) which is presented in section 2.2. It seems that often the theory is
considered to be too obvious to explain. So, only the hypothesis is stated and then the analyses
begin. For my purpose this is not sufficient. Without a clear theoretical formulation, testing
health selection effects seems to be a merely exploratory endeavor. Evaluating the results of
my study would also be harder, because no expectations could be formulated. One additional
thing that cannot be done without a theory is making predictions as to where health selection
effects should not appear. Yet, such a prediction is as important as the conventional health
selection hypothesis. If the hypothesis is not supported by the data it is unclear what kind of
conclusions can be drawn for different contexts, for different indicators of health inequalities, or
for different measures of social positions. It is therefore key to make an attempt at developing
a theory of health selection. Formulating such an explicit theory of health selection will be
one of the contributions of this thesis. It allows me to investigate the conditions under which
the health selection hypothesis is valid and where its limitations lie. This will help guiding
research on health inequalities in the future. It also shows that health inequalities due to health
selection depend on a wide range of social conditions and can be modified by such conditions.
It will thus become clear that health selection is a form of Unnatural Selection.
Where do we start in the quest for a theory of health selection? Usually, it is best to start with
the simplest theory available. We can evaluate if it can explain the phenomenon under scrutiny
and afterwards adjust or rephrase the theory using additional or less restrictive assumptions.
This is how I will proceed here. The easiest way to link health and labor market rewards is
by human capital theory. I will elaborate the human capital model, its assumptions and its
implications for the study of health selection. Then, I will evaluate whether the theory is
sufficient for my purposes or not.

29
 2. Theory

2.4. Health Selection - from Hypothesis to Theory

2.4.1. Human Capital Theory

One of the most well known theories in labor economics is human capital theory. Human
capital theory is an extension of neoclassical theories of the labor market. It knows two actors:
employers demanding labor and workers supplying it.

“Neoclassical labor economics consists, in brief, of the marginal productivity

theory of demand-based on profit maximizing behavior of employers-and a supply
theory based on utility maximization of workers.” (Cain 1976, 1216)

For the employers it is important to note that they are assumed to be profit maximizing
rational, unitary actors. In equilibrium of the labor market employers will pay workers an hourly
wage equal to the marginal profit they get from the last hour of work the worker does. If for
some reason marginal productivity is lower than hourly wage, the employer either reduces the
wage or reduces the hours of employment for the worker until marginal productivity and wages
are equal again. One can see that both work hours and wages are seen as completely flexible
in this model.
Looking at the worker we are also assuming utility maximizing rational actors which draw their
utility from the wages they earn at work and their leisure time. Workers will offer so many
hours of work on the labor market until the utility they gain from the income of their last hour
of work is less than the utility they would gain from the same hour in leisure. When this is
the case, is determined by their preferences for leisure and income. Workers are therefore only
willing to work for a minimum of wage per hour which is called the reservation wage. So far
this is all in line with traditional neoclassical labor market theory. Human capital theory makes
one important addition to the model.
Workers are not assumed to be uniformly productive. Instead their productivity depends, as
the name suggests, on their human capital. This consists of e.g. education, training, and
experience at work. Human capital is seen as an investment in analogy to physical capital. This
investment can be made either by the worker or the employer. For her investment in human
capital the worker usually forgoes income for a period of time and might have additional costs
like tuition fees. These costs need to be compensated in form of higher wages. The employer
is willing to cover these additional labor costs due to the higher productivity, which in turn
results from education, training or experience. Put differently, workers will only invest in their
human capital if they expect the returns to this investment in form of wages to be equal or
higher than the costs of the investment.
Human capital can be further distinguished into general human capital, occupational specific
human capital (Kambourov & Manovskii 2009), and firm specific human capital (Hashimoto
1981, Becker 1994, Lazear 2003). The first increases productivity of the worker regardless of

30
 2. Theory

whom he works for. The second type increases productivity of the worker in all firms and for
all employers as long as the worker is working in the occupation or profession he was trained
in. Last, firm specific human capital only increases the productivity of workers as long as they
work for the firm they were trained in.
Workers are expected to be less interested in investing in firm specific human capital, because
they will lose their investment as soon as they change the employer (for a better paying job or
if they are fired, for example). On the other hand employers have little interest in financing
general human capital, because the worker can easily transfer the human capital when quitting
the firm, as it is as useful to him at any other firm. In this case, the employer can easily lose
the investment made.
Human capital theory argues that there can be short-term imperfections so that some workers
earn more or than their stock of human capital would predict. Market forces will ensure however
that all workers have the same income2 - given same preferences for leisure and income - over
their life course. Higher educated persons get their human capital investments back through
higher wages and persons who invested less had a longer period of time earning wages or had
less costs.

2.4.2. Human Capital Theory and Health

Now we have reviewed the basic idea of human capital theory. There are uncounted numbers of
extensions and modifications. For my purpose the basics are enough. What has not been said
is how health fits into this model. So far only human capital has been mentioned. Fortunately,
Gary Becker himself developed a model which incorporates the effort a person spends on his
work as one additional determinant of wages. Effort will be the important link between an
individual’s health and her labor market rewards. How this is accomplished will be shown in
this section. In the end, we will have a model which predicts the effect of health on labor
market rewards. The model also states the basic conditions under which health selection is to
be expected. It is also the starting point to look for gender differences in health effects on the
labor market.

Becker (1994) sees health as a human capital investment. It can increase productivity
in the same way as training or education can (Becker 1994, Mushkin 1962). The analogy
between health and education with regard to productivity has been termed health capital theory
(Stern 1983, 42). Health capital increases the time a person can spend working and earning
wages and commodities (Grossman 1972).

2
Though not necessarily the same accumulated wages. They will differ if costs other than lost time to
work are needed to gain more human capital. Tuition fees should make such a difference. Paying a
student loan will need higher overall wages to compensate the investment, but will lead to the same
total net income over the life course.

31
 2. Theory

One of the most well-known phenomena connecting health selection and labor market processes
is the so called healthy-worker-effect (McMichael, Spirtas & Kupper 1974). The healthy-
worker-effect hypothesis states that in order to participate in the working force an individual
has to have a minimum level of good health. Those who do not reach this level cannot
participate in the labor force. The sick therefore become, stay, or already are unemployed or
not-employed. On aggregate this leads to the empirical observation that those in the labor
force are more healthy than those who are not part of the labor force. The magnitude of the
healthy-worker-effect can vary among different sub-populations of the working force (Li &
Sung 1999). There is also some evidence for health selection between occupations. Those who
are worse off with their health choose other jobs which are less endangering to their health (Li
& Sung 1999, Ostlin 1988).

After these introductory examples, I will now go into detail about how health affects a
person’s productivity, performance, and labor market rewards.
Health constitutes an important resource for a person. Effort spend on the job costs energy and
creates the need for recovery. Several studies show that there is an inverse relationship between
health, exhaustion and need for recovery (Sluiter, van der Beek & Frings-Dresen 1999, Sluiter,
Frings-Dresen, van der Beek & Meijman 2001, Sluiter 2003). Other studies show a strong
association between subjective health and sickness absence (Ferrie, Kivimäki, Head, Shipley,
Vahtera & Marmot 2005, Eriksson, von Celsing, Wahlström, Janson, Zander & Wallman 2008,
Roelen, Koopmans & Groothoff 2010).
I will argue that the state of health of workers is an important factor for their labor market
outcomes. Later, I will develop the theory further allowing for differences in the health effect
between men and women.
Most theories deal with wage effects of certain health characteristics of workers. I will just refer
to rewards in general, because my empirical analyses refer to job positions. It is also the most
general assumption that fulfilling certain tasks within a job are rewarded by certain benefits,
whatever their nature (Sørensen 1983, 205). It is thereby a more general approach and allows
for an easier adaption to different labor market contexts.
Human capital theory argues that labor market rewards increase with work hours, human
capital, and the effort spent per hour of work (Becker 1985, S44).
The effort persons can spend during their hours at work is limited, because their overall
time budget is limited (Galama & van Kippersluis 2010, 10). The effort at work depends on
the overall amount of energy available, the effort needed for other activities outside the labor
market, and the ability to regenerate energy. All three factors can be directly or indirectly
influenced by a person’s health status.
The overall amount of energy depends on a person’s physical and mental resources to deal with
problems. If a person is taken ill, or has a long-standing physical impairment, a part of the
energy a person has is spent on fighting the illness, coping with the impairment, and recovery

32
 2. Theory

in general. This leaves less energy for all other activities, including paid work.
Illnesses and health impairments often demand a lot of time and attention by the stricken
person. Doctor visits, hospital stays, therapy sessions or longer time needed for day-to-day
activities are just some examples of how impaired health may lead to increased off-labor-market
efforts that are (in)directly related to health problems.
Sleep, leisure time, and relaxation are important for recovering energy spend on different
activities on and off the labor market. If a person does not have enough time or support to
recover the energy needed, the person either has to spend overall less energy or has to “borrow”
energy (Becker 1977, 30).
Energy can be borrowed from time to come, so that one has to deal with reduced energy
in the future. This may be both in the close, e.g. fatigue at weekend (Demerouti, Blanc,
Bakker, Schaufeli & Hox 2009), or in the distant future, e.g. burnout, or chronic health
problems (Paringer 1983). For example, within the German context Schnitzlein (2011) finds
that taking less holidays impairs subsequent subjective health. For another empirical study see
e.g. de Croon, Sluiter & Frings-Dresen (2003).
If a period of illness occurs, persons therefore have to decide whether to use energy now at
the expense of less energy in the future or to reduce effort at work now. Both options carry a
risk. The problem is that usually not enough information exists to make a completely informed
decision. Due to this lack of information I argue that persons rely on what they know or think
about work, recovery, and health. This knowledge is usually based on experiences at work,
on how absenteeism is received there, and on knowledge about regaining strength, long-term
fatigue, illnesses, etc. The decision will therefore be influenced by health related behavior and
values learned and adopted through socialization in childhood, school, and at work. It is thus
feasible to expect that the possibility of borrowing energy is used to different extents by men
and women. Why this is the case and why this might be important for the relationship between
health and LMR is a separate question which I will address in section 2.6.1.
I add a second way of borrowing energy. Energy can also be borrowed from other persons in
which case they have to take over part of the work a person has to do. In case of illnesses or
reduced overall health the demand for recovery is usually greater than in times of good health.
In fact, increased sickness absence is one way to deal with a high demand for recovery. Another
way would be to externalize non-labor-market-activities to other persons. A partner could take
over housework, childcare, or other chores which would allow for more time to recover. In this
way one would indirectly borrow energy from one’s partner.

The effect of less effort can be seen in either a reduction of work hours or a reduction of
effort per hour. Especially the second option is often the case if health impairments are not too
severe or if economic constraints do not allow a reduction of work hours (through sick days).
Sometimes this reduction might be even unwittingly if rather latent factors like mental health
problems, lack of sleep or related phenomena are the cause of the health situation. In any case

33
 2. Theory

the worker will get less work done in the same amount of time or the quality of the work is
reduced. Overall productivity can therefore be severely reduced even if work hours are not
reduced. This should lead to a reduction in labor market rewards, especially if the health state
is impaired for a longer period of time (for a similar argumentation, see Grossman 1976)3 . This
is the basic model of health, effort, and labor market rewards. Becker (1985, S44) provides a
formalization of his general model that allows to link effort to income. The general is equation
is:

I = αeσ t, 0 < σ < 1 (2.1)

I stands for income, α represents human capital of the individual, e is effort per hour, t are
the hours of work on the labor market, and σ stands for the effort intensity of the job. The
effort intensity modifies the relation between income and effort. This means that the return
to effort depends on how effort intensive a job is. This interesting proposition is discussed
in detail in section 2.4.3. For my purposes the model can be generalized to consider all
forms of labor market rewards instead of income. In addition, I will develop the hypothesis
that the effort-per-hour a person can show at work is a function of his health (H) and the
non-labor-market effort (EN LM ) required of him. Therefore we can make a small modification
to end up with this equation:

LM R = αe(H, EN LM )σ t (2.2)

I will pick up this formalization later in the theory building (section 2.4.3 and 2.6.2) as a way
to systematize the propositions. It will also be picked up in the methods section to show how
the different aspects of the theory are measured and influence the regression equation (see 4.9.2).

At this point, we can explain why health inequalities due to health selection should appear
on the labor market. Human capital theory states that healthier individuals are more productive
and therefore receive higher rewards or better positions, which in turn results in labor market
related health inequalities. The simplicity of the theory is a great advantage. However, so far
the theory is gender-neutral. It makes no distinctions between men and women. But is this
a reasonable simplification of reality? In the next section, I will argue that it is indeed too
simplifying for the research purpose at hand. I will first draw on theories of social inequalities
and gender in general, which will raise concerns about the neglect of gender in human capital

3
Of course, this argument can also be turned around. Workers earn more per hour, because they are
healthier and can spent more effort on work (Becker 1985, Galama & van Kippersluis 2010).

34
 2. Theory

theory. I will then go into more detail explaining how gender matters for health selection and
how the theory of health selection must be altered to be plausible.

2.4.3. Decreasing Returns to Effort

Drawing on human capital theory one special assumption should be elaborated a bit further.
This is the notion of decreasing returns to each unit of effort at work. For this purpose, it is
best to start with an analogy from sports and then with an example from a hypothetical job.
If we have walked for 1 km it will take us only a little effort to go another kilometer. But if we
take part in the very popular Vierdaagse Nijmegen walking tour in the Netherlands we have to
walk 4 times 50 kilometers to be among the finishers. At the end of each day we have walked
50 km. Going another km in the evening will take considerably more effort than after the first
short 1 km walk. So, walking 1 km can take very different amounts of effort depending on how
much we have already walked. Now we can turn the argument around: For each unit of effort
we can walk a longer distance if we have walked only 1 km than if we have already walked
50 km. This means that each additional unit of effort will get us just a little bit less far than
the unit before. If we transfer this notion to effort spent at work we can come up with the
following example:
A person standing at the assembly line putting together parts for a TV can increase her output
of pieces put together by working faster. Working faster costs more effort, because the person
has to move more and faster, and she needs to concentrate more to make no mistakes. If the
person is basically just standing next to the assembly line and only occasionally picks up two
parts to put them together, she might increase her output considerably by just working a little
faster and spending just a little more effort. If she is already working really fast and taking few
breaks, then the same increase in effort will increase the output only a little. At some point
additional effort will hardly be noticeably, so that the maximum productivity is reached. Again,
we see that a unit of effort yields more gain in productivity if a person has so far spent less
effort on the job than if she has already spent a lot of effort.
We should also take into account that the slope of the decrease in productivity is different
for different jobs. At the assembly line each additional unit of effort has a strong impact on
productivity. In contrast, if you are a door man you might still influence how well you are doing
your job by putting more effort into it. But effort should have much less effect for such a job
than for a job at the assembly line.
Summing it up, we can say that productivity depends on effort, but the marginal effect of
effort is reduced with amount of effort already spent and increases with the effort-intensity of
the job. In formal terms the argument is:

LM R = αe(H, EN LM )σ t (2.3)

35
 2. Theory

Now this section points out that the following always holds: 0 < σ < 1. σ represents the
effort-intensity. It is bounded by zero and one which leads to a decreasing return to effort. We
can see that if we take the first partial derivative with respect to effort as the marginal effect
of effort on labor market rewards:

δLM R(α, t, σ)
= σe(σ−1) αt (2.4)
δe

Rewriting this equation we get:

1
σe(σ−1) αt = σ αt (2.5)
e(1−σ)

Here we see that the marginal effect of effort on labor market rewards is a product of the
effort-intensity and more importantly depends inversely on the level of effort. The effect of
effort depends on the level of effort.

2.5. Gender Differences in Health Effects

So far we have a fairly simple model of health selection that does not know any gender
component. While the simplicity is convenient, the discussion of health inequalities should refer
back to broader sociological debates about social inequalities. One very important topic in the
analysis of social inequalities is of course gender inequality. In Germany, the study of systematic
inequalities between men and women became more widespread in the 1970s. Back then the
label womens studies was often used (Gottschall 2000, 14). Today, it has been established in
the scientific community that social inequalities cannot be reduced to the analysis of class. At
least gender and race need to be considered as well4 . For the purposes of the analyses race is
excluded for the moment, because a discussion would be beyond the scope of this thesis.
The starting point of my argument is the observation that even after decades of improvement of
women’s education and their legal rights and status women still face a disadvantage compared
to men in several societal areas. Again, I will have to narrow my focus of research and will
consider only labor market disadvantages. When I use the phrase disadvantage I imply that
women as a group face structural barriers on the labor market, which are not an individual
phenomenon but exist throughout society (Gottschall 2000, 12).

2.5.1. Subjective Performance Evaluations

After this very general view on structural inequalities between men and women in society I will
focus on a very narrow topic. The focus is necessary to address one mechanism in labor market

4
In this context it should be mentioned that other potentially structuring categories of analyses exist as
well even if they do not have the same dispersion as gender and race. Among them are for example
disability, sexual orientation and trans-gender or religion.

36
 2. Theory

rewards determination that should not be seen as gender neutral for my thesis. The subject of
inquiry is how employers - as one main actor in the process of labor market rewards setting -
evaluate their employees and attribute rewards to their accomplishments. Standard human
capital theory assumes they have perfect unbiased information on productivity. In this section
we will see that this assumption is hardly feasible and should be modified for the purpose of
explaining gender differences in health effects.

In my theory I assume that workers receive their labor market rewards according to the
perceived effort, not the actual effort. In the following part I explain why.
Employers want to maximize their profit and therefore want to pick the most qualified candidate
for a job or promotion. If criteria which can be directly compared5 are at hand, employers
will be able to assess productivity rather objectively. The (subjective) evaluation of effort in
contrast is much more subject to subconscious prejudices, feelings, and errors in judgment. As
it happens, rewards are most often awarded according to subjective evaluation, not objective
measures of performance (Prendergast & Topel 1996, MacLeod 2003). Surprisingly, one
study by Alexander & Wilkins (1982) does not even find a statistical significant association
between subjective and objective measures of performance among vocational and rehabilitation
counselors. Seemingly irrational judgments about performance can also be found among
colleagues at the same level. In one study, considering someone for cooperation at work is
only dependent on actual competence of those persons if interpersonal feelings are positive. If
they are negative competence does not play a role (Casciaro & Lobo 2008). Therefore, in the
following I will go into detail on gender specific problems of subjective evaluations.
This rather intuitive claim about subjective evaluations becomes especially important if we
have reason to believe that there are systematic biases in evaluation towards certain social
groups. There is substantial research from the fields of (organizational) psychology which backs
the claim about the subjectivity of performance evaluation and a related gender bias. The
argument below holds in case we are talking about an employer as an actual individual, but also
for the employer seen as a firm or as another type of organization. In classical organizational
theory the subject of investigation is often conceptualized as gender neutral. The rationale
behind this approach is that the defining elements of an organization in the classical view is
the rationalization of all social processes (Wilz 2002, chapter 1).
Analog to the critique of neoclassical theories of the labor market we can reasonably doubt the
adequacy of a gender neutral approach here. Recent empirical studies and the development of
theories of gendered organizations also cast serious doubt on the classical view (Wilz 2002,
44-46). For the purpose at hand, this means that performance evaluation can be seen as
subjective regardless whether the employer is an individual or an organization.

5
Such as work hours, tenure, piece-rates, or credentials.

37
 2. Theory

A review of the literature allows the statement that subjective evaluation often shows
considerable gender (or race) biases. This is important because of the finding that women
are assigned to jobs where effort is easier to monitor while men’s jobs encourage effort by
efficiency wages (Bielby & Bielby 2002, 197). Kanter (1977, 216) writes that such differences
in surveillance are integral part of male company culture:

“The Token6 does not have to work hard to have her presence noticed, but she
does have to work hard to have her achievements noticed.”

Kanter (1977, 219) states also that persons in Token positions are under much stronger scrutiny,
have to perform higher, and need to make bigger efforts at fitting in, not standing out too
much, keeping themselves constantly under control which is very taxing.
In the following I will give some examples of studies which warrant a claim of subjective
evaluation that might result in a gender bias.
If performance pay based on objective measures like piece rate or commissions is considered
the gender pay gap is reduced (Madden 2012). This finding indicates that other pay schemes
based on subjective evaluation contribute to the gender bias. Cohen & Huffman (2007) can
show that having female members in high status management positions lowers the gender
wage gap among general employees. Elvira & Saporta (2001) can show that given same
productivity blacks receive less favorable ratings from their supervisors than whites in a large US
company. Racial wage differentials disappear when job allocation and subjective performance
evaluation are controlled for (Elvira & Saporta 2001, 587). Among managers, it was found that
supervisors were less likely to attribute performance to ability in case of female managers than
in case of male managers (Greenhaus, Parasuraman & Wormley 1990). In a study by Igbaria
& Baroudi (1995) a gender bias in performance evaluation could not be found. On the other
hand, there was a gender bias in the supervisors’ perception of chances for promotion, favoring
men. This result is replicated by a meta-analysis in the study of Roth, Purvis & Bobko (2012).
An experiment conducted by Maas & Torres-González (2011) shows that women expect to
be rated lower if the proportion of raters is predominantly male which corresponds to the
finding from field research that women are less favorably evaluated the smaller the proportion
of women is among the raters (Sackett, DuBois & Noe 1991). Bohnet, Bazerman & Geen
(2012) suggest that joint evaluation of workers or applicants could overcome gender bias in
subjective evaluations as it makes comparison, contrasting and calibration of evaluation easier.
These new findings corresponds to older views in the economic literature on internal labor
markets. Doeringer & Piore (1985, 137-140) argue in the context of internal labor markets,
that discriminatory practices at entry level are closely related to

1. the informality of the screening and recruitment process.

2. variations in the process due to changes in job nature.

6
Token refers to persons in very clear minority position (less than 15 % of the whole population).

38
 2. Theory

3. statistical nature of the screening process.

4. reliance on limited channels of recruitment.

All of these factors might impair women’s chances in comparison to men’s.

However, it has to be conceded, that some studies do not find clear gender effects in the
evaluation of jobs (Grams & Schwab 1985, Graves & Powell 1995, Bowen, Swim & Jacobs 2000).
Also, in the context of more standardized ways of evaluation (and possibly affirmative action),
race or gender effects of evaluation might be reduced (Powell & Butterfield 1997, 125).
Overall, the theoretical argument and the empirical research allow the conclusion that - given
the same actual effort - a disadvantaged group will receive less favorable evaluations. They are
considered as if they had spent less actual effort. Combined with the classical human capital
theory approach of wage and labor market rewards determination this leads to the conclusion
that a discriminated group will receive less labor market rewards for the same actual effort than
the reference group.

2.5.2. Women’s Disadvantage on the Labor Market

For certain highly qualified positions on the labor market special skills, training, and qualifica-
tions are needed. In addition, a lot of effort has to be put into acquiring and keeping such a
position as they are very demanding and are granted only to high performing individuals. There-
fore health capital is very important for obtaining and keeping these jobs, because the applicant
or incumbent needs good health to invest a lot of effort to show the performance required of her.

What follows is an argument explaining how gender differences in the health effect with
regard to highly qualified positions might arise. The main assumption is that women face a
certain disadvantage in gaining a high status job and in earning wages. The use of a very broad
meaning of disadvantage is deliberately chosen, because the origin of the disadvantage does
not play a role for my theory. I will review some theories and studies which suggest different
ways of how such a disadvantage can come into existence without giving precedence for one or
another theory. Each of the mechanisms described in the theories is deemed to be sufficient,
but not necessary, to establish a disadvantage.
By disadvantage I mean that members of a certain group - for whatever reasons - need more
overall effort than members of another group to get the same reward. My use of the word
disadvantage should not be understood as normative in the sense that the stated differences
between groups are just or unjust, good or bad. The definition has only analytical, but no
normative implications. All discussions about the best way to explain the gender pay gap or
the gender gap in high status jobs are better suited for other studies, as are the normative
implications of these gender gaps.

39
 2. Theory

Among the many theories explaining the gender pay gap and the gender gap in managerial
positions I will only highlight a few. I refer to theories on discrimination, social networks at
work, personal characteristics, and the household context.
Highly qualified and managerial jobs are usually held by men. A lot of studies have
claimed that women face gender discrimination when trying to obtain such a position (Eagly
& Carli 2007, chapter 5). Relative to men their overall performance has to be higher to
compensate discrimination. One of the results of such discrimination is that women need
to be especially good at their job and need to spend considerably more effort on their work
than men to get promoted to such a position. This can be explained with the concepts
of taste for discrimination as proposed by Becker (1971) or with statistical discrimination
(Phelps 1972, Thurow 1975). For example, in the case of statistical discrimination (supposedly)
higher levels of sickness absence of women in general could lead employers to promote mostly
men which are expected to be less absent from work (regardless of whether this is the case
for the individual worker). Statistical discrimination can work through different skill levels or
different signals for skill levels between men and women (Bjerk 2008). An overall higher rate
of quitting by women (Frederiksen 2008) might also lead to reduced opportunities of women
for promotion.
Another approach to theorizing discrimination is the theory of taste for discrimination. Here
the assumption is that discriminating practices in themselves have a certain indirect utility gain
for the employer. They simply prefer not to employ or work with members of certain social
groups even if it means that their profit suffers from such a decision.
In a seminal qualitative study in the 1970s Kanter (1977) provides groundbreaking theory and
empirical material for the analysis of gender relations at the workplace. Since then many studies
with similar intent have been published inspired by Kanter’s approach. In her theory the taste
for discrimination can be explained e.g. by a company and leadership culture dominated by
men through their numerical superiority7 Kanter (1977):

“Quite apart from the content of particular jobs and their location in the
hierarchy, the culture of corporate administration and the experiences of men in it
were influenced by this fact of numerical dominance, by the fact that men were
the many.” [Emphasis in the original] (Kanter 1977, 206)

Token status in this sense means clear numerical minority status within a certain context.
This implies negative consequences for promotion chances and rewards in general. In these
contexts persons in minority or token status need to outperform the majority group. They

7
The author attributes the culture of dominance as an environment for women not due do femaleness,
but to rarity and scarcity of women in the company (Kanter 1977, 207). A group with a ratio of 85:15
is classified as skewed, 100:0 uniform, 65:35 majority vs. minority, and 50:50 is balanced (Kanter 1977,
208). Kanter (1977, 210-211) argues that minorities or tokens get more attention, are contrasted
(possibly in a exaggerated way) against the majority and generalizations about this group are more
often made.

40
 2. Theory

have to rely heavily on good health to be able to show such a performance. This leads to the
hypothesis that in a male dominated context women’s health should be more important for
labor market success than men’s health. Although the theory was developed from a gender
perspective where women are the minority group, Kanter (1977) states that the theory is
generalizable to other types of groups and is symmetric. This means that in a female dominated
context men’s health should be more important than women’s health for labor market success,
because under these circumstances men are in a token position.

Building on theories of discrimination and male dominated company culture, most studies
agree that wage or promotion gaps cannot be explained by differences in relevant criteria like
education, experience or occupation alone. At this point it is useful to take a small detour to
the issue of the gender wage gap and then come back to the gender promotion gap, because
they are interlinked. However, the gender wage gap is better known and has been investigated
more often.
There are several ways of presenting the gender wage gap, each having a different interpretation
and way of estimating it. As Kunze (2008, 74) puts it: “There is no undisputed method of
measuring the gender wage gap”. It should be noted that this can easily become a political
issue. I try to avoid this and will explain later how.
First, there is the overall gender pay gap which is usually presented as the percentage of
wage-per-hour that women earn less than men.
Following the model of labor market rewards by Becker (1985), wage differences will arise if
men and women have different levels of accumulated human capital. Often this is controlled
for in a regression and the remaining difference between men and women is then called the
unexplained gender wage gap (Kunze 2008, 68) or the adjusted gender wage gap. Taking
only part of human capital variables into account can lead to big differences in the estimation
of the gender wage gap (Weichselbaumer & Winter-Ebmer 2005). Taking these ideas into
account, in the methods part it will be discussed which controls for differences in human capital
endowments between men and women are chosen for the empirical analysis (section 4.9).
Sometimes it is argued that occupations dominated by men require heavier work which is
physically more straining and has to be compensated by higher wages to attract sufficient labor
supply (Viscusi 1978, Viscusi & Moore 1987, Filer 1985, Okamoto & England 1999). This
leads to the hypothesis that controlling for the physical (and also psychological strain) in an
occupation might explain the gender wage gap. However, empirical investigations cast doubts
on this theory. Jacobs & Steinberg (1990) and Kilbourne, Farkas, Beron, Weir & England
(1994) show that occupations with physical strain do not necessarily pay higher wages. Hersch
(1998) uses a different method and does find compensating differentials for both genders.
Similar results are found by Cousineau, Lacroix & Girard (1992). Another study found that
both men and women receive compensation differentials. While for men these compensations
are higher for risks of fatal injuries, for women they are higher for non-fatal injuries (Leeth &

41
 2. Theory

Ruser 2003). Macpherson & Hirsch (1995) can show that job hazards and the environment do
affect rewards significantly, but reduce the effect of female-percentage in a job on wages only
slightly. For Germany a study by Liebeskind (2004) cannot find any substantial reduction of
the gender wage gap after adjusting for physical strain. The overall results are inconclusive,
therefore compensating differentials cannot be disregarded and occupational strain will be taken
into account in this thesis. This is of special importance as health as a predictive factor might
correlate with the strain of occupation.
Another important part of the gender wage gap is occupational segregation. This refers to the
fact that a lot of occupations are dominated by either men or women. Occupational segregation
can impact differences in wages, likelihood of promotion and job security between men and
women (Busch & Holst 2012). For Germany there are also other studies that find that gender
specific occupational segregation and in some cases segregation by industry is one factor in
explaining the gender wage gap (e.g. Leuze & Strauß 2009, Achatz, Gartner & Glück 2005, Hinz
& Gartner 2005, Liebeskind 2004). Consequently, the industry of the employer will also be
taken into account in all empirical models on gender specific health selection mechanisms.

Similar to the gender wage gap the small proportion of women among high status jobs and
leadership positions can be seen as a combination of differences in human capital endowments,
occupational segregation, and unexplained factors which might present barriers or “glass
ceilings”. These unexplained factors are what is targeted by the theories of Eagly & Carli
(2007) or Kanter (1977).
Even among female dominated occupations promotions are sometimes more likely to go to
men than to women. It should be noted that the famous “glass ceiling” effect is contested in
the literature. The “glass ceiling” hypothesis states that it gets constantly harder for women
to get promoted the higher the desired position is. The more authority is at stake the stronger
the disadvantage. Some authors however point out that disadvantages start earlier and strong
underrepresentation of women in the absolute top management can also be explained by
constant disadvantage or even decreasing disadvantage (Eagly & Carli 2007).
Linked to this research are studies which find opposite results. Higher percentages of
women in an occupation result in an increased chance of promotion for men in one country
(Hultin 2003), but they lead to a decreased chance of promotion for men in a study in the US
(Maume 1999).
This points to a theory called “glass-escalator” theory (Williams 1992), which is opposed to
some of the predictions derived from Kanter’s (1977) theory. Hultin (2003) finds that men have
markedly higher promotion chances than women in female dominated occupations. In male
dominated occupations there are no significant differences in chances of promotion between
men and women. The author draws on theoretical explanations for the “glass-escalator” effect
developed by Williams (1992). Coworkers welcome men in female dominated occupations as
bearers of potential prestige and pay-raise giving them a better standing within the occupational

42
 2. Theory

field. Supervisors, who are in the majority men, also favor men over women due to homophily.
Clients who are used to women in female dominated jobs want to be treated by women rather
than men. Since a lot of female dominated occupations involve high interaction with clients
(e.g. care-jobs, nursing, social work) men are promoted to higher or supervisory positions
which require less client interaction to circumvent client disapproval (Hultin 2003, 36-37).
The finding that women have the same promotion opportunities as men in male dominated
occupations is interpreted not as disproving the argument of discrimination or exclusion, but
rather pointing to a different mechanism of exclusion which works at the occupation entry
level, but not afterwards (Hultin 2003, 54). However, not every study finds the glass-escalator
effect (e.g. Kullberg 2012, Snyder & Green 2008).

At this point we must acknowledge research which indicates that the choice of occupation
is influenced strongly by gendered socialization. Referring back to socialization recognizes
that social structures, processes, norms, and values influence parents in the upbringing and
education of their children. Thus parents do not choose their way of raising their children
independently of society (Hurrelmann 1994). Women are influenced in a different way than
men during childhood in their role formation: “Playing doctor or playing nurse” (Cain 1976,
1236). Such childhood influences present another disadvantage for women in attaining the
same labor market rewards as men.
This leads us to another important field that can induce a disadvantage for women on the labor
market. These are gendered responsibilities in the household and in the family. It should be
taken into account that women might choose occupations which are less effort intensive, because
they need to spend more effort on non-labor-market activities. Gendered division of household
labor can thus lead to a gendered division of the labor market (Becker 1985, Polachek 1981).
The household and family context has been shown to play a significant role in determining career
chances (Stone & Lovejoy 2004). Women are much more constrained by non-labor-market
activities than men (Dempsey 2000, Bianchi, Milkie, Sayer & Robinson 2000). These activities
include childcare, housework, and care for family members all of which are often less flexible than
men’s chores (Hook 2010). Family responsibilities constitute a strain on women’s resources,
because they cannot spend these resources on their job. The resources concern both time and
effort. In addition, women might suffer from employers’ expectation about the potential of these
extra burdens, which increases tendencies for statistical discrimination (Ortiz & Roscigno 2009).
Most notably potential or actual drop-outs from the job due to pregnancy are often reasons
not to hire a woman for a leadership position early in her career (on the wage penalty for
motherhood and marriage, see Budig & England 2001, Loughran & Zissimopoulos 2009). Time
off to care for children can have a similar effect. Maternity leave depreciates human capital, a
situation men (usually) do not face (Staff & Mortimer 2012). Summing up, we can say that ex-
pected and actual family strains constitute a major disadvantage for women on the labor market.

43
 2. Theory

A third important factor determining the difference between men and women in obtaining
high status positions are social contacts and networks (Eagly & Carli 2007, 144). Social
contacts and networks can facilitate three things.
First, information can be easier transported than through formal channels alone (Granovetter
1973, Burt 1995, Lin 1999). This holds true for directly transmitted information or information
given by third parties, sometimes in the form of recommendations (Franzen & Hangartner 2005).
Employers are better informed about the performance and productivity of men than of women.
They also have better access to information about characteristics of the male worker like
personality, reliability, or productivity at a former job, which are hard to certify (Flap &
Völker 2001, 163).
Second, trust can be facilitated via social networks (Burt 1995, 15). This means that although
employers do not have more direct information about men than about women, they do have
higher trust in men due to recommendations or personal contact. Both trust and additional
information reduce the costs of evaluating and hiring workers (Gorman, Marsden, Kalleberg
& Berg 2001, 107-109). This makes it more convenient for employers to give preferential
treatment to those persons for whom informal information is available.
Third, social contacts and networks can facilitate personal preferences for working with certain
persons which are not grounded in professional performance but in personal feelings of sympathy,
affection or homophily (Mount, Sytsma, Hazucha & Holt 1997, McPherson, Smith-Lovin &
Cook 2001, McDonald 2011). Reskin & McBrier (2000) can show that open recruitment
processes reduce disadvantages of women in contrast to recruitment through informal networks.
Networks of incumbents of managerial positions, which systematically keep women out, are
referred to as old boy networks (Oakley 2000, 328-329). Homophily is also at the bottom of a
potential taste for discrimination of employers against women.
All three ways in which social networks at work operate can lead to a systematic disadvan-
tage for women, because their networks are far more focused on private and family context
than men’s networks (Campbell 1988, Moore 1990, McDonald, Lin & Ao 2009, McDonald 2011).

Women’s values could also be a cause for disadvantage, although empirical validity of this
claim is contested (Burke & McKeen 1994). It is well known that men favor higher wages
while women are more focused on other kind of non-pecuniary rewards (e.g. Marini, Fan, Finley
& Beutel 1996). This can include a rewarding content of a job, or the possibility to have both
career and family (Cinamon & Rich 2002). More generally - and referring to the comprehensive
value framework of Schwartz - men show higher scores on values like power and achievement,
while women tend to value benevolence and universalism higher (Schwartz & Rubel 2005).
Higher values in the self-enhancement sphere could lead to higher priorities for advancement at
work and a stronger focus on achieving goals at work at the expense of other activities aside
from work. Therefore men have an advantage regarding values compared to women. As values
are assumed to be personal characteristics which are quite stable, methodological approaches

44
 2. Theory

which can control for time constant unobserved heterogeneity like a fixed-effects approach are
valuable in controlling for differences in values.
Escriche (2007) argues8 that women’s preference for family over career will persist as a minority
preference even if the prevalence of the preference drops. A minority preference in this sense
is a preference held by a clear numerical minority of the reference society. The idea is that
these preferences will not die out because those who have them do not assimilate. Instead they
invest more in teaching their children their values and preferences. The additional investment
is made due to their knowledge that society will most likely not teach their kids the desired
values. At the same time those in a majority position will invest less in teaching their children
their preferences, because they know it is likely that the children will adopt the majority prefer-
ences anyway. This presents a rational-choice approach to utility maximization with regard to
preference transmission from one generation to the next. Specifically, Escriche (2007) claims
that a traditional role model for women (less career oriented than men) has the characteristics
of such a minority preference and will remain as such in the prediction of the theoretical model.
This would lead to the fact that women will continually face statistical discrimination on the
labor market, because employers cannot differentiate between career or family oriented workers.
However, they want to hire preferably career oriented persons. Additionally, discriminatory
practices and lack of success on the labor market can lead to less taste for work which leads to
a vicious circle of less and less success on the labor market (Cain 1976, 1223).
If women (have to) make a trade off (e.g. in favor of family responsibilities) at the expense of
higher wages or promotion they have a relative disadvantage. They could for example choose
more flexible working hours, a job closer to home, or a better work environment instead of more
traditional rewards. It can be said that individual behavior based on preferences for work or
family (Hakim 1998, Hakim 2002) and decisions in interaction with structural constraints can
reinforce the disadvantage of women on the labor market. However, one should be cautious
to adopt a “blaming the victim” stand in this matter as the interplay of individual decisions,
values, and constraints are often too complex for such simple conclusions (Gottschall 2000,
277-279).
A related subject are bargaining skills and self-assessment. Women tend to be more modest
in their bargaining about wages and job positions. They also are more likely than men to
underestimate their own performance. This will lead to a weaker position when bargaining for
wages or promotions. Values, preferences, and bargaining behavior are theoretical mechanisms
that explain why women earn less for the same work or why they are less often promoted than
men (Stuhlmacher & Walters 1999, Kray, Galinsky & Thompson 2002, Stevens, Bavetta &
Gist 1993).

8
based on a model of intergenerational transmission by Bisin & Verdier (1998).

45
 2. Theory

2.5.3. The Result of Disadvantage and Subjective Evaluation

In the previous sections we have established that performance evaluation is subjective and
disfavors certain groups. We then saw that for several reasons women are a disadvantaged
group on the labor market.
Given a bias in performance evaluation, women need to show more absolute effort to gain
the same evaluation of their effort. To ensure that women can spend more effort for a longer
period of time it is important that they are in good health. To be precise women’s health
should be more important for getting in a high status position than men’s health, because
they have to spend relatively more effort on an already demanding job which makes it highly
dependent on their health capital.

So far I have given intuitive reasoning for gender differences in the impact of health on
labor market rewards. For many purposes this might suffice. However, as the argument is just
an extension of the basic human capital model not a refutation, it makes sense to formalize
the argument above in the way Becker does for the human capital theory. This makes it
comparable to the original and gives another perspective on the same problem under more
formalized conditions.

I = αeσ t, 0 < σ < 1 (2.6)

To recapitulate what was defined above: I is income, α is human capital, e is the effort per
hour, σ the effort intensity of the job and t the hours of work. I replace income by labor
market rewards to make a more general case. Effort also becomes a function of health (H) and
non-labor-market effort (EN LM ). The important alteration of the equations shown above is
that I take the theory on discrepancies between actual and perceived effort into account. I will
therefore replace effort in the equation by perceived effort ϵ for men (m) and women (f) which
depend on a perception factor τ that has a gender bias.

ϵf = e(H, EN LM ) ∗ τf (2.7)
ϵm = e(H, EN LM ) ∗ τm (2.8)

In the argument above I have argued that the same actual effort is on average evaluated as
being less for women than men. That means the same actual effort leads to a smaller perceived

46
 2. Theory

effort. So, assuming a gender bias in perception, we conclude that:

0 < τf < τm < 1 (2.9)

The new model before substituting actual effort for perceived effort is:

LM R = αϵσ t (2.10)

If we now take the first partial derivative with respect to perceived effort as the marginal effect
of effort on labor market rewards we get:

δLM R(α, t, σ)
= σϵ(σ−1) αt (2.11)
δϵ

Rewriting this equation we get:

1
σϵ(σ−1) αt = σ αt (2.12)
ϵ(1−σ)

To get health into the equation, we have to refer back to actual effort instead of perceived
effort, because it is the actual effort that is a function of health. Inserting actual effort instead
we get:

1 1
σ αt = σ αt (2.13)
ϵ(1−σ) (e(H, EN LM ) ∗ τ )(1−σ)

We know that 1 − σ is by definition positive as σ only takes values between 0 and 1. Comparing
the effect for men and women - given the same health status - it holds for any possible σ:

1 1
σ αt > σ αt | H (2.14)
(e(H, EN LM ) ∗ τf )(1−σ) (e(H, EN LM ) ∗ τm )(1−σ)

It follows that a change in health results in higher return to labor market rewards for women
than for men.

2.5.4. Visibility of Health Problems

I have assumed that a reduction in effort will be noticed by the employer and will lead to
reduced labor market rewards. But do employers really notice if you are a bit tardy at work? Or
is it only a problem as soon as you faint at work, because you are so ill? It is worth thinking the
different possibilities through and see how they will alter our conclusions about the influence of
health on labor market rewards.
Introducing effort as an important determinant for labor market outcomes assumes that
employers can (one way or another) assess or monitor the effort an individual worker makes
(Becker 1985, S43). Thus, making a distinction between health effects, which are directly

47
 2. Theory

visible to the employer and those who are only indirectly visible seems appropriate. One obvious
kind of health effect that is directly visible are days of sickness absence. An employer can count
these days and calculate how much productivity of each of the workers in the firm is lost due
to sick days off. The other obvious effect would be a reduction in work hours by the worker.
The latter usually automatically reduces the labor market rewards of the worker, at least as
long as he is paid by the hour.
However, one can also conceive of health influencing productivity via the effort per hour (section
2.4). This is especially important in a labor market environment like Germany which has rigid
labor market contracts and where full-time employment is not only normal, but also the norm
(Pierenkemper & Zimmermann 2009). If a reduction in work hours is not an option, reduced
effort per hour could be the only way to react to a health impairment. In a severe case this
would lead to a behavior, which is referred to as presenteeism in the literature. This means
that an individual goes to work although he or she is in fact ill and should stay at home for
recovery9 . Presenteeism increases among other things with the degree of job insecurity and
with high job demands (Demerouti et al. 2009, Johns 2010).
Reduction of effort per hour is a lot harder for employers to monitor. If the employment is
not based on piece work, reduced effort per hour can be seen as a health effect that is only
indirectly visible (if at all).
This problem appears in the academic literature as well. In a review article by Edington &
Schultz (2008) about the impact of health risks on productivity, the authors report only studies
using presenteeism and absenteeism as measures of productivity. Other measures are usually
not available for practical reasons.
Effort that is not directly visible is much more prone to errors in judgment, and to less certain
evaluations (discussed in detail in section 2.5.1). The employer would have to assess the output
a worker produces in a certain amount of time. If the work is highly specialized the employer
might even lack the means to monitor or control the effort a worker spends at work completely
(see e.g. Sørensen 1996). Another reason why workers are not monitored is that employers
might want to avoid negative consequences of monitoring and surveillance which were found in
some recent studies (Dittrich & Kocher 2011, Hasan, Akif, Subhani & Imtiaz 2012, O’Donnell,
Ryan & Jetten 2013) like reduced quality of work or a breach of trust.
This raises the question of how much workers are monitored. Control systems implemented
at work should make reduced productivity more visible to the employer. It constitutes an
information advantage (or a reduction in an information disadvantage if you will) which enables
the employer to make informed decisions.
The employer will then be able to react better to such a loss in productivity. The stronger
such systems of controlling employees are, the stronger the link between health and labor
market rewards should get. However, in the last decades new forms of management have arisen.

9
Some studies show that presenteeism might even be more economical harmful than its antonym
absenteeism (Collins, Baase, Sharda, Ozminkowski, Nicholson, Billotti, Turpin, Olson & Berger 2005).

48
 2. Theory

Foremost the High Performance Management has risen as the most important post-fordistic
paradigm of work organization and worker control (Butler, Felstead, Ashton, Fuller, Lee, Unwin
& Walters 2004). Here responsibility for workers actions, their performance and output is
often delegated from higher levels of authority like supervisors to the employees themselves.
From an employee perspective German sociologists have created the concept and term of
“Arbeitskraftunternehmer” or “entreployees” which neatly describes the consequences of High
Performance Management for employees (Voß & Pongratz 1998, Pongratz & Voß 2003).
Another insight from the distinction between visible and non-visible forms of health problems is
that one might expect differences between the effects of individual health status and sickness
absence on labor market rewards. At last, the theoretical discussion is not developed enough to
allow a directed hypothesis in this area. The analyses will thus be of more exploratory nature.
One clue we might get is that there are studies reporting that men are more often faced with a
culture of presenteeism. The study by Watts (2009, 520,525) about managers in constructions
sites reveals that men build a culture of competitive presenteeism which requires them to be
at work at all costs. This means that sickness absence is penalized stronger than for women.
Therefore one could expect that job status is more strongly affected by sickness absence in the
case of men than in the case of women. Piha, Laaksonen, Martikainen, Rahkonen & Lahelma
(2010) also show that higher education and occupational class are associated with reduced
sickness absence among municipal employees in Helsinki. These associations are markedly
stronger for men than for women. Avdic & Johansson (2013) conduct a study that provides
an indirect test of preferences for sickness absence. They conclude that men indeed have a
higher preference for few days of sickness absence than women. The theoretical argument is
that women’s double role in household and work requires them to invest more in health to
fulfill their duties. This results in longer recovery time in case of sickness, ergo longer sickness
absence given same illnesses.
These more specific studies suggest that sickness absence is a good indicator for a visible health
problem. which might yield interesting gender patterns in my empirical analyses.

2.6. Health Behavior as an Explanatory Factor

As mentioned in the discussion of the human capital model there are two main actors involved
in the health selection process: employers and employees on the labor market. So far we
have looked at the first actor and modified the human capital model from their perspective
introducing gender biased subjective performance evaluation as a modifier of labor market
rewards determination.
Now, we can take a closer look at the other actor, the employee. Do employees have something
to contribute to the discussion of possible gender differences in health effects? I argue that
they do. Their health behavior can be an important modifying factor in the human capital
model. One main assumption which has been upheld so far will be challenged: Is health always

49
 2. Theory

related to productivity? Especially a gender perspective makes room for doubt to this claim.
In the following we will see why.

The starting point of the argument is an external health shock affecting a worker. What
should now be under scrutiny is the worker’s reaction towards this health shock. Two possibilities
arise:

1. The worker can reduce her effort and/or working hours to deal with her health condition.
This means accepting a (short-term) reduction of labor market rewards (with possible
longterm consequences) in favor of a short-term improvement of the health status.

2. The worker can show the same amount of effort regardless of his health condition. This
means accepting a (short-term) reduction of health (with possible long-term consequences)
in favor of keeping the labor market rewards.

Which option the worker chooses depends on his preferences (for a related argument, see
Johns 2010, Cropanzano, Rupp & Byrne 2003). Preferences for health come into play as well
as preferences for career (to have a broad category encompassing preferences for labor income,
working hours, etc).
Preferences are mostly generated through socialization processes as discussed in section 2.5.2.
This highlights the role of the parents, school, and early adulthood for later life decisions on
health related behavior (Hurrelmann 1994). The stronger the preference for health is, the
higher the chance that the person chooses option 1. The higher the preference for career is
the more likely the person is to choose option 2. It is important to note that these preferences
should be seen in relation to each other. So the higher the preference for subjective health
compared to the preference for career the more likely a person is to choose option 1.
In reality options 1 and 2 form a continuum. One pole represents complete focus on recovery,
the other complete focus on the job, disregarding recovery. It follows that the higher the
preference for career is - relative to preference for health - the more time and energy will be
used for the job instead of recovering from a health shock.
If a person chooses to (partially) neglect recovery it can have different consequences. First,
it is possible that she fully recovers despite the lack of focus on recovery. In this case one could
say the person is lucky. Second, the person needs to borrow energy or health from future points
in time or the person needs to externalize other non-labor-market efforts to other persons.
Recovery will take longer. There might even be long-term negative health consequences. This
can be directly visible or rather be a cumulative process (which might lead to burn-out or
chronic conditions of mental or physical health).

50
 2. Theory

2.6.1. Gender Differences in Health Behavior

The preferences for health and the respective health behavior do not vary randomly within
the population. A lot of studies have shown that women put more emphasis on maintain-
ing a healthy lifestyle and treating illnesses or mental problems with more care than men
do (Dean 1989, Wickrama et al. 1999, Cockerham, Hinote & Abott 2006, Stefansdottir &
Vilhjalmsson 2007). Men are often considered to be less reactive towards health problems.
The problem is finding evidence not only for gender differences in preference for health, but
for health in relation to career. Some empirical findings directly indicate that such particular
gender differences exist.
Women see themselves as responsible for matters of health in the family, acquire more knowledge
and follow through with the consequences (Faltermaier 2008, 41). Differences in health behavior
as an explanatory factor for differences in sickness absence are not uncommon (Zok 2008, 119).
Collins et al. (2005) find that women are more likely to be absent from work given a chronic
health condition than men are. Self-reported impairment of work due to a chronic health condi-
tion was not gender sensitive. In a study by Fried, Melamed & Ben-David (2002) the results show
that women are more susceptible to stress from noise than men are. In a study by Sandanger,
Nygard, Brage & Tellnes (2000) women showed absolute higher prevalence of sickness absence
given a mental health problems, but relatively to the prevalence of mental health problems a
lower prevalence of sickness absence than men. This could either be interpreted as women
reacting less to mental health problems or men recognize or report fewer mental health problems.

2.6.2. Consequences of Differences in Health Behavior on Health

Effects
Let us say we accept the assumption that men have lower preference for health than women.
It follows from the argument in section 2.6 that we should expect less reaction in form of
recovery and at the expense of work effort from men than from women. If you do no adjust
your effort after a health shock then your labor market success should not be affected. Labor
market success becomes thereby invariant to health. The association between health and labor
market success is reduced the stronger the preference for career is (relative to preference for
health). Since men have a higher preference for career than women, their health should have a
lower effect on their labor market success than women’s health has.
I assume that there are no gender specific learning processes with growing age. I make this
assumption to rule out the possibility that men and women adapt over time in their reactions
to health behavior. The adaption could lead to a possible convergence of health behaviors or
to a divergence depending on success and preferences. This assumption could be relaxed if age

51
 2. Theory

specific analyses10 were possible. However, the number of observations even in the SOEP is
too small for such an analysis.

Again we can easily show a formalized version of the argument above. We simply assume
that effort is only a function of health for women, but not for men:

ef = e(H, EN LM ) (2.15)
em = e(EN LM )⊥H (2.16)

Jumping to the calculation of marginal effects of effort on health we get for women:

1
αtσ (2.17)
(ef (H, EN LM ) ∗ τf )(1−σ)

and for men:

1
αtσ (2.18)
(em (EN LM ) ∗ τm )(1−σ)

The equation shows that health does not play a role for LMR in the case of men, but it
does for women. This is a theoretical conclusion that is matched by findings in the literature
(McDonough & Amick 2001, 136).

2.7. Open and Closed Positions

So far I have used alterations of human capital theory to explain the link between health and
labor market rewards. Now, I will turn to an area of the labor market where we have reached
the limits of human capital theory and must take a new perspective to give sound theoretical
background to the research question.
The question which arises is: What happens if there is no competition, no market mechanisms
regulating labor supply? Is the link between health and labor market rewards the same? This
part addresses the question where we should look for health selection and even more important,
where we should not look. Asking these questions helps to locate health selection mechanisms
and enables researchers to answer questions of health selection versus social causation more
carefully.
Before I can answer the question I have to provide a complementary theory to human capital
theory. I will use the sociological theory of open and closed positions. It is a major departure
from classical thinking without discarding all the insights gained by human capital theory. I will

10
E.g. through interaction-terms in the regression analysis or analyses of subgroups.

52
 2. Theory

describe its foundations, assumptions, and conclusions about labor market processes. This lays
the groundwork for answering the question about health and labor market rewards in open and
closed positions.

2.7.1. Open and Closed Positions - Weber’s Theory of Social Closure

The theory of open and closed positions has its roots in a more general theory of social closure
that goes back to Max Weber. In his opus magnum Economy and Society, Weber (1976)
distinguishes between open and closed social relations:

“Eine soziale Beziehung (gleichviel ob Vergemeinschaftung oder Vergesellschaf-

tung) soll nach außen ’offen’ heißen, wenn und insoweit die Teilnahme an dem an
ihrem Sinngehalt orientierten gegenseitigen sozialen Handeln, welches sie konstitu-
iert, nach ihren geltenden Ordnungen niemand verwehrt wird, der dazu tatsächlich
in der Lage und geneigt ist. Dagegen nach außen ’geschlossen’ dann, insoweit und
in dem Grade, als ihr Sinngehalt oder ihre geltenden Ordnungen die Teilnahme
ausschließen oder beschränken oder an Bedingungen knüpfen.” (Weber 1976, 23)

In this definition open relations are relations to which access is not regulated beyond the
point that a person has to be able and willing to be part of the social relationship. A closed
relation is one that regulates the access to a social relationship in a special way. The process of
regulating and implementing restrictions to the access to social relationships is what is meant
by social closure (Parkin 2003, 3).

Weber sees the development of closed social positions as a process which happens under
different circumstances and in different societies. He claims that the desire to ward off
contestants for a social position rises with the degree of competition. He argues that the main
way of achieving closure is using an ascriptive characteristic like race or gender to exclude
members of this group from access to the social position.

“Mit wachsender Zahl der Konkurrenten im Verhältnis zum Erwerbsspielraum

wächst hier das Interesse der an der Konkurrenz Beteiligten, diese irgendwie
einzuschränken. Die Form, in der dies zu geschehen pflegt, ist die: daß
irgendein äußerlich feststellbares Merkmal eines Teils der (aktuell oder potentiell)
Mitkonkurrierenden: Rasse, Sprache, Konfession, örtliche oder soziale Herkunft,
Abstammung, Wohnsitz usw. von den anderen zum Anlaß genommen wird,
ihren Ausschluß vom Mitbewerb zu erstreben. Welches im Einzelfall dies
Merkmal ist, bleibt gleichgültig: es wird jeweils an das nächste sich darbietende
angeknüpft.”(Weber 1976, 201)

53
 2. Theory

Interestingly, we can find analogies to the theory of open and closed positions in Durkheim’s
(1922) work. He notes that the division of labor becomes anomic if the division of labor
does not match the natural talents and skills of persons. He states that certain careers are
completely closed 11 or harder to access for disadvantaged persons.

“ [...] grâce à la persistance de certains préjugés, une certaine faveur s’attache

aux uns, une certaine défaveur aux autres, qui est indépendante de leurs mérites.”
(Durkheim 1922, 371)

With practices descended from the caste system as an example Durkheim states that due
to prejudices certain people are evaluated with favor and others with disfavor. Important to
note is that these evaluations are regardless of actual merit of the person. This claim very
closely corresponds to the definition of rent given by Sørensen (1996). In Durkheimian terms
rent arises if social inequalities do not exactly match natural inequalities. At this point we
have to be careful with the analogy. Sørensen builds this theory of rent on a notion of rewards
deviating from what they would be under perfect competition. We do not know exactly if
perfect competition as a reference point for Sørensen is equal to natural inequalities which is
Durkheim’s point of reference for the evaluation of existing social inequalities.

2.7.2. Open and Closed Positions - Sørensen’s theory

In this part I will explain the theory of open and closed positions or social closure on the labor
market developed by Aage B. Sørensen (1983). As discussed above the theory is closely linked
to the theory of segmented labor markets, which is a term used by Cain (1976) in his overview
of challenges to neo-classical views on the labor market processes in the 1970s (for a short
overview of segmented labor market theories, see section A.3 in the appendix).
Sometimes it is discussed as a sociological alternative to human capital theory. This is a bit
misleading. The theory of open and closed position is rather pointing out the limitations
of human capital theory and describes ways of determining labor market rewards where the
assumptions of human capital theory do not hold. Sørensen (1983, 207) even explicitly points
out the strengths of human capital theory:

“For the open position scenario, this allocation process is well described in standard
economic theory.”

The intriguing feature about Sørensen’s theory is that it is flexible enough to integrate
theories of segmented labor market and neoclassical labor market theories without losing
11
Durkheim uses the term fermé.
“Même aujourd’hui et chez les peuples les plus cultivés, il y a des carrières qui sont ou
totalement fermées, ou plus difficiles aux déshérités de la fortune.” (Durkheim 1922, 372)
.

54
 2. Theory

precision. In addition, it analyzes exactly those processes that are responsible for heterogeneous
ways of wage determination on the labor market. How his theory is applied to the analysis of
health selection is explained in the next section. Before health selection is addressed, I will give
a brief outline of the theory of open and closed positions.

As the name indicates Weber’s (1976) distinction between open and closed positions is
at the heart of Sørensen’s (1983) theory of open and closed positions12 . Open positions are
competitive jobs, closed are non-competitive jobs, possibly in internal labor markets. Closed
positions are only available when vacated by the previous holder of the position. Unlike in closed
positions, incumbents of open positions can be replaced at any time (Sørensen 1983, 206).
These positions are not to be understood as a dichotomy, but as two poles of a continuum.
Sørensen’s (1983) theory of open and closed positions on the labor market has several impli-
cations for the selection process of employees, because it leads to the question under which
circumstances health selection is unlikely to be a driving factor of health inequalities.

In open positions market forces dictate reward of and access to jobs. Open positions are
assumed to be unrelated to each other. They do not exist independently of their incumbents.
A person holding an open position can be replaced at any time (Sørensen 1983, 206-207).
Competition between employees is high, wages and all other benefits are derived exclusively from
a worker’s productivity. Employers choose their workers without interferences of institutional or
legal regulation and treat workers as if they were exchangeable (Eliason 1995, 248).
This means that the occupancy of a positions does not grant the holder any benefits except
what she gains through her productivity. The incumbent is in direct competition with other
workers, unemployed, and non-employed as the matches between employer and employee are
definite and usually short. Human capital theory aptly describes these processes of allocation
to jobs on labor market (see section 2.4).
Closed positions on the other hand are not allocated as described by the human capital
model. A human capital approach would suggest that a worker is paid according to her
productivity which is determined by experience, education and the amount of effort she invests
in her work (e.g. Becker 1985, Mincer 1974). In closed positions, however, payment and other
rewards are linked to the job and not to the worker (Sørensen 1983, 211). This means that the
incumbent of a certain position gets the same reward regardless of her performance on the job
(Sørensen 1983, 209). It is also important to note that in closed positions incumbents can hold
their positions as long as they want. The match between person and position is indefinite unless
the incumbent chooses to leave. A new person can therefore only get into such a position if
it has been voluntarily vacated. The mechanism of choosing a new incumbent is not based

12
Weber (1976) uses open and closed position not only for the analysis of labor market positions, but all
social positions. The focus in this thesis is on labor market positions.

55
 2. Theory

on market mechanisms, but on authority decisions by an employer13 which results in career

tournament or job competition as described below (Sørensen 1983, 206). Closed positions
are deeply embedded in relationship to other positions so that they exist independently of the
persons in the positions.

“[...] new matches cannot be established when changes in individual performance,

or the availability of a candidate with qualifications believed to be superior to
the incumbent, would make it desirable.” (Sørensen 1983, 206)

It should not be concluded from the argument so far that there is no competition for closed
positions. The contrary is true. There can be fierce competition for openings of jobs in a closed
position. The outcome of an application process is interdependent on the performance of other
contestants (possibly co-workers). Even if one receives extra training and shows more effort it
might not be enough, because in a ranking one is still not the first. It is a winner-takes-it-all
situation. The individual will therefore take the effort of others into account when deciding how
much effort to spend himself (Sørensen 1983, 208-209). The other important difference is that
the competition is over as soon as the vacancy is filled. Sometimes this kind of competition is
referred to as career tournament (Inkson 2004). The notion of career tournament implies that
persons within a firm compete for promotion at each step of hierarchy. One needs to win each
round to progress to the next level. Those who are promoted early and have steadily been
promoted have higher chances to get promoted even further than colleagues who have more
unclear career trajectories (Rosenbaum 1979). The tournament model implies closed positions
and vacancy chains as mechanisms of job allocation.
Related to the concept of open and closed positions is the distinction between wage competition
and job competition proposed by Thurow (1975). As the name suggests in labor markets where
wage competition is preeminent workers compete though wages, lowering them to get a job. In
the job competition framework, wages are seen as fixed. Workers compete for vacant jobs. It
presents a more general model than the career tournament model, because it can be applied to
whole parts of the labor market, and not only to within-company promotions.

2.7.3. Reducing Supply, Increasing Demand - Mechanisms of Social

Closure
I have described many of the differences between open and closed positions so far. What we
do not know yet is how closure is achieved. For this purpose I will first present the two major
mechanisms which bring about social closure. Second, I will give some examples of concrete
strategies facilitating the mechanisms in the context of the labor market.

13
Employer in the broadest sense. This can be a supervisor, head of Human Resources, or the actual
owner of the business.

56
 2. Theory

Social closure on the labor market (and on product markets as well) usually modifies either
supply or demand, so that a state of equilibrium, as predicted by neoclassical economics, does
not come about. It presents a kind of market imperfection. Closure strategies often try to
limit the supply (of labor or products) so that - given constant demand - prices will go up and
above the competitive price.
The other possibility is that closure strategies aim at increasing the demand for a certain asset
(a product or labor) beyond the normal state it would reach in a perfectly competitive market.
If supply is kept constant this leads to an increase in prices (wages)14 (Weeden 2002).

On the labor market there are several general strategies to achieve reduction of supply or
increase of demand. I will only mention one here, because it will be the base for an empirical
measure of closure I use15 . This strategy is professionalism and credentialism. In the appendix,
I added the most common alternative strategies (A.2).

2.7.3.1. Credentialism and Professions

Weeden (2002, 61) defines credentialism as “[...] the use of familiar symbols or markers of
knowledge (e.g., grad levels, diplomas) conferred by formal educational institutions to monitor
entry into occupations.” This definition picks up Weber’s idea that the access to a certain
social relationship (in this case a certain occupation) is restricted through educational titles or
similar means regardless of the fact if somebody is capable of doing the job or not. Credentials
are either seen as a formalized proof of ability to fulfill certain tasks or as arbitrary ways to join
a particular group (Weeden 2002, 61).
An aspect similar to credentialism are professions. Abbott (1988, 7) concludes that most
scholars agree that a profession is an occupation which requires a special, but abstract skill
acquired through training. Dependent on different definitions of what drives professionalism,
professions are either seen as a social group or as an occupational class. The latter case
emphasizes external rewards to professions (Abbott 1988, 14). In addition professional orga-
nizations can lobby their own occupation to increase demand for their services or products
and to advocate better working conditions (Weeden 2002, 65). They can also try to channel
the demand for a certain service or product to their group by advocating that only they can
deliver it (Weeden 2002, 66). It is a key aspect that expert knowledge of professionals makes
it hard for consumers, the government, the public or other third party outsiders to assess the
quality of work. This allows the extraction of a rent. In the empirical assessment of closure and

14
Weeden (2002) argues that there are two additional mechanisms through which social closure influences
outcomes. The one is channeling demand to the group and the other signals quality of service
(Weeden 2002, 60). While this is a plausible argument I think these two additional mechanisms can be
categorized as sub-mechanisms of increasing demand for the asset provided by the group. Going into
detail is not necessary at this point. A simpler version of the argument serves just as well.
15
Unionization might be the most important indicator of closure on the labor market in Germany. However
the data set I use does not identify individual or employer bound unionization.

57
 2. Theory

health effects one way of measuring closure will be knowledge intensity of occupational groups.
This is not exactly the same as professionalism or credentialism, but is close to both concepts.
It is an indicator for how easy or hard it is to assess the actual productivity of a worker in
such an occupation. Also, it represents a reduction in supply of labor as a lot of workers are
excluded from competition due to a lack of occupation or profession specific knowledge and skills.

2.8. Open and Closed Positions and Health

In the preceding section, I have used the theory of open and closed positions to argue that
some positions on the labor market are not allocated by market mechanisms. In these closed
positions human capital theory has only limited power of explanation. This section provides
the link between open and closed positions and the theory of health selection. In the following
sections the impact of social closure on the relationship between health and labor market
rewards is further differentiated.

A relationship between labor market outcomes and health can so far only be established under
the assumption that labor market outcomes, like wages or promotions, are awarded according
to performance or productivity. The theory of social closure on the labor market suggests
that sometimes jobs constitute non competitive positions. This means that productivity or
performance is not the factor which decides whether some rewards are granted by the employer
or not. Other factors like age or tenure might come into play, regardless of the actual company-
specific human capital. For instance promotion to a high status job could occur not due to the
good performance of a worker, but as a result of bureaucratic regulations16 . These regulations
state that a worker is designated for promotion at a certain time without a clear link to observed
performance. If the rewards do not depend on the productivity, then it follows that they should
be unrelated to health.
The two hypotheses which follow from this section are: In jobs with high potential for social
closure (e.g. in the public sector) health does not affect the probability of job loss or high
status attainment.
On the other hand, in jobs which are highly competitive (outside competition) health has a
substantial influence on labor market outcomes.
A formalization of the argument would give two options. In a fully closed position the original
model we were using

LM R = αϵσ t (2.19)

16
in the labor contract, laws for civil-servants, or collective agreement.

58
 2. Theory

would transform to:

LM R = αϵ0 t (2.20)

In this case labor market rewards are determined regardless of effort. The effort intensity is
reduced to zero. If we allow for a more continuous view on closure we can introduce a factor
which represents the openness of a job, meaning an inverted scale of the social closure factor c.
If c takes the value 0 the position is completely closed. If it takes the value 1 it is completely
open.

LM R = αϵσ∗c t (2.21)

Deriving the marginal effect of health on labor market rewards we get:

1
c ∗ σϵ(σ∗c−1) αt = c ∗ σ αt (2.22)
(e(H, EN LM ) ∗ τ )(1−σ∗c)

Here we can clearly see that if c takes the value 0 the marginal effect of effort and therefore
of health will be zero as well. The higher the value of c the stronger the effect of effort, and
in consequence health, on labor market rewards. The formula therefore leads to the same
conclusion as the intuitive argument: The more closed a position is the smaller should be the
impact of health on labor market rewards.

2.8.1. Open and Closed Positions: Incumbents vs. Applicants

I have treated the relationship between health and labor market rewards as if labor market
rewards can be awarded in a continuous way like wages. However for high status jobs this
assumption does not hold. It might be true that the probability of attaining such a position
can be altered continuously with regard to health status. The question whether one does or
does not get a reward cannot be defined in such a way. One cannot get a promotion just a bit
or a little bit more. It’s all or nothing.

2.8.1.1. How the Choice of Labor Market Outcome Affects the Role of Health

A distinction can be made between the classical measure of labor market generated income
which is continuous and a dichotomous reward like a high status job17 . Such a distinction
only makes sense if we expect a non-perfect market. If all jobs were completely open than the
distinction between high and low status jobs would be irrelevant as all relevant information
about the importance or value of a job would be included in the wages. However, if we accept

17
Which is of course often associated with an increase in wages, but not necessarily and the degree to
which it is varies a lot.

59
 2. Theory

that some jobs (especially high status jobs) cannot just be created and eliminated at will, then
a decision to give such a position to a person implies uncertainty and risk for the employer. In
open positions hiring someone is not a risk for the employer, because he can fire the person at
any time.
High status jobs can be seen as closed positions, because they provide important functions,
supervision, and expert knowledge. All of these cannot simply be created and abolished anytime
the employer wants. Other jobs in companies depend on these jobs. Incumbents have strong
bargaining position, because of the importance of the position. Hiring someone in a closed
position involves the risk of hiring someone who is not up to the task which is contrary to
employers’ preference for maximizing profit. This is especially problematic as the candidates
for a job will highlight their strengths and try to hide their weaknesses (Williamson 1973,
319). Such a strategy improves their bargaining position as applicants. Of special relevance is
this behavior in fields and for positions where productivity cannot directly be measured and
employers need to look for indirect signs of quality in the applicant. Not all candidates might
overstate their potential usefulness to the employer. Nevertheless, it is sufficient to assume
that some persons will act opportunistic in such a situation to warrant uncertainty for the
employer (Williamson 1981, 553). The employer can never be sure, that the person under
review is actually capable of doing what he claims. Therefore special attention will be paid
to supposed signals of performance. That encourages statistical discrimination, which in turn
facilitates gender differences in health effects (Sørensen 1983, 210).
Analyzing job status also allows to distinguish between the position of the applicant and the
incumbent and how this modifies the role of health (section 2.8.1.2). As I will argue below
it makes a systematical difference whether one applies for a high status job, or whether one
already occupies such a position. The latter is much more to the benefit of the employee.

2.8.1.2. Incumbents vs. Applicants

If we accept the argument that incumbents and applicants have a structurally different bargain-
ing position, we can raise another question. Does it make a difference for the role of health
if I want to attain a position or if I want to keep it? At first thought one might think that
the mechanisms should be the same. Highly productive workers get the job, workers with low
productivity lose it. End of story. However, the preceding sections about different mechanisms
of allocation and rewards in open and closed positions warrant a further investigation into the
matter.
Regarding health effects, I argue that it is necessary to make a distinction between the role of
an incumbent and an applicant. In the following, I will explain why and how this allows me to
formulate more precise hypotheses.
I assume that employers have a preference to reduce risk of productivity loss. This preference

60
 2. Theory

is higher than the preference for maximizing profit in the long run, as most employers18 cannot
or are not willing to risk short-term productivity losses which might arise from incompetent
employees in important positions. Most employers cannot try out 5, 10, or even more employees
for a job until they find the perfect match. The backside of the coin is that they might hire
on average lower performing employees, but with less variance in performance. Assessment
centers are a way of circumventing this problem to some degree. But an assessment center also
presents a large investment on the side of the employer and requires a lot of know-how to be
effective. A lot of employers might not be able to handle this. Given employers’ preference to
reduce risks it seems plausible that selection criteria for open positions are less strict than for
closed positions. In open positions hiring an unproductive employee can be corrected after a
short period of time, in closed positions this is by definition very hard as the match is indefinite
(Sørensen 1983, 209).
Therefore health should play a minor role in getting into jobs in open positions. Employers do
not need strong signals of high performance, because they can easily get rid of employees who
are ill and show low performance. The other side of the argument is that if you are occupying
an open position, health is of major importance. Employers can replace you at any time, so
reduced performance will be penalized immediately.
For closed positions the argument goes the other way around. The criteria for entry into a
position are also an indicator of the openness of a position. So, if demands in form of e.g.
diplomas, tests, work experience, and most important for my example physical fitness are high
then the position has a high degree of closure (Doeringer & Piore 1985, 47). This suggests
that health is very important when applying for a job in a closed position. The importance of
health in the application process can further be supported by job search models. In this view
workers have to make an investment in their search (usually in form of reduced income). If
they are ill their search will be less effective, raising opportunity costs and reducing the chance
of finding a good job. This includes the application for promotion. It will also reduce their
performance in interviews or other forms of assessment (Paul & Moser 2009, 268). They will
have less energy for preparation or to look for alternatives. They might be in a worse position
for negotiations as well.
We can therefore view applying workers as being in a situation of career tournament (Rosenbaum
1979) or job competition (Thurow 1975). They need to show the absolute highest performance
to get the job. They will have to take the performance of others into account, because it will
influence their own ranking in the view of the possible employer. This makes it very important
that they are in good health. However, once occupying a closed position performance and
rewards of the incumbent are detached. Job loss, pay cuts, and demotions are a minor risk.
Therefore the performance they show does not need to be high. They have no further need to
be ranked as number one. This makes it less important that they are completely fit. Illnesses
can be easier compensated, and consequently health is of little importance to the question of
18
Who lead medium and small businesses.

61
 2. Theory

whether they keep their position or not.

The last point has to be differentiated. If the person occupying a closed position has
aspirations to move further up the job ladder, things look different. Even if further aspirations
exist, job loss and pay should still be detached from health. But if another promotion or a better
job is desired, then incumbents of closed positions find themselves again in a job competition
or career tournament and the process starts again only this time with different contestants.
This potential for climbing the ladder within a firm or bureaucracy has been argued to be one
of the strongest incentives for high performance that employers can use for employees in closed
positions (Sørensen 1983, 211). In addition, such incentives based on seniority promotions can
be efficient for employers if they fear costs of labor turnover (Carmichael 1983). One might
even expect that the higher on the career ladder a person enters such a tournament the more
important personal health is, because the fiercer the competition for the ever reduced number
of closed positions will be. Such a mechanism can be superimposed if social contacts, not
performance is the main criterion for ranking. The literature reports that social contacts are
an important factor for the highest of managerial positions. Then a strategic position in the
old boy networks (Oakley 2000) is more important than physical or mental health.
In conclusion, we can say that health is important if you want to get into a closed position
(applicant) - regardless of whether you are in an open or closed position at the moment. The
important part is that you enter the competition for a position. Health is also important if
you want to keep a job in an open position (incumbent). On the other hand health is of no
consequence if you want to attain an open position or want to keep a job in a closed position
(see table 2.1 for a schematic overview).

2.8.1.3. Open and Closed Positions and Discriminated Groups

The theory of open and closed positions can be further modified allowing for disadvantaged
groups. For this purpose, I make a distinction between the degree of closure of a position
towards different groups. The distinction between applicant and incumbent will also be upheld.
Discrimination increases the degree of closure of a position for a certain group if they try to get
into such a position (applicant). This corresponds to an argument made already by scholars
from segmented labor market theory. One way of discrimination in closed positions is entry
discrimination. Hiring standards, screening criteria, and recruitment practices can be used to
directly or most often indirectly select the discriminated from the main group (Doeringer &
Piore 1985, 137-140).
However, once in the position, it will be less closed, because the employer has reduced
opportunity costs of firing the incumbent as long as he or she has a taste for discrimination
against the incumbent. The argument holds for employers with a taste for discrimination,
because this taste should persist after employing a disadvantaged person. If discrimination

62
 2. Theory

Table 2.1.: Expected Effect of Health Depending on the Type of Labor Market Position
Status of individual Applicant Incumbent
Type of labor market position
Open no yes
Closed yes no

Table 2.2.: Change of the Degree of Closure of a Labor Market Position Depending on
Discrimination
Discrimination no Discrimination
Incumbent more open/none none
Applicant more closed none

does not persist the position of the incumbent is not influenced. This shows that it makes a
difference whether statistical discrimination is the relevant mechanism. If it is, discrimination is
expected to vanish after hiring, because then the employer can assess the individual and does
not have to draw on averaged information from the individual’s group. In the case of taste for
discrimination the taste should persist and so should the discriminatory practice (see table 2.2
for a schematic overview).
We can conceptualize this as the overlaying of two processes of social closure. Discrimination
is one. The other is a not nearer defined process of job closure. This can be an occupational
closure mechanism as described in section 2.7.3 or a high status job, which is closed due to its
high degree of specialization and responsibility involved. The two processes of social closure
reinforce each other.
On the other hand, membership in a discriminated group means that certain positions are
closed against the discriminated individual. When occupying such a position (e.g. a high status
job) it is less closed for all other non-discriminated persons, because, as mentioned above,
employers have reduced opportunity costs of firing such a person.
Here we can refer back to the argument that women have a harder time getting into a high
status job than men. They have to compensate the fact that they are discriminated against by
showing more effort to get into a closed position. In addition, when occupying a high status
job, health is more important for women, because they are more likely to lose their position
again.

2.8.2. The Benefits of the Theory of Open and Closed Positions for
a Theory of Health Selection
At this point, I will briefly review how the theory of open and closed position enriches a theory
of health selection. My argument will be on two levels. First, I will make a general comment
on the usefulness of contextualizing health selection theory. Second, I will briefly review the

63
 2. Theory

concrete hypotheses derived from a combination of the theory of open and closed positions
and health selection theory.

The theory of health selection states that employers select healthy workers to receive higher
labor market rewards, because they are more productive. This assumes that there is competition
between workers for jobs and between employers for productive workers. The theory of open
and closed positions now proposes that not all social positions are competitive. In fact, closed
position are exempt from competition. If there is no competition employers lose their ability
for selection according to productivity criteria. Workers’ health can therefore no longer be
a selection criterion. This results in the fact that within closed positions health inequalities
cannot be generated through health selection processes. Still there can be the chance for
selection between open and closed positions. Health inequalities between open and closed
positions can therefore be potentially explained by health selection.
The theory of open and closed positions is therefore useful to highlight in which areas of society
or specifically of the labor market we can expect health selection processes to take place and
where we would not expect them.

In addition to this more general benefit, there are several theoretical propositions developed
that apply to the labor market. We expect that health should not be a selective factor for
high status jobs in closed positions. In the empirical example this will be exemplified with
a differentiation between public and private sector, professionalized (knowledge intensive)
occupations, and male vs. female dominated occupations.
Selection processes play a role when workers try to attain a high status job, but health should
be of lesser importance when retaining such a position. Such a differentiated hypothesis could
not have been derived from the basic theory of selection only based on Becker’s effort model.
Last, considering discriminatory practices as a form of social closure overlaying and modifying
other existing forms of closure leads to the hypothesis, that women might actually be selected
out of high status while men are not. Women face the risk of being selected out, because
taste for discrimination lowers opportunity costs of lay-offs for employers. On the other hand
men might face health selection while trying to attain a high status job to a lesser degree than
women do.

As I did several times before in the theory part, I let a formalized version follow the intuitive
argument. The formal equation of the health selection theory was:

LM R = αϵσ∗c t (2.23)

64
 2. Theory

Status of applicant and incumbent is not yet considered. Neither is the interaction of closure
and discrimination. Recall that c stands for the degree of openness of a job, ranging between 0
(closed) and 1 (open). Now we introduce the interaction of status of applicant (A) into the
equation. A takes the value 1 if the person is an applicant and 0 if the person is an incumbent.

LM R = αϵσ∗|A−c| t, A ∈ {0, 1} (2.24)

This formulation makes effort very important for applicants if the position is very closed (e.g.
|1 − 0.1| = 0.9) and unimportant if the person already holds the position (|0 − 0.1| = 0.1). The
opposite is true if the position is open. Now it is only a simple step to introduce discrimination
(d) as a modifying factor of closure.

1
LM R = αϵσ∗|A−c∗d| t, 1 ≥ d ≤ (2.25)
c

The last equation shows that the higher the discrimination is the more open a position is as
an incumbent, and the more closed it is as an applicant.

Another field for interesting applications of closure theories is research on health inequalities
related to educational attainment. Here the focus would be on the extend in which school,
university, apprenticeship, or other forms of institutionalized education can develop selective
mechanism which are related to the performance of young persons in these institutions. This
performance could then be linked to their health status, so that health inequalities generated
by health selection seem to be very plausible in this context. Certain educational institutions
are subject to more or less social closure, so that a heterogeneity in the selective function of
health for educational attainment could be expected.
We can see that the limits of what kind of research could be stimulated by a combination of
the health selection theory and the theory of open and closed positions have not nearly been
reached in this thesis.

2.9. Summary, Assumptions and Hypotheses

2.9.1. Summary and Hypotheses

In this part, I will sum up the arguments made so far. I put a special emphasis on highlighting
the assumptions underlying the theory and I list them separately with reference to the sections
where they are elaborated.

65
 2. Theory

The theoretical developments aim to explain how structural determinants modify micro-
mechanisms of health selection on the labor market. I build on an effort and labor market
rewards model by Gary Becker which implies that labor market rewards are awarded according
to productivity and that effort has a decreasing return on productivity (Assumption 2 & 3).
I expand the theory by arguing that effort in the model is a function of health (Assumption
1). Paying employees only up to their productivity is a rational behavior of employers in a
functioning labor market under the implicit assumption that employers want to maximize their
profit (Assumption 4).
Concluding from the assumptions 1-4 we can state the first hypothesis which motivates the
whole study:

H1: Health influences labor market rewards. In consequence, better health increases the
chance of high job status attainment.

In evaluating the productivity of a worker, employers are biased towards underrating women’s
performance compared to men’s (Assumption 5). Further, I assumed that men and women
have different preferences for health in relation to career (Assumption 7) and that workers
adjust their effort at work after a health shock dependent on this relation (Assumption 6).
Given higher preference for health in relation to career among women I deduct that women are
more reactive to health shocks in reducing effort at work than men. I also state the auxiliary
assumption that there are no gender specific learning processes in effort adjustment after health
shocks (Assumption 10). Another way of testing the general health selection hypothesis is to
see if the effect of a negative health status gets stronger the longer the person is within this
state, because longer periods of bad health are harder to compensate (Assumption 16). The
respective hypotheses are:

H2a: Health has a stronger influence on labor market rewards for women than for men. The
effect of health on the chance of high job status attainment is stronger for women than
for men.

H2b: The longer a negative health condition lasts, the stronger the impact is on job status.

I assume that employers also have a preference for risk avoidance which outweighs the
preference for profit maximization (Assumption 8). Drawing on the theory of open and closed
positions, I propose that the higher the degree of social closure, the higher is the employers
risk when employing someone with at least partly unknown productivity (Assumption 9).
High status jobs are positions with a high degree of social closure, which in turn constrains
demotions and layoffs (Assumption 11). Jobs in the public sector (especially high status jobs)
are treated as fully closed positions (Assumption 12). Taking into account the employers’
tendency for risk avoidance health is of greater importance for applicants to closed positions
and for incumbents of open positions. The following hypotheses are derived:

66
 2. Theory

H4: The chance of dropping out of a high status job is not related to health.

H5: In the public sector health does not affect the chance of high status job attainment.

For women as a disadvantaged group a social position is reduced in its degree of closure if
the person is an incumbent and the degree of closure is increased if the person is an applicant
(Assumption 13). This leads to a hypothesis that is in explicit competition to hypothesis 4:

H6: Dropping out of a high status job is related to women’s health, but not to men’s health.

It is expected that high degrees of closure lead to decreasing importance of health for
job status. This is exemplified with the degree of knowledge intensity of occupations as an
occupational closure mechanism. Two further hypotheses can be formulated with respect
to social closure in occupations and health selection. In occupations with a high degree of
gender specific closure two alternative hypotheses arise. The homohily-hypothesis is based
on the assumption that homophily exists in female dominated occupations (Assumption 14a).
The alternative hypothesis (glass-escalator hypothesis) rests on the assumption that there is a
“glass-escalator” effect for men in female dominated occupations (Assumption 14b).

H7a: The higher the degree of knowledge intensity, the lower the effect of health on job status.

H7b: Women’s health is less important for high status job attainment in female dominated
occupations than in mixed or male dominated occupations.

H7c: Men’s health is less important for high status job attainment in female dominated
occupations than in mixed or male dominated occupations.

The last hypothesis refers to a different way of measuring health. It states that for the
employer visible health problems might follow a different logic, because consequences for
productivity are easier to judge for an employer. Specifically, I postulate that the days of
sickness absence are easy to monitor. Given a male culture of presenteeism (Assumption 15)
days of sickness absence should have more negative effects for men than for women, contrary
to the pattern of effects of general health.

H8: The effect of the number of days of sickness absence on the chance of high status job
attainment is stronger for men than for women.

The preceding hypotheses are all modifications of the general health selection hypothesis.
Social causation pathways and mechanisms are not discussed in detail in the theory chapter.
However, at a later stage of the empirical analyses I will test social causation mechanisms as
well. The respective hypotheses are:

H9a: Incumbents of high status jobs have better health due to higher resources, better
employment and working conditions.

67
 2. Theory

H9b: Incumbents of high status jobs have better health due to social-psychological factors like
fear of job loss and satisfaction with work.

H9c: Incumbents of high status jobs have better health, which has to be attributed to common
background factors leading to spurious correlation.

H9d: Incumbents of high status jobs have better health, which cannot be explained by health
selection, resources and working conditions, and background factors. This is a social
status effect on health.

68
 2. Theory

2.9.2. Set of Assumptions

Assumption 1: The effort a person can show at work depends on the person’s health. In this sense effort
is an (unknown) positive function of health (section 2.4.2).

Assumption 2: Labor market rewards are awarded according to perceived productivity. Perceived produc-
tivity is determined by human capital, perceived effort, effort-intensity of the job, and
work hours (section 2.4.2). High job status is seen as a labor market reward.

Assumption 3: Effort has a decreasing return on productivity (section 2.4.3).

Assumption 4: Employers want to maximize their profit (section 2.5.1).

Assumption 5: Employers perceive a certain amount of effort made by a woman as less than the same
amount of effort made by a man (section 2.5.2).

Assumption 6: All individuals have a known preference for health and career. The adjustment of effort
at work depends on the relation of the preference for health and work (section 2.6).

Assumption 7: Women have relatively higher preferences for health in relation to career than men
(section 2.6.1).

Assumption 8: Employers’ preference for risk avoidance is higher than their preference for profit maxi-
mization (section 2.8.1).

Assumption 9: The more a job is subject to social closure the higher is the employer’s risk when hiring
someone for the job (section 2.8.1).

Assumption 10: There is no gender specific learning process in dealing with health shocks (section 2.6.1).

Assumption 11: High status jobs are subject to social closure. Employers cannot create and abolish these
positions (which means promoting or demoting an employee) at any given time nor can
they hire or fire their incumbents at will. High status jobs are not completely closed
positions either. That means that promotions, demotions and layoffs are still possible,
but are constrained (section 2.8).

Assumption 12: In the public sector high status jobs are fully closed positions where Assumption 11 does
not hold (section 2.8).

Assumption 13: Members of disadvantaged groups experience modified degrees of social closure of
positions. As an incumbent the position is more open, as an applicant the position is
more closed (section 2.8.1.2).

Assumption 14a: There is gender specific homophily within occupations (section 2.5.2).

69
 2. Theory

Assumption 14b: Men in female dominated occupations have an exceptional status, allowing them to reach
high job status more easily (section 2.5.2).

Assumption 15: Men are subject to a culture of presenteeism at work which requires them to be present
at work even when they are ill. Deviations from this norm are sanctioned (section 2.5.4).

Assumption 16: Longer periods of bad health are harder to compensate than short periods (section 2.4.2).

70
3. Review of the Empirical Literature

3.1. Overview and Development of a “Good Practice”

Approach
In this part of the dissertation, I will review the empirical literature which tries to test the
health selection hypothesis. Studies from the fields of medical sociology, public health, medical
psychology, health economics and epidemiology are taken into account.
This review is not exhaustive, but still captures the most important literature. It includes a
large table listing all studies discussed with several relevant criteria as an overview.

For this review I searched for studies dealing with the health selection hypothesis. In addition,
I looked for economic literature which does not use the term, but still estimates effects of health
on labor market outcomes. I restricted my search to labor market outcomes and socio-economic
status (SES) as the dependent variables. In other areas health selection might also be present
(marriage, educational attainment, migration, residential mobility, etc.), but due to a necessary
reduction of complexity I will focus only on labor market and SES.

Overall, my search lead to a total of 76 studies, which claim that they test the health
selection hypothesis. First, we can take a look at the range of outcomes the literature dealt
with. Most of the studies use unemployment (19), or employment (15) as an outcome. Another
big part of the literature (mostly from economics) focuses partly or entirely on wages (17). 14
studies dealt with occupation or occupational classes and 8 studies with household income.
The other studies had different (additional) dependent variables in their study design. Among
them are for example financial strain, wealth, hours worked, job loss, tenure, or temporary
employment.
22 studies used a measure of SES, though it is not always measured in the same way. A lot of
studies relied on occupational classifications or household income, sometimes supplemented by
educational attainment.

I also looked at how health is measured. The most frequently used indicator of health (28) is
self-rated health or subjective health (SRH). A lot of studies refer to the standard 5-point-scale
SRH question, but not all do so.

71
 3. Review of the Empirical Literature

Mental health is the specific aspect of health, which is used most often in the literature (20).
However, under the umbrella term of mental health there the studies employ a wide range of
mental problems and measures as their outcomes. This ranges from depression and mood to
mental health from the SF-12 score. Assorted physical symptoms (13) or specific diseases (4)
are widely used as well. What is true for mental health is true here as well. No standardized
measure was used. Another block of the literature asked for limitations in every day or work
activities or physical mobility (11). Other health measure that are used are hospitalization
(4), accidents or injuries (3), and sickness absence (5). A topic on its own is the literature on
differentials in mortality. Most studies in mortality do not discuss the health selection issue.
Those who do (8) cannot, for obvious reasons, use mortality as a predictor of labor market
outcomes or SES. It is rather several indirect ways in which they try to assess health selection,
arguing that negative labor market outcomes and mortality might both be influenced by bad
health prior to death.

It is also of interest to look in which country the studies are conducted. The country presents
the context of the labor market and the general level of competitiveness and solidarity in the
society. Not surprisingly given the size and scientific impact, we can see that most studies are
conducted in the United States (23). After the US the Scandinavian countries and Finland
are clearly overrepresented given their modest size (SE 10, NO 8, FI 6, DK 1). The reason
is probably that the issue of health inequalities is considered to be very important in these
countries. Researchers also have access to register data as in almost no other countries, which
explains the huge number of observations in their studies. A lot of studies are done in the
United Kingdom (12), in Germany there are 5, same as in the Netherlands. The other studies
are conducted in various European countries, Canada, and Australia. Regarding data sets, all
German studies used the SOEP, and several of the UK studies used the British Household
Panel Study (BHPS) which is fairly similar to the SOEP in design.

One other important feature is whether studies directly tested health selection or whether
they inferred its existence or absence indirectly. Most studies conduct direct tests of the health
selection hypothesis (59). 32 studies claim to assess health selection versus social causation,
although not all studies conduct a direct test of this causal relationship.
I propose that a “good practice” approach would be to assess both hypotheses in one study,
and conduct a direct test of both hypotheses. There are 20 studies which fulfill the “good
practice” standard. As the “good practice” studies are the core studies in the field, I will go
into more detail to describe each of them.

72
 Table 3.1.: Overview of Studies Involving Health Selection

Study Year HS DT VS NumbObs Country SES LM Measure SRH Health Measure Econ Journal Age Long GP
Aittomäki, Mar- 2012 yes yes yes 211,639 FI yes unemploment,wages no sickness absence no SSM 17-66 yes yes
tikainen, Laakso-
nen, Lahelma &
Rahkonen (2012)
Arrow (1996) 1996 yes yes no yes DE no unemployment no sickness leave, no SSM 18-64 yes no
chronic disability
Bartel & Taubman 1979 yes yes no 5000 US no wages no diseases, symptoms yes TRES no
(1979)
Bartley (1991) 1991 no no no 500000 GB no unemployment no mortality no JSP 15-64 yes no
Baum & Ford 2004 yes yes no 12000 US no wages no obesity yes HE 14-40 yes no
(2004)
Berkowitz & John- 1974 yes yes no 900 US no wages no limitations yes JHR 25-64 no no
son (1974)
Black, Devereux & 2005 yes yes no 5000 NO no wages no birthweight yes NBER 16-74 yes no
Salvanes (2005)
Böckerman & Il- 2009 yes yes yes 19206 FI no unemployment yes no yes HE yes yes

73
makunnas (2009)
Buddelmeyer & Cai 2009 yes yes yes 1769 AUS no Income poverty yes yes IZA 18-64 yes yes
(2009)
Burgard, Brand & 2007 yes no no 6115 US no jobloss yes depression yes JHSB 15-35 yes no
House (2007)
Cai & Kalb (2006) 2006 yes yes yes 9000 AUS no employment yes no yes HE 15-64 yes yes
Cai (2010) 2010 yes yes yes 4669 AUS no employment yes yes LE 25-64 yes yes
3. Review of the Empirical Literature

Cardano, Costa & 2004 yes yes no 127384 IT yes occ. classes, no hospitalization no SSM 25-49 yes no
Demaria (2004) MIOM scale,
unemployment,
non-employment
Carlsen, Dalton, 2008 yes yes no 380000 DK no unemployment no cancer survivor no EJoC 10-60 yes no
Diderichsen &
Johansen (2008)
Chandola et al. 2003 yes yes yes 10308 UK no employment grade, no physical, mental no SSM 35-67 yes yes
(2003) financial depriva-
tion
 Table 3.1 – Continued from previous page
Claussen, Bjørndal 1993 yes yes yes 5000 NO no unemployment no mental health no JECH 16-70 yes yes
& Hjort (1993)
Contoyannis & 2001 yes yes no 1296 US no wages yes mental health yes EE yes no
Rice (2001)
Crichton, Stillman 2011 yes yes no 85390 US no employment, wages no injury yes ILRR 15-69 yes no
& Hyslop (2010)
Dahl (1993a) 1993 yes no yes 2200000 NO yes occupation no mortality no EJPH 20-64 yes no
Dahl (1993b) 1993 yes no no 420000 NO no occupations no mortality no JECH 20-64 yes no
Dahl & Kjærsgaard 1993 yes no no 2215 NO no employment, occu- no mortality limita- no SSM 25-66 yes no
(1993) pation tions
Duguet & 2012 yes yes no 3618 FR no job loss, employ- no chronic illnesses ac- yes LAMETA 19-59 yes no
Le Clainche (2012) ment cidents
Eaton, Muntaner, 2001 no yes yes 907 US yes labor/HH-income, no depression no JHSB 16-64 yes yes
Bovasso & Smith job percentiles,
(2001) social benefits
Elstad & Krokstad 2003 yes yes yes 9189 NO no occ. class yes no no SSM 25-59 yes yes
(2003)
Elovainio, Ferrie, 2011 yes yes yes 8312 UK yes employment grade, no cardiometabolic no AJE yes yes

74
Singh-Manoux, promotion biomarkers, child-
Shipley, Batty, hood hospitaliza-
Head, Hamer, tion, birth weight
Jokela, Virtanen,
Brunner, Marmot
& Kivimäki (2011)
Ettner, Frank & 1997 yes yes no 4626 US no wages,employment, no psychatric disor- yes NBER 18+ no no
Kessler (1997) hours of work ders
3. Review of the Empirical Literature

Fox, Goldblatt & 1985 no no yes 500000 GB yes no no mortality no JECH 15-64 yes no
Jones (1985)
Fox (1990) 1990 no no no US yes no no mental illness no JHSB no
Frijters, Johnston 2010 yes yes no 35000 AUS no employment no mental health SF36 yes IZA 22-64 yes no
& Shields (2010)
Gambin (2005) 2005 yes yes no 200000 EU no wages yes chronic illness yes HEDG yes no
Garcı́a-Gómez, 2010 yes yes no 5500 GB no employment yes limitations yes LE 16-59 yes no
Jones & Rice
(2010)
Haan & Myck 2009 yes yes yes 4420 DE no employment yes no yes JHE 30-59 yes yes
(2009)
 Table 3.1 – Continued from previous page
Haas (2006) 2006 yes yes no 2805 US yes wages, wealth, ocu- no birthweight no JHSB 16-64 yes no
pation
Haas, Glymour & 2011 yes yes no 6155 US no wages yes childhood health no JHSB 25-50 yes no
Berkman (2011)
Halleröd & Gustafs- 2011 yes yes yes 2976 SE yes SIOPS, HH-income no limitations diseases no SSM 31-63 yes yes
son (2011)
Hammarström & 2005 yes yes no 1083 SE yes blue collar no weight smoking al- no SSM 16-30 yes no
Janlert (2005) cohol etc.
Harkey, Miles & 1976 yes no yes 16569 US no HH-income no limitations no JHSB 6-65 no no
Rushing (1976)
Haveman, Wolfe, 1994 yes yes yes 4640 US no wages, hours of no limitations yes JHE 25-65 yes yes
Kreider & Stone work
(1994)
Heponiemi, Elo- 2007 yes yes yes 90000 FI no unemployment no mental health sym- no JPR 20-45 yes yes
vainio, Mander- potoms digestive
backa, Aalto, sytsem
Kivimäki &
Keskimäki (2007)

75
Hofoss, Dahl, El- 2012 no no no 2261076 NO no income deciles no mortality no EJPH 25-66 yes no
stad & Cvancarova
(2012)
Huurre, Rahkonen, 2005 yes yes yes 1262 FI yes occupation, educa- no psychosomatic dis- no SPPE 16-32 yes yes
Komulainen & Aro tion tress
(2005)
Jäckle & Himmler 2010 yes yes no 14100 DE no wages yes no yes JHR 18-65 yes no
(2010)
3. Review of the Empirical Literature

Jones, Rice & 2010 yes yes no 1135 GB no retirement yes limitations yes EM 50-65 yes no
Roberts (2010)
Jusot, Khlat, 2008 yes yes no 5807 FR no unemployment yes obesity smoking no JECH 30-54 yes no
Rochereau &
Serme (2008)
Ki, Sacker, Kelly & 2011 yes yes no 7171 GB yes employment yes no no JECH 30-64 yes no
Nazroo (2011)
Kivimäki, Vahtera, 2003 yes yes no 886 FI no employment yes mental symptoms no AJCP yes no
Elovainio, Pentti & sickness
Virtanen (2003)
 Table 3.1 – Continued from previous page
Klein-Hesselink & 1992 yes no yes 1000 NL yes unemployment, no chronic illness, no IJE 30-50 yes no
Spruit (1992) education, occupa- symptoms, depres-
tion, HH-income sion
Koskela, Luoma & 1976 yes yes no 1789 FI no turnover no diseases symptoms no SJWEH 15-74 no no
Hernberg (1976)
Lawrence (1948) 1948 yes no yes 1010 US yes subjective class no chronic illness no yes no
Lee (1982) 1982 yes yes yes 2800 US no wages yes limitations yes IER 45-59 yes yes
Leino-Arjas, Li- 1999 yes yes no 586 FI no unemployment no mental health no BMJ 40-59 yes no
ira, Mutanen, health behavior
Malmivaara &
Matikainen (1999)
Lichtenstein, Har- 1992 yes no yes 758 SE yes occupation, educa- yes chronic illness no SSM 26-87 no no
ris, Pedersen & Mc- tion, material re-
Clearn (1992) sources
Lundberg (1991) 1991 no yes no 2957 SE yes occupation no symptoms, abs- no ESR 20-64 yes no
cence
Lundborg, Nilsson 2011 yes yes no 1100000 SE no wages no hospitalization yes IZA 30-59 yes no
& Vikström (2011)

76
Lundin, Lundberg, 2010 yes no yes 49321 SE no unemployment no mortality no JECH 20-54 yes no
Hallsten, Ottosson
& Hemmingsson
(2010)
Magee (2004) 2004 yes yes no 19000 CA no jobloss no illness, disability no SSM 16-69 yes no
Manor, Matthews 2003 yes no yes 11405 UK yes occupation yes absence no SSM 23-33 yes no
& Power (2003)
Mastekaasa (1996) 1996 yes yes no 2119 NO no unemployment no mental health no JCASP yes no
3. Review of the Empirical Literature

(chronic) illness
McDonough & Am- 2001 yes yes no 5378 US no employment yes no no SSM no
ick (2001)
Meerding, IJzelen- 2005 yes yes no 560 NL no productivity yes eq5d pcs-12 mcs- no JCE no no
berg, Koopman- 12 mental symp-
schap, Severens & toms
Burdorf (2005)
Montgomery, 1996 yes yes no 2256 GB no unemployment no height, social ad- no JECH 22-32 yes no
Bartley, Cook & justment
Wadsworth (1996)
Mulatu & Schooler 2002 yes yes yes 705 US yes HH-income, educa- distress, sleep no JHSB 41-88 yes yes
(2002) tion, occupation
 Table 3.1 – Continued from previous page
Mullahy & Sindelar 1991 yes yes no 4800 US no wages, employ- no alcoholism yes TAER 30-59 no no
(1991) ment, HH-income
Ostlin (1988) 1988 yes no no 10800 SE no occupation no illness symptoms no JECH 24-74 yes no
Palloni, Milesi, 2008 yes yes yes 8512 GB yes education yes birthweight chronic yes CDEWP 7-46 yes yes
White & Turner conditions
(2009)
Paul & Moser 2009 yes yes yes 18000 META no unemployment, re- no mental health no JVB yes yes
(2009) employment
Pavalko, Gong & 2007 yes no yes 5066 US no employment no mobility no JHSB 14-44 yes no
Long (2007)
Pelkowski & Berger 2004 yes yes no 13957 US no wages, hours of no symptoms, limita- yes QREF 25-55 yes no
(2004) work tions
Perrott & Syden- 1935 no no yes 12000 US yes HH-income no chronic illness no AJS no no
stricker (1935)
Power, Matthews 1996 yes yes yes 17414 GB yes occ. group yes no no BMJ 7-33 yes yes
& Manor (1996)
Schmitz (2011) 2010 yes no yes 23734 DE no unemployment yes mental health hos- yes LE 28-58 yes no
pital

77
Schurer (2008) 2008 yes yes no 32224 DE no unemployment yes doctor hospital yes REP 40-60 yes no
Schuring, Burdorf, 2007 yes yes no 4446 EU no employment yes mental chronic dis- no JECH 16-65 yes no
Kunst & Macken- ability
bach (2007)
Smith (1999) 1999 yes no no 10236 US yes wealth yes no yes AER yes no
Stansfeld, Clark, 2011 yes yes yes 9377 GB yes occ. class tenure no depression, psycho- no SPPE 7-42 yes yes
Rodgers, Caldwell logical distress
& Power (2011)
3. Review of the Empirical Literature

Stewart (2001) 2001 yes yes no 5817 CA no unemployment no limitations yes JHE 16-64 yes no
Timms (1998) 1998 yes no yes 15000 SE yes occ. class no mental health no SSM 0-32 yes no
van de Mheen, 1998 no yes yes 2800 NL yes education, occ. yes mortality symtoms no 15-59 yes yes
Stronks, Looman class chronic
& Mackenbach
(1998)
Virtanen, Janlert 2012 yes yes no 1070 SE no temporary yes sleep, symptoms no AOEH 16-42 yes no
& Hammarström
(2013)
Virtanen, Janlert 2012 yes yes no 1083 SE no unemployment yes mood no PH 16-42 yes no
& Hammarström
(2012)
 Table 3.1 – Continued from previous page
Wagenaar et al. 2012 yes yes no 7100 NL no unemployment yes symptoms no JOEM 15-65 yes no
(2012)
Warren (2009) 2009 no yes yes 2394 US yes no yes symptoms, sickness no SF yes yes
abscence
Note: HS = Health Selection; DT = Direct Test; VS = Social causation versus Health Selection; LM = Labor Market; GP = “Good Practice”
Countries:US = United States; NL = Netherlands; DE = Germany; GB/UK = Great Britain/United Kingdom; NO = Norway; FI = Finland; AUS =
Australia; IT = Italy; DK = Denmark; FR = France; EU = European Union; SE = Sweden; CA = Canada
Journals: SSM = Social Science & Medicine; JECH = Journal Epidemiology and Community Health; JHSB = Journal of Health and Social Behavior;
JHE = Journal of Health Economics; HE = Health Economics; EJPH = European Journal of Public Health; EE = Empirical Economics; IZA = IZA
Working Paper Series; SF = Social Forces; JOEM = Journal of Occupational and Environmental Medicine; PH = Public Health; AOEH = Archives of
Occupational and Environmental Health; SPPE = Social Psychiatry and Psychiatric Epidemiology; AER = American Economic Review; REP = Ruhr
Economic Papers; LE = Labour Economics; AJS = American Journal of Sociology; QREF = Quarterly Review of Economics and Finance; JVB = Journal
of Vocational Behavior; CDEWP = Center for Demography and Ecology Working Papers; TAER = The American Economic Review; JCE = Journal of
Clinical Epidemiology; JCASP = Journal of Community & Applied Social Psychology; ESR = European Sociological Review; BMJ = BMJ Public Health;
IER = International Economic Review; SJWEH = Scandinavian Journal of Work, Environment & Health; EM = Economic Modeling; JPR = Journal of
Psychosomatic Research; AJCP = American Journal of Community Psychology; NBER = Working Paper of the National Bureau of Economic Research;
IJE = International Journal of Economics; AJE = American Journal of Epidemiology

78
3. Review of the Empirical Literature
 3. Review of the Empirical Literature

3.2. The “Good Practice” Studies in Detail

3.2.1. Studies from Medical Sociology and Psychology,

Epidemiology, Public Health, and Sociology of Health
Chandola et al. (2003) use the Whitehall II study from the UK to assess whether British white
collar civil servants’ health deteriorates when their job status does or whether it is the other
way around. Regarding the labor market outcome the study is closest to my measurement of
job status. It captures not mobility between occupations (which is often horizontal), but rather
exclusively vertical mobility within occupations. The study uses a structural equation modeling
approach with a cross-lagged panel structure. The health variables of interest are physical
and mental health from the SF-36 and GHQ questionnaire measured as latent variables. They
run their analyses separately for men and women, but do not control for any other factors in
their model. The authors find almost no effect of health on employment grade, but rather
robust results for the reversed causality. To sum it up, health selection is of minor importance
according to their study. The measurement of health was well done, their dependent variable is
similar to the one used in this study. The modeling is complex and catches the problem of
reversed causality. However, they do not control for spurious correlation through third factor
variables, and their sample comes from civil servants only.
Claussen, Bjørndal & Hjort (1993) use a sample of long term unemployed persons from 4
municipalities in northern Norway. They estimate the effect of health on chances of continuous
unemployment and recovery through re-employment. The health items they use are a checklist
of somatic symptoms, anxiety, and depression, and a score from the General Health Question-
naire (GHQ). Additionally, self-reported doctors’ diagnoses were used. All health indicators
suggested that there are health selection processes from unemployment to re-employment.
Most substantial results came from the GHQ measure, anxiety, and doctor’s diagnosis for
psychiatric disease or personality disorder. They also found some evidence that re-employment
led to recovery, but the effects were substantially lower than the health selection effects. A
strength of the study is that they used a number of important variables suggested by human
capital theory as controls. These include gender, SES, social networks, education, and labor
market experience. This reduces the chance of estimating spurious effects of health. The
limitations are clearly that they only have two time points, and the sample is restricted to a
small area in Norway, making generalizations difficult.

Power, Matthews & Manor (1996) use the British 1958 Birth Cohort study with 17,414
participants to analyze health selection and social causation. The study follows the subjects
from ages 7 to 33. They look at the influence of childhood health on subsequent adult SES
and of parents’ SES on adult health. They analyze health selection and social causation in
separate models looking at unadjusted odds-ratios. The choice of SES indicator is occupational

79
 3. Review of the Empirical Literature

group. They only consider persons in poor health in their health selection analysis disregarding
the possibility that good health might also lead to social mobility albeit in a positive way. The
health selection effect is characterized as negligible. Social causation is preferred as the way
to explain health inequalities in adult life. The study lacks controls for possible third factor
explanations and comparability of effects across models is questionable.

Eaton et al. (2001) look at the association of SES and depressive syndrome over the life
course. They estimate the effect of childhood SES on young adult depression, and of parent’s
depression on early adulthood SES. Additionally, the effect of early adulthood SES on depressive
syndromes in later life are examined, as well as early adulthood syndrome on later life SES. In
later life, job percentile according to the Nam method were used, as well as income percentile,
and financial dependence. The sample was taken from residents in east Baltimore in the US
resulting in a sample of only 907 subjects. Depression was measured by the Diagnostic Interview
Schedule. Results show almost no association of depression with SES over the life course,
neither as a result of SES nor as a selective factor. The strength of the study is the life course
perspective allowing for different mechanisms and stages of the SES-depression association.
It is also noteworthy that they use relatively fine grained measures of SES instead of broad
categories. The limitations lie again in the sample both restricted to a part of Baltimore and
the sample size. Only a very limited number of control variables (such as gender) were used.
Additionally, model selection was based on statistical significance of indicators, not theory
based, which yields problems of both statistical and theoretical nature.

The study by Elstad & Krokstad (2003) analyzes health inequalities in a longitudinal study
in Nord-Trondelag. They try to test social causation versus health selection with a time
lag of 10 years for a sample of 9,189 men aged 25-49 and 35-49 respectively. They use
multinomial logistic regression to evaluate the effect of subjective health on mobility between
three occupational groups, and into and out of employment. They use separate models, not
a joint estimation. Further they only use age as a control variable. Other ways of exclud-
ing third factor explanations are not used. The authors admit that occupational mobility
might be underestimated by the crude measurement. They conclude that widening health
inequalities among continuously employed persons should be attributed to social causation
while the widening health gap between employed and non-employed men is mostly a result of
health related mobility. The most obvious limitations of the study are its regional and gender
restrictions. Further improvement could have been made by controlling for third factors in
the regression models. The comparability of the coefficients between models is also problematic.

Halleröd & Gustafsson (2011) conduct a study in Sweden following subjects over 16 years.
They analyze different aspects of SES - similar to the study of Eaton et al. (2001) - and
their relationship with health. Their health indicator is morbidity measured by the report of

80
 3. Review of the Empirical Literature

longstanding illness, handicap or weakness. They use latent growth curve models to analyze
the association between initial SES and health, and trajectories of SES and health in further
life. They can show that both health selection and social causation are at work in their sample.
Initial SES is measured by occupation and household income. It predicts subsequent morbidity
and vice versa. Interesting is that they also test changes in SES and health as predictors
of changes in the other, respectively. This is a feature which latent growth curve modeling
allows them to do and constitutes a stronger test of the hypotheses, because it excludes time
constant unobserved factors as explanations for the association. They find strong evidence
for the association of changes in health with changes in SES. Their method does not allow,
however, to determine the direction of causality in this regard. They only control for gender and
age. The strength is the representative survey over a period of 16 years. Their complex model
allows to conduct stronger tests of their hypotheses than other studies do. The only drawbacks
are the small number of controls (and no stratification across gender) which is probably due
to the already complex model, and that they cannot establish a direction of causality in the
analysis of change of SES and health. Still, together with the study of Chandola et al. (2003)
this presents one of the best studies among the “good practice” studies with regard to data
and modeling.

A wide range of different diseases, and mental disorders among health care professionals in
Finland was used by Heponiemi et al. (2007) to assess health selection and social causation with
regard to unemployment. They linked several official register data of health care professionals
with employment statistics and hospital discharge registers to obtain their sample. They adjust
their logistic regression for age, income, marital status, and educational level, running separate
analyses for men and women. Unemployment had no consequences on hospital admissions.
Selection was found among women for diseases, and among men for mental disorders. They
refer to their analysis as cross lagged panel analysis. The strength of the study lies in the use
of objective data from registers and its huge sample size suggesting high quality of data. The
offside is that only health care workers were analyzed, other occupations were not included.
Especially with regard to health consequences it seems unlikely that health care workers expe-
rience of health problems can be seen as representative in its effect for the population in general.

Mulatu & Schooler (2002) use a nationally representative sample from the US to estimate
reciprocal effects of health and SES. They use a non-recursive SEM approach with health
and SES from 20 years ago as instruments for contemporary health and SES, controlling
for age, race, and gender. They allow the relationship to be mediated by behavioral factors
and find that weight and sleeping behavior mediate part of the health selection effect, while
health behavior and psychological distress mediate part of the effect of SES on health. Rather
broad categories of education, income, and occupation were used as measures of SES. The
indicators for health are self reported serious physical illness or injury, and the impact of health

81
 3. Review of the Empirical Literature

problems on the individual’s life. The strength of the study lies in the adequate modeling
of reversed causality and the analysis of mediating factors. Problematic are prior health and
SES as instrumental variables to identify the model under the equilibrium assumption and the
lack of temporal order in the association of SES and health. The sample size is rather small
with about 700 subjects included in the analyses. SES is measured by fairly time constant
factors like education or broad occupational classifications, allowing for little variation over time.

A meta-analytic study on the relationship of mental health and unemployment was conducted
by Paul & Moser (2009). They found a substantial difference in mental health between
unemployed and employed favoring the employed. The selection effect they found was smaller,
characterized by them as weak (less than half the size of the social causation effect). However,
the selection effect is almost as strong as the causation effect if only longitudinal studies
are taken into account, which is a feasible approach given the complex nature of the causal
relationship. Therefore my interpretation of the health selection effect would differ from the
authors’ view with regard to relative strength. A special feature of their meta-analytic approach
is that they can assess mediators of the association of mental health and unemployment. It
shows that e.g. high unemployment protection reduces the association, pointing to important
aspects of social context.

The study by Stansfeld et al. (2011) follows a similar design as the study of Eaton et al. (2001).
They assess the effect of childhood and early adulthood SES on depression and anxiety and the
reversed causality. Their sample comes from the 1958 Birth Cohort in England, Scotland, and
Wales. SES was measured by manual vs. non-manual occupational class. In adulthood, it was
also assessed by home owner status. Depression and anxiety were measured by the Revised Clini-
cal Interview Schedule. Their results point to a process of health selection and give some smaller
support to the social causation hypothesis. The study does not assess the reversed causality in a
SEM framework. Their SES measure is very crude and the analyses are only adjusted for gender.

Warren (2009) conducts a study using longitudinal data from Wisconsin to assess reciprocal
causality between SES and health. His approach is again similar to Eaton et al.’s (2001).
He follows the subjects over the life course from childhood to adulthood. He uses cross
lagged panel analysis in a SEM framework to test the hypotheses. SES was measured by
earnings, occupational education, and wealth. Health is measured as a mix of self-rated health,
muscular-skeletal symptoms, and depression. The author finds no health selective mechanisms
in his sample. Social causation, however, seems to play a role in the generation of health
inequalities. The measurement model and design of the studies are well done. However,
the model does not include any control variables, except for separate analyses for men and
women. The use of occupational education as indicator for SES can also be seen with skepticism.

82
 3. Review of the Empirical Literature

van De Mheen, Stronks & Mackenbach (1998) present the fourth study of “good practice”
with a similar design to Eaton et al. (2001). Their sample comes from the Dutch city of
Eindhoven and consists of a cross-sectional analysis with a follow up study. They measured SES
as highest education and occupational level (EGP) of the respondent and head of household.
Health indicators that were used included self-reported health, complaints, chronic conditions,
and mortality. They report substantive reductions in health gradients when controlling for
background SES, but no effect of health on subsequent occupational orientation or educational
achievement. They find the social causation hypothesis supported in contrast to the health
selection hypothesis. The modeling of reversed causality is less refined than in the SEM
approaches used by other studies. Their measures of SES are very broad and the link to health
selective mechanisms is unclear.

Palloni et al. (2009) use longitudinal data from Great Britain to look at the effects of
childhood health on educational and occupational attainment in adulthood. They restrict their
sample to boys. Birth weight is used as a very early measure of health. In addition they measure
chronic conditions in childhood and adolescence. SES is measured as self-reported occupational
ranking on a six point scale. Using a SEM approach they cannot find any indication of health
selection from children’s health status to occupational level in adulthood. While the study takes
an interesting life course perspective, their measure of SES is very broad and the proposed
hierarchy of occupational status is unconvincing.

Huurre et al. (2005) again use an approach which is similar to Eaton et al. (2001). Their
study is based on a sample of 1262 men and women in Tampere in Finland. They look at
the reverse effects of psychosomatic distress and SES over the life course. SES is measured
as manual vs. non-manual occupation, in midlife as educational attainment. Psychosomatic
distress was measured as a latent variable captured by 4 indicators of psychosomatic symptoms.
They use a SEM cross-lagged panel approach to estimate the relative effect sizes. The authors
find that both health selection and social causation play an important role. Health selection
from young age to education was especially important for men. It shows that no one model can
explain the association of SES and psychosomatic health. Overall the study was well conducted,
but lacks certain control variables to capture common background factors. The sample is also
restricted to one city in Finland reducing the possibility for generalization.

In a UK study 8,312 men and women were repeatedly screened for several biomarkers
indicating health status. In addition, childhood conditions were asked in form of hospitalization
and birth weight (Elovainio et al. 2011). Occupational class and promotions were assessed in
adult life. In this form the study explicitly tests health selection versus social causation. They
use separate models for the test of health selection and social causation. The authors adjust
their analyses for age, gender, and previous measures of the dependent variable. Other measures

83
 3. Review of the Empirical Literature

to exclude third factor explanations were not taken. They can show that childhood conditions
are associated with lower occupational status. On the one hand, adult health problems are not
associated with likelihood of promotion. On the other hand, adult downward mobility with
regard to occupation increases cardio-metabolic problems. The authors conclude that health
selection operates in younger years while social causation becomes more important in midlife.
A strength of the study is the use of a wide range of health indicators checking for sensitivity
of the results. However, third factor explanations are not controlled for and comparability
across models is difficult. Having said that, the authors used multiple imputation to deal with
missing values which is rarely done, but very commendable, because it is appropriate in most
observational studies.

The interesting study of Aittomäki et al. (2012) uses a cross-lagged SEM approach to assess
the dual causal relationship between SES and health. The data they use are register data from
Finland with a large nationally representative sample. They use sickness allowance days as a
measure of health problems. They assume that health has a direct impact on labor market
advantage measured by individual income and unemployment, but only an indirect effect on
future economic resources. This theoretical consideration is close to the argument made in
this thesis giving precedence to market generated inequalities as the outcome of differences
in health. They find that both health selection and social causation are supported by their
data. The strength of the social causation parameters are stronger, though. A novel feature
of their study is that they consider the impact of possible degrees of measurement error in
their variables on the estimates. They find that assuming a higher degree of measurement
error increases effects. This is not surprising, because it conditionally reduces variances in the
predictors leading to inflated coefficients. A better approach would be to vary the assumed
degree of measurement error unequally across measures based on theoretical considerations.
The strength of their analysis is the nature of the data and their complex statistical model,
which follows careful theoretical arguments.

3.2.2. Studies from Health Economics

Böckerman & Ilmakunnas (2009) analyze the relationship between unemployment and self-rated
health in Finland using the European Community Household Panel for Finland. Health is
measured via the standard self-rated health survey item. They use a difference-in-difference
estimation technique to assess whether unemployment affects health or the other way around.
In their diff-in-diff analysis and in various other model specifications they cannot find a sub-
stantial effect of unemployment on health. However, subjective health proves to be a predictor
of subsequent unemployment. The authors do not use causation or selection hypothesis as
terms - this is rarely done in economics -, but their results clearly support the health selection
hypothesis for Finnish unemployed. The strength of the study lies in its modeling techniques,

84
 3. Review of the Empirical Literature

which control for time constant unobserved heterogeneity, and some observed time variant
variables. On the other hand a SEM was not used.

A cross-sectional data set from Australia is the basis for Cai & Kalb’s (2006) analysis on the
relationship between labor force participation and health. The health item used is the common
self-rated health survey questions. They estimate simultaneous equation models for men and
women separately. Their models control for a wide range of labor market and household related
variables. They find a clear and strong effect of health on labor force participation, which is
stronger for women and older study subjects. Social causation is also found in the parallel
equation. Labor force participation is beneficial to subjective health status. The results point
to a clear feedback mechanism in which bad health reduces labor force participation, and being
out of the labor force reduces health status. The strength of their study lies in the usage of
high quality data and suitable modeling techniques including controls for spurious correlation.
The study is one of the best when it comes to modeling the complex relationship between
health and labor force participation. The cross-sectional nature of the study leaves room for
improvement with regard to modeling of temporal priority.
In a further study Cai (2010) uses longitudinal data to reanalyze his previous study. He
introduces a random effect structure to control for time constant unobserved heterogeneity.
Results show that health has a strong influence on labor force participation. The reverse causal
direction holds only for men.

Haan & Myck (2009) use the German Socio-economic Panel Study (SOEP) to assess the
relationship between health and non-employment. They use self-rated health as a health
indicator and pool unemployed and non-employed as a labor market risk group. They estimate
a simultaneous hazard equations model, controlling for various confounders in different specifi-
cations. Similar to the results of Cai & Kalb (2006) they find a positive feedback mechanism
between employment and health. Both social causation and health selection mechanisms can
be detected. The strength of the study is clearly the use of high quality data with up-to-date
SEM methods controlling for confounders and allowing for temporal ordering. The authors,
however, seem to unnecessarily restrict their sample to men aged 30-59, probably the most
over-studied group in the literature on labor market processes.

The Panel Study of Income Dynamics (PSID) is the basis for the analyses conducted by
Haveman et al. (1994). They look at the interrelationship between wages, work hours, and
health. Their health indicators are self-reported health and health related work limitations.
They estimate a complex three equation simultaneous equations model. They find that health
limitations have a negative impact on work hours and wages while work hours have no effect
on health. Their complex model tries to capture the difficult inter-relationship between health
and labor market outcomes. However, they do not make use of the panel structure of the data

85
 3. Review of the Empirical Literature

and they restrict their sample to white men aged 25-65 who are head of the household. This
refers again to the most over-studied group in labor economics.

Lee (1982) also employs a simultaneous equations approach for wages and health. The
health indicator is the self-rated health item and a report of health limitations. They use one
wave of the National Longitudinal Survey of men aged 45-59. They find evidence that wages
and health affect each other in a feedback mechanism. They introduce SEM at an early stage
of the research literature, but use only cross-sectional data on men aged 45-59.

Summing up, we can note the following insights from the literature review. Most studies
which look explicitly at SES and health either find no support for health selection or conclude
that social causation is more important. A clear exception are studies on employment.
Regardless of the health indicator, various studies from different countries, using different
methods found a clear link between health and labor force participation or unemployment.
The majority of those studies agree that the major part of this effect can be attributed to
a health selection process. This means that the recent state of art allows us to draw the
conclusion that participation on the labor market depends strongly on individuals’ health. The
labor market is the most important source for income for most households and plays a crucial
role for social integration. These results alone disqualify any claims that health selection is
generally a negligible factor when looking at health inequalities.
Looking at the studies in the review we can see that different dimensions and indicators of
health are used when assessing health inequalities. However, there seems to be no clear pattern
whether some dimensions provide stronger evidence for social causation or health selection.
The use of varying indicators thus presents a problem, because it remains unclear what role
the health measure actually plays. There are too few studies, and too many different health
measures to make a statement about any clear trends at this point.
Education is used several times as a measure of SES. It has the advantage of being universally
available. On the other hand, education changes very little after a certain age, leaving little
room for health selective processes. Employment is another measure which is often used.
In its broadest sense it can also be applied to all individuals and is very responsive to other
socio-economic conditions and to health. Occupation, or occupational group is the measure of
SES which is used most often. Most studies use broad categories with six or less categories,
sometimes adding a non-employment category. There are several possible reasons for this
common use of occupation. First, it has a long tradition in health inequalities research to look
for occupational differences in health. Second, occupation is rather stable, yet can change
more than education, especially downward mobility is possible. Third, income can be seen as a
derivate of occupation. In this sense occupation would be a more fundamental cause of health
than income. Fourth, besides education, occupation is the measure which allows the easiest
estimates of intergenerational mobility in addition or instead of intragenerational mobility.

86
 3. Review of the Empirical Literature

One concern I have with several studies reviewed here is that they dismiss third factor
explanations too easily. If health has common correlates with e.g. skill, or personal characteristics
like locus of control, any association might be spurious. The responsible factors for the
association might lie in childhood. If interactions of acquired childhood characteristics with
changing environment over the life course is considered it is even unclear whether methods like
fixed-effects can account for such spurious correlation. It is advisable to conduct sensitivity
analyses, which indicate how strong a common background factor must be correlated with
health and SES to account for estimated effects in the study (for such an approach see Do,
Wang & Elliott 2013).

87
4. Methods
In this chapter I present the methodological approaches I choose to adress the formulated
hypotheses. I explain my notion of causality, explain how I measure health, and describe various
regression and decomposition models I employ.

4.1. Health Selection vs. Social Causation - The Issue of

Causality

4.1.1. The Counterfactual Model of Causality

In this thesis, I propose different theoretical explanations of how health selection processes can
generate health inequalities. I apply the theory to health differences between incumbents of
high status and of regular jobs. The theory makes clear statements about causal mechanisms.
The question is now, which methods are adequate for testing these causal relationships?
I will address this question in two steps. First, I will define the concept of causality I am using
and describe what problems arise when testing causal claims based on this notion of causality.
Second, I will describe which specific problems apply to my empirical example of high status job
attainment on the German labor market using SOEP data. I will show caveats and how I will
deal with them. Finally, I will sum up and discuss in what sense “effects” from the statistical
models should be understood as causal in this thesis.

Discussing causality is important, because sociologists have changed their way of treating
the matter of causation (Bernert 1983). In my thesis I use the model often referred to as the
counterfactual model of causality. In economics it is clearly the dominant framework of causality
(Angrist & Pischke 2010). A similar trend can be observed for sociology (Gangl 2010). Another
term for the model is the Rubin causal model (RCM) after Rubin (1974) who introduced it
to non-experimental data. Rubin calls the model potential outcomes. I will use the terms
counterfactual model or counterfactual argument throughout the thesis.
I chose the counterfactual model for two reasons. First, it reflects the current state of the art
regarding causality in (quantitative) social sciences (e.g. Angrist & Pischke 2010, Gangl 2010).
Second, it has a clear theoretical definition of causality and is able to transfer this definition
unambiguously into statistical models which can be applied to quantitative data analysis.

88
 4. Methods

4.1.2. The Counterfactual Argument

When talking about treatment and outcome in this section the most basic criterion which
needs to be fulfilled is the temporal relationship. The treatment has to come before the
outcome, otherwise it cannot have a causal effect on the outcome (Ward 2009). Following
the counterfactual model of causality researchers ask the “what if” question. The interest lies
in the effect Y of a certain treatment X on a particular unit i. To answer the question if X
causes Y one asks “what would have happened to unit i if it had (not) been treated with
X?”. The actual outcome, that is the outcome that really happened, is compared with the
counterfactual outcome that would have happened if the treatment had been withheld - all
other things being equal. One thus tries to compare two different versions of reality which
vary only in regard to whether unit i is given the treatment or not (Kaufman & Poole 2000,
102). This is done for every unit under observation. The difference between the actual and the
counterfactual effect is called unit effect (Winship & Sobel 2001, 14).
For the purpose of my study one would ask for example: What would happen to a certain
worker’s job status if he was not in poor health, but was in good health? The difference in job
status between the actual outcome (poor health) and the potential outcome (good health) is
the unit effect.
Due to the fact that one of the outcomes is unobservable, the unit effect is by nature not
measurable. If one unit has been treated it cannot be compared with its untreated status since
it has already been treated. This problem is known as the “Fundamental Problem of Causality”
(Holland 1986, 947).
The good news is that compared to others sciences (like e.g. medicine) in social sciences the
single unit effect is rarely of high interest. The focus lies on the average treatment effect (ATE)
which is the average of all unit effects. This means that social scientist do not ask: What
happens to the career prospects of a specific worker if his or her health changes? Rather they
ask: What happens on average to the career prospects of workers if his or her health changes
by a certain degree? The equation to calculate the ATE is (Winship & Sobel 2001, 18):

 (Yt i − Yc i)
δ̄ = E[Yt − Yc ] = E[Yt ] − E[Yc ] = (4.1)
i∈P
N

where P denotes the population under observation, the index t stands for treatment while c
stands for non-treatment or control. δ can only be estimated, never measured or calculated
directly. For the calculation the unobserable counterfactual outcomes are needed. To make a
consistent estimate of δ a sufficient assumption is that the treatment effect on the treated
(ATT) is the same as the treatment effect on the untreated (ATU) and that the non-treatment
effect is the same for the treated and the untreated:

89
 4. Methods

E[Yt |X = t] = E[Yt |X = c]; E[Yc |X = t] = E[Yc |X = c] (4.2)

The assumption implies that for example the job status of those workers who are actually
in poor health is affected in the same way as would the job status of those workers who are
actually in good health if they were in a different health state.
The assumption of equation 4.2 can be achieved through randomized assignment of the
treatment as done in experiments (Winship & Sobel 2001, 22). If equation 4.2 holds true, the
ATE can be calculated as:

δ̄ = E[Yt |X = t] − E[Yc |X = c] (4.3)

This means that if health status were assigned to workers randomly, the difference in job
status between the groups of the treated (good health) and the untreated (bad health) would
be equal to the ATE of health on job status.
If random assignment cannot be guaranteed1 this estimation might be biased for two possible
reasons:

1. The baseline difference in the outcome Y is unequal to zero.

2. There is a difference in the treatment effect between the treated and the controls.

Bias 1 means that the treatment group has a higher level of the outcome variable Y (e.g. job
status) than the non-treatment group before the treatment. So the difference in Y measured
after treatment cannot be attributed to the treatment, but to some other causes prior to the
treatment. For example, workers in good health might for different reasons be in better jobs
before they experience a change in health than those workers who are in bad health. The
difference then does not stem from the treatment, but from some other cause (e.g. difference
in education as a common cause for job status and health).
Bias 2 states that the ATE is estimated incorrectly if the treatment has a different effect for
the treated than for the untreated. This is often the case in processes of self selection into
the treatment2 . The second bias does not have to be a problem for a researcher. It depends
if one seeks to estimate the average treatment effect on the whole population or if one is
interested in the treatment effect on the actually treated (ATT). In the latter case bias 2 can
be admitted freely, but ignored as the conclusion which is drawn is limited to the group of the
treated (Winship & Sobel 2001, 23-24). For this study it means that I could state for example
that a change of health status (treatment) does have a negative causal effect on workers’ job
1
As it is certainly the case with health due to practical and ethical reasons.
2
Those who benefit more are more likely to choose the treatment.

90
 4. Methods

status. But I can only say that it has this effect for those workers who actually experience a
change in health. I do not know if a change in health would have the same effect on those
who were not treated (and did not change their health status).

A weaker form than random assignment to overcome bias 1 is ignorability. It relaxes the
assumption that the treatment is independent of all variables. It only states, that the treatment
is at least independent of the potential outcomes (Winship & Sobel 2001, 26-27):

(Yt , Yc )⊥X (4.4)

Unfortunately ignorability is rarely given in social sciences and, as will be shown later, is not
given in this study. Rather the probability is dependent on other variables Z. But it can be
argued that given the probability to get the treatment (P(Z)), ignorability holds:

(Yt , Yc )⊥X|P (Z) (4.5)

This is also called the conditional independence assumption (Dawid 2000, 419). It would mean
for example that health (the treatment) is not independent of job status (the outcome). Due
to some other factors the workers in regular jobs have worse health than workers in high status
jobs. At the same time these factors (e.g. education) influence the chance of being a regular
worker. Therefore regular jobs and health are not independent.
However, one can estimate the probability to be in a certain health state dependent on these
factors, which link job status and health. Given this probability3 , health can be seen as if it
were randomly assigned with regard to job status.
The model to estimate P(Z) has to be complete for the ignorability assumption to hold.
This will be a problem if there are important unmeasured variables which can bias P(Z).
As Rubin (2005, 324) states: In the end “[...] causal inference is impossible without
assumptions.” Consequently, I will make an effort to guarantee that the conditional independence
assumption holds true, so that causal inference is possible from the estimates in my thesis.

4.2. Applying the Counterfactual Model

After this general discussion about causality, I will go into more detail about problems that
arise when trying to estimate health selection effects on the labor market. I will first discuss
the direction of causality and related problems, then spurious correlation and problems of
measurement. Last, I will review in how far I can test causality and under what kind of
limitations the estimates in the analysis section should be understood as causal.

3
This means comparing those with the same probability to be in a certain health state, but differing in
the actual fact that some are in this certain health state and others are not.

91
 4. Methods

4.2.1. Direction of Causality

The most obvious problem with job status and health is the direction of causality. Usually
health inequalities are explained by the social causation approach (see section 2.1.3). For my
case that means explaining how resources or working conditions of workers in high status jobs
are beneficial to their health, generating an overall higher health status for incumbents of high
status jobs.
So how do I make sure that my estimates of health influencing job status are not merely
artifacts produced by reversed causality, meaning job status influencing health?
The first step in excluding reverse causality is that all analyses presented in this thesis - if not
stated otherwise - look at the effect of health at time point t on job status one year later
(t+1). This ensures that the treatment (change in health) comes before the outcome (change
in job status). However, this does not solve the problem of anticipation effects. It could
be that a change in job status is expected or anticipated by the individuals (for a theory of
anticipation, see Tavory & Eliasoph 2013). This might lead to the fact that the positive or
negative psychological effects of status change already influence the individual at an earlier
point in time. In addition, an adaption to the new situation with regard to resources might
occur earlier than the actual change. In some cases it might be preceded by e.g. an increase in
wages.
The second problem of reversed causality due to anticipation will be addressed in the following
manner. I will introduce three variables into the regression equations which are supposed to
capture the psychological effects of an anticipation of a job change. These variables measure
satisfaction with work, worries about own economic situation, and worries about job security.
In combination with a measure for wages they should be adequately capturing any anticipation
effects.

4.2.1.1. Closed Positions and Anticipation Effects

In my theory I argued that a distinction between open and closed positions on the labor market
can yield important insights in the interaction of context and health selection.
The analysis of closed positions has an additional advantage. If the estimated health effects
are due to the anticipation of the event - following the social causation instead of the health
selection hypothesis - there should be no differences between open and closed positions regarding
the health effect.
This means that a certain worker would anticipate that he or she will receive a certain (positive
or negative) labor market reward in the future. For my analysis this would just need to
be an anticipation of a reward within the next year which is a plausible assumption. If the
worker expects the rewards the psychological reaction could precede the actual reward. The
methodological interesting part about a distinction between open and closed position is the
following. There is no reason to assume that the effect of job loss or high status attainment

92
 4. Methods

on health is different for incumbents of open or closed positions. Therefore the reaction should
be the same as well. If the reaction is the same, then the estimated effects should be the same
for open and closed positions.
With this in mind, I can counter the argument that the estimates in my model are simply
anticipation effects of future labor market rewards. If we find that open positions show health
effects and closed positions do not, it speaks against the anticipation hypothesis. If health
effects are found both in open and closed positions, brushing off the anticipation hypothesis
could be a mistake.
Bringing anticipation effects into the formal equation, we would view health as a function of
the anticipation of the labor market reward:

LM R = α(e(H(ALM R ), EN LM ) ∗ τ )σ∗|A−c∗d| t (4.6)

4.2.1.2. Testing Causality in One Model

Another important step in estimating causal effects of health on job status is to consider
both directions of causality at the same time. In section 3.1, I argued that one criterion for a
“good practice” study is exactly this simultaneous test of the health selection and the social
causation hypothesis. This allows the evaluation of the hypotheses in a framework, which
makes them comparable with regard to the magnitude of the estimated effects. The question
can be addressed whether social causation effects play a bigger role than health selection
effects in creating health inequalities between job status. Section 4.7 describes the respective
statistical model.

4.2.2. Spurious Correlation

Another problem with establishing a causal relationship between health and job status is that
there might be third factors influencing both health and job status. For example, the positive
relationship between health and high job status could be caused by the education differential
of the two groups. Individuals with higher education take more or more efficient care of their
health and are therefore in better health. They are also much more likely to occupy a high status
job. Another example would be the unobservable rate of health deterioration and unobserved
skills or productivity which might cause the gradient in the relationship of job status and health
(Case, Fertig & Paxson 2005). The correlation between health and job status could therefore
be spurious.
Of course, spurious correlation is hardly a new problem in social sciences where experimental
data are unavailable. What has to be done to uphold the conditional independence assumption
is to control for all important factors which might influence both health and job status. Relevant
factors for labor market rewards are captured in my theoretical model. For each construct

93
 4. Methods

suitable indicators are chosen to make the conditional independence assumption as plausible as
possible. In section 4.9.2 the control variables are listed and described.

4.2.3. Are the Estimated Effects Causal Effects?

So, can we interpret the effects estimated in these models as causal in nature? The study
design and the statistical model of cross-lagged panel with fixed-effects adequately captures
possible confounders from spurious correlation, simultaneity, and reversed causation. Therefore,
I think that the proposed estimates represent the strongest test of causality available for the
subject under investigation. Using different modeling approaches throughout the thesis further
increases the reliability of the results, because they present a form of robustness check. The
effects of health on job status, which are estimated are therefore treated as causal estimates in
the rest of the thesis. The known limitations are explicated here and are not repeated every
time in the results part.

4.3. Measuring Health - Methodological and Theoretical

Implications
Estimating the impact and relevance of individuals’ health on their labor market outcomes
is the main purpose of this study. For this purpose it is necessary to address the question
of how health is measured. In the public health literature on the measurement of health in
standardized surveys two very general distinctions are made. First, it is distinguished between
self-reported measures and measures reported by third-parties. In the first case the respondents
of the survey answer questions regarding their health, illnesses, injuries, or well-being on their
own. The interviewer (usually) has no way of controlling the answers a person gives. Reports
by a third party are usually made by either the interviewer or a doctor through tests, blood or
saliva samples or through medical records. These are then used to measure the health status
of the person. In most surveys which do not have a special focus on health, self-reported
measures are the standard, as they are much easier and cheaper to acquire, require less expert
knowledge and are not invasive.
The second distinction is between so called objective and subjective measures of health. An
objective measure should return the same results for two persons in the same state of health
every time. In practice objective health measures can be and are subject to reporting bias.
However, reporting bias is a much bigger issue with subjective measures. Subjective measures
leave the response (to a survey question) to the perspective of the respondent. For example, if
a person is asked to rate her health, the answer “good” might mean something very different
for person A than it does for person B. This has two implications. First, it is a reasonable
assumption that subjective health measures are subject to higher degrees of measurement error

94
 4. Methods

than objective measures. Measurement error can lead to bias in estimates of health effects
usually underestimating the true effect of health (Bollen 1989). Second, subjective measures
might be systematically biased due to different reporting behavior of certain social groups (e.g.
men and women). This would make comparisons of subjective health measures between these
groups problematic. As the comparison of health effects between men and women is a central
element of my thesis, I will deal in detail with this possible reporting heterogeneity with regard
to subjective health in the next sections.

I will give an overview over the literature stating the problems regarding self-reported and
subjective health measures in survey research, present solutions offered so far, and discuss these
studies critically with regard to transferability to my case. In the end, I will present my own
approach to the problem which is a confirmatory factor analysis (CFA). I will present the idea,
the concept, the estimation techniques necessary for this approach, and discuss its advantages
and limitations.

4.3.1. Literature on the Measurement of Subjective Health

The gold standard of external reliability in health measures is usually mortality. Therefore
a lot of studies deal with the question whether self-rated health can predict mortality. It
is generally agreed upon in the literature that self-rated health is a very good predictor for
mortality (Jylhä, Guralnik, Ferrucci, Jokela & Heikkinen 1998, Quesnel–Vallée 2007, Dowd &
Zajacova 2007, Huisman, Van Lenthe & Mackenbach 2007, Singh-Manoux, Dugravot, Shipley,
Ferrie, Martikainen, Goldberg & Zins 2007, Singh-Manoux, Gueguen, Martikainen, Ferrie,
Marmot & Shipley 2007, Subramanian & Ertel 2008, Jylhä 2009, Nishi, Kawachi, Shirai, Hirai,
Jeong & Kondo 2012). However, these studies report different degrees of predictive power of
self-rated health on mortality. Sometimes the predictive power of self-rated health on mortality
is found to vary with SES (Singh-Manoux, Dugravot, Shipley, Ferrie, Martikainen, Goldberg
& Zins 2007, Huisman, Van Lenthe & Mackenbach 2007, Dowd & Zajacova 2007) or gender
(Jylhä et al. 1998, Nishi et al. 2012). Sometimes no such interactions are found (Subramanian
& Ertel 2008), or the results are not conclusive regarding the direction of the interaction
(Jylhä 2009).
A review of the most important studies can be found in the appendix in section A.4. It can
be summarized with two main conclusions. First, most studies use a sort of index of objective
health measures as “true” health, with the exception of the study by Shmueli (2003). Second,
there is no common agreement on reporting heterogeneity. Studies with different datasets
from different countries report varying degree of reporting heterogeneity or none at all. It
seems therefore wise to test reporting heterogeneity in the SOEP, as it is unclear what can be
expected.

95
 4. Methods

4.3.2. Measurement of Health in this Study

As already stated, I use the “social model” of health using the WHO definition. The concrete
way I measure health is similar to the approach of Shmueli (2003). My intent is to use
self-reported health measures which are available over several points in time to measure the
latent variable I call latent subjective health (LSH) or short latent health or simply health.
Latent subjective health determines the observed variables self-rated health (SRH), satisfaction
with health (SAT), and worries about own health (WAH). As my research design is mainly of
longitudinal nature, change in latent subjective health is therefore the main variable of interest.
The definition of change I use in this context is:
“Change is regarded as the occasion-specific deviation of a test or item score from a stable
person-specific level” (Eid 1996, 66).
The problem I am facing is that in contrast to most psychological studies my indicators
(SRH, SAT, WAH) were not constructed to measure a latent variable as a set of items. I
combine these three indicators out of necessity. It is necessary to take this into consideration
when evaluating the model fit, because the model fit will be worse than in psychological studies
that use indicators explicitly designed and pretested to measure a certain latent trait.

4.4. Confirmatory Factor Analysis

The methodological idea behind my approach of measuring health has a long tradition in
psychometric research. Some concepts cannot be measured directly. Instead, we look for
indicators which represent the concept. These indicators are prone to varying degrees of
measurement error. This might be random measurement error or systematic measurement error
(Bollen 1989). Choosing a latent variable approach we assume that our concept (in the following
health) is a latent metric variable which is normally distributed. In survey research health is
usually measured on a five-point-scale which captures the self-rated health in an ordinal fashion.
This indicator is a hierarchical categorization of the latent linear health variable with both a
loss of information, random measurement error, and potentially systematical measurement
error. If health is measured with more than one indicator, it is possible to estimate the latent
metric variable health by confirmatory factor analysis (CFA).
Figure 4.1 captures the concept. Health is measured by the indicators Self-rated health (SRH),
satisfaction with health (SAT), and worries about own health (WAH). Each indicator is seen as
the dependent variable in a regression model of the indicator on the latent variable with an
intercept. The following equations define the latent variable health (H):

96
 4. Methods

W AHi = αW AH + λW AH Hi + ϵi,W AH
SRHi = αSRH + λSRH Hi + ϵi,SRH
SATi = αSAT + λSAT Hi + ϵi,SAT

These equations do not take into account that the indicators used are ordinal and not
metric. For the time being I will stick to this simplification, because it makes the argument
for my methodological approach easier to understand without loss of generality. In the next
section, I will explain the necessary modifications for incorporating indicators containing ordinal
information. I will use an analysis of the correlation matrix of the three observed variables
generated by polychoric correlation (see 4.4.1).
For the model above to be identified an additional assumption has to be made. Usually, either
one of the factor loadings (λ) is fixed (to 1) or the variance of the latent variable is fixed (to
1). It does not matter which one is fixed, the models are equivalent with different scaling
parameters.
Referring to the literature I interpret the estimated latent variable as the “true health” I am
really interested in.

97
 Figure 4.1.: Health as a Latent Variable

VARHealth
Health

λSAT λSRH λWAH

98
4. Methods

SAT εSAT SRH WAH εWAH

εSAT

αSAT αSRH αWAH

Intercept
 4. Methods

4.4.1. CFA with Categorical Dependent Variables

Polychoric correlation is used if Pearson correlation, which assumes interval scaled variables,
cannot be used. In much of social sciences research variables are measured on ordinal (likert)
scales. Such is the case with the subjective health indicators in the SOEP. The basic assumption
is that the observed ordinal variables are determined by latent unobserved variables, which are
continuous and normally distributed. Based on the probability distribution of the contingency
table polychoric correlation is an estimate of the correlation between the two latent variables
determining the two ordinal observed variables if this correlation could be calculated. The
probabilities of the contingency table for two variables x and y with m1 and m2 categories are
calculated as (Holgado–Tello, Chacón–Moscoso, Barbero–Garcı́a & Vila–Abad 2010, 155):

 ai  bj (x2 −2ρxy+y 2 )
1 −
2(1−ρ2 )
P [X = i, Y = j] = pij =  e dxdy (4.7)
ai−1 bj−1 2π 1 − ρ2

ai and bj are the cutoff values of the latent unobserved variables to decide the distribution
of the observed ordinal variables. They are estimated in a first step as:

ai = Φ−1
1 (Pi̇ ) (4.8)
bj = Φ−1
1 (Pj̇ ) (4.9)

Pi̇ and Pj̇ are the observed cumulative marginal proportions of the contingency table of the
two observed ordinal variables x and y (Flora & Curran 2004, 467-468).
Maximum likelihood offers a way of estimating the correlation by maximizing the following
likelihood function:

m1 
 m2
lnL = nij log(pij ) (4.10)
i=1 j=1

This way of estimating the correlation matrix between the subjective health variables takes
into account their ordinal nature. This is especially important as one item (worries about
health) has only three categories which is too few to simply use Pearson correlation and treat
the data as if it was continuous and normally distributed (Holgado–Tello et al. 2010, 154).
This procedure results in a polychoric correlation matrix.
The polychoric correlation matrix should not be analyzed with standard maximum likelihood.
Rather, research showed that a weighted least squares (WLS) approach also called asymptotically
distribution free (ADF) is more appropriate (Flora & Curran 2004, Holgado–Tello et al. 2010).
The fitting function for WLS is:

99
 4. Methods

FW LS = [s − σ(Θ)]′ W −1 [s − σ(Θ)] (4.11)

In STATA 12.1 this WLS approach is applied by using the sem option method(adf).
However, analyzing polychoric correlation matrices cannot be combined with a WLS estimation
technique in STATA 12.1. Therefore, these analyses are the only ones where another statistics
package is used. The software package for the estimation of the confirmatory factor analysis is
MPlus 7 (Muthén & Muthén 2012).

4.4.2. Measurement Invariance

So far, I have discussed the appropriate way of measuring subjective health. I argued that a CFA
approach suits the theory best. But there is another important reason to choose CFA. It allows
testing whether there is reporting heterogeneity between men and women. This is an issue
raised in the literature without a conclusive answer (see section A.4). Using a confirmatory
factor analysis approach the question whether SRH means the same for men and for women can
now be addressed using statistical methods. What I try to establish is measurement invariance
(MI). Plainly speaking, establishing different forms of measurement invariance can answer the
question whether and to what degree we are measuring the same concept for two groups.
We can also determine if coefficients or scores can be compared between the groups or not
(Bollen 1989). Measurement invariance is defined as:

“Statistically, this means that the mathematical function that relates latent variables
to the observations must be the same in each of the groups involved in the
comparison. This idea has become known as the requirement of measurement
invariance.” (Borsboom 2006, S176)

Another definition is that

“measurement invariance (equivalence) concerns whether scores from the

operationalization of a construct have the same meaning under different conditions.”
(Kline 2010, 251) [emphasis in the original]

More informally speaking we can ask: Are we measuring the same thing in different groups?
Does subjective health as measured by the indicators in the SOEP relate to the same construct
in the same way for men and women? Do women on average rate their health lower? Is the
curve relating indicator and latent variable steeper for men? These questions will be addressed.
If these questions can be answered, then we can also address one of the issues from section
4.3.1. This refers to the question whether a comparison of effects from or on subjective
health between men and women actually reflects substantive differences or only differences in

100
 4. Methods

measurement.

There are five types of measurement invariance, which can be put in the order of the
restrictiveness of their assumptions.
The first step is configural invariance (Kline 2010, 252) or invariance of form (Bollen 1989,
356). This means that the variables defined in the model, their measurement, and the structural
relationships among them are theoretically the same. It does not imply that the strength of
these relationships is the same. Put another way, we ask whether the constructs we define are
measured by the same indicators for different groups. This step is most often skipped in the
analysis if the theory dictates configural invariance. Only violations of the model fit in this
least restrictive form of measurement invariance might change that.

The next step is to test for metric invariance (Kline 2010, 253). This is done by constraining
the factor loadings of the model to be equal across groups. If this still yields a satisfactory
model fit than we can conclude that the latent variable is constructed giving the same weight
to the respective indicator. Showing that metric invariance holds is an important step, because
it is a sufficient condition for comparing associations of the observed variable under scrutiny
with other variables between the groups. This means that if I could establish metric invariance
across gender we could compare correlations or regression coefficients of health with other
labor market variables between men and women.
It should be noted that the unstandardized factor loadings are constrained to be equal, not the
standardized ones. An issue with categorical dependent variables is that metric measurement
invariance (section 4.4.2) cannot be tested apart from structural invariance, at least not without
certain restrictions. The analysis of measurement invariance in the results part are nevertheless
based on categorical dependent variables, because some research shows that treating these
likert-scale items like continuous variables in group comparisons will result in misguided conclu-
sions (Lubke & Muthén 2004). However, there will be no separate test for metric measurement
invariance. Only structural invariance will be tested. As structural invariance implies metric
invariance, this constitutes an even stronger test of comparability than required.

If we also want to compare the absolute level or means of a latent variable then we need
to assume that the incepts of the indicators are equal across groups. This leads to structural
invariance or invariance of intercepts (Bollen 1989, 365-366). If structural invariance can be
accepted according to the model fit, we can conclude that there are no level differences in the
indicators between groups, so that any remaining differences in the latent or observed variables
should be due to substantive level differences, which cannot be explained by measurement or
response differences. This means, that a given score on the latent variable will lead to the
same response pattern on the observed indicators for both groups. There are no distortions
through certain group specific response behaviors.

101
 4. Methods

Tests of invariance can also be made for the variance of the error terms and the variance
of the latent variable. These highly restrictive models are however not necessary for me to
test whether subjective health measures the same thing for men and women in the SOEP.
Consequently, I will not apply them here.
The elegance and usefulness of the approach of testing measurement invariance in CFA is the
following. Many different studies in the literature have found different theoretical arguments
and empirical findings to support or reject the hypothesis about comparability of subjective
health status across gender. Measurement invariance relies on a latent variable approach
which is in accordance with most of the theoretical approaches in the literature, but not with
their empirical strategy. It allows to test the hypothesis stated, without relying on indexes of
“objective” or “true” measures of health.
The reason why a confirmatory factor analysis is almost never applied might be that a lot
of studies on reporting heterogeneity come from health economics. For reasons unclear,
econometrics and psychometrics rarely deal which each other’s approaches, although they can
without a doubt be seen as the two driving forces in the development of applied statistical
methods in the social sciences4 . Perhaps the research questions and theoretical approaches are
too different. Still I think that this is a loss, because the research question should determine
the method used, rather than the discipline.

4.4.3. Comparing Fit in CFA Models - Tests and Model Fit Indices
4.4.3.1. The Problem of Model Fit Assessment Using CFA

The previous section spoke a lot about testing the data for the feasibility of different kinds
of measurement invariance. But what are good tests and evaluation criteria for confirmatory
factor analysis models? This question has and still does stir a lot of analytical and simulation
based research (Kline 2010, 191). It is far beyond the scope of this dissertation to give a
comprehensive review. As of today there is no one accepted criterion or set of criteria. I will
explain which criteria are most often used and which I will use and what they signify. In the
empirical application I will state whether or not certain criteria have been met. It should be
kept in mind that all test statistics and fit indexes have their general and specific draw backs.
That a model meets certain criteria does not prove that the model is right. It just lends support
to the argument that it is the correct model (Kline 2010, 192). It is far easier to show that a
certain model is misspecified than showing that it is correctly specified.

Fit statistics are only good indicators of average model fit, not of specific parts of the model.
And each fit statistic reports one aspect of model fit. Fit statistics cannot tell you how and
where you might have misspecified your model. Also, with good model fit does not necessarily

4
Biostatistics also built the foundation for a lot of statistical methods used in social sciences nowadays,
but it is not part of the social sciences.

102
 4. Methods

come good predictive power of the model. And fit statistics have nothing to say about the
interpretation or theoretical meaningfulness of a model (Kline 2010, 192-193).

Generally two classes of fit statistics can be distinguished: Model test statistics and fit
indexes. Although they are treated separately, most fit indexes are a function of test statistics
or vice versa (Kline 2010, 196). A test statistic tests whether the covariance matrix, which
results from the model specified by the researcher, systematically deviates from the empirical
covariance matrix found in the data. If such a test is found to be statistically significant, it
means that it is unlikely that the deviations between theoretical and empirical covariance matrix
are simply by chance or due to sampling error. One problem is that such an accept-support
approach will generally yield more support for models with low statistical power and less support
for models with high statistical power (Kline 2010, 193-194). If the sample is large (as it is
using SOEP data) additional views on the model fit should be taken into account, because it is
likely that a model is rejected because of large sample size and not because of poor model
specification (Kline 2010, 198).

Fit indexes are not yes-or-no decisions in determining whether a model should be rejected or
not, but rather continuous indicators of goodness or badness-of-fit of the model. They also
indicate how closely the model resembles the actual data (Kline 2010, 195). One important
group of fit indexes used in the literature are comparative fit indexes. These indexes compare
the specified model with the baseline model, which is usually a model where there is complete
independence of all variables. It means that one compares the specified model with probably
the worst possible model and sees how it fares. This might not give any indication whether the
model is good from an absolute standpoint.
Some indexes are adjusted for their (lack of) parsimony. That means that in one way or another
the degrees of freedom in the model are taken into account. This pays tribute to the fact
that more parsimonious models with higher degrees of freedom will ceteris paribus always have
worse fit (Kline 2010, 196).
Absolute fit indexes are used to assess how much of the empirical covariance is explained by
the model. They are similar to an R2 statistic in ordinary regression (Kline 2010, 195).
So, are there any golden rules for model fit at all? The closest thing to golden rules for model
fit assessment was the study undertaken by Hu & Bentler (1999). Still other studies point out
that their cut-off criteria do not hold under all circumstances, some of them are quite real-
istic in real research. Therefore it is safe to say, that there are no golden rules (Kline 2010, 198).

4.4.3.2. Fit Statistics Used

For my purposes the following fit statistics are picked and interpreted jointly:

103
 4. Methods

1. likelihood ratio χ2 test

2. Root Mean Square Error of Approximation (RMSEA)

3. Comparative Fit Index (CFI)

4. Standardized Root Mean Square Residual (SRMR)

The RMSEA is a parsimony-adjusted badness-of-fit index where zero stands for the best fit.
If the degrees of freedom are equal or larger than the χ2 statistic of the model than RMSEA is
zero. For all other models RMSEA is calculated by the formula (Kline 2010, 205):


χ2M − dfM
RM SEA =
dfM (N − 1)

With the point estimate of RMSEA comes a 90%-confidence interval. Ideally the lower
bound should be zero. A rule of thumb is that a RMSEA ≤ 0.05 is a good fit and that the
upper bound should not be above 0.10 (Kline 2010, 206).

The comparative fit index is calculated by the formula (Kline 2010, 208):

χ2M − dfM
CF I = 1 −
χ2B − dfB

It assesses the χ2 -statistic of the model against the χ2 -statistic of the baseline model
(independence model), correcting for the degrees of freedom, penalizing complex models. If
CFI is 1 it indicates best fit. 0 would indicate that the model is as poor as the baseline model.
The more it goes in direction of 0 the poorer the fit. CFI is robust to sample size which makes
it a valuable information for my thesis (Kline 2010, 207).

The SRMR relies on the transformation of the predicted and the empirical covariance matrix
into a correlation matrix. It measures the average absolute correlation residual. The cut-off
criterion proposed by Hu & Bentler (1999) is 0.08.

4.5. The Unidirectional Approach: Fixed-Effects Logistic

Regression
After I have discussed issues of causality and measurement of health I will now turn to the
statistical methods used for the analysis of health effects on job status.
The theory of health selection predicts that persons in good health will have a higher chance

104
 4. Methods

of getting into or remaining in a high status position on the labor market. In addition to
social causation and health selection explanations the association of health and job status
could also be due to third factors influencing both health and job status. Certain attitudes
might lead to a healthier lifestyle and a better performance on the labor market. To avoid such
biases in the estimation I will control for a wide range of theoretically relevant factors in my
statistical models. Section 4.9.2 describes these variables and the rationale for choosing them.
In this part, I will explain how time constant unobserved heterogeneity can be controlled in the
models using a fixed-effects regression approach. Time constant unobserved heterogeneity is
a technical term for all factors influencing e.g. health or job status which are constant over
time for an individual. What remains are only factors which change over time. Kunze (2008,
66) points out that a fixed-effects approach for the estimation of wages has been a common
approach since the study of Mincer & Polachek (1978). This section presupposes knowledge
about standard OLS and logistic regression.

High status job is a dichotomous dependent variable. One way to treat such a variable
adequately in a statistical model is to use a logistic function to link the probability of having a
high status job (from now on success in contrast to failure) to the predictor variables. Trying
to eliminate time constant unobserved heterogeneity (individual intercepts) works as follows.
Similar to linear fixed-effects regression the starting point is an effects model, where a linear
unobserved variable is estimated (Greene 2003, 690):

yit∗ = Xit β + ui + ϵit

yit = 1|yit∗ > 0,
yit = 0|yit∗ ≤ 0

For the random-effects approach u and e are assumed to be uncorrelated. For the fixed-effects
approach, which is considered here, there is no assumption about the functional relationship
between the two error terms. Using a rather complicated argument, it can be derived, that
given a certain number of successes to the overall number of tries T, the conditional distribution
of the dependent variable does not depend on the individual error term ui . Formally speaking
the following conditional probability

P (yi1 = y1 , ..., yiT = yT |Xi , ui )

P (ni = n|Xi , ui )

105
 4. Methods

can lead to a likelihood function which allows to estimate the coefficients β of the model
(Wooldridge 2002, 491):

N
 T
  T

L(β) = log(exp( yit Xit β)[ exp( αt Xit β)]−1 )
i=1 t=1 a∈R t=1

The actual estimation process is implemented using STATA 12.1 with the command
xtlogit and the fe option.

An alternative to the complicated fixed-effects-logit model is a linear probability approach.

The simple linear probability model is defined as (Wooldridge 2002, 454):

P (y = 1|X) = Xit β

This is a linear approximation of a non-linear process. It should be noted, that by construction

the model produces heteroscedasticity in the error-term, so that robust estimation techniques are
warranted, but also easily implemented. The linear probability model can now be transformed
like the common linear fixed-effects regression:

N 
T N 
T
′ ′
 
β̂F E = ( Ẍit X¨it )−1 ( Ẍit ÿit )
i=1 t=1 i=1

ÿit = Ẍit β + ϵ̈it

The advantage is that it is easier to estimate, and that marginal effects are directly estimated,
which is not possible in fixed-effects-logit model, where marginal effects depend on the not
observed individual intercepts. These models can be estimated by using the STATA 12.1
command xtreg with the fe and robust option.
In the context of the cross-lagged panel fixed-effects linear probability model (see section 4.7)
the estimation is done via STATA’s sem package.

4.6. Applicants vs. Incumbents - The Use of Survival

Analysis
In section 2.8.1 I proposed two hypotheses which stated that health selection might depend on
whether a worker applies for a position or already holds it. In the one case it is the goal of the
worker to move upwards on the job ladder. In the other case he or she wants to avoid moving
downwards through demotion or being fired.
The fixed-effects-logit approach presented above cannot make this distinction in an empirical

106
 4. Methods

test. Despite all its benefits one limitation is that a promotion and a demotion are treated the
same in the statistical analysis. That means that - all else being equal - it does not matter
for the estimation of the health coefficient whether a person is first in a regular job and then
moves upward in the hierarchy to a high status position or whether it is the other way around.
In addition, all those who stay in one of the two states for the whole period of observation are
excluded from the analysis, because only change in job status is analyzed.
The method which is most appropriate to assess the hypotheses on applicants and incumbents
are survival analytic models or so called event-history analysis. There is no one method for
this purpose. There are dozens of different possible model specifications. What they have
in common is that their basic level of analysis is the event time. This is the time from the
beginning of the study (or the first observation of the individual) until the event occurs, until
the study ends, or until the individual leaves the study in which case this observation is said to
be right-censored (Singer & Willett 1993, 157).

Deciding which model to estimate is not easy. Generally, a rough distinction between
parametric, semi-parametric, and non-parametric approaches can be made. I will choose a
non-parametric approach which does not make any assumption about the functional form
of either the analysis time or of the hazard-function. There is no theory indicating any kind
of functional assumption, so that a non-parametric approach seems the safest and most
conservative choice.
I follow the model presented by Singer & Willett (1993) as a discrete-time survival analytical
model. The discrete-time hazard (h), indicating the probability for a randomly chosen individual
that the event happens at a certain time period j, given that the event has not already happened,
is defined as (Singer & Willett 1993, 163):

hj = P r(T = j|T ≥ j) (4.12)

T stands for the random variable which indicates the period in which the event happens.
Introducing a matrix of observable characteristics (Z) and defining the probability for individuals
yields the following definition (Singer & Willett 1993, 164):

hij = P r(Ti = j|Ti ≥ j, Z1ij = z1ij , Z2ij = z2ij , ..., ZP ij = zP ij ) (4.13)

P stands for the number of predictors in the model. This is identical to the number of
independent variables in the model. Using a logistic function to link these probabilities and the
predictors the discrete-time hazard model is (Singer & Willett 1993, 166):

1
hij = (4.14)
1+ e−[(α1 D1ij +α2 D2ij +...αJ DJij )+(β1 Z1ij +β1 Z1ij +...+βP ZP ij )]

107
 4. Methods

J is the last time period observed in the whole sample. D stands for dummies which index
time periods (d1ij = 1, if f j = 1, etc.). α are the intercept parameters, β the slope parameters
describing the effect of the variables in the model on the hazard function on a logistic scale.

hij
log( ) = (α1 D1ij + α2 D2ij + ...αJ DJij ) + (β1 Z1ij + β1 Z1ij + ... + βP ZP ij ) (4.15)
1 − hij

This equation shows that the scale of the model will be log-odds, the same as it was with
the fixed-effects-logit approach (Singer & Willett 1993, 167).
The likelihood function that allows the estimation of the relevant parameters is derived in the
following way. The likelihood is simply the product of the probabilities of observing the events
and censored individuals (c) as the data presents it (Singer & Willett 1993, 170):

n

L= [P r(Ti = ji )]1−ci [P r(Ti > ji )]ci (4.16)
i=1

This can be transformed to:

n ji −1 ji
  
1−ci
L= [hij (1 − hij )] [ (1 − hij )]ci (4.17)
i=1 j=1 j=1

Taking the logarithm and skipping some substitutions and transformations, we get (Singer
& Willett 1993, 170):

ji
n 
 hij
l= [log( )yij + log(1 − hij )] (4.18)
i=1 j=1
1 − hhij

where yij stands for the event occurring or not.

The models estimated in section 5.6 are all applications of this non-parametric discrete-time
survival analysis. They are implemented by transforming the SOEP into a survival data set and
estimating the model with STATA 12.1 using the command logit allowing for repeated events
within an individual by calculating standard errors corrected for within individual-correlated
error terms and allowing for hetereoskedastic error distribution with the option cluster.

4.7. Testing Selection vs. Causation

In my literature review part I argued that one feature of a “good practice” study is that health
selection is directly tested against a social causation explanation. Following my own advice I
will use a method which has been used in a couple of “good practice” studies (e.g. Chandola

108
 4. Methods

et al. 2003, Eaton et al. 2001, Stansfeld et al. 2011). The method can be called cross-lagged
panel model within a structural equation modeling (SEM) framework. The idea behind this
approach is to see whether health (at time point t) has a significant impact on job status at
time point t+1 beyond what is explained already by job status and control variables at time
point t. The same goes the other way around for the impact of job status at t on health at
t+1. The concept is captured in figure 4.2.
I modify the model used in the literature by controlling for unobserved time constant hetero-
geneity with a fixed-effects approach (compare section 4.5). I use standard demeaning strategy
to make the estimated coefficients independent of the individual intercepts. As this is a linear
regression approach and job status is a dichotomous variable this results in a linear probability
model (LPM, see again section 4.5). Taken all together the model could be described with
the monstrous term cross-lagged panel fixed-effects regression including a linear probability
model. I will simply use the term “cross-lagged fixed-effects” (CL-FE) as it contains the most
important features of the model. For the interpretation of the coefficients it is key, however,
to remember that the equation with job status as dependent variable is treated as a linear
probability model.
The advantage of ignoring non-linearity in the model is that marginal effects are estimated.
From the marginal effects we can get an impression of the absolute strength of the effects.
Such an assessment is harder in a logistic regression model.

Formally the conceptual model is described by two equations:

ÿit+1 = z̈it β + Ẍδ1 + ϵ̈it

z̈it+1 = ÿit γ + Ẍδ2 + µ̈it

The dots indicate that the variables have been demeaned to exclude individual intercepts from
the equation. X is a vector of control variables. The error terms of the two equations are
allowed to be correlated to captured any remaining influence which might else lead to estimates
biased due to spurious correlation.
A slight deviation from the standard approach in the literature is that the effects of the
variables at t on variables t+1 are constrained to be equal across all waves. This assumption is
also found in common fixed-effects regression. This disallows time trends in effect sizes. As
time trends are not the subject of investigation here, a simplification of the model will lead to
an easier interpretation. Leaving the coefficients to be freely estimated does not change any of
the presented results in substantive terms.

109
 4. Methods

Figure 4.2.: A Cross-Lagged Model with Fixed-Effects

t t+1

Δ Subjective Health Δ Subjective Health ε-1

Δ Controls

Δ Job Status Δ Job Status ε-1

4.8. Decomposition of Health Inequalities

In the literature the question is often posed to what degree health inequalities can be explained
by different factors. In my case, I ask to what degree the overall differences in health between
normal and high status job workers can be attributed to:

a) spurious correlation due to common background factors

b) materialist-environmental factors

c) social-psychological factors

d) health selection

At the same time, I attempt a decomposition into time varying and time constant factors.
I will show that the approach which is taken by several studies (Kroh, Neiss, Kroll & Lampert
2012, Richter, Moor & Lenthe 2012, Vonneilich, Jöckel, Erbel, Klein, Dragano, Siegrist &
Knesebeck 2012, Christensen, Labriola, Lund & Kivimäki 2008, Ball, Crawford & Mishra 2006,
Thrane 2006, Warren, Hoonakker, Carayon & Brand 2004, Cohen & Hamrick 2003, Stronks,
Mheen, Bos & Mackenbach 1997) does not yield the correct decomposition. In addition, I am
able to provide confidence intervals for the decomposition. I do not develop a new method, I
simply apply the well known path analysis and calculation of indirect effects based originally on
Wright (1934). The calculation of the standard errors for these indirect effects follows Sobel’s
method (Sobel 1987), which can be implemented in most recent major statistical software
packages. I use STATA 12.1’s sem and estat teffects commands for this purpose.
The argument below is made for the case of linear regression for simplicity’s sake. The same

110
 4. Methods

problem and solution are also true for logistic regression and probit regression. In these cases
things are additionally complicated by the fact that the scaling of the coefficients changes
when additional variables are considered in the model. This is a problem which is also rarely
addressed in applied research (Mood 2010, Karlson, Holm & Breen 2012).
I begin by outlining the approach often taken in the health inequalities research literature
and then show how it can be improved. Further, in the subsequent section I provide - to the
best of my knowledge - the first attempt in the health inequalities literature at a decomposition
of an effect (not a variable!) into time constant and time varying parts. I provide an estimate
of the mediation effect of all time constant factors including a confidence interval. Respective
statistics are provided for time varying factors.
Before I start, I must stress that this decomposition approach makes no claims to causal
interpretation. This is a difference to some of the previous approaches presented. Although
in some cases models are estimated which could imply causation from health to some of
the mediators these “effects” should not be interpreted causally. What is done is merely a
decomposition of the health effect according to factors with which health and job status are
associated. In some instances I will use the term mediation. This is not meant as mediation in
the causal sense, but in a technical sense. It simply describes how the model is estimated not
how the causal direction runs. For example, it makes no sense to claim that health during the
labor market period has an influence on education, which was attained before entering the la-
bor market. And of course age as a dependent variable is only reasonable for technical purposes.

Traditional approaches in the wider field of medical sociology, public health, and epidemiology
have tried to decompose health inequalities in the following way. First, a bivariate regression
(sometimes adjusted for age) is estimated. Then successively additional variables are added
to the model. The difference in the new estimated coefficient of the variable of interest (e.g.
income) in comparison to the bivariate case is interpreted as health inequalities due to income,
which can be explained by other factors such as health behavior or education. This allows
statements like: 25% of income related health inequalities can be traced back to differences in
education, and 40% to differences in health behavior between income groups. The rest of the
health effect (35%) remains unexplained or is the direct effect of income on health.
These statements are scientifically interesting and useful, because they address substantial
questions of research in health inequalities and can be easily understood by those who are
not trained in statistics, reaching a wider audience. It follows that I want to retain the
ability to make such statements when modifying existing approaches. My argument is divided
into three steps. First, I show that the reported percentage points are only in very specific
cases the correct estimates. Second, I provide a measure of uncertainty of these estimates
in form of a confidence interval, which is well known from conventional parameter estimates.
Third, I show such statements can be made for time constant factors and for time varying factors.

111
 4. Methods

Intuitively the argument is as follows. The change in the coefficient of the independent
variable of interest (x1 ) can only be attributed to other factors in the way presented above if the
additional variables which are introduced are uncorrelated in the sample or if the variables are
completely unrelated to the outcome (y4 ), in which case their contribution to the explanation
of health inequalities is of course zero. In this case each change actually presents the portion
of the coefficient attributable to the newly introduced (set) of variables (always 0). As soon as
the newly introduced variables (y1 , y2 , y3 ) are correlated in the sample with x1 the order in
which they are added to the model becomes a determinant for the change in the coefficient of
x1 on the outcome (y4 ). In most cases the change in the coefficient of x1 will be different if
the order is altered.
Several studies recognize this and consequently use every possible combination of three variable
blocks (only y1 or y2 or y3 , y1 and y2 , y1 and y3 , y2 and y3 , all three). They interpret the
change of the coefficient if only one of the three blocks is included as the raw mediation. The
change in the coefficient from the model which includes already two variable blocks to the model
with all variable blocks is interpreted as the independent contribution of the variable set which
is included last. The difference between the two values is characterized as the contribution due
to correlation with the two other variable sets. This is a better way of addressing the issue, but
it is unfortunately still not correct. It recognizes the problem, but fails to provide the correct
solution.
If y1 and y2 and y3 are all uncorrelated in the sample their own direct effects on the outcome
will be the same regardless whether only one of the variables, two or all three are included in the
model. However, if they are correlated the coefficients of both x1 and the other variables in the
model (e.g. y1 and y2 , with y3 excluded) are biased in the reduced model. x1 is conditioned on
one variable, but not using the correct coefficient. This results in a biased conditional estimate.
Therefore the difference in effects of x1 between a reduced model and the full model will under
these circumstances not yield the correct mediation effect.
A fictional example might look like this: If first y1 is introduced in the equation and this reduces
the coefficient by 25% and then y2 which results in an additional reduction of 10% we just
know that jointly the reduction is 35%, but the individual contribution could be anything. It is
not even restricted to the bounds of 10% and 35%. Conditioning on y2 could lead to a sign
switch in the coefficient of y1 leaving it with an actual negative5 contribution, or the coefficient
could be close to zero after controlling for y2 which leads to no independent contribution. In
almost all empirical cases, we cannot tell beforehand. We have to calculate the indirect effects
through path analysis.
At this point it should also be noted that if only one variable or one set of variables is used for
mediation, the traditional method is still valid. This means that the difference between the full
model and the model only including x1 can be interpreted as the mediating effect of all set of

5
This would indicate that health inequalities are actually increased when controlling for y2 . See table 4.1.

112
 4. Methods

variables together.

After this rather intuitive argument I will show formally that the common approach in the
literature is incorrect. The model that includes x1 and all additional sets of variables has
to be seen as the correctly specified model. If it is not, then all further claims about any
relationship become void. The full model provides an estimate of all direct effects on the
outcome conditioned on all other variables. The actual mediation effect is now the product of
the regression coefficient of e.g. y1 on x1 times the coefficient of x1 in the final model.
The full model is:

y4 = αy4 + γdirect x1 + β1 y1 + β2 y2 + β3 y3 + ey4 (4.19)

The additional equations which are implicit, but should be made explicit are:

y1 = αy1 + γ1 x1 + ey1 (4.20)

y2 = αy2 + γ2 x1 + ey2 (4.21)
y3 = αy3 + γ3 x1 + ey3 (4.22)

The total effect of x1 on y4 is simply the covariance divided by the variance of x1 or it could
be written as the sum of all indirect effects and the direct effect (Bollen 1989):

cov(x1 , y4 )
γx1 total = = γdirect + γ1 β1 + γ2 β2 + γ3 β3 (4.23)
var(x1 )

A model which now excludes one of the mediators (y3 ) would look like this:

` x1 + β`1 y1 + β`2 y2 + e`y4

y4 = αy4 + γdirect (4.24)

The claim that would need to be made if the approach in the literature was correct is that:

` − γdirect = γ3 β3
γdirect (4.25)

Only in this case would we get a correct estimate of the effect of x1 on y4 which is mediated
by y3 . However, rearranging equation 4.23 and inserting it in 4.25 yields:

` = γx1 total − (β`1 γ1 + β`2 γ2 )

γdirect (4.26)
γdirect = γx1 total − (β1 γ1 + β2 γ2 + γ3 β3 ) (4.27)
` − γdirect = γ3 β3 + γ1 (β1 − β`1 ) + γ2 (β2 − β`2 )
γdirect (4.28)

113
 4. Methods

So we can see that the difference in the two coefficients estimated the conventional way
usually does not yield the correct mediating effect. Only if either both γ1 and γ2 are zero
or both β1 = β`1 and β2 = β`2 is true does the conventional method estimate the correct
mediating effect. This means that either x1 has to be completely unrelated to both y1 and y2
making them effectively useless mediators, or y3 is uncorrelated with y1 , y2 , and x1 (compare
Wooldridge 2009, 92) making it again an entirely useless mediator.

We can see that the conventional approach fails to give us the correct answer with regard to
strength of mediation. In addition, it is not clear in what direction the results will be biased. It
could be either a downward or an upward bias. The sign could even switch depending on the
signs and the relative sizes of the terms γ3 β3 and γ1 (β1 − β`1 ) + γ2 (β2 − β`2 ). It seems therefore
best to use a path model to estimate mediator effects. Fortunately this technique has been
around in applied social science research for decades now and is easy to implement in various
software packages. In addition, the standard errors for the indirect effects are usually provided
using the delta-method (Sobel 1987). This allows us to calculate a confidence interval for the
proportion which is explained by the mediator. The proportion for the jth mediator with k total
mediators is calculated as (confidence interval in brackets):

γj βj γj βj
P ropyj = = k (4.29)
γtotal j=1 (γj βj ) + γdirect
γj βj − 1.96 ∗ se(γj βj ) γj βj + 1.96 ∗ se(γj βj )
[ k ; k ] (4.30)
j=1 (γj βj ) + γdirect j=1 (γj βj ) + γdirect

We should also note that the estimate of the mediation effect is not bounded by 0 and 1 (or
100%). All values are theoretically possible. Table 4.1 explains what the different value ranges
mean:

114
 4. Methods

Table 4.1.: Meaning of Different Value Ranges of Decomposition - in Percent

Value of P ropyj Meaning

0-1 The remaining effect is X∗100% smaller than the total ef-
fect. Conditional health inequalities are reduced compared to
unconditional health inequalities.

>1 The remaining effect is (X-1) times the size of the total effect
and the sign has switched. Conditional health inequalities are
inverted when compared to unconditional health inequalities.

<0 The remaining effect is |X| times larger than the total effect.
Conditional health inequalities are increased in comparison to
unconditional health inequalities.

4.8.1. Indirect Effects Through Time Constant Factors

In this part, I will explain how I estimate which contribution time constant factors make to the
explanation of health inequalities. Such an estimate can answer the important question as to
what degree health inequalities are actually subject to change during the period of observation
(which is mainly the adult work life). If time constant factors mediate a lot of the cross-sectional
observed health inequalities this means that there is little change and reduced chances for
interventions trying to reduce health inequalities. It also means that job status in itself does
not have a strong beneficial or harmful influence on health. If time constant factors are mainly
responsible for health inequalities in job status, interventions need to take place before the
period of observation which means before individuals actually enter the labor market. Else only
the symptoms can be treated, not the cause of health inequalities. On the other hand if time
constant factors play a minor role, health inequalities might be more susceptible to change and
intervention.

A decomposition of the health inequalities into time constant and time varying factors is of
course only possible with longitudinal data. I start by considering the general panel regression
model and the estimate of gross health inequalities through a cross-sectional OLS regression:

yit = Xit γOLS + eit (4.31)

yit = Xit γF E + ui + ϵit (4.32)

ui represents the time constant factors which might partly be unobserved. We want to
explain to which degree γOLS can be explained by ui . The problem is that we do no observe ui .
To solve this problem, I have to use a trick. First, I consider the fixed-effects transformation

115
 4. Methods

of the equation which is usually used to analyze within-person variance. It works through
demeaning all variables. That means subtracting the individual mean from each observation.

yit − ȳi = (Xit − X̄i )γF E + (ui − ūi ) + (ϵit − ϵ̄i ) (4.33)

Usually this is just a stage in the transformation where it can be shown that the unobserved
time constant part of the equation drops out. Then the difference between γOLS and γF E
could be interpreted as the part of γOLS attributable to time constant factors. However, we
would not be able to get a measure of uncertainty in form of a confidence interval this way.
One small rearrangement and a reinterpretation of the equation will give us the opportunity to
estimate the indirect effects of health on job status via time constant factors with confidence
intervals.

yit − ȳi − (Xit − X̄i )γF E − (ϵit − ϵ̄i ) = (ui − ūi ) (4.34)
yit − Xit γ − ϵit = ui (4.35)
ȳit − γXit γ − ϵ̄it = ūi (4.36)
ui = ūi (4.37)
ϵ̄it = 0 (4.38)

The panel model explicitly decomposes the error-term into a time constant component (ui )
which is basically the mean of the error within the person, and one error component which is
the time dependent deviation from this error term with an intra-individual expectation of zero.
Note that the equation ϵ̄it = 0 is true by definition. Combining and rearranging equations 4.36,
4.46, and 4.38 we get:

yit = Xit γF E + (X̄i ∗ −γF E ) + ȳi + ϵit (4.39)

This equation can be interpreted as follows: The dependent variable depends on X times
the coefficient from the fixed-effects-transformation (γF E ) plus the means of X times the
fixed-effects-coefficient which has the opposite sign (−γF E ), plus the mean of the dependent
variable without a coefficient (which is equal to a coefficient fixed at 1) and a time varying
residual term.
From this we can calculate the indirect effect of Xit through time constant factor ui . It is
the effect which is mediated through X̄i and ȳi . The total effect is then the indirect effect
through time constant factors γT C plus the indirect effect through time varying factors γT V
plus the direct effect of the time varying part of Xit which is equal to the fixed-effects-estimate

116
 4. Methods

(γdirect = γF E ). Mediation through other time varying variables is the same as normal mediation
explained in the previous part. To keep things simple, it is disregarded here.

117
 Figure 4.3.: Indirect Effect of Health Through Time Constant Factor u
(a) Before FE transformation - u is unknown

Individual Level
u

!! !
1

Health Job Status

!!" !

Level of Observations

118
(b) After FE transformation - u is estimated through means
4. Methods

!!
Individual Level
−!!" ! !!
!! !!
!! ! !!

!! !
1

Health Job Status

!!" !

Level of Observations
 4. Methods

In the following, I will analytically proof my approach. The approach is also presented
graphically in figure 4.3.
Proposition 1: The indirect association of X and Y through time constant X and time
constant Y is an unbiased estimator of the indirect association of X and Y through the
unobserved time constant factor u.

E(γ̂ȳ − γ̂X̄ γ̂F EX ) = γu (4.40)

I denote averages within individuals (over time) with bars (Greene 2003, 293).
Ti
i=1 Xit
X̄i = (4.41)
Ti
Ti
i=1 yit
ȳi = (4.42)
Ti

Where Ti is the number of observations per individual which is allowed to vary over individuals
making the whole method feasible for both balanced and unbalanced panels. Large COV,
and VAR indicate population parameters while small cov, and var indicate sample parameters.
Assumptions:

yit = Xit γF EX + ui + eit (4.43)

E(ēi ) = 0 (4.44)
ui = ūi (4.45)

It follows that:

ui = ȳi − X̄i γF EX (4.46)

For more details on this see Greene (2003, 288). The definition of the true association of X
with u in the population is:

COV (X, u)
γu = (4.47)
V AR(X)

This is also the mediated effect of the independent variable through time constant factors.
Indexes are left out for convenience. Proof:

cov(X, ȳ) cov(X, X̄)

E(γ̂ȳ − γ̂X̄ γ̂F EX ) = E( − γ̂F EX ) (4.48)
var(X) var(X)

119
 4. Methods

In the first step we just rewrite the covariance in terms of expected values:

E(X ȳ) − E(X)E(ȳ) − E(X X̄)γ̂F EX + E(X)E(X̄)γ̂F EX

= E( ) (4.49)
var(X)

Then we take the expected value of the fraction. The expected value of the expected values in
the term yields again the expected values. The expected value of the fixed-effects coefficient
estimate is the fixed-effects population coefficient (γF EX instead of γ̂F EX̄ ). The expected value
of the sample variance of X (var(x)) is the population variance of X (VAR(X)).

E(X ȳ) − E(X)E(ȳ) − E(X X̄)γF EX + E(X)E(X̄)γF EX

= (4.50)
V AR(X)

Now I rearrange the equation, so that in the next step equation 4.46 can be used.

E(X(ȳ − X̄γF EX )) − E(X)E(ȳ − X̄γF EX )

= (4.51)
V AR(X)

I substitute according to equation 4.46. The reformulation of the covariances terms will lead
us to the parameter of interest.

E(Xu) − E(X)E(u)
= (4.52)
V AR(X)
COV (X, u)
= = γu (4.53)
V AR(X)

which is what I wanted to proof. This is the analytic proof for Proposition 1.
What that means is that as long as we have panel data we can use the individual averages of
all independent variables and the dependent variable to estimate the indirect association of a
specific independent variable with the dependent variable which is due to factors that are time
constant during the period of observation. The proof does not make any assumptions about
whether the panel is balanced or unbalanced, so the method works in both cases.
For the case of health inequalities this allows us to assess to what degree health inequalities are
changeable over time, and to what degree they are already fixed by the time the observation
starts.

The proportion of the effect due to time constant factors is thus:

γT C (γȳ + γX̄ γF E )
P ropyT C = = (4.54)
γtotal (γȳ + γX̄ γF E ) + γT V + γdirect

Confidence intervals are calculated analog to the previous section using the delta-method.

120
 4. Methods

4.9. Data, Variables, and Measurement of Controls

After the detailed discussion of the statistical models, I turn to the description of the dataset
and the measures I use.

4.9.1. The Socio-Economic Panel Study - SOEP

I use data from the German Socio-Economic Panel Study (SOEP) to conduct my analyses.
The SOEP is a representative household survey with annual interviews since 1984. The survey
currently includes more than 20,000 individuals in more than 10,000 households (see Wagner,
Frick & Schupp 2007). The data for the regression analyses contains the waves P (1999) to
BB (2011). I restrict my sample in the following way:

• Only currently employed persons

• No persons in vocational or educational training

• Aged between 18 and 64

• No self-employed

As a measure of health problems, which are directly visible to the employer, I use days of
sickness absence in the last year. I take the logarithm of the (measure + 1), because I believe
that the first days of sickness absence will have a stronger impact on labor market rewards
than later increases. That means that a change from zero to 5 days has a stronger influence
than a change from 20 to 25 days of sickness absence.
An individual is defined as being in a high status job if he or she reports that he or she has a
job that required either highly specialized skills or supervisory tasks, or both.

4.9.2. Control Variables

My theory of health selection builds on a simple yet elegant model of labor market rewards by
Becker (1985). It states that LMR are not only determined by health, but other factors as well.
Those are the following:

1. Human capital (α)

2. Labor market effort (t)

3. Non-labor market effort (EN LM )

4. Effort intensity of the job (σ)

in addition I will control for:

121
 4. Methods

5 Demographic factors and period effects

6 Anticipation of status change (ALM R )

I will briefly describe each indicator used to capture these constructs. Table 4.2 lists all those
variables as an overview. In the results section (5.1), tables 5.3 and 5.4 describe the sample
statistics of the control variables.
Human capital (α) is measured mainly by the years of education an individual has received.
It could be argued that within the German context it makes more sense to control for certain
educational titles. As it is not the aim of this thesis to estimate any educational effects, a
more parsimonious measurement seems appropriate. In addition, the years of tenure with an
employer and years of full-time labor market experience are important indicators for human
capital, both general/occupational and company-specific human capital. As another measure
the current wage is used as an indirect measure of job quality and skill.
Labor market effort (t) is measured through work hours which are agreed upon in the contract
(if this figure is not available, the actual hours of work are used) and the amount of reported
overtime. Effort intensity (σ) relies on a measure of physical and psychological strain of the
job based on the occupational classification (Kroll 2011), as well as firm size and the industry
of the employer.
Non-labor market effort (EN LM ) is captured by the household size, the number of children
below 16 in the household, the hours spent on childcare and housework each weekday, and the
marital status.
Demographic factors are controlled through age, age squared, and a dummy for East Germany.
Anticipation effects (ALM R ) are covered by an item on worries about job security and an
item that captures worries about the own economic situation and satisfaction with the own work.

The theoretical formulation for these measures is:

LM R = α(e(H(ALM R ), EN LM ) ∗ τ )σ t (4.55)

122
 Table 4.2.: Description of Control Variables

Variable Description Reference to construct in theory

Anticipation of job loss
Worries about job security Dummy variable indicating strong vs. low/no worries ALM R
... own economic situation Dummy variable indicating strong vs. low/no worries ALM R
Satisfaction with work 11-point scale ALM R
Human capital
Education Years of formal education α
Job tenure Years of job tenure α
Labor market experience Years of full-time labor market experience α
Wage Log gross labor income (monthly) α
Labor market effort and effort intensity
Industry Industry of employer (NACE Rev.1.1), 10 categories: Agriculture and Fishing; Mining and Manufacturing; σ
Energy and Water Supply; Construction; Wholesale, Hotel and Restaurant; Transportation and Information; σ
Financial Intermediate; Real Estate, Law Counseling; Public Administration; Public and Private Services σ
Firm size Size of firm, 5 categories: <5; 5-19; 20-199; 200-2,000; >2,000 σ
Work hours Contractual hours of work per week t
Overtime Reported hours of overtime per week t
Effort intensity (physical) Variable indicating relative physical strain of occupational group (deciles) σ
Effort intensity (psychological) Variable indicating relative psychological strain of occupational group (deciles) σ
Non-labor market effort

123
Number of HH members Number of persons in the household EN LM
Number of children Number of children in the household EN LM
4. Methods

Marital status A persons marital status, 3 categories: married; single; other EN LM

Housework and Childcare Hours of housework and childcare per day EN LM
Youngest HH member Age of youngest person in the household EN LM
Demography and period effects
Period effects Dummy variables for three year periods
Age Age in years
Age2 Age squared
East Germany Dummy variable indicating if a person lives in East Germany
 4. Methods

4.10. Measuring Social Closure on the Labor Market

In this part, I will explain how I operationalize the concept of open and closed positions. More
precisely, I measure the degree of closure at the occupational level in two ways.

4.10.1. “Glass-Escalators” and Token Positions

The theory of Kanter (1977) states that individuals in so called token positions face difficulties
on the job as soon as they are aiming for promotion. At the same time the theory of Williams
(1992) claims that men are easier promoted in female dominated jobs, riding a figuratively
speaking “glass-escalator”.
In my study I will use the same measurement as in Hultin’s (2003) study. She divides the
occupations into three types. Those with less than 30% women are defined as male dominated.
Those occupations with more than 70% women are called female dominated occupations. All
occupations in-between are referred to as mixed occupations.
I use data from the German Federal Statistical Office (FSO) on gender proportions in occupa-
tional groups of the KLDB-92 classification of the FSO on the 3-digit level. These proportions
were calculated based on the Mikrozensus, a yearly 1% sample of the German population.
Table 4.3 shows how men and women are distributed among those three categories of occu-
pations. The results are valid for the restricted sample described in section 4.9, weighted to
make the descriptives representative for the whole population which falls under the mentioned
restrictions. 68% of the men work in male dominated occupations, 24% in mixed occupations,
and only a little less than 8% work in female dominated occupations. 56% of the women work
in female dominated, 34% work in mixed, and around 10% work in male dominated occupations.
The distribution is a little more skewed for men than for women, but all in all nothing out of
the ordinary, given the definition of the groups.

Table 4.3.: Proportion of Men and Women According to Hultin’s Classification of Occupa-
tions - in Percent

Male dominated Mixed Female dominated

Men 68.32 23.90 7.77
Women 10.13 33.72 56.15

4.10.2. Professionalism - Knowledge Intensity of Occupations

One way of merging the closure mechanisms of credentialism and professionalism in one
empirical measure is knowledge intensity of occupational groups. To the best of my knowledge

124
 4. Methods

this has not been used so far within the research on occupational closure. The idea, the
concept, and the measurement are taken from Tiemann (2010). All descriptions presented
here refer to this study where the whole concept and measure is described in more detail. The
intent is to define occupations as knowledge intensive if they have high requirements on their
workers regarding the following four activities (Tiemann 2010, 11):

• Development and Research

• Investigation and Documentation

• Training and Teaching

• Organizing flow of work or operational procedures of other persons

The level of aggregation for the measure of knowledge intensity is occupational groups based
on a 3-digit coding of the occupational classification of the FSO, same as for proportion of
women in occupations described in the previous section. The measure is based on a survey of
employees from the years 1999 and 2006 conducted by the Institute for Labor Market and and
Occupational Research (IAB). Questions about the activities stated above were used to code
a variable of knowledge intensity which ranges from 0 to 1. It represents the proportion of
employees in an occupational group which reported high demands with regard to knowledge
intensive activities. Zero means that no employee reported high demands, 1 means all employees
report high demands with regard to knowledge intensive activities. For a detailed list of the
occupations and their degree of knowledge intensity see the study by Tiemann (2010).

125
5. Results

5.1. Sample Description

In this part, I will give an overview of the sample statistics of the sample used for estimation.
First, I show the sample size for the years under investigation and show how the sample is
affected by the restrictions made in section 4.9 and by the requirements of the various methods
chosen for estimating health selection effects (tables 5.1 and 5.2). Second, I provide tables
with sample statistics for all variables in the models (tables 5.3 and 5.4). In the following
section I will show the association between health and job status.

126
 Table 5.1.: Sample and Sample Restrictions for Women
(1) (2) (3) (4) (5) (6)
Full Sample Aged <18 or >64 Not or Unemployed Still in Training Self-Employed Civil Servants
1999 7,279 5,998 3,407 3,375 3,143 2,988
2000 12,762 10,296 5,843 5,794 5,390 5,115
2001 11,629 9,317 5,386 5,321 4,969 4,710
2002 12,329 9,922 5,957 5,884 5,449 4,994
2003 11,728 9,301 5,575 5,521 5,134 4,728
2004 11,452 8,988 5,422 5,361 4,962 4,579
2005 11,012 8,552 5,197 5,137 4,738 4,363
2006 11,854 9,049 5,538 5,485 5,035 4,651
2007 11,111 8,413 5,316 5,258 4,845 4,467
2008 10,436 7,831 5,099 5,043 4,649 4,298
2009 10,987 8,158 5,408 5,351 4,922 4,548
2010 10,012 7,412 4,917 4,876 4,490 4,155

127
2011 11,260 8,291 5,486 5,452 5,018 4,648
5. Results

Total 143,851 111,528 68,551 67,858 62,744 58,244

 Table 5.2.: Sample and Sample Restrictions for Men
(1) (2) (3) (4) (5) (6)
Full Sample Aged <18 or >64 Not or Unemployed Still in Training Self-Employed Civil Servants
1999 6,806 5,852 4,337 4,298 3,843 3,542
2000 11,814 9,902 7,306 7,251 6,414 5,843
2001 10,722 8,901 6,550 6,502 5,744 5,225
2002 11,563 9,539 7,047 6,994 6,080 5,403
2003 10,883 8,817 6,417 6,355 5,526 4,939
2004 10,567 8,459 6,168 6,116 5,270 4,714
2005 10,093 7,945 5,804 5,762 4,998 4,462
2006 10,811 8,327 6,092 6,047 5,204 4,646
2007 10,121 7,697 5,771 5,744 4,989 4,471
2008 9,509 7,146 5,425 5,384 4,726 4,258
2009 10,048 7,477 5,599 5,552 4,852 4,373
2010 9,115 6,717 5,053 5,016 4,396 3,952

128
2011 10,076 7,289 5,495 5,462 4,792 4,341
5. Results

Total 132,128 104,068 77,064 76,483 66,834 60,169

 5. Results

Table 5.3.: Sample Statistics for Women

Summary Statistics
Mean Stand. Dev. Min Max
Health -.0271 .717 -2.42 1.25
Log. Days of Sickness Absence 1.13 1.32 0 5.9
High Status Job .123 .328 0 1
Log. Knowledge Intensity -1.36 .679 -3.67 -.135
Male Occupation .104 .305 0 1
Mixed Occupation .325 .469 0 1
Female Occupation .571 .495 0 1
Public Sector .287 .452 0 1
Years of Education 12.2 2.4 7 18
Years of Full-Time Experience 12 10.1 0 49
Log. Wage per Hour 2.38 .512 .0077 5.63
Years of Tenure 9.24 8.87 0 47
Contractual Hours of Work 29.4 11.5 1 80
Hours of Overtime 1.53 2.8 0 23.1
Firmsize: <5 .119 .324 0 1
Firmsize: 5-19 .201 .401 0 1
Firmsize: 20-199 .293 .455 0 1
Firmsize: 200-2,000 .206 .405 0 1
Firmsize: >2,000 .18 .384 0 1
NACE 1.1: Agriculture and Fishing .0081 .0897 0 1
NACE 1.1: Mining and Manufacturing .162 .368 0 1
NACE 1.1: Energy and Water Supply .006 .0774 0 1
NACE 1.1: Construction .0163 .127 0 1
NACE 1.1: Wholesale, Hotel and Restaurant .208 .406 0 1
NACE 1.1: Transportation and Information .0336 .18 0 1
NACE 1.1: Financial Intermediate .048 .214 0 1
NACE 1.1: Real Estate, Law counseling .0915 .288 0 1
NACE 1.1: Public Administration .359 .48 0 1
NACE 1.1: Public and Private Services .0675 .251 0 1
Degree of Psychological Strain 5.05 3.1 1 10
Degree of Physical Strain 4.96 2.53 1 10
Household Size 2.83 1.18 1 14
Number of Children in HH .495 .801 0 7
Married .617 .486 0 1
Single .216 .411 0 1
Other .167 .373 0 1
Hours of Housework and Childcare 3.7 3.79 0 20
Age of youngest HH Member 26.1 17.3 0 64
Great Worries About Job Security .15 .357 0 1
Great Worries About own Economic Situation .223 .416 0 1
Satisfaction with Work 7 2.03 0 10
Age 41.8 10.7 18 64
Age2 1861 887 324 4096
East Germany .241 .428 0 1

129
 5. Results

Table 5.4.: Sample Statistics for Men

Summary Statistics
Mean Stand. Dev. Min Max
Health .0143 .6989 -2.419 1.249
Log. Days of Sickness Absence 1.108 1.299 0 5.9
High Status Job .2836 .4508 0 1
Log. Knowledge Intensity -1.109 .5799 -3.665 -.1349
Male Occupation .677 .4676 0 1
Mixed Occupation .2449 .43 0 1
Female Occupation .0781 .2684 0 1
Public Sector .159 .3657 0 1
Years of Education 12.35 2.635 7 18
Years of Full-Time Experience 18.85 11.32 0 49
Log. Wage per Hour 2.642 .5163 .0077 6.405
Years of Tenure 11.05 9.983 0 50.1
Contractual Hours of Work 38.88 7.61 .6 80
Hours of Overtime 2.819 4.143 0 23.1
Firmsize: <5 .0521 .2222 0 1
Firmsize: 5-19 .1557 .3625 0 1
Firmsize: 20-199 .3084 .4618 0 1
Firmsize: 200-2,000 .2411 .4278 0 1
Firmsize: >2,000 .2428 .4288 0 1
NACE 1.1: Agriculture and Fishing .0161 .126 0 1
NACE 1.1: Mining and Manufacturing .3989 .4897 0 1
NACE 1.1: Energy and Water Supply .0177 .1318 0 1
NACE 1.1: Construction .1047 .3061 0 1
NACE 1.1: Wholesale, Hotel and Restaurant .1145 .3184 0 1
NACE 1.1: Transportation and Information .0717 .258 0 1
NACE 1.1: Financial Intermediate .0437 .2044 0 1
NACE 1.1: Real Estate, Law Counseling .0754 .2641 0 1
NACE 1.1: Public Administration .1148 .3188 0 1
NACE 1.1: Public and Private Services .0424 .2016 0 1
Degree of Psychological Strain 5.658 2.689 1 10
Degree of Physical Strain 5.919 3.075 1 10
Household Size 2.977 1.295 1 14
Number of Children in HH .621 .9295 0 8
Married .6499 .477 0 1
Single .2416 .4281 0 1
Other .1085 .311 0 1
Hours of Housework and Childcare 1.432 1.698 0 20
Age of youngest HH Member 24.02 17.16 0 64
Great Worries About Job Security .1687 .3745 0 1
Great Worries About own Economic Situation .2022 .4016 0 1
Satisfaction with Work 6.983 1.982 0 10
Age 41.87 10.72 18 64
Age2 1868 906.4 324 4096
East Germany .227 .4189 0 1

130
 5. Results

5.2. Health and Job Status - Descriptive Evidence

Figure 5.1 shows the absolute amount of health inequalities in subjective health between
incumbents of high status jobs and incumbents of regular jobs. The two sub-figures a) and b)
present unweighted results. Sub-figures c) and d) show results weighted with SOEP weights and
sub-figures e) and f) show results weighted with SOEP weights and a weight which balances age
differences between the groups. The weight for age was constructed using entropy balancing
with STATAs ado package ebalance (Hainmueller & Xu 2011). The last row therefore
presents the age adjusted health inequalities.

The figures show two main results. First, the health difference between regular and high
status jobs is stronger for women than for men in the private sector. In the public sector no
such gender differences can be observed. Second, weighting the sample with SOEP weights and
equalizing age in the two groups increases the overall differences. Especially, the adjustment
for age seems important. Workers in high status jobs are on average older and have a lower
health status due to their higher age. Controlling for age gives us the social differences which
cannot be attributed to age. For women in the private sector the adjusted differences are 0.24
on the latent health score which is about one third of a standard deviation. In high status jobs
the number of individuals rating their own health as good or very good is 14 percentage higher
than in regular jobs. For men these differences are 0.18 and 11 percentage points. For the
public sector they are 0.17 and 12 percentage points for both men and women. All adjusted
differences between status groups are significant, the difference in difference between men and
women in the private sector is significant on the 10% level.

131
 Figure 5.1.: Health Inequality due to Job Status - Subjective Health

(a) Latent Health (b) (Very) good SRH

Private Sector, Women Private Sector, Men Private Sector, Women Private Sector, Men

-0.04 -0.01 0.56 0.58

0.11 0.10 0.66 0.65

-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Public Sector, Women Public Sector, Men Public Sector, Women Public Sector, Men

-0.07 -0.07 0.54 0.54

0.05 0.07 0.61 0.63

-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Mean of latent Health Proportion of Respondents in good Health
Regular Status High Status Job Regular Status High Status Job

(c) Latent Health - SOEP weights (d) (Very) good SRH - SOEP weights
Private Sector, Women Private Sector, Men Private Sector, Women Private Sector, Men

-0.07 -0.03 0.54 0.57

0.12 0.08 0.65 0.64

132
-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Public Sector, Women Public Sector, Men Public Sector, Women Public Sector, Men
5. Results

-0.09 -0.08 0.53 0.54

0.05 0.05 0.62 0.62

-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Mean of latent Health Proportion of Respondents in good Health
Regular Status High Status Job Regular Status High Status Job

(e) Latent Health - SOEP + age weights (f) (Very) good SRH - SOEP + age weights
Private Sector, Women Private Sector, Men Private Sector, Women Private Sector, Men

-0.12 -0.09 0.51 0.53

0.12 0.08 0.65 0.64

-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Public Sector, Women Public Sector, Men Public Sector, Women Public Sector, Men

-0.13 -0.12 0.50 0.51

0.05 0.05 0.62 0.62

-.2 -.1 0 .1 .2 -.2 -.1 0 .1 .2 0 .2 .4 .6 .8 0 .2 .4 .6 .8

Mean of latent Health Proportion of Respondents in good Health
Regular Status High Status Job Regular Status High Status Job
 5. Results

Now we can take a look at the differences in days of sickness absence (see figure 5.2). It is
similar to the results from health inequalities in subjective health in the way that using SOEP
and age weights increases the overall health inequalities. For women in the private sector the
age adjusted health inequalities are about two days. That means women in high status jobs
report on average two days less of sickness absence. For men in the private sector those in
high status jobs have an average absence of 5.23 days a year whereas men in regular jobs are
absent from work 9.79 days a year. This is a difference of more than 4 days, more than twice
the difference of women. In the public sector the overall amount of days of sickness absence is
higher than in the private sector. The relative differences between regular and high status jobs
are only slightly higher. They are 2.6 days for women and 4.3 days for men.

133
 Figure 5.2.: Health Inequality due to Job Status - Sickness Absence

(a) Days of Sickness Absence (b) Days of Sickness Absence - SOEP weights
Private Sector, Women Private Sector, Men Private Sector, Women Private Sector, Men

7.94 9.19 8.37 9.34

6.79 5.06 6.80 5.23

0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15

Public Sector, Women Public Sector, Men Public Sector, Women Public Sector, Men

11.08 10.63 11.72 10.48

9.08 6.18 9.65 6.56

0 5 10 15 0 5 10 15 0 5 10 15 0 5 10 15
Days of Sickness Absence Days of Sickness Absence
Regular Status High Status Job Regular Status High Status Job

(c) Days of Sickness Absence - SOEP + age weights

Private Sector, Women Private Sector, Men

134
8.76 9.79

6.80 5.23
5. Results

0 5 10 15 0 5 10 15

Public Sector, Women Public Sector, Men

12.23 10.88

9.65 6.56

0 5 10 15 0 5 10 15
Days of Sickness Absence
Regular Status High Status Job
 5. Results

I want to address one more question descriptively: Can we say there are strong health
inequalities between incumbents of high and regular status jobs? The absolute values presented
above tell us little about that. Health inequalities have no natural scale where the size of the
inequalities can be easily assessed. The unit is not intuitive as it is with income inequalities.
To solve this problem, I use an anchor to compare the degree of health inequalities. I compare
health inequalities due to job status with health inequalities due to education. To be more
precise I use the common CASMIN classification to build three educational groups (primary,
secondary, tertiary) and compare the differences in the health variables between primary and
tertiary education. This is useful, because health inequalities due to education have been
found to be robust, relatively strong and can therefore serve as an anchor. Figure 5.3 shows
the comparison of the magnitude of the effects in all four relevant groups. While there is
some variation we can see that on average inequalities between job status are about 60-100
percent the size of inequalities between primary and tertiary education. That means that the
differences in subjective health between high and low status jobs is about the same magnitude
as the difference between persons with a university degree and those who only received basic
schooling. So we can conclude, that while health inequalities investigated in this thesis are
not exceptionally large, they are still quite sizable and of comparable importance as health
inequalities between educational groups, at least with regard to their gross size.

Key results from this section are: There are sizable health inequalities between
high and low status jobs, both for subjective health and for sickness absence. These
inequalities are larger in the private sector. Within the private sector they are larger
for women than for men.

135
 Figure 5.3.: Size of Health Inequalities - Job Status Compared to Education

(a) Latent Health

Private Sector, Women Private Sector, Men

0.24 0.18

0.24 0.17

0 .1 .2 .3 .4 .5 0 .1 .2 .3 .4 .5

Public Sector, Women Public Sector, Men

0.16 0.27

0.17 0.17

0 .1 .2 .3 .4 .5 0 .1 .2 .3 .4 .5
Mean of latent Health
Education Job Status

(b) (Very) good SRH

Private Sector, Women Private Sector, Men

0.16 0.13

0.14 0.11

0 .1 .2 .3 .4 .5 0 .1 .2 .3 .4 .5

136
Public Sector, Women Public Sector, Men
5. Results

0.12 0.18

0.11 0.11

0 .1 .2 .3 .4 .5 0 .1 .2 .3 .4 .5
Proportion of Respondents in good Health
Education Job Status

(c) Days of Sickness Absence

Private Sector, Women Private Sector, Men

3.36 6.17

1.96 4.57

0 2 4 6 8 10 0 2 4 6 8 10

Public Sector, Women Public Sector, Men

3.16 6.40

2.58 4.32

0 2 4 6 8 10 0 2 4 6 8 10
Days of Sickness Absence
Education Job Status
 5. Results

5.3. Health as a Latent Variable - Confirmatory Factor

Analysis

5.3.1. Establishing Measurement Invariance

As discussed in sections 4.3 and 4.3.1 there has been serious doubt whether self-rated health is
a reliable indicator for the analysis of health inequalities. To address this problem four questions
will be answered in this part.

1. Are answers on items of subjective health in the SOEP comparable across gender in the
population?

2. Are these items also comparable between workers in high and regular job status and
between private and public sector workers?

3. Are these items comparable across time?

4. Is a latent health variable created by a factor score a good predictor of mortality?

Figure 5.4 reports the results from a confirmatory factor analysis (CFA) where the latent
variable health is determining the observed variables self-rated health (SRH), satisfaction with
health (SAT) and worries about health (WAH) for the year 2000 in the SOEP. All observed
variables are ordinal in their nature. The measurement model and the results are reported with
unstandardized coefficients in figure 5.4.
I restricted the model, so that intercepts of indicators are equal across groups. This means
that I test for structural invariance between genders. I do not test separately for metric
invariance as this is highly complicated when dealing with categorical dependent variables. As
structural invariance implies metric invariance, establishing structural invariance is sufficient.
The χ2 statistic with 13 degrees of freedom is 53 and highly significant, indicating a bad model
fit. However, as already explained in the methods sections this statistic is very sensitive to
the number of observations which is very high in my sample (> 8000). Therefore the other
fit indexes should also be scrutinized. CFI is 0.999, TLI 1.000 indicating that model fit is
almost as good as the saturated model. RMSEA is 0.021 [CI: 0.015;0.027], and the confidence
interval is well within the range of acceptable values. Overall the model fit is still very good,
and structural invariance between genders can be established for my sample. This means that
in addition to associations with other variables, means of the health variables can be compared
between men and women. This will be done in the next section where descriptive evidence on
health inequalities between job status is provided. Further restrictions on the model to test for
equivalence of error-term variance are possible, but unnecessary for any of my further analyses.
Therefore, I will not conduct them.
The first question of this section can be answered now. The items SRH, SAT, and WAH
in the SOEP can be compared across gender in my sample. Any further analyses rely in

137
 5. Results

their conclusions on these tests. The confirmatory factor analysis addresses the critique that
differences in effect sizes might be due to different meanings that these items carry for men
and women. Regardless of whether this argument is theoretically sound, it has no empirical
relevance for my data set.

138
 Figure 5.4.: Confirmatory Factor Analysis - Gender

χ^2 = 52.99
df = 13
CFI = 0.999 F= 0.880
TLI = 1.000 Health M=0.819
RMSEA = 0.021 [0.015;0.027]

0.896 1 0.731

F=1 N=1
F=1

139
SAT M = 1.034 SRH WAH H = 1.013
M = 1.037
5. Results

Sat1 = -2.207
Sat2 = -1.869 SRH1 = -2.131
Sat3 = -1.506
Sat4 = -1.221
SRH2 = -1.224 WAH1 = -1.109
Sat5 = -0.736 SRH3 = -0.225 WAH2 = 0.303
Sat6 = -0.446 SRH4 = 1.141
Sat7 = -0.005
Sat8 = 0.687
Sat9 = 1.208

Intercept

Note: F = Women; M = Men

 5. Results

Now we can turn to the question whether the health items can be compared over the years
under observation. In my study these are the years 1999-2011. As this seems to be of lesser
importance in the literature I will just briefly summarize the results of figure 5.5, which presents
the unstandardized results and fit statistics of the models. Structural invariance can also be
accepted in my sample despite the significant χ2 statistic of 824 with 156 degrees of freedom.
The high χ2 is mainly due to the sample size of over 100,000 observations. The fit indexes are
still very good, only slightly worse than the fit of the structural invariance model for gender.
The CFI is 0.999, TLI 1.000, RMSEA 0.019 [CI: 0.018;0.021].
It is therefore save to conclude, that all health indicators can be compared over time for my
sample.

The last measurement invariance test I am going to conduct is between workers in different
job status (see figure 5.6). Again I use the year 2000 as a reference. Comparability of subjective
health across job status is essential to allow causal interpretation of estimated effects in
my regression models. If subjective health turned out not to be invariant it could not be
distinguished whether estimated coefficients came from systematic measurement error or an
actual causal relationship.
Structural invariance leads to a model with a significant χ2 statistic of 82 with 13 degrees of
freedom. However, similar to invariance over time and across gender, we can see that all other
model fit indicators are very good. CFI and TLI are 0.998 and 0.997 respectively. The estimate
of RMSEA is 0.031 [0.025;0.038] with the confidence interval indicating that it is very unlikely
that the true parameter is above the critical value of 0.05. Such excellent fit indicators let me
conclude that subjective health is comparable across job status.

Key results from this section are: Self-rated health items from the SOEP are
comparable across gender, time, and job status. Differences in health effects can be
substantively interpreted, because reporting heterogeneity is not an important issue
in the data set.

140
 Figure 5.5.: Confirmatory Factor Analysis - Time

1999 = 0.874
2000 = 0.971
χ^2 = 824.39 2001 = 0.932
df = 156 2002 = 0.916
CFI = 0.999 2003 = 0.895
2004 = 0.895
TLI = 1.000 2005 = 0.883
RMSEA = 0.019 [0.018;0.021] Health 2006 = 0.889
2007 = 0.851
2008 = 0.871
2009 = 0.919
2010 = 0.916
2011 = 0.969

0.895 1 0.668

1999 = 1
1999 = 1
2000 = 0.945 1999 = 1 2000 = 1.011
2001 = 0.970 2000 = 0.959 2001 = 1.031
2002 = 0.986 2001 = 0.968 2002 = 1.067
2003 = 0.996 2002 = 0.984 2003 = 1.083
2004 = 1.001

141
2003 = 0.990 2004 = 1.047
SAT 2005 = 1.001 SRH 2004 = 0.987 WAH 2005 = 1.090
2006 = 1.004 2005 = 0.999 2006 = 1.056
2007 = 1.027
5. Results

2006 = 0.993 2007 = 1.079

2008 = 1.029 2007 = 1.006 2008 = 1.071
2009 = 0.991 2008 = 0.990 2009 = 1.036
2010 = 0.992 2009 = 0.984 2010 = 0.994
2010 = 0.989

Sat1 = -2.283
Sat2 = -1.881 SRH1 = -2.181
Sat3 = -1.484 SRH2 = -1.236 WAH1 = -1.087
Sat4 = -1.185 SRH3 = -0.214 WAH2 = 0.336
Sat5 = -0.729
Sat6 = -0.437
SRH4 = 1.242
Sat7 = 0.042
Sat8 = 0.793
Sat9 = 1.417

Intercept
 Figure 5.6.: Confirmatory Factor Analysis - Job Status

χ^2 = 82.13
df = 13
CFI = 0.998 N = 0.879
TLI = 0.997 Health H = 0.882
RMSEA = 0.031 [0.025;0.038]

0.881 1 0.701

N=1 N=1
N=1

142
SAT H = 1.044 SRH WAH H = 1.078
H = 1.006
5. Results

Sat1 = -2.309
Sat2 = -1.954 SRH1 = -2.274
Sat3 = -1.548
Sat4 = -1.237
SRH2 = -1.282 WAH1 = -1.122
Sat5 = -0.744 SRH3 = -0.223 WAH2 = 0.321
Sat6 = -0.445 SRH4 = 1.198
Sat7 = 0.014
Sat8 = 0.729
Sat9 = 1.246

Intercept

Note: N = Normal Job; H = High Status Job

 5. Results

5.3.2. External Reliability - Latent Health as a Predictor of Mortality

Table 5.5.: Latent Health Factor Score as Predictor of Mortality

(1) (2)
Men Women
Latent Health 0.405 0.479
(-19.88) (-13.61)
R2 all Variables 0.471 0.423
R2 latent Health 0.372 0.280
Subjects 13135.000 13020.000
Deaths 914.000 723.000
Exponentiated coefficients; t statistics in parentheses

The previous section showed that my measure of health as a latent variable is internally
reliable. It yields consistent scores for all relevant groups. Now, I address the external reliability.
Is this latent health score a useful variable? The gold standard for such an assessment is a
mortality analysis. This is important, since a general health indicator, which does not predict
mortality, is not very useful.
To assess whether the latent health variable estimated from the confirmatory factor analysis
model, as defined above, is a good predictor of mortality, I ran Cox proportional hazard models
with latent health as an independent variable and years of education, number of doctor visits,
nights spend in the hospital, occupational class according to the EGP-classification as control
variables. Mortality estimates are obtained for all individuals above the age of 30, reported in
hazard-ratios. The results show that latent health is a highly significant predictor even after con-
trolling for some standard-demographic variables and some more objective measures of health.
This holds for both men and women. Using Royston’s (2006) method for estimating a statistic
similar to the R2 in linear regression, I come up with a contribution of latent health to the
explanation of the variation in the hazard of mortality of 0.37. All other controls merely explain
an additional 0.1. These are the estimates for men. For women the contribution of latent health
is a little less, 0.28. Other variables explain 0.14 of the variation in the mortality hazard. Still
it is clear that the factor score is a very important predictor of mortality, the gold standard of
external reliability. The factor score of latent health is internally reliable as shown in the confir-
matory factor analysis, and the score is also externally reliable as shown in the mortality analysis.

Key results from this section are: The constructed subjective health variable
shows good external reliability through strong prediction of mortality in the data set.

143
 5. Results

5.4. Basic Analyses

In this part, I will conduct a first test of the hypotheses made in the theory chapter. At this
stage it will be only considered whether there is an influence of health on high status job
(Hypothesis H1) and if this effect of health is stronger for men than for women (Hypothesis
H2a). Hypothesis H5 is also scrutinized, comparing the effects of health in private and public
sector. The analyses are also repeated to address the question whether a health measure visible
to the employer, like days of sickness absence, shows a different pattern (Hypothesis H8). For
this purpose a fixed-effects-logit approach is chosen (see section 4.5). The effect of change
in health at time point t on change in job status at time point t+1 is estimated. All models
control for the variables described in section 4.9.2 and for unobserved time constant factors
thanks to the fixed-effects method.
What is estimated is therefore only a direct effect of health on job status which is not mediated
by other labor-market, or non-labor-market factors or by unobserved personal characteristics like
skill or motivation. In this restrictiveness this constitutes are very strong test of the hypothesis
that subjective health influences job status. However, only the uni-directional case is considered.
Effects subsumed under the social causation hypothesis are discussed in a later section.
The analyses are split up between public and private sector and between men and women
yielding four basic configurations. Figure 5.7 gives an overview of the effects of health on job
status. The dots indicate the point estimate measured as log-odds. The lines indicate the 95%
confidence intervals. If a confidence interval line crosses the line at zero, the effect is not sta-
tistically significant on the 5% level. If the line is not crossed the effect is statistically significant.

We can see that a change in health increases the chance of a positive change in job status
for women in the private sector. This was to be expected from hypothesis H1. Surprisingly,
health does not play a role for men in the private sector. It was expected that the effect should
be stronger for women, but not that there is no effect for men (H2a).
In the public sector, we find no effects of health on job status. This confirms the hypothe-
sis, which states that in the public sectors due to different rules for promotion, health does
not play a role for high status attainment (H5). Promotion due to age or tenure are not
related to effort and therefore not health-related. So the promotion regime seems impor-
tant in determining whether health selection takes place or not. In case employers do not
select on effort-relevant criteria there will be no health inequalities generated by health selection.

Now we can take a look at how the results work if we use a visible indicator as a selective
factor. Do more days of sickness absence reduce the chance of getting into a high status
job? And are the gender differences in the health effect the same as they were with subjective
health?
Figure 5.8 shows the effect of log. days of sickness absence on job status. We can see that

144
 5. Results

Figure 5.7.: Change in Health as a Predictor of Change in Job Status

Women, Private Sector Women, Public Sector

-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5
Men, Private Sector Men, Public Sector

-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in table A.1 in the appendix.

from a gender perspective the results are the opposite of the results from the regressions with
subjective health as an independent variable. In the public sector sickness absence does not
predict a change in job status. However, in the private sector men who show an increase in
days of sickness absence have a reduced chance of changing from regular jobs into high status
jobs. For women we do not find a statistically significant association. Note however, that the
direction of the effect for women is the same as for men. This stands in contrast to the effect
of general health for men that was basically zero, and not the same direction as the female
effect.
The hypothesis that sickness absence has more severe consequences for men than for women
with regard to job status can be confirmed for the private sector. This supports the idea that a
culture of presenteeism penalizes absence from work for men (H8).

Key results from this section are: Women’s health is a predictor of subsequent
job status in the private sector. For men in the private sector and for both men and
women in the public sector the effects are close to zero. Sickness absence predicts job
status for men in the private sector, and only to a smaller (non significant) degree for
women. In the public sector sickness absence does not matter for job status.

145
 5. Results

Figure 5.8.: Change in Sickness Absence as a Predictor of Change in Job Status

Women, Private Sector Women, Public Sector

-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2
Men, Private Sector Men, Public Sector

-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in table A.2 in the appendix.

5.4.1. Other Model Specifications

Figure 5.9 shows the same analyses as figure 5.7, but with some different specifications of
the model. We can see that the effect of health for women in the private sector - the only
substantial effect in the full model specification -is gradually reduced when taking additional
sets of variables as controls into the regression. However, the overall trend is not very strong.
In all other settings differences between models are existent, but the differences are not huge.
In no case does one model specification produce a completely different result. All in all the
results seem to be very stable and robust to the sets of control variables.
The same is true for different specifications of the models with sickness absence. They are
presented in figure 5.10 and are rather robust. Controlling for an anticipation effect has the
strongest impact on the estimate coefficient of sickness absence.

Key results from this section are: Effect estimation is not sensitive to the intro-
duction of control variables.

146
 5. Results

Figure 5.9.: Different Model Specifications of Results - Health

Women, Private Sector Women, Public Sector

Baseline
Human Capital
LM Effort
NLM Effort
Effort Intensity
Anticipation
-.5-.4-.3-.2-.1 0 .1 .2 .3 .4 .5 -.5-.4-.3-.2-.1 0 .1 .2 .3 .4 .5
Men, Private Sector Men, Public Sector
Baseline
Human Capital
LM Effort
NLM Effort
Effort Intensity
Anticipation
-.5-.4-.3-.2-.1 0 .1 .2 .3 .4 .5 -.5-.4-.3-.2-.1 0 .1 .2 .3 .4 .5

95% Confidence-Interval Point Estimate

Figure 5.10.: Different Model Specifications of Results - Sickness Absence

Women, Private Sector Women, Public Sector

Baseline
Human Capital
LM Effort
NLM Effort
Effort Intensity
Anticipation
-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2
Men, Private Sector Men, Public Sector
Baseline
Human Capital
LM Effort
NLM Effort
Effort Intensity
Anticipation
-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2

95% Confidence-Interval Point Estimate

147
 5. Results

5.4.2. Taking a Closer Look at Closure in the Private Sector: Glass

Escalator and Professionalism
So far we have seen that health selection does not operate within the public sector with regard
to job status. This was expected from the theory (H5). Now, I will take a closer look at
differences within the private sector. The estimates from the last section are averaged estimates
for the private sector. We can make differentiations of occupations according to different
kinds of social closure. In particular, I will look at the share of women in an occupation and
at the degree of knowledge intensity of an occupation. The first measure allows me to test
the “glass-escalator” hypothesis (H7c) versus the homophily-hypothesis (H7b). The second
measure allows to assess in how far credentials or professionalism modify health selection
processes (H7a).
As described in more detail in section 4.10, I distinguish between female-dominated, male-
dominated, and mixed occupations. I take the natural logarithm of knowledge intensity as an
interaction term with health into the regression.
The results are shown in figure 5.11. We can see that there is a significant logarithmic
relationship between knowledge intensity and the health effect on job status. For women who
work in occupations with low knowledge intensity the health effects are stronger and statistically
significant. While professionalism and specialization get stronger health becomes less important
for job status among women. For men the association of the effect size and knowledge intensity
is weak, not significant and at no point is health a substantial predictor of job status.
There seems to be no clear interaction pattern for sickness absence with knowledge intensity.
Figure 5.12 shows the effect of log. days of sickness absence on job status depending on
knowledge intensity. For men the peculiar finding is that there is no overall trend of the health
effect with rising knowledge intensity, but in between the effects become significant. This is due
to the fact that the number of observations is higher in the middle range of knowledge intensity
than for the very high or very low values. For women the trend is inversed and insignificant.
It seems that a high degree of knowledge intensity does protect workers from competition, at
least competition related to effort and health (H7a). However, for men this type of closure
does not make any difference in their health effect. This supports the hypothesis that men do
not adjust their effort to their health status. Else we would expect to see some differences in
the effects between different types of occupational closure.

148
 5. Results

Figure 5.11.: Health Effect Depending on Knowledge Intensity of Occupation

Women Men
1.5
1
Health Effect
.5
0
-.5

0 .5 1 0 .5 1
Knowledge Intensity
95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in table A.3 in the appendix.

Figure 5.12.: Effect of log. Days of Sickness Absence Depending on Knowledge Intensity
of Occupation

Women Men
.4
.2
Health Effect
0
-.2

0 .5 1 0 .5 1
Knowledge Intensity
95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in table A.4 in the appendix.

In figure 5.13 we have an assessment of the “glass-escalator” hypothesis for the question
at hand. Is health selection stronger for women in female or in male dominated occupations?
And is this effect symmetric - so are the results for men the opposite?
The results for women show that only in female dominated occupations there is a strong and

149
 5. Results

significant effect of health on job status. For mixed occupations the effect is almost zero. For
male dominated occupations the effect is close to zero and not significant. This supports the
“glass-escalator” hypothesis (H7c) and not the homiphily hypothesis (H7b).
For men again we find that there is no substantial health effect in any of the three types of
occupations. Regardless of token or majority status, health does not seem to play a role for a
change in job status for men.
When looking at days of sickness absence we see that it is mostly women and men in male
dominated occupations who are influenced in their job status (see figure 5.14). The point
estimate for women is much stronger than for men. The standard error is also very large due to
the fact that there are very few women in the sample working in a male dominated occupation1 .
The strong effect for women points to a token position in these occupations. The culture
of presenteeism puts pressure on men, but even more pressure on women. This lends strong
support to the idea that male work culture demands presenteeism and punishes absenteeism
(H8). Interesting is that in female dominated occupations for both genders we get estimates
of similar size to the effect of sickness absence for men in male dominated occupations. In
female dominated jobs it seems that there is a slight selection effect regardless of gender. Due
to sampling uncertainty the conclusion is not as strong as for male dominated occupations.
Overall it is interesting that women face health selection in female dominated occupations,
but not in other kinds of occupation. This is an important finding for two reasons. First, it
shows that the context of occupation mediates health effects. Social context does play a role
in determining whether health selection produces health inequalities or not. Second, women are
most often employed in female-dominated occupations. This is almost a truism, but relevant.
Most women - but not all - work in an environment where job status is linked to their health.
And most jobs in those occupations are traditionally not well paid and offer little possibilities
for advancement anyway. Health selection in this case applies to those already worse off on the
labor market, not the better offs2 .
Key results from this section are: Subjective health is a selective factor only
for women in female dominated jobs supporting the “glass-escalator” hypothesis.
Sickness absence negatively affects job status in male dominated occupations, possibly
stronger for women.
The higher the knowledge intensity is, the lower the subjective health effect for women.
The trend is opposite for sickness absence. For men there is no interaction of sickness
absence and knowledge intensity.

1
If the analyses are repeated using a fixed-effects-LPM point estimates are also significant.
2
In other settings this might be different.

150
 5. Results

Figure 5.13.: Health Effect Depending on Male or Female Dominated Occupations

Women Men
> 70%

30-70%

< 30%
-.8 -.6 -.4 -.2 0 .2 .4 .6 .8 -.8 -.6 -.4 -.2 0 .2 .4 .6 .8

95% Confidence-Interval 90% Confidence-Interval

Note: The complete results of the regression can be found in tables A.5 and A.6 in the
appendix.

Figure 5.14.: Effect of log. Days of Sickness Absence Depending on Male or Female
Dominated Occupations

Women Men
> 70%

30-70%

< 30%
-.6 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 -.6 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5

95% Confidence-Interval 90% Confidence-Interval

Note: The complete results of the regression can be found in tables A.7 and A.8 in the
appendix.

151
 5. Results

5.5. Effects of Long Term Health Conditions

The previous analyses looked at the effect of contemporary changes in health. In this part,
I will take a look at long standing health conditions to test whether a long standing health
condition has a stronger impact on job status than a short-term health condition (Hypothesis
H2b). For this purpose I construct a categorical variable indicating how long a person has
reported less good or poor health. I count how many years in a row each individual reports
poor health. Then I categorize the variable into four categories:

1. not in poor health

2. 1 up to 2 years in poor health

3. 2 up to 4 years in poor health

4. 4 years and more in poor health

The reference category for the analysis are those individuals who do not report poor health.
The method is fixed-effects-logit regression, as before. All models control for the same set of
variables as all previous models. In figure 5.15 we see the results of long-term health conditions
on job status. For women in the private sector we see that the longer the health condition
lasts the stronger the effect on job status is. For 1 year in poor health the effect is almost zero.
For 2-4 years it is about 0.25, but not significant. Long term spells of 4 years and more have a
severe effect on the chance of getting into a high status job. The effect is about 0.45 which is
equal to an odds-ratio of 1.56.
For men in the private sector only 4 years and more in poor health have a negative impact
on job status. The point estimate is about 0.22 and significant. The other categories are not
significantly different from the reference category and show no substantial effect. These results
are in line with the prediction (H2a), that the health effect should be less for men than for
women. It seems that only long lasting severe conditions influence men’s chance of getting
into a high status job. This would support the idea that men usually do not adjust their effort
to their health status. Very longstanding health conditions seem to be the exception.
In the public sector we do not find the same effects as in the private sector. For men, long-
standing poor health has a negative impact, but the effect is not significant3 . The conclusion
that long standing health conditions influence men’s job status in the public sector should take
the large standard error into account. In conclusion, the results suggest that hypothesis H2b
can be seen as accepted for the private sector and has to be rejected for public sector.

Key results from this section are: The longer the period of poor health lasts the
stronger the negative effect is on job status. This gradient is stronger for women
than for men, but cannot be established with certainty in the public sector.
3
probably due to smaller sample size.

152
 5. Results

Figure 5.15.: Change in Long Term Bad Health as a Predictor of Change in Job Status

Women, Private Sector Women, Public Sector

1 year

2-3 years

4+ years
-.8 -.6 -.4 -.2 0 .2 .4 .6 -.8 -.6 -.4 -.2 0 .2 .4 .6
Men, Private Sector Men, Public Sector
1 year

2-3 years

4+ years
-.8 -.6 -.4 -.2 0 .2 .4 .6 -.8 -.6 -.4 -.2 0 .2 .4 .6

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in table A.9 in the appendix.

5.6. Health Effects and the Status of Applicants vs.

Incumbents
Up to this point, I have treated change in job status as if changing from a normal job into a
high status job means the same as changing from a high status job into a normal job. This
implies treating a promotion and a demotion as if the same mechanisms are at work. Is it
a feasible assumption that climbing up the job ladder relates in the same way to health as
tumbling down the job ladder?
The theory in section 2.8.1.2 states that this would be an oversimplification of health selection
mechanisms if open and closed positions are considered. High status jobs are defined as closed
positions. Therefore it is expected to be hard to get into such a position, because the applicants
face a job competition or career tournament situation. The positive effect of a closed position
for an employee is that incumbents are protected from competition. Health should therefore be
among the criteria on which employers select their employees when looking for new incumbents
of a high status job. However, once occupying such a position health should no longer affect
job status. The incumbents can keep their position regardless of performance and health. Thus,
I use survival analysis to estimate the effects of health on the separate events, high status

153
 5. Results

attainment and loss of high status job, to test hypotheses H4, H5, and H6.

The results presented in this section are based on a non-parametric discrete event-history
or survival analysis described in section 4.6. The models estimate the “risk” to change job
status from normal to high status and vice versa. All employees in normal jobs are treated as
applicants, all those in high status jobs as incumbents. The event for applicants is the change
into a high status job, for incumbents the event is changing into a regular job. After the event
takes place or an observation is censored, the analysis ends. As in the previous sections the
results are reported in log-odds. All models control for the same set of covariates as in the
previous models.
Figure 5.16 shows the results of the analysis stratified by gender and by private and public
sector. Just as a reminder: In the regular analysis health showed substantial effects only for
women in the private sector.
For women in the private sector only applicants are selected according to their health status
as proposed by hypothesis H4. Incumbents of high status jobs have a similarly higher risk of
dropping out of their position with increasing health (sic!), which is a very surprising result.
Taking the low significance level into account only marginally reduces the concern about the
finding, because the size of the effect is quite sizable. Men in the private sector do not face
health selection as either incumbents or applicants, which supports hypothesis H4, but partially
speaks against hypothesis H1.
For women in the public sector health does not play a role for job change neither as incumbents
nor as applicants. The same holds true for men in the private sector. In the public sector there
is a positive health effect on high status attainment as an applicant, and a negative effect
as an incumbent, both insignificant due to small sample size. These results are supporting
hypothesis H5 for women, stating that there is no health selection in the public sector. For
men the picture remains unclear.

If we look at the effect of sickness absence in figure 5.17 we can see that the results are
similar to the results from health as an independent variable. In the fixed-effects-logit analysis
we could only find a significant effect of sickness absence on status change for men in the
private sector. Sickness absence only matters for applicants for high status jobs, not for
incumbents, which supports the theory. The effect vanishes once a person already is inside a
closed position. In all other constellations there are no significant effects of sickness absence
on the chance of either getting or losing a high status job. For incumbents of high status jobs
in the public sector, the number of events was too low to report reliable estimates. However,
the low number of drop-outs indicates that losing such a job is so rare, that it probably cannot
be related to days of sickness absence.

154
 5. Results

Taken together the results reveal two things. First, the differentiation between incumbent
and applicant makes sense not only in theoretical terms, but reveals more detailed structures
in the empirical analyses as well. This counts for both visible and non-visible measures of
health. Second, the results are very similar to the analyses using a fixed-effects-logit approach.
This speaks in favor of the stability and robustness of the models chosen. For women in the
private sector the point estimates are almost the same in the survival analysis models as in the
fixed-effects-logit models.

Key results from this section are: Subjective health has a positive impact on high
status attainment for women in the private sector. The effects for women in the
public sector are almost zero as are the effects for men in the private sector. For
men in the public sector the results suggest strong selection effects, but sampling
uncertainty is high.
Sickness absence matters for male applicants in the private sector. All other respective
effects are negligible.

Figure 5.16.: Health Effect on Status Change Depending on the Position of Applicant or
Incumbent

Women, Private Sector Women, Public Sector

Applicants

Incumbents
-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5
Men, Private Sector Men, Public Sector
Applicants

Incumbents
-.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5 -.5 -.4 -.3 -.2 -.1 0 .1 .2 .3 .4 .5

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in tables A.10 and A.11 in the
appendix.

155
 5. Results

Figure 5.17.: Effect of Sickness Absence on Status Change Depending on the Position of
Applicant or Incumbent

Women, Private Sector Women, Public Sector

Applicants

Incumbents
-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2
Men, Private Sector Men, Public Sector
Applicants

Incumbents
-.2 -.15 -.1 -.05 0 .05 .1 .15 .2 -.2 -.15 -.1 -.05 0 .05 .1 .15 .2

95 % Confidence-interval Point Estimate

Note: The complete results of the regression can be found in tables A.12 and A.13 in the
appendix.

5.7. Testing Health Selection vs. Social Causation

Up until this point, I have only looked at unidirectional approaches to modeling health selection
effects. Now, I will take a further step and evaluate health selection versus social causation in
one model. I will do this once in a cross-sectional perspective and once from a longitudinal
fixed-effects approach. This chapter therefore reexamines hypotheses H1, H2a, and H5 with a
new method, and allows the first test of hypothesis H9d from the social causation perspective,
testing for a direct social status effect on health. In the next section, hypotheses H9a - H9c
can also be tested using a decomposition approach that is based on the same models as in this
section.

Figure 5.18 shows the effects of the cross-lagged fixed-effects regression, which is the next
step from a unidirectional fixed-effects approach. The model simultaneously estimates the
effects of health at time point t on job status at t+1 and job status at t on health at t+1. It
controls for the same variables as the models presented before and for time constant unobserved
effects. The error terms of the dependent variables are allowed to be correlated.

156
 5. Results

We can see that the health effects are barely affected by allowing for reversed causality. The
pattern of the results is the same as it has been with the unidirectional fixed-effects approach.
Health plays a role for women in the private sector, but not in the public sector. For both men
in the private and in the public sector health has no substantial impact on the probability of a
change in job status. So the interpretation that health effects are only found for women in
context of competition holds true for this model specification as well.

Figure 5.18.: Health Effects - Allowing for Reversed Causality in a Cross-Lagged Model
with Fixed-Effects

Women, Private Sector Women, Public Sector

-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3
Men, Private Sector Men, Public Sector

-3 -2 -1 0 1 2 3 -3 -2 -1 0 1 2 3

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in tables A.14 and A.15 in the appendix.

But what about social causation effects? Finding or not finding health selection effects in
the model does not tell us whether social causation effects also exist. These results of the
simultaneous equation, which tests the effects of job status on health, are presented in figure
5.19. While there are health selective effects for women in the private sector no social causation
effects can be found for this constellation. Direct social status effects cannot be found for men
either. However, for women in the public sector high status jobs seem to increase health. This
is the only social causation effect which can be found.
Note, however that it is only the direct effect that was estimated. What I have not considered
so far is that social causation might work indirectly via observed job characteristics, which are
controlled for in the model. To see whether there is such an indirect social causation effect I
divide the control variables into three groups. The first group are background characteristics
which are not directly influenced by a change in job status. This includes all variables concerning
household characteristics, characteristics of the occupation and employer, years of education

157
 5. Results

Figure 5.19.: Effects of High Status Jobs on Health - Allowing for Reversed Causality in a
Cross-Lagged Model with Fixed-Effects

Women, Private Sector Women, Public Sector

CLFE w/o C

CLFE w BG

CLFE w C

CLFE w A
-.1 -.05 0 .05 .1 -.1 -.05 0 .05 .1
Men, Private Sector Men, Public Sector
CLFE w/o C

CLFE w BG

CLFE w C

CLFE w A
-.1 -.05 0 .05 .1 -.1 -.05 0 .05 .1

95% Confidence-Interval Point Estimate

Note: w/o C = without controls; w BG = with background characteristics (group 1); w C = with BG and job controls (group 2); w A = with
anticipation effect (group 3). The complete results of the regression can be found in tables A.14 and A.15 in the appendix.

and years of full-time employment. The second group consists of variables measuring job
characteristics which might change with job status. These are hours of work, overtime, and
log. wage per hour. Third, the anticipation variables as social-psychological mediators are
grouped together. Table 5.6 gives an overview.

158
 Table 5.6.: Variable Groups for Assessing Mediated Effects

Variable Mechanism Group

Anticipation of Job Loss
Worries About Job Security social causation - social-psychological 3
... own Economic Situation social causation - social-psychological 3
Satisfaction with Work social causation - social-psychological 3
Human Capital
Education third factors - background characteristics 1
Job Tenure social causation - environmental-materialistic 2
Labor Market Experience third factors - background characteristics 1
Wage social causation - environmental-materialistic 2
Labor Market Effort and Effort Intensity
Industry of Employer third factors - background characteristics 1
Firm Size third factors - background characteristics 1
Work Hours social causation - environmental-materialistic 2

159
Overtime social causation - environmental-materialistic 2
5. Results

Effort Intensity (Physical) third factors - background characteristics 1

Effort Intensity (Psychological) third factors - background characteristics 1
Non-Labor Market Effort
Number of HH Members third factors - background characteristics 1
Number of Children third factors - background characteristics 1
Marital Status third factors - background characteristics 1
Housework and Childcare third factors - background characteristics 1
Youngest HH Member third factors - background characteristics 1
Demography and Period Effects
Period Effects basic control variable -
Age basic control variable -
Age2 basic control variable -
East Germany basic control variable -
 5. Results

This division has the following advantages. If background characteristics mediate the effect,
this lends credibility to the explanation that health inequalities between job status are at least
partly explained by spurious correlation or third factors. If group two or three of the control
variables mediate the effect, this points to a social causation explanation based on either
environmental-material or social-psychological grounds. The remaining effects from above
present social causation effects not explained by these factors.
Surprisingly, taking control variables into the model does not influence the estimation of the
impact of job status on health. Regardless of whether we look at women or men, public or
private sector, the effects remain constant. This indicates that confounders can probably only
be found in a cross-sectional analysis, because any existing health inequalities which might
be explained by third factors are explained by time constant factors. For this reason I ran
the whole models one more time without using a fixed-effects approach. Figure 5.20 shows
the results of these models. Indeed background characteristics which are rather stable over
time can explain most of the effect that high job status has on health. For all contexts job
status effects become insignificant and are reduced in magnitude when taking these background
factors into account. Both groups of social causation variables further reduce the effect of job
status but not as much as the background factors do.

160
 5. Results

Figure 5.20.: Effects of High Status Jobs on Health in Cross-Section - Allowing for Reversed
Causality in a Cross-Lagged Model

Women, Private Sector Women, Public Sector

CL w/o C

CL w BG

CL w C

CL w A
-.04 -.02 0 .02 .04 .06 .08 -.04 -.02 0 .02 .04 .06 .08
Men, Private Sector Men, Public Sector
CL w/o C

CL w BG

CL w C

CL w A
-.04 -.02 0 .02 .04 .06 .08 -.04 -.02 0 .02 .04 .06 .08

95% Confidence-Interval Point Estimate

Note: w/o C = without controls; w BG = with background characteristics (group 1); w C = with BG and job controls (group 2); w A = with
anticipation effect (group 3). The complete results of the regression can be found in tables A.14 and A.15 in the appendix.

The last estimate, which should be interpreted, is the correlation of the error terms of
the dependent variables. A high correlation would indicate that there is at least one factor
determining health and high job status, which has not been modeled correctly. This could be an
omitted variable or an unspecified path in the model. Whereas there is a significant correlation
of about 0.4 in the basic cross-sectional approach, the correlation in the final fixed-effects-
models is very low (0.04) and not significant. This supports my view that the cross-lagged
panel fixed-effects model adequately captures the causal mechanisms that are at work be-
tween health and job status. It is another indicator for the robustness of the results in my thesis.

For days of sickness absence the pattern is very similar to subjective health. Allowing for
reversed causality does not have a strong influence on the estimated effects. Figure 5.21
presents the results from the cross-lagged fixed-effects model. Again there is only a significant
effect for men in the private sector.
The reversed causal direction is presented in figure 5.22. Again, we see that there is no direct
effect of job status when the set of control variables in introduced. If the controls are excluded
stepwise we get the results in figure 5.21. The figure shows that the control variables do not
have a substantial influence on the estimation of the coefficient of job status.
Only when we take a cross-sectional perspective we can that there is a effect of job status if
the major control variables are not in the model (see figure 5.23). This association is however

161
 5. Results

not present in the longitudinal analysis.

Key results from this section are: The cross-lagged approach yields the same
results as the unidirectional analysis of health selection. In the private sector, subjective
health is important for women, and sickness absence is important for men. A direct
effect of job status on health can be found for women in the public sector, but not
for any other constellation.

Figure 5.21.: log. Days of Sickness Absence Effects - Allowing for Reversed Causality in a
Cross-Lagged Model with Fixed-Effects

Women, Private Sector Women, Public Sector

-.6 -.4 -.2 0 .2 .4 .6 -.6 -.4 -.2 0 .2 .4 .6

Men, Private Sector Men, Public Sector

-.6 -.4 -.2 0 .2 .4 .6 -.6 -.4 -.2 0 .2 .4 .6

95% Confidence-Interval Point Estimate

Note: The complete results of the regression can be found in tables A.16 and A.17 in the appendix.

162
 5. Results

Figure 5.22.: Effects of High Status Jobs on log. Days of Sickness Absence - Allowing for
Reversed Causality in a Cross-Lagged Model with Fixed-Effects
Women, Private Sector Women, Public Sector
CLFE w/o C

CLFE w BG

CLFE w C

CLFE w A
-.15 -.1 -.05 0 .05 .1 .15 -.15 -.1 -.05 0 .05 .1 .15
Men, Private Sector Men, Public Sector
CLFE w/o C

CLFE w BG

CLFE w C

CLFE w A
-.15 -.1 -.05 0 .05 .1 .15 -.15 -.1 -.05 0 .05 .1 .15

95% Confidence-Interval Point Estimate

Note: w/o C = without controls; w BG = with background characteristics (group 1); w C = with BG and job controls (group 2); w A = with
anticipation effect (group 3). The complete results of the regression can be found in tables A.16 and A.17 in the appendix.

Figure 5.23.: Effects of High Status Jobs on log. Days of Sickness Absence in Cross-Section
- Allowing for Reversed Causality in a Cross-Lagged Model
Women, Private Sector Women, Public Sector
CL w/o C

CL w BG

CL w C

CL w A
-.25 -.2 -.15 -.1 -.05 0 .05 -.25 -.2 -.15 -.1 -.05 0 .05
Men, Private Sector Men, Public Sector
CL w/o C

CL w BG

CL w C

CL w A
-.25 -.2 -.15 -.1 -.05 0 .05 -.25 -.2 -.15 -.1 -.05 0 .05

95% Confidence-Interval Point Estimate

Note: w/o C = without controls; w BG = with background characteristics (group 1); w C = with BG and job controls (group 2); w A = with
anticipation effect (group 3). The complete results of the regression can be found in tables A.16 and A.17 in the appendix.

163
 5. Results

5.8. Comparing the Health Effect to Other Predictors

In this part, I take a short look at the relative strength of the health effect in comparison to
the influence of other predictors of job status.
As a reference for comparison I will use the influence of hours of work, overtime, and the
amount of housework and childcare. These are useful variables for comparison, because they
are factors that can be influenced by the workers, perhaps to compensate health problems.
It allows fictional statements such as: How many hours do you have to work overtime to
compensate my poor health? As a second point of reference I will use the baseline probability
of being in a high status job. This allows us to judge whether the overall chance of job status
change is affected in a sizable manner by individuals health. The comparison is based on the
cross-lagged fixed-effects models from section 5.7.

Table 5.7 shows the x-standardized effects of health, housework and childcare, work hours,
overtime, and age. These effects can be interpreted as the increase in the probability (in
percentage points) of attaining a high status job if the independent variable changes by one
standard deviation. Standardizing the results makes them comparable with regard to the
magnitude of their effect despite the differences in scales. We can see that one standard
deviation in health changes the probability of job status change by 0.24 percentage points for
women in the private sector. I will only report relative strength for this subgroup, because all
other effects are neither substantial nor significant. For women in the private sector this effect
size is equal to the effect of housework and about half the size of work hours and overtime. The
effect of age is by far the strongest - almost 10 times the size of the health effect - pointing to
very gerontocratic structures in promotion schemes even in the private sector. A comparison
to the public sector shows that the effects of age in the public sector are more than double in
size. For men they are doubled again.
The baseline probability of being in a high status job is 11% for women in the private sector.
This means that, relative to the baseline probability, one standard deviation change in health
increases the probability of job status change by about 2.5%. These are not huge effects.
However, they are still sizable and not of lesser importance than other important predictors of
labor market rewards like the hours of work.

Table 5.8 presents the same comparison for the effect of sickness absence. It makes most
sense to interpret the relative size of the coefficient for men in the private sector, because
it is the only substantial effect. We can see that the magnitude of the effect is the same as
the effect of subjective health for women. It is approximately 40 % smaller than the effects
of housework and childcare, overtime, and work hours. Compared to the baseline probability
the effect is considerably smaller than the respective effect for women’s subjective health, as
the overall baseline probability is considerably higher for men than for women. So while for

164
 5. Results

women in the private sector subjective health increases the probability of change in job status
by 2.5% per standard deviation, for men in the private sector a standard deviation of log.
days of sickness absence increases the probability of change in job status by a little more than
1%. Again age is a far more dominant factor in determining the probability of job status change.

Key results from this section are: For women in the private sector subjective
health is a little less important than work hours, overtime, and about as important as
childcare and housework for job status. The increase in probability is about 2.5%.
For men in the private sector sickness absence is half as important as work hours,
overtime, and about as important as childcare and housework for job status. The
increase in probability is only about 1%.

165
 Table 5.7.: Relative Size of the Health Effect Compared to Other Predictors of Job Status - in Percentage Points
Women - Private Women - Public Men - Private Men - Public
Subjective Health 0.246 0.047 -0.031 0.288
Housework/Childcare -0.291 -0.061 -0.202 0.185
Hours of Overtime 0.439 0.518 0.190 -0.048
Hours of Work 0.433 0.831 0.319 0.141
Age 2.259 5.437 6.101 10.083
Baseline Value of High Status Job 11.11 18.36 29.21 33.75
Note: All effects are standardized.

166
5. Results

Table 5.8.: Relative Size of the Effect of Sickness Absence Compared to Other Predictors of Job Status - in Percentage Points
Women - Private Women - Public Men - Private Men - Public
log. Days of Sickness Absence -0.013 -0.033 -0.299 0.121
Housework/Childcare -0.247 0.055 -0.177 -0.130
Hours of Overtime 0.635 0.703 0.434 -0.009
Hours of Work 0.597 0.691 0.409 0.218
Age 2.104 3.719 6.305 10.632
Baseline Value of High Status Job 9.87 16.47 25.32 30.12
Note: All effects are standardized.
 5. Results

5.9. Decomposing Overall Health Inequalities

In this last part of the analyses, I try to attribute the overall health inequalities to different
explanatory approaches. These are:

1. Health selection

2. Social causation
a environmental-materialist
b social-psychological
c direct social causation

3. Third factors: Background characteristics

This presents a new perspective and also allows to test hypotheses H9a - H9c. I use the
cross-lagged fixed-effects models in a slightly modified version (see section 4.8). This implies
treating job status as if it were linear. Such a linear probability model has the advantage of
reporting marginal effects as coefficients. The gross health inequalities which are decomposed
in this section are measured by the cross-sectional association of health at time t with job status
at time point t+1. This cross-sectional estimate of health inequalities does not presuppose any
direction of causality, although technically health is used as the predicting variable.
In table 5.9 we can see to what degree the overall health inequalities, which can be found in
a cross-sectional perspective, can be explained by taking specific variable sets as mediators.
For women in the private sector a quarter of the inequalities can be explained by observed
background characteristics, whereas demography plays almost no role (ca. 3%, group 1).
Controlling for materialist-environmental social causation variables (group 2) does not explain
much of the original health inequalities (3%). Taking social-psychological anticipation variables
in account further explains about 6%, an estimate which is significantly positive according to
the computed confidence interval (1.16; 11.29), but substantively small. State dependence,
meaning the job status one year before (at time t) has a huge explanation power of about
55%. This suggests that time constant factors are extremely important as is also shown by the
explanatory contribution of time constant factors, which is about 73% or almost three quarters,
a highly significant effect. All time varying factors together only add up to 5% of the explana-
tory power, a non-significant estimate. Evaluated on their own only the social-psychological
dimension has a time varying influence, similar to the overall explanation power of 6%. This
means that except for this group of variables things that change over the period of observation
actually do nothing to mediate health inequalities. The remaining effect, after all mediators
are taken into account, is the health selective part of health inequalities which amounts to
about 21%, a significant estimate. Figure 5.244 gives a visual impression of the relative contri-
4
Women in the private sector are the only group in which such a graphical representation is possible.
All other estimates include negative values that are hard to represent in a graph.

167
 5. Results

bution of time varying, time constant factors and health selection for women in the private sector.

Figure 5.24.: Women in the Private Sector: Proportion of Health Inequalities due to
Factors which are...

21.24%

5.52%

73.24%

Time Constant Time Varying

Health Selection

For women in the public sector the picture looks different. Background factors and
environmental-materialist factors are significant mediators of about 31% and 6% respectively.
State dependency is even a little higher than for women in the private sector (61%).
Using a longitudinal approach we can see that almost all health inequalities (99%) can be
traced back to time constant factors, time varying factors or selection do not play a role. This
is an extreme case where any change during the period of observation does not influence health
inequalities any more.

For men in the private sector demography is actually hiding part of the overall health
inequalities (17%). Background characteristics play a major role in mediating health inequal-
ities with 41%. Environmental-materialist and social psychological factors both contribute
significantly with about 8%. State dependency amounts to 62% of the health inequalities.
Longitudinally we see that time constant factors make up more than 90% of the health in-
equalities. Time varying factors in their sum contribute nothing, although individually change
in state (7%) and social-psychological factors (10%) play a significant mediator role. Health

168
 5. Results

selection is not significant.

Cross-sectionally the results are very similar for men in the public sector to men in the
private sector. Environmental-materialist and social psychological factors have a similar size,
but are not significant.
In the longitudinal perspective it is revealed that changes in demographics actually hide health
inequalities. This is a sign for gerontocratic promotion structures which run counter to the fact
that health deteriorates with age. Time constant factors make up 70% of health inequalities,
time varying factors (except for demography) are not significant. Health selection is also not
significant.
We can derive two major conclusions from this analysis. First, health selection is important
only for women in the private sector. In this context, it explains about 20% of overall health
inequalities. Second, explanations of health inequalities between regular and high status jobs
vary widely between men and women and public and private sector. Depending on the context
a different combination of background factors, social causation factors, and health selection
determines the degree of health inequalities. Health selection factors are embedded in their
social context. Their contribution to health inequalities cannot be judged to be high or low in
general. Researchers have to make an effort to specify the contribution (or lack thereof) for
the special case and the special context. Claiming insignificance of the contribution of health
selection without a detailed analysis might be very misleading as this decomposition shows.

169
 5. Results

Table 5.9.: Decomposition of Health Inequalities - Subjective Health

(1) (2) (3) (4)
Women - Private Women - Public Men - Private Men - Public
CS DEMO -3.75 -4.80 -17.35 -18.10
[-9.32;1.82] [-14.64;5.03] [-27.02;-7.68] [-33.26;-2.94]
CS BG 21.93 33.80 41.56 44.63
[13.85;30.01] [16.95;50.66] [30.05;53.07] [26.69;62.56]
CS E-M 2.91 6.34 8.68 2.87
[-0.91;6.73] [0.40;12.28] [4.80;12.57] [-2.11;7.84]
CS S-P 6.22 7.58 8.43 6.70
[1.16;11.29] [-1.05;16.21] [4.20;12.65] [-1.10;14.50]
CS STATE 55.22 61.99 62.95 59.83
[32.09;78.35] [20.52;103.47] [42.65;83.25] [23.46;96.20]
TV DEMO 16.88 -27.46 -48.92 -157.98
[-6.77;40.53] [-83.89;28.97] [-122.81;24.97] [-305.29;-10.67]
TV BG -11.26 30.72 45.32 158.49
[-38.90;16.38] [-15.19;76.64] [-40.59;131.22] [-4.43;321.42]
TV E-M -8.82 1.12 -3.93 5.56
[-16.10;-1.55] [-26.37;28.61] [-10.86;2.99] [-11.19;22.31]
TV S-P 6.20 -8.34 9.50 7.94
[0.03;12.36] [-19.20;2.53] [4.26;14.74] [-2.99;18.87]
TV STATE 2.52 0.54 7.33 2.16
[-1.05;6.10] [-4.73;5.81] [3.68;10.98] [-3.78;8.10]
Total TC 73.24 99.34 92.50 70.93
[31.35;115.13] [23.46;175.23] [59.72;125.28] [10.42;131.44]
Total TV 5.51 -3.41 9.29 16.18
[-19.89;30.91] [-47.91;41.09] [-11.78;30.35] [-28.80;61.16]
Health Selection 21.24 4.07 -1.78 12.89
[0.72;41.76] [-31.67;39.80] [-16.49;12.92] [-16.55;42.33]
Note: CS: cross-sectional; TV = time varying; TC = time constant; DEMO = demography;
BG = background characteristics; E-M = environmental materialist factors; S-P = social
psychological factors; STATE = state dependency

170
 5. Results

Table 5.10 contains the same results for the decomposition of health inequalities measured by
log. sickness absence. Decomposing the inequalities in sickness absence yields the following re-
sults. For women in the private sector the cross-sectional perspective shows that background
characteristics and social-psychological factors reduce inequalities tremendously (about 100%
and 50% respectively) while environmental-materialist factors actually invert health inequalities.
State dependency and demographic factors do not play a significant role. The longitudinal
analyses show that both time constant and time varying factors are strongly associated with
explaining health inequalities, but with opposing signs. Both estimates show huge ranges of
the confidence-interval. Individually environmental-materialist and social-psychological factors
have the same direction of effect as in the cross-sectional analysis. Health selection does not
play a significant role.

Health inequalities in sickness absence of women in the private sector can be decomposed
in small to associations with background characteristics, and in large part to state dependency.
This is reflected in the longitudinal analysis as well. Time constant factors significantly reduce
health inequalities, time varying factors and health selection do not play a role. Time varying
social-psychological factors actually lead to a small increase in health inequalities.

The cross-sectional perspective on men in the private sector reveals, that a fifth of
health inequalities can be attributed to background factors, a small portion (ca. 5%) to
environmental-materialist factors, and a big portion to state dependency (ca. 60%). The
longitudinal perspective shows some unusual results. First, on its own, change in background
characteristics partly explains health inequalities. Then a change in state dependency makes
also a significant, although small contribution (5%). That would mean that a change in job
status at time point t partly mediates the fact that health at t influences job status at t+1.
Taken together time constant factors account for almost 34 of health inequalities. However, both
health selection (7%) and other time varying factors (20%) also make a significant contribution.
Health selection is, relatively to other factors, less important in explaining health inequalities in
sickness absence than in general health.

For men in the public sector we can see that mostly state dependency (60%), background
characteristics (25%) and to a small degree environmental-materialist factors (4%) are as-
sociated with health inequalities. Interesting is that in the longitudinal perspective, no one
time varying factor is significantly associated with health inequalities, together they explain
an estimated third of health inequalities, with two thirds explained by time constant factors.
Health selection has no independent contribution.

171
 5. Results

Key results from this section are: Health selection explains health inequalities
only in the private sector. For subjective health it explains about a fifth of overall
health inequalities between female high and regular status job incumbents. For men
and for sickness absence the contribution is only about 7%. Time constant factors
are by far the most important explanatory force for health inequalities between job
status. This indicates that during labor market trajectories job status only has limited
independent impact on health inequalities, and health inequalities due to job status
do not change much over time.

Table 5.10.: Decomposition of Health Inequalities - log. Days of Sickness Absence

(1) (2) (3) (4)
Women - Private Women - Public Men - Private Men - Public
CS DEMO 8.20 0.00 0.35 -0.86
[-3.55;19.94] [-2.43;2.42] [-0.22;0.92] [-3.13;1.41]
CS BG 107.47 18.58 20.64 25.90
[26.08;188.86] [5.28;31.89] [17.85;23.44] [18.64;33.15]
CS E-M -225.20 -3.01 6.26 3.56
[-282.01;-168.39] [-8.92;2.89] [4.77;7.76] [1.39;5.74]
CS S-P 45.48 3.34 0.79 1.41
[19.08;71.89] [-0.20;6.88] [0.17;1.42] [-0.22;3.05]
CS STATE 35.01 62.63 61.71 60.96
[-204.46;274.47] [26.57;98.69] [53.87;69.56] [42.19;79.73]
TV DEMO 25.78 -3.38 0.25 -7.68
[-22.61;74.17] [-16.82;10.07] [-1.85;2.34] [-29.31;13.96]
TV BG -106.12 12.09 11.41 38.56
[-334.85;122.61] [-20.82;45.00] [2.65;20.18] [-0.41;77.53]
TV E-M -204.59 -5.01 1.03 0.85
[-309.33;-99.86] [-22.81;12.79] [-0.50;2.55] [-3.43;5.14]
TV S-P 50.93 -4.90 1.13 1.26
[16.30;85.56] [-9.39;-0.42] [0.31;1.95] [-0.92;3.43]
TV STATE 0.68 -0.33 5.69 1.60
[-4.46;5.81] [-5.92;5.27] [2.67;8.70] [-4.43;7.62]
Total TC 247.92 104.94 73.66 63.12
[-171.59;667.42] [42.32;167.56] [59.36;87.95] [28.77;97.46]
Total TV -233.33 -1.53 19.50 34.60
[-478.44;11.79] [-34.77;31.72] [9.16;29.84] [5.64;63.55]
Health Selection 85.43 -3.42 6.84 2.29
[-124.82;295.68] [-34.32;27.48] [1.32;12.37] [-8.82;13.40]
Note: CS: cross-sectional; TV = time varying; TC = time constant; DEMO = demography;
BG = background characteristics; E-M = environmental materialist factors; S-P = social
psychological factors; STATE = state dependency

172
 5. Results

5.10. Summary of Results

Before coming to the major conclusion a summary of the results is useful. I will structure the
summary around central questions and hypotheses. The summary will cut across the different
subsections of the results chapter to synthesize the findings from different model specifications.
The summary is supposed to conclude how health selection works in a non-technical manner
by discussing and answering 4 major questions of the thesis. At the end of this section table
5.11 presents a schematic overview of the hypotheses and shows whether they can be accepted
or not based on the analyses.

5.10.1. Does Health Selection Contribute to Job Status Related

Health Inequalities?
In the descriptive part of the analyses we could see that there are health inequalities between
workers who have regular jobs and workers who have high status jobs. These inequalities
are stronger in the private than in the public sector and stronger among women than among
men. The main question was whether health selection contributed to these health inequalities.
The estimates of the fixed-effects-logit models and of the cross-lagged fixed-effects models
suggested that we can only identify a contribution of health selection to health inequalities for
women in the private sector. This mainly confirmed the hypotheses proposed in the theoretical
part of the thesis that health effects are stronger for women. The fact that there was no health
effect for men in the private sector supports the hypothesis that in contrast to women, men do
not adjust their effort at work if faced with impaired health. As expected, in the public sector
health did not influence job status. This is in line with the argument that a more bureaucratized
promotion scheme leaves little room for health related performance effects on job status.
What could also clearly be seen by the results of the survival analysis is that health selection
works at the entrance to high status jobs. For women in the private sector, the group where a
health selection effect could be established, only the health of applicants to high status jobs
works selective. There was no downward health related mobility, only hindrance of upward
mobility. This makes sense if we view high status jobs as closed positions where firing the
employee is not possible without substantial costs to the employer. Occupying high status jobs
makes workers invulnerable to health related reduction of performance.

The answer is: Health selection contributes to health inequalities only for women in
the private sector (20%). For sickness absence the contribution is strongest in male
dominated jobs. Note that social causation explains even less of health inequalities.
The major parts are explained by common background factors.

173
 5. Results

5.10.2. Does Social Context Matter?

The preceding paragraph already stated that health selection does not occur in the public
sector. This is one important way of showing that social context in the form of rules, contracts
or traditions influences the way health inequalities are generated. Far from being a natural law
it lies well within the power of society to shape employment relations in a way that health is
not an important selective factor.
Is the state as an employer necessary to insure that health selection does not contribute to
health inequalities? Or is there evidence that in certain parts of the private sector there are
also no selection effects? For this purpose different forms of occupational closure in the private
sector were investigated. Jobs which have a low knowledge intensity are not well guarded
against competition from outside. On the other hand, the more professionalized an occupation
is the less important health selective factors are. So health selection can be stronger or less
strong depending on the type of occupation one works in.
This could also be seen with regard to gender proportions in jobs. The “glass-escalator”
hypothesis stated that in jobs with a high percentage of women, men should have an easier
time getting a promotion than women. Therefore health should be primarily important for
women in female dominated jobs. And indeed health is a selective factor for women, but not
for men, in female dominated occupations.
In male dominated jobs women seem to have a token status with regard to sickness absence. A
culture of presenteeism leads to moderate selective effects for men and strong selective effects
for women. The effects of women are subject to a high uncertainty due to small numbers and
should be treated with caution.
These results highlight the importance of social context for health selection processes. In light
of these results it seems unreasonable to reduce the process of health selection to something
biological or natural.
In table 5.11 we can see if different overlaying contexts, positions and social roles produced
health selection or not. If the cells are colored white, it indicates that the theory would expect
a closed position and thus no health selection, gray indicates the opposite. The large X show
where health selection effects could be found. Minus stands for no substantial health selection
effects. The table provides visual evidence that certain circumstances disallow health selection.
Or put differently, it shows that we need to search for an intersection of open positions as
context for health selection to find competition and health selection. The table also shows that
for both men and women there is one unexpected lack of health selection each. For women
we have no health selection on the visible health measure5 , for men no health selection on
the non-visible measure. This shows that gender plays an important role, not always on the
individual level, but also on the occupational level.

5
Except in male dominated occupations.

174
 5. Results

Table 5.11.: Health Selection in Open and Closed Positions

The Situation is...

Sector Public Sector Private Sector
Occupation Open Occupations Closed Occupations
Health is ... visible non-visible visible non-visible visible non-visible
Worker is... Male Incumbent - - - - - -
Applicant - - X - - -
Female Incumbent - - - - - -
Applicant - - (X) X - -
Note: X= effect found; - = no effect found; Gray Cells = HS is expected; White Cells = HS is not expected

The answer is: Health selection can only be found in open positions where compe-
tition is possible. Gender is a moderator on the individual and on the occupational
level. Context is highly important to health selection processes.

5.10.3. What is More Important: Health Selection, Social Causation

or Third Factors?
So what relative contribution does health selection make to the explanation of health inequal-
ities? Concluding from the two preceding paragraphs we can say that in non-competitive
environments and for men the contribution is probably negligible, because we cannot even
establish substantial effects in the models.
But what contribution does health selection make for women in the private sector? The effects
were statistically significant, but are they also important? A decomposition of the overall health
inequalities suggests that about one fifth of the health inequalities can be attributed to health
selection. This does not make health selection the strongest factor, but hardly so little that
it can be safely ignored. Third factors make up about half of the health inequalities. Social
causation in the form of social-psychological factors can explain about a quarter of the effect.
The comparison of cross-sectional and fixed-effects models indicates that it is mostly time
constant characteristics of incumbents of high status jobs, which gives them an advantage
over incumbents of regular jobs. However, for women in the private sector the only major time
varying factor is health selection. This is interesting, because it suggests that at a certain point
in time health inequalities do not change anymore, except through selection processes.
For men and in the public sector time constant factors, and to a smaller degree social-
psychological factors are the driving force of health inequalities. Here health selection is actually
a factor which can be ignored.
For sickness absence health selection plays a certain role in male dominated occupations, but
overall time constant factors prevail here as well.

175
 5. Results

The answer is: Common background factors are by far the most important determi-
nant of job status related health inequalities. Health selection plays a substantial role
in certain contexts. Social-psychological factors play a small role for men. Even taken
together health selection and social causation are far less important than background
factors.

5.10.4. Are There Gender Differences in Health Effects?

What do the results tell us about gender differences in health selection? The general hypothesis
was that health should have a stronger impact on women’s job status than on men’s. Overall
this was supported. The theory offered two explanations for this gender difference. One was
gender specific disadvantage in performance evaluation which necessitated a stronger reliance
on health on the part of women. The other explanation stated that men simply do not adjust
their effort at work to a change in health status.
I think that the results favor the second explanation. There was absolutely no direct effect of
subjective health on job status for men, whereas the first explanation would merely expect a
reduced effect in comparison to women. On the other hand, as soon as men were absent from
work their job status becomes subject to change, decreasing the chance of upward mobility.
This indicates that a deviance from the presenteeism culture of male work environments is also
penalized for men.
The fact that in various contexts of occupational closure it was only women’s health effect
which is moderated also speaks in favor of a general invariance of effort with regard to subjective
health among men.
However, we should be careful at this point not to overstate these interpretations. A direct
test of explanation 1 versus explanation 2 was neither conducted nor is it possible with the
data used in the thesis. It should also be mentioned that the two explanations are not absolute
statements and are not mutually exclusive.

The answer is: Gender matters for health selection. Men and women are not affected
in the same way by their health. For sickness absence gender matters especially as an
interaction of the gender proportion on the occupational level and individual gender.
This suggests that simple separated regression analyses will not always account for all
gender differences.

176
6. Conclusion

6.1. Overview
I started my thesis by noting that health is important to individuals across societies, and is
consequently the subject of government policies. Yet, I wondered whether it is more than a
goal in itself, a resource which allows us to be successful in society. This would also imply that
health inequalities could be the result of advantages or disadvantages individuals face due to
their health state. I elaborated on the idea that health selection might be a driving force in
labor market processes and the generation of health inequalities. Now it is time to reflect on
the theoretical arguments and the empirical analyses to see how my theory fares and what kind
of contribution it can make to research on health inequalities.
I start by summarizing the theoretical and methodological approach of the thesis highlighting
the new features developed in this context. In the next section a very brief summary of the key
findings1 and of some unanswered questions follows.
The third part will mention the most important caveats and limitations of the study known to
me. This is done both for reasons of scientific rigor and as a connection to the last part of the
conclusion that suggests different fields of further research, which derive from both unresolved
questions and limitations of this thesis.
I will then go on and conclude what my findings mean for the theory and how the theory could
be generalized beyond the context of the labor market. Finally, I make some suggestions for
future research.

The theoretical questions of the thesis were: Through what mechanisms does health selection
influence health inequalities in society? And, are these mechanism natural or social processes?
The line of argument criticized widespread claims that health selection is a pseudo-biological or
natural way of explaining health inequalities and that it could be empirically disregarded. My
critique was based on the claim that there is actually no coherent theory of health selection.
Consequently the mechanisms lying behind the hypothesis remain opaque in most studies. It
became clear that health selection should be interpreted as a social process. In addition, the
circumstances, namely the degree of competitiveness and socialized health behavior, should
have a major influence on health selection.
For the example of job status it was concluded that in open positions, like the private sector,
1
A longer version can be found at the end of the previous chapter.

177
 6. Conclusion

health selection matters while in closed positions, like the public sector, health selection is
negligible. At the same time structural factors like labor market disadvantages of women and
socialized differences in health behavior lead to stronger health selection effects for women than
for men. Another important distinction with regard to closed positions is between incumbents
and applicants to certain positions. Incumbents of closed positions do not face health selection
while applicants to closed positions do.
My statistical approaches for testing the hypotheses included several novelties. First, the
comparability of measures of subjective health across the important groups was tested. The
latent variable approach I chose addressed the debate in the scientific literature about the
comparability of self-rated health across gender, time, and status empirically without having to
rely on purely theoretical arguments. The models of the confirmatory factor analyses showed
that comparability in the SOEP is not a problem for the groups under scrutiny. Second,
a cross-lagged panel model with fixed-effects was introduced complementing the standard
fixed-effects logistic regression and survival analytic models. The cross-lagged panel model takes
time constant unobserved factors and observed time varying factors into account. Both might
lead to spurious correlation in the relationship between health and job status. In addition, it
takes reversed causality into account allowing both health to influence job status and vice versa.
The variety of models chosen reflects varying requirements according to different sub-questions
and can be seen as a robustness check of the results. Third, a decomposition approach of
health inequalities was undertaken. In contrast to previous attempts in the literature, this
decomposition approach explicitly models the part of the health inequalities between regular
and high status jobs associated with time constant factors. To the best of my knowledge it
is also the first study which provides a confidence interval not only for mediating observed
variables, but also for the time constant factors as a whole.

6.2. Key Results and Unresolved Questions

The study presented here is the first to investigate the effect of health on job status for Germany.
Generally, in the literature there are very few studies linking health and job status, favoring
wages, employment, or SES depending on the disciplinary background. My thesis is also one
of the few studies making a systematic comparison of an - for the employer - unobservable
(subjective) health measure, self-rated health, with an observable (objective) measure, days of
sickness absence.

If we want to use a less scientific language, the empirical examples from the German labor
market can be broken down to the following statements.

178
 6. Conclusion

• If you are a woman, health is an important asset if you want to get a good job in the
private sector.

• If you work in an occupation which is dominated by men, do not call in sick too often or
your chances of a good job are dwindling.

• Do not expect a change in job status to matter too much for your health. At the point
in your life when you compete for such jobs, the harm is already done. What matters for
health and inequalities in health happens before.

A more detailed look on the analyses yields the following picture:

Health selection was a substantial factor in explaining health inequalities between persons of
high and normal job status. These substantial effects could only be found in open positions on
the labor market (private sector, low knowledge intensity). There are considerable differences
between genders. With regard to subjective health only women appear to be selected according
to health. In female dominated occupations a “glass-escalator” effect of health can be seen
favoring men over women. For sickness absence, it seemed at first that the reverse was true.
Only men seemed to be selected by the number of days of sickness absence. However, more
detailed analyses showed that it is mainly in male dominated occupations where selection with
regard to sickness absence occurs. The social context in which health selection takes place is
therefore very important. Taking reversed causality into account did not change these results.
It is also important to note that applicants to closed positions seemed to be subject to stronger
selective effects than incumbents of such positions.
The strongest explanation of health inequalities between regular and high status jobs was due
to time constant factors. These are most likely determined before the period of observation,
maybe even before the entry into the labor market. For women health selection with regard to
subjective health made up the greater part of the remaining variation. Actual changes in job
status could therefore only account for one sixth of health inequalities not explained by time
constant factors. This makes health selection much more influential for women in the private
sector. For all other defined groups health selection did not play a role. Social causation in
form of social-psychological factors can explain most of the time varying health inequalities for
men in the private sector.

Some questions remain unanswered:

Is the difference between incumbents and applicants in selective effects due to an actual differ-
ence in effects or due to a lack of statistical power of the estimator? Replication with larger
samples or under circumstances of more events could provide clues to answer this question.
Is the difference in health selection effects between men and women due to different health
behavior, disadvantage or both? The data does not allow to test this directly. The fact that
there is virtually no health selective effect for men regardless of model specification or context

179
 6. Conclusion

speaks in favor of the health behavior explanation. However, the “glass-escalator” effect of
men in female dominated occupations suggests that disadvantage might also play a part. We
are therefore left with an unsatisfactory conclusion: Both mechanisms are probably at work,
but we cannot tell which one is more important.

6.3. Caveats and Limitations

My thesis has some potential problems and limitations. The most important of these shall be
mentioned now.

1. Restriction to the German labor market

2. No life course perspective

3. No integration of social causation and health selection theory

4. Health as an exogenous variable; no natural experiment

The first limitation is that I restricted my analyses to individuals who are part of the German
labor market. Also excluded are self-employed persons and all those who are not yet or not
anymore employed. The results presented here are therefore not fit for generalization to
other labor markets, self-employed, retired persons, or those still in the educational system.
Nonetheless, the theory allows to make predictions for other groups as well and the integration
of the theory of open and closed position is a good starting point to state context sensitive
hypotheses about health selection in other contexts.

The second challenge is that the theory so far only throws a spotlight at a random point in
the life course of an individual. It does not take into account the career and health trajectory
nor the outlook of the individual. A theoretical framework which allows the integration and
the age dependency of such relations would yield a more coherent view on health selection
processes. One might find that health selection is of special importance during certain periods
of life while during others it is of little importance. A life course perspective would also allow
to integrate selective mechanisms before entry into the labor market. The theory of Galama &
van Kippersluis (2010) shows how such a life course perspective could look like. However, it
does not deal with heterogeneous and context sensitive selection mechanisms. This could be a
starting point for integrating the theories.
A life course perspective could also overcome the third major limitation of the thesis. I did not
provide an integration of social causation and health selection hypotheses. Such an integrated
theory would imply that reciprocal causation over the life course is explicitly modeled. The
relative importance of health selection and social causation in explaining health inequalities

180
 6. Conclusion

might also vary over the life course. Galama & van Kippersluis’s (2010) theory integrated with
the theory of health selection could be a good foundation for a reciprocal theory of health
inequalities.
Yet another closely related issue is that I treat health as an exogenous variable throughout the
thesis. As soon as a life course perspective is taken this assumption is not feasible anymore.
Decisions about health behavior and health related labor market decisions will influence future
health. At least to a certain degree this will be reflected and taken into account by individuals.
At this point health becomes endogenous to decisions and events in the individuals’ lives at an
earlier point in the life course. Strategic investments in health should therefore be a part of
future developments of the health selection theory.
Certain methodological schools of thought would argue that the absence of a natural experiment
in this thesis is also a weak point in the argument. For the empirical analyses no external
quasi-random variation was used to ensure that the estimated effects are causal in the sense of
the counter-factual argument. One such approach would have been an instrumental variable
estimation approach. However, finding such an instrument is hard, and the results are sometimes
unclear and limited in time and context with regard to their actual interpretation in terms of
causality (Stowasser, Heiss, McFadden & Winter 2011, 10).
In defense of the approach taken, I want to make two points. First, the methods chosen go a
long way in making the conditional independence assumption plausible. In my opinion, it is
the strongest test possible given the subject of inquiry. Second, a focus solely on topics where
natural experiments occur would lead away from answering theoretically interesting research
questions towards forming research questions around natural experiments. Such an approach
to research designs will not take into account whether a research question is important or not.
Consequently, certain fields in social sciences would become completely understudied. This
holds true for health research and research on subjective health in particular, but to a certain
extend it also applies to labor market research.

6.4. Generalizing the Theory of Health Selection

How can we generalize the theory of health selection to bring it into accordance with the findings
from the study? At this point, it seems useful to construct a list of necessary conditions which
need to be fulfilled so that a health selection mechanism might contribute to the explanation
of health inequalities.
One sufficient condition would be the existence of formal health criteria for entering a social
positions. In such cases health selection is made explicitly. If a person fails a certain criteria he
or she cannot attain the position. These positions will only be occupied by individuals who
show a certain health profile. For example, pilots need to pass physical examinations before
and during their employment which are often regulated and demanded by law. Civil servants in
Germany need to pass a medical check up, where among other things the BMI should not be

181
 6. Conclusion

higher than a certain threshold.

Another sufficient condition for health selective mechanisms are health related discriminatory
practices. This is most likely the case for stigmatized diseases or health limitations. One could
think of HIV, physical disabilities, or certain conditions like epilepsy. Here the question of
visibility arises. Is there a direct or an indirect way for employers or other institutions, which
select individuals, to recognize such a stigmatizing health impairment?
If none of these two sufficient conditions are present, the following two conditions are condi-
tionally necessary:

1. competition through performance

2. performance depends on health

These four conditions (two sufficient, two conditionally necessary) lead to the following
logical set of conditions for health selection. Formal health criteria (F) are a sufficient, but not
a necessary condition for health selection. Health discrimination (D) and visibility (V) are also
jointly sufficient, but not necessary. The third way is a selection according to performance (P),
given that performance is dependent on health (P|H). Performance depends on health if effort
is adjusted according to health status, and the individuals are face a situation of competition
(C). Note that these conditions include self-selection, because if either performance does not
depend on health or there is no competition, than there would be no incentive for self-selection.
In all three cases a selecting actor needs to exist (A).
So we can state the following logical condition for health selection processes. A social context
(SC) allows health selection (HS) if it has one or more of the four elements listed above.

SC ∈ HS → A ∧ (F ∨ (D ∧ V ) ∨ (P |H ∧ C)) (6.1)

From this logical scheme various sub-theories and hypotheses can be derived about the
relationship of different health conditions and various social positions or social outcomes.
The next section elaborates a little more on possible further research which is based on a
generalization of the theory of health selection.

6.5. Further Research

Is there potential for further research after the investigations of this study? I think so and will
give some examples below. There are a lot of areas which can be studied using the approach
of this thesis. They can be categorized as follows:

1. Theoretical developments

2. Non-labor-market applications

182
 6. Conclusion

3. Generalization to other contexts

One important step in the theoretical development would be to integrate the individual level
with the macro level. In particular, an investigation should be started into how overall social
inequalities influence health inequalities generated by health selection. Here, work by authors
like Wilkinson & Pickett (2009) could be picked up. They investigated the relationship of
health inequalities and overall health status of societies and social inequalities in these societies.
However, the argument was dominated by a social causation explanation of health inequalities.
From the theory of health selection we could derive a preliminary hypothesis that socially
unequal societies have stronger selection mechanisms at least with regard to selection on social
characteristics. These societies might in general be more competitive. This could well lead
to stronger selection in the educational system or on the labor market with regard to health
of the individuals. A complementary explanation of the association between social inequality,
level of health, and health inequalities on a societal level might arise from such results. It is
definitely an area which promises great research opportunities and that could use parts of the
theoretical insights from this thesis to develop a theory that allows for micro-macro interaction
to predict societal outcomes of health selection.

A different field of research is the comparison of educational systems with regard to their
selectivity not only on social characteristics, but also on health characteristics. A lot of
research has compared educational systems with regard to their moderating function on the
effect of children’s and adolescents social background on their educational achievement. In
particular, there are several studies dealing with the impact of educational systems on children’s
and adolescent’s health (Zambon, Boyce, Cois, Currie, Lemma, Dalmasso, Borraccino &
Cavallo 2006, Hurrelmann, Rathmann & Richter 2011, Richter, Rathman, Gabhainn, Zambon,
Boyce & Hurrelmann 2012). There are also studies estimating the impact of different health
conditions on educational achievement. But studies focusing on the interaction between
educational system and the degree of health related selectivity are rare. Thus, the theory of
health selection could be fruitful in making predictions which could then be tested empirically
across different school or university systems.

The last area for further research is a search for context sensitivity of health selection. The
present study already made several such tests and showed that e.g. the occupational context,
or gender matter for health selection. However, the study was restricted to an analysis of the
German labor market. Given that the claims of health selection theory are fairly general it is
possible to look at other labor market constellations in other countries or at other points in
time. It could be investigated whether the predictions hold in different contexts or whether
under certain conditions health selection theory fails to provide useful hypotheses. It would
be very interesting to see how the theory fares in countries where the general level of public
health is much lower than in most of the OECD countries. Does this lead to stronger selection

183
 6. Conclusion

processes due to more severe conditions among the working population? Or are working and
living conditions at the same time so much more important for determining health outcomes
that health selection does not contribute anything in the explanation of health inequalities?
In any case, there are countless possibilities to use other data sources from different contexts to
further test and develop the health selection therory. My thesis provided the first stepping stone.

184
Bibliography
Abbott, Andrew. 1988. The System of Professions: An Essay on the Division of Expert Labor.
1 ed. University Of Chicago Press.

Achatz, Juliane, Hermann Gartner & Timea Glück. 2005. “Bonus oder Bias?” KZfSS Kölner
Zeitschrift für Soziologie und Sozialpsychologie 57(3):466–493.

Aittomäki, Akseli, Pekka Martikainen, Mikko Laaksonen, Eero Lahelma & Ossi Rahkonen.
2012. “Household economic resources, labour-market advantage and health problems – A
study on causal relationships using prospective register data.” Social Science & Medicine
75(7):1303–1310.

Akerlof, George A. & Janet L. Yellen. 1986. Efficiency Wage Models of the Labor Market. 3
ed. Cambridge University Press.

Alexander, Elmore R. & Ronnie D. Wilkins. 1982. “Performance Rating Validity: The Rela-
tionship of Objective and Subjective Measures of Performance.” Group & Organization
Management 7(4):485–496.

Anderson, Michael & Michael Marmot. 2012. “The Effects of Promotions on Heart Disease:
Evidence from Whitehall.” The Economic Journal 122(561):555–589.

Angrist, Joshua D & Jörn-Steffen Pischke. 2010. “The Credibility Revolution in Empirical
Economics: How Better Research Design is Taking the Con out of Econometrics.” The
Journal of Economic Perspectives 24(2):3–30.

Aparicio, Ainhoa. 2011. “The Effect of Product Market Competition on Job Instability.” IZA
Discussion Paper 5669.

Arrow, O. J. 1996. “Estimating the influence of health as a risk factor on unemployment::

a survival analysis of employment durations for workers surveyed in the German Socio-
Economic Panel (1984-1990).” Social Science & Medicine 42(12):1651–1659.

Avdic, Daniel & Per Johansson. 2013. Gender differences in preferences for health-related
absences from work. Technical report IFAU-Institute for Evaluation of Labour Market and
Education Policy.

185
 Bibliography

Azfar, Omar & Stephan Danninger. 2000. “Pofit-Sharing, Employment Stability, and Wage
Growth.” Industrial and Labor Relations Review 54:619.

Ball, Kylie, David Crawford & Gita Mishra. 2006. “Socio-economic inequalities in women’s
fruit and vegetable intakes: a multilevel study of individual, social and environmental
mediators.” Public Health Nutrition 9(05):623–630.

Bartel, Ann & Paul Taubman. 1979. “Health and Labor Market Success: The Role of Various
Diseases.” The Review of Economics and Statistics 61(1):1–8.

Bartley, Mel. 1988. “Unemployment and health: selection or causation - a false antithesis?”
Sociology of Health & Illness 10(1):41–67.

Bartley, Mel. 1991. “Health and Labour Force Participation: ‘Stress’, Selection and the
Reproduction Costs of Labour Power.” Journal of Social Policy 20(03):327–364.

Bartley, Mel & Ian Plewis. 1997. “Does Health-Selective Mobility Account for Socioeconomic
Differences in Health? Evidence from England and Wales, 1971 to 1991.” Journal of
Health and Social Behavior 38(4):376–386.

Bartley, Mel, Jane Ferrie & M. Scott Montgomery. 2006. Health and labour market disadvantage:
unemployment, non-employment and job-insecurity. In Social determinants of health, ed.
Michael Marmot & G. Richard Wilkinson. Oxford: Oxford Univ. Press pp. 78–96.

Bauer, Ullrich, Uwe H. Bittlingmayer & Matthias Richter, eds. 2008. Health
Inequalities: Determinanten und Mechanismen gesundheitlicher Ungleichheit (Gesundheit
und Gesellschaft). 2008 ed. VS Verlag für Sozialwissenschaften.

Baum, Charles L & William F Ford. 2004. “The wage effects of obesity: a longitudinal study.”
Health Economics 13(9):885–899.

Becker, Gary S. 1977. “A Theory of the Production and Allocation of Effort.” NBER Working
Paper Series 184.

Becker, Gary S. 1985. “Human Capital, Effort, and the Sexual Division of Labor.” Journal of
Labor Economics 3(1):S33–S58.

Becker, Gary S. 1994. Human Capital: A Theoretical and Empirical Analysis with Special
Reference to Education. 3 ed. University of Chicago Press.

Becker, Gary Stanley. 1971. The economics of discrimination. University of Chicago Press.

Benı́tez-Silva, Hugo & Huan Ni. 2008. “Health status and health dynamics in an empirical
model of expected longevity.” Journal of Health Economics 27(3):564–584.

186
 Bibliography

Berkowitz, Monroe & William G. Johnson. 1974. “Health and Labor Force Participation.” The
Journal of Human Resources 9(1):117.

Bernert, C. 1983. “The career of causal analysis in American sociology.” British Journal of
Sociology p. 230–254.

Bianchi, Suzanne M., Melissa A. Milkie, Liana C. Sayer & John P. Robinson. 2000. “Is Anyone
Doing the Housework? Trends in the Gender Division of Household Labor.” Social Forces
79(1):191–228.

Bielby, William T. & Denise D. Bielby. 2002. Telling Stories About Gender and Effort: Social
Science Narratives About Who Works Hard for the Money. In The New Economic Sociology:
Developments In An Emerging Field, ed. Mauro F. Guillén & Paula England. Russell Sage
Foundation pp. 193–217.

Bisin, Alberto & Thierry Verdier. 1998. “On the cultural transmission of preferences for social
status.” Journal of Public Economics 70(1):75–97.

Bjerk, David. 2008. “Glass Ceilings or Sticky Floors? Statistical Discrimination in a Dynamic
Model of Hiring and Promotion.” The Economic Journal 118(530):961–982.

Black, Sandra E., Paul J. Devereux & Kjell Salvanes. 2005. “From the Cradle to the Labor
Market? The Effect of Birth Weight on Adult Outcomes.” National Bureau of Economic
Research Working Paper Series No. 11796.

Blacksher, Erika, Elizabeth Rigby & Claire Espey. 2010. “Public Values, Health Inequality,
and Alternative Notions of a “Fair” Response.” Journal of Health Politics, Policy and Law
35(6):889–920.

Blane, D., S. Harding & M. Rosato. 1999. “Does social mobility affect the size of the
socioeconomic mortality differential? Evidence from the Office for National Statistics
Longitudinal Study.” Journal of the Royal Statistical Society Series A 162(1):59–70.

Blane, David. 1985. “An assessment of the Black Report’s explanations of health inequalities.”
Sociology of Health & Illness 7(3):423–445.

Blane, David, George Davey Smith & Mel Bartley. 1993. “Social selection: what does it
contribute to social class differences in health?” Sociology of Health & Illness 15(1):1–15.

BMAS. 2013. Lebenslagen in Deutschland - Der vierte Armuts- und Reichtumsbericht der
Bundesregierung. Bundesministerium für Arbeit und Soziales.

Bohnet, Iris, Max H. Bazerman & Alexandra Van Geen. 2012. “When Performance Trumps
Gender Bias: Joint Versus Separate Evaluation.” SSRN eLibrary .

187
 Bibliography

Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. John Wiley & Sons.

Bommier, Antoine & Guy Stecklov. 2002. “Defining health inequality: why Rawls succeeds
where social welfare theory fails.” Journal of health economics 21(3):497–513.

Borg, V & T S Kristensen. 2000. “Social class and self-rated health: can the gradient be
explained by differences in life style or work environment?” Social science & medicine
(1982) 51(7):1019–1030.

Borsboom, Denny. 2006. “When does measurement invariance matter? Commentary.” Medical
Care 44(11):S176–S181.

Bourdieu, Pierre. 2012. Ökonomisches Kapital, kulturelles Kapital, soziales Kapital. In Handbuch
Bildungs-und Erziehungssoziologie. Springer p. 229–242.

Bowen, Chieh-Chen, Janet K. Swim & Rick R. Jacobs. 2000. “Evaluating Gender Biases on
Actual Job Performance of Real People: A Meta-Analysis1.” Journal of Applied Social
Psychology 30(10):2194–2215.

Bowling, A. & J. Windsor. 2008. “The Effects of Question Order and Response-Choice on
Self-Rated Health Status in the English Longitudinal Study of Ageing (ELSA).” Journal
of Epidemiology and Community Health 62(1):81–85.

Buddelmeyer, Hielke & Lixin Cai. 2009. “Interrelated Dynamics of Health and Poverty in
Australia.” IZA Discussion Paper .

Budig, Michelle J. & Paula England. 2001. “The Wage Penalty for Motherhood.” American
Sociological Review 66(2):204–225.

Bundestag, Deutscher. 2011. Enquete-Kommission ‘Wachstum, Wohlstand, Lebensqualität’.

Wege zu nachhaltigem Wirtschaften und gesellschaftlichem Fortschritt in der Sozialen
Marktwirtschaft.

Burgard, Sarah A., Jennie E. Brand & James S. House. 2007. “Toward a Better Estimation of
the Effect of Job Loss on Health.” Journal of Health and Social Behavior 48(4):369 –384.

Burke, R.J. & C.A. McKeen. 1994. “Career development among managerial and professional
women.” Women in management: Current research issues p. 65–79.

Burt, S. Ronald. 1995. Structural holes: The social structure of competition. Cambridge,
Mass.: Harvard Univ. Press.

Busch, Anne & Elke Holst. 2012. “Occupational sex segregation and management-level wages
in Germany: What role does firm size play?” IZA Discussion Paper .

188
 Bibliography

Butler, P., A. Felstead, D. Ashton, A. Fuller, T. Lee, L. Unwin & S. Walters. 2004. “High
performance management: a literature review.” Learning as Work Research Papers (1).

Böckerman, Petri & Pekka Ilmakunnas. 2009. “Unemployment and self-assessed health:
evidence from panel data.” Health Economics 18(2):161–179.

Cai, Lixin. 2010. “The relationship between health and labour force participation: Evidence
from a panel data simultaneous equation model.” Labour Economics 17(1):77–90.

Cai, Lixin & Guyonne Kalb. 2006. “Health status and labour force participation: evidence from
Australia.” Health Economics 15(3):241–261.

Cain, Glen G. 1976. “The Challenge of Segmented Labor Market Theories to Orthodox Theory:
A Survey.” Journal of Economic Literature 14(4):1215–57.

Campbell, Karen E. 1988. “Gender Differences in Job-Related Networks.” Work and

Occupations 15(2):179–200.

Cardano, Mario, Giuseppe Costa & Moreno Demaria. 2004. “Social mobility and health in the
Turin longitudinal study.” Social Science & Medicine 58(8):1563–1574.

Carlsen, Kathrine, Susanne Oksbjerg Dalton, Finn Diderichsen & Christoffer Johansen. 2008.
“Risk for unemployment of cancer survivors: A Danish cohort study.” European Journal of
Cancer 44(13):1866–1874.

Carmichael, Lorne. 1983. “Firm-Specific Human Capital and Promotion Ladders.” The Bell
Journal of Economics 14(1):251–258.

Casciaro, Tiziana & Miguel Sousa Lobo. 2008. “When Competence Is Irrelevant: The Role of
Interpersonal Affect in Task-Related Ties.” Administrative Science Quarterly 53(4):655–
684.

Case, Anne, Angela Fertig & Christina Paxson. 2005. “The lasting impact of childhood health
and circumstance.” Journal of health economics 24(2):365–389.

Chandola, Tarani, Mel Bartley, Amanda Sacker, Crispin Jenkinson & Michael Marmot. 2003.
“Health selection in the Whitehall II study, UK.” Social Science & Medicine 56(10):2059–
2072.

Chase, Ivan D. 1991. “Vacancy Chains.” Annual Review of Sociology 17:133–154.

Christensen, Karl Bang, Merete Labriola, Thomas Lund & Mika Kivimäki. 2008. “Explaining the
social gradient in long-term sickness absence: a prospective study of Danish employees.”
Journal of Epidemiology and Community Health 62(2):181–183.

189
 Bibliography

Cinamon, Rachel Gali & Yisrael Rich. 2002. “Gender Differences in the Importance of Work
and Family Roles: Implications for Work–Family Conflict.” Sex Roles 47(11-12):531–541.

Claussen, B., A. Bjørndal & P. F. Hjort. 1993. “Health and re-employment in a two year
follow up of long term unemployed.” Journal of Epidemiology and Community Health
47(1):14–18.

Claussen, Bjorgulf, Jeroen Smits, Oyvind Naess & George Davey Smith. 2005. “Intragenerational
mobility and mortality in Oslo: Social selection versus social causation.” Social Science &
Medicine 61(12):2513–2520.

Cockerham, C. William. 2005. “Health Lifestyle Theory and the Convergence of Agency and
Structure.” Journal of Health and Social Behavior 46(1):51–67.

Cockerham, C. William, P. Brian Hinote & Pamela Abott. 2006. “Psychological distress,
gender, and health lifestyles in Belarus, Kazakhstan, Russia, and Ukraine.” Social Science
& Medicine 63:2381–2394.

Cohen, Philip N. & Matt L. Huffman. 2007. “Working for the Woman? Female Managers and
the Gender Wage Gap.” American Sociological Review 72(5):681–704.

Cohen, S. & N. Hamrick. 2003. “Stable individual differences in physiological response

to stressors: implications for stress-elicited changes in immune related health.” Brain,
Behavior, and Immunity 17(6):407–414.

Collins, James J, Catherine M Baase, Claire E Sharda, Ronald J Ozminkowski, Sean Nicholson,
Gary M Billotti, Robin S Turpin, Michael Olson & Marc L Berger. 2005. “The assessment
of chronic health conditions on work performance, absence, and total economic impact for
employers.” Journal of Occupational and Environmental Medicine / American College of
Occupational and Environmental Medicine 47(6):547–557.

Contoyannis, Paul & Nigel Rice. 2001. “The impact of health on wages: evidence from the
British Household Panel Survey.” Empirical Economics 26(4):599–622.

Cousineau, Jean-Michel, Robert Lacroix & Anne-Marie Girard. 1992. “Occupational Hazard and
Wage Compensating Differentials.” The Review of Economics and Statistics 74(1):166–
169.

Crichton, Sarah, Steven Stillman & Dean Hyslop. 2010. “Returning to work from injury:
longitudinal evidence on employment and earnings.” Industrial and Labor Relations Review
64:765.

Cropanzano, Russell, Deborah E. Rupp & Zinta S. Byrne. 2003. “The relationship of emotional
exhaustion to work attitudes, job performance, and organizational citizenship behaviors.”
Journal of Applied Psychology 88(1):160–169.

190
 Bibliography

Crossley, Thomas F. & Steven Kennedy. 2002. “The reliability of self-assessed health status.”
Journal of Health Economics 21(4):643–658.

Dahl, E. & P. Kjærsgaard. 1993. “Social mobility and inequality in mortality An assessment of
the health selection hypothesis.” The European Journal of Public Health 3(2):124–132.

Dahl, Espen. 1993a. “High mortality in lower salaried Norwegian men: the healthy worker
effect?” Journal of Epidemiology and Community Health 47(3):192–194.

Dahl, Espen. 1993b. “Social inequality in health—The role of the healthy worker effect.” Social
Science & Medicine 36(8):1077–1086.

Dawid, A. P. 2000. “Causal Inference without Counterfactuals.” Journal of the American

Statistical Association 95(450).

de Croon, Einar M, Judith K Sluiter & Monique H W Frings-Dresen. 2003. “Need for recovery
after work predicts sickness absence: a 2-year prospective cohort study in truck drivers.”
Journal of Psychosomatic Research 55(4):331–339.

Dean, Kathryn. 1989. “Self-care components of lifestyles: The importance of gender, attitudes
and the social situation.” Social Science & Medicine 29(2):137–152.

Demerouti, Evangelia, Pascale M. Le Blanc, Arnold B. Bakker, Wilmar B. Schaufeli & Joop
Hox. 2009. “Present but sick: a three-wave study on job demands, presenteeism and
burnout.” Career Development International 14(1):50–68.

Dempsey, Kenneth C. 2000. “Men and women’s power relationships and the persisting
inequitable division of housework.” Journal of Family Studies 6(1):7–24.

Dittrich, Dennis & Martin Kocher. 2011. “Monitoring and Pay: An Experiment on Employee
Performance Under Endogenous Supervision.”.

Do, D. Phuong, Lu Wang & Michael R. Elliott. 2013. “Investigating the relationship between
neighborhood poverty and mortality risk: A marginal structural modeling approach.” Social
Science & Medicine 91:58–66.

Doeringer, Peter & Michael J. Piore. 1985. Internal labor markets and manpower analysis.
Armonk N.Y.: M.E. Sharpe.

Dowd, Jennifer Beam & Anna Zajacova. 2007. “Does the Predictive Power of Self-Rated Health
for Subsequent Mortality Risk Vary by Socioeconomic Status in the US?” International
Journal of Epidemiology 36(6):1214–1221.

Dowler, Elizabeth & Nick Spencer. 2007. Challenging Health Inequalities: From Acheson to
Choosing Health. The Policy Press.

191
 Bibliography

Dragano, Nico & Johannes Siegrist. 2009. Die Lebenslaufperspektive gesundheitlicher Ungleich-
heit: Konzepte und Forschungsergebnisse. In Gesundheitliche Ungleichheit: Grundlagen,
Probleme, Perspektiven, ed. Matthias Richter & Klaus Hurrelmann. VS Verlag pp. 181–
194.

Duguet, Emmanuel & Christine Le Clainche. 2012. “The Impact of Health Events on Individual
Labor Market Histories: The Message from Difference-in-Differences with Exact Matching.”
SSRN eLibrary .

Durkheim, Èmile. 1922. De La Division Du Travail Social. Paris: Librairie Félix Alcan.

Eagly, Alice H. & Linda L. Carli. 2007. Through the Labyrinth: The Truth About How Women
Become Leaders. Mcgraw-Hill Professional.

Eaton, William W., Carles Muntaner, Gregory Bovasso & Corey Smith. 2001. “Socioeconomic
Status and Depressive Syndrome: The Role of Inter- and Intra-generational Mobility,
Government Assistance, and Work Environment.” Journal of Health and Social Behavior
42(3):277–294.

Edington, Dee W. & Alyssa B. Schultz. 2008. “The total value of health: a review of literature.”
International Journal of Workplace Health Management 1(1):8–19.

Eid, Michael. 1996. “Longitudinal confirmatory factor analysis for polytomous item responses:
Model definition and model selection on the basis of stochastic measurement theory.”
Methods of Psychological Research Online 1(4):65–85.

Elias, Peter. 1994. “Job-Related Training, Trade Union Membership, and Labour Mobility: A
Longitudinal Study.” Oxford Economic Papers 46(4):563–578.

Eliason, Scott R. 1995. “An Extension of the Sørensen-Kalleberg Theory of the Labor Market
Matching and Attainment Processes.” American Sociological Review 60(2):247–271.

Elovainio, Marko, Jane E. Ferrie, Archana Singh-Manoux, Martin Shipley, G. David Batty,
Jenny Head, Mark Hamer, Markus Jokela, Marianna Virtanen, Eric Brunner, Michael G.
Marmot & Mika Kivimäki. 2011. “Socioeconomic Differences in Cardiometabolic Factors:
Social Causation or Health-related Selection? Evidence From the Whitehall II Cohort
Study, 1991–2004.” American Journal of Epidemiology 174(7):779–789.

Elstad, Jon Ivar & Steinar Krokstad. 2003. “Social causation, health-selective mobility, and the
reproduction of socioeconomic health inequalities over time: panel study of adult men.”
Social Science & Medicine 57(8):1475–1489.

Elvira, Marta M. & Ishak Saporta. 2001. “How does Collective Bargaining Affect the Gender
Pay Gap?” Work and Occupations 28(4):469–490.

192
 Bibliography

Emerson, Michael. 1988. “Regulation or deregulation of the labour market : Policy regimes
for the recruitment and dismissal of employees in the industrialised countries.” European
Economic Review 32(4):775–817.

EPEH. 2007. Closing the Gap: Strategies for Action to tackle Health Inequalities in Europe
A European project from 2004 to 2007. European Partners For Equity In Health.

Eriksson, Hans-G, Anna-Sophia von Celsing, Rolf Wahlström, Lotta Janson, Viktoria Zander &
Thorne Wallman. 2008. “Sickness absence and self-reported health a population-based
study of 43,600 individuals in central Sweden.” BMC Public Health 8:426.

Eriksson, Ingeborg, Anna-Lena Undén & Stig Elofsson. 2001. “Self-rated health. Comparisons
between three different measures. Results from a population study.” International Journal
of Epidemiology 30(2):326–333.

Escriche, Luisa. 2007. “Persistence of Occupational Segregation: the Role of the Intergenera-
tional Transmission of Preferences.” The Economic Journal 117(520):837–857.

Ettner, Susan L., Richard G. Frank & Ronald C. Kessler. 1997. “The Impact of Psychiatric
Disorders on Labor Market Outcomes.” National Bureau of Economic Research Working
Paper Series No. 5989.

Faltermaier, T. 2008. Geschlechtsspezifische Dimensionen im Gesundheitsverständnis und

Gesundheitsverhalten. In Fehlzeiten-Report 2007, ed. Bernhard Badura, Helmut Schröder
& Christian Vetter. Berlin, Heidelberg: Springer Berlin Heidelberg pp. 35–45.

Ferrie, J E, M Kivimäki, J Head, M J Shipley, J Vahtera & M G Marmot. 2005. “A comparison

of self-reported sickness absence with absences recorded in employers’ registers: evidence
from the Whitehall II study.” Occupational and Environmental Medicine 62(2):74–79.

Filer, Randall K. 1985. “Male-Female Wage Differences: The Importance of Compensating

Differentials.” Industrial and Labor Relations Review 38:426.

Flap, Henk & Beate Völker. 2001. “Goal specific social capital and job satisfaction: Effects of
different types of networks on instrumental and social aspects of work.” Social Networks
23(4):297–320.

Flora, David B. & Patrick J. Curran. 2004. “An Empirical Evaluation of Alternative Methods of
Estimation for Confirmatory Factor Analysis With Ordinal Data.” Psychological methods
9(4):466–491.

Fox, A J, P O Goldblatt & D R Jones. 1985. “Social class mortality differentials: artefact,
selection or life circumstances?” Journal of Epidemiology and Community Health 39(1):1
–8.

193
 Bibliography

Fox, John W. 1990. “Social Class, Mental Illness, and Social Mobility: The Social Selection-Drift
Hypothesis for Serious Mental Illness.” Journal of Health and Social Behavior 31(4):344–
353.

Franzen, Dr Axel & Dominik Hangartner. 2005. “Soziale Netzwerke und beruflicher Erfolg.”
KZfSS Kölner Zeitschrift für Soziologie und Sozialpsychologie 57(3):443–465.

Frederiksen, Anders. 2008. “Gender differences in job separation rates and employment stability:
New evidence from employer-employee data.” Labour Economics 15(5):915–937.

Fried, Yitzhak, Samuel Melamed & Haim A Ben-David. 2002. “The joint effects of noise, job
complexity, and gender on employee sickness absence: An exploratory study across 21 orga-
nizations - the CORDIS study.” Journal of Occupational and Organizational Psychology
75(2):131–144.

Frijters, Paul, David W Johnston & Michael A Shields. 2010. “Mental health and
labour market participation: Evidence from IV panel data models.” Discussion paper
series-Forschungsinstitut zur Zukunft der Arbeit .

Galama, Titus & Hans van Kippersluis. 2010. “A Theory of Socioeconomic Disparities in
Health over the Life Cycle.” RAND Working Paper .

Gambin, Lynn Maria. 2005. “The impact of health on wages in Europe – Does gender matter?
(DRAFT).”.

Gangl, Markus. 2010. “Causal Inference in Sociological Research.” Annual Review of Sociology
36:21–47.

Garcı́a-Gómez, Pilar, Andrew M Jones & Nigel Rice. 2010. “Health effects on labour market
exits and entries.” Labour Economics 17(1):62–76.

Giesecke, Johannes. 2006. Arbeitsmarktflexibilisierung und Soziale Ungleichheit:

Sozio-ökonomische Konsequenzen befristeter Beschäftigungsverhältnisse in Deutschland
und Großbritannien. 1 ed. VS Verlag für Sozialwissenschaften.

Giesecke, Johannes & Martin Groß. 2007. Flexibilisierung durch Befristung. In Atypische
Beschäftigung - Flexibiliserung und soziale Risiken, ed. Bernd Keller & Hartmut Seifert.
Berlin: edition sigma pp. 83–105.

Goldberg, E. M. & S. L. Morrison. 1963. “Schizophrenia and Social Class.” The British Journal
of Psychiatry 109(463):785–802.

Goldman, Noreen. 2001. “Mortality differentials: selection and causation.” Internation

Encyclopedia of the Social and Behavioral Sciences p. 10068–10070.

194
 Bibliography

Gorman, Elizabeth, Peter V. Marsden, L. Arne Kalleberg & Ivar Berg. 2001. Social networks,
job changes, and recruitment. In Sourcebook on Labor Markets: Evolving Structures and
Processes. New York: Kluwer Academic/Plenum pp. 467–502.

Gottschall, K. 2000. Soziale Ungleichheit und Geschlecht. Vol. 13 Leske+ Budrich.

Graham, Hilary. 2009. Understanding Health Inequalities. McGraw-Hill International.

Grams, Robert & Donald P. Schwab. 1985. “An Investigation of Systematic Gender-Related
Error in Job Evaluation.” Academy of Management Journal 28(2):279–290.

Granovetter, S. Mark. 1973. “The Strength of Weak Ties.” American Journal of Sociology
78(6):1360–1371.

Graves, Laura M. & Gary N. Powell. 1995. “The Effect Of Sex Similarity On Recruiters’ Evalu-
ations Of Actual Applicants: A Test Of The Similarity-attraction Paradigm.” Personnel
Psychology 48(1):85–98.

Greene, William H. 2003. Econometric Analysis, 5/e. Pearson Education India.

Greenhaus, Jeffrey H., Saroj Parasuraman & Wayne M. Wormley. 1990. “Effects of Race on
Organizational Experience, Job Performance Evaluations, and Career Outcomes.” Academy
of Management Journal 33(1):64–86.

Grossman, Michael. 1972. “On the Concept of Health Capital and the Demand for Health.”
Journal of Political Economy 80(2):223–255.

Grossman, Michael. 1976. The Correlation between Health and Schooling. In Household
Production and Consumption, ed. Nestor Terleckyj. NBER pp. 147–224.

Gunasekara, Fiona Imlach, Kristie Carter & Tony Blakely. 2012. “Comparing self-rated health
and self-assessed change in health in a longitudinal survey: Which is more valid?” Social
Science & Medicine 74(7):1117–1124.

Haan, Peter & Michal Myck. 2009. “Dynamics of health and labor market risks.” Journal of
Health Economics 28(6):1116–1125.

Haas, Steven A. 2006. “Health Selection and the Process of Social Stratification: The Effect of
Childhood Health on Socioeconomic Attainment.” Journal of Health and Social Behavior
47(4):339 –354.

Haas, Steven A., M. Maria Glymour & Lisa F. Berkman. 2011. “Childhood Health and
Labor Market Inequality over the Life Course.” Journal of Health and Social Behavior
52(3):298–313.

195
 Bibliography

Hainmueller, Jens & Yiqing Xu. 2011. “Ebalance: A Stata Package for Entropy Balancing.”
SSRN eLibrary .

Hakim, Catherine. 1998. “Developing a Sociology for the Twenty-First Century: Preference
Theory.” The British Journal of Sociology 49(1):137–143.

Hakim, Catherine. 2002. “Lifestyle Preferences as Determinants of Women’s Differentiated

Labor Market Careers.” Work and Occupations 29(4):428–459.

Halleröd, Björn & Jan-Eric Gustafsson. 2011. “A longitudinal analysis of the relationship
between changes in socio-economic status and changes in health.” Social Science &
Medicine 72(1):116–123.

Hamermesh, Daniel S. 1987. “The Costs of Worker Displacement.” The Quarterly Journal of
Economics 102(1):51 –75.

Hammarström, Anne & Urban Janlert. 2005. “Health selection in a 14-year follow-up study–A
question of gendered discrimination?” Social Science & Medicine 61(10):2221–2232.

Harkey, John, David L. Miles & William A. Rushing. 1976. “The Relation Between Social
Class and Functional Status: A New Look at the Drift Hypothesis.” Journal of Health
and Social Behavior 17(3):194–204.

Hasan, D., S. Akif, D. Subhani & M. Imtiaz. 2012. “Top management’s snooping: Is sneaking
over employees’ productivity and job commitment a wise approach?”.

Hashimoto, Masanori. 1981. “Firm-Specific Human Capital as a Shared Investment.” The

American Economic Review 71(3):475–482.

Haveman, Robert, Barbara Wolfe, Brent Kreider & Mark Stone. 1994. “Market work, wages,
and men’s health.” Journal of Health Economics 13(2):163–182.

Heponiemi, Tarja, Marko Elovainio, Kristiina Manderbacka, Anna-Mari Aalto, Mika Kivimäki
& Ilmo Keskimäki. 2007. “Relationship between unemployment and health among health
care professionals: Health selection or health effect?” Journal of Psychosomatic Research
63(4):425–431.

Hersch, Joni. 1998. “Compensating Differentials for Gender-Specific Job Injury Risks.” The
American Economic Review 88(3):598–607.

HHS. 2011. HHS Action Plan to Reduce Racial and Ethnic Health Disparities. Washington,
D.C.: U.S. Department of Health and Human Services.

Hinz, Thomas & Hermann Gartner. 2005. “Geschlechtsspezifische Lohnunterschiede in Branchen,

Berufen und Betrieben.” Zeitschrift für Soziologie 34(1):22–39.

196
 Bibliography

Hofoss, Dag, Espen Dahl, Jon Ivar Elstad & Milada Cvancarova. 2012. “Selection and mortality:
a ten-year follow-up of income decile mortality in Norway.” The European Journal of Public
Health pp. 1–5.

Holgado–Tello, Francisco Pablo, Salvador Chacón–Moscoso, Isabel Barbero–Garcı́a & Enrique

Vila–Abad. 2010. “Polychoric versus Pearson correlations in exploratory and confirmatory
factor analysis of ordinal variables.” Quality & Quantity 44(1):153–166.

Holland, P. W. 1986. “Statistics and causal inference.” Journal of the American Statistical
Association 81(396):945–960.

Hook, Jennifer L. 2010. “Gender inequality in the welfare state: sex segregation in housework,
1965-2003.” American Journal of Sociology 115(5):1480–1523.

Hoven, Hanno & Johannes Siegrist. 2013. “Work characteristics, socioeconomic position and
health: a systematic review of mediation and moderation effects in prospective studies.”
Occupational and Environmental Medicine .

Hu, Li-tze & Peter M. Bentler. 1999. “Cutoff criteria for fit indexes in covariance structure
analysis: Conventional criteria versus new alternatives.” Structural Equation Modeling: A
Multidisciplinary Journal 6(1):1–55.

Huisman, Martijn, Frank Van Lenthe & Johan Mackenbach. 2007. “The Predictive Ability of
Self-Assessed Health for Mortality in Different Educational Groups.” International Journal
of Epidemiology 36(6):1207–1213.

Hultin, Mia. 2003. “Some Take the Glass Escalator, Some Hit the Glass Ceiling? Career
Consequences of Occupational Sex Segregation.” Work and Occupations 30(1):30–61.

Hurrelmann, Klaus. 1994. International Handbook of Adolescence:. Westport: Greenwood.

Hurrelmann, Klaus, Katharina Rathmann & Matthias Richter. 2011. “Health inequalities and
welfare state regimes. A research note.” Journal of Public Health 19(1):3–13.

Huurre, Taina, Ossi Rahkonen, Erkki Komulainen & Hillevi Aro. 2005. “Socioeconomic status
as a cause and consequence of psychosomatic symptoms from adolescence to adulthood.”
Social psychiatry and psychiatric epidemiology 40(7):580–587.

Idler, E. L. & S. V. Kasl. 1995. “Self-Ratings of Health: Do they also Predict change in
Functional Ability?” The Journals of Gerontology Series B: Psychological Sciences and
Social Sciences 50B(6):S344–S353.

Igbaria, Magid & Jack J. Baroudi. 1995. “The Impact of Job Performance Evaluations on Career
Advancement Prospects: An Examination of Gender Differences in the IS Workplace.”
MIS Quarterly 19(1):107–123.

197
 Bibliography

Inkson, Kerr. 2004. “Images of career: Nine key metaphors.” Journal of Vocational Behavior
65(1):96–111.

Jacobs, Jerry A. & Ronnie J. Steinberg. 1990. “Compensating Differentials and the Male-Female
Wage Gap: Evidence from the New York State Comparable Worth Study.” Social Forces
69(2):439–468.

Jacobsen, J., G. Skillman & Joyce Jacobsen. 2004. Labor Markets and Employment
Relationships: A Comprehensive Approach. Blackwell Publishers.

Johns, Gary. 2010. “Presenteeism in the workplace: A review and research agenda.” Journal of
Organizational Behavior 31(4):519–542.

Jones, Andrew M, Nigel Rice & Jennifer Roberts. 2010. “Sick of work or too sick to work?
Evidence on self-reported health shocks and early retirement from the BHPS.” Economic
Modelling 27(4):866–880.

Jungbauer-Gans, Monika, Christiane Gross, Matthias Richter & Klaus Hurrelmann.

2009. Erklärungsansätze sozial differenzierter Gesundheitschancen. In Gesundheitliche
Ungleichheit: Grundlagen, Probleme, Perspektiven. VS Verlag pp. 77–98.

Jusot, F, M Khlat, T Rochereau & C Serme. 2008. “Job loss from poor health, smoking and
obesity: a national prospective survey in France.” Journal of Epidemiology and Community
Health 62(4):332 –337.

Jylhä, Marja. 2009. “What is self-rated health and why does it predict mortality? Towards a
unified conceptual model.” Social Science & Medicine 69(3):307–316.

Jylhä, Marja, Jack M. Guralnik, Luigi Ferrucci, Jukka Jokela & Eino Heikkinen. 1998. “Is Self-
Rated Health Comparable across Cultures and Genders?” The Journals of Gerontology
Series B: Psychological Sciences and Social Sciences 53B(3):S144–S152.

Jäckle, Robert & Oliver Himmler. 2010. “Health and Wages Panel Data Estimates Considering
Selection and Endogeneity.” Journal of Human Resources 45(2):364–406.

Kalleberg, Arne L. & Aage B. Sorensen. 1979. “The Sociology of Labor Markets.” Annual
Review of Sociology 5:351–379.

Kambourov, Gueorgui & Iourii Manovskii. 2009. “Occupational Specificity of Human Capital.”
International Economic Review 50(1):63–115.

Kanter, Rosabeth Moss. 1977. Men And Women Of The Corporation. Basic Books.

Karlson, Kristian Bernt, Anders Holm & Richard Breen. 2012. “Comparing Regression Coef-
ficients Between Same-sample Nested Models Using Logit and Probit A New Method.”
Sociological Methodology 42(1):286–313.

198
 Bibliography

Kaufman, J. S & C. Poole. 2000. “Looking Back on Causal Thinking in the Health Sciences.”
Annual Review of Public Health 21(1):101–119.

Kenkel, Donald S. 1991. “Health Behavior, Health Knowledge, and Schooling.” Journal of
Political Economy 99(2):287–305.

Ki, Myung, Amanda Sacker, Yvonne Kelly & James Nazroo. 2011. “Health selection operating
between classes and across employment statuses.” Journal of Epidemiology and Community
Health 65(12):1132–1139.

Kilbourne, Barbara Stanek, George Farkas, Kurt Beron, Dorothea Weir & Paula England. 1994.
“Returns to Skill, Compensating Differentials, and Gender Bias: Effects of Occupational
Characteristics on the Wages of White Women and Men.” American Journal of Sociology
100(3):689–719.

Kivimäki, Mika, Jussi Vahtera, Marko Elovainio, Jaana Pentti & Marianna Virtanen. 2003.
“Human Costs of Organizational Downsizing: Comparing Health Trends Between Leavers
and Stayers.” American Journal of Community Psychology 32(1-2):57–67.

Klein-Hesselink, D. John & Ingeborg P. Spruit. 1992. “The Contribution of Unemployment to

Socioeconomic Health Differences.” International Journal of Epidemiology 21(2):329–337.

Kline, Rex B. 2010. Principles and Practice of Structural Equation Modeling, Third Edition.
Third edition ed. The Guilford Press.

Koskela, Riitta-Sisko, Kalevi Luoma & Sven Hernberg. 1976. “Turnover and health selection
among foundry workers.” Scandinavian journal of work, environment & health p. 90–105.

Kray, Laura J, Adam D Galinsky & Leigh Thompson. 2002. “Reversing the Gender Gap in
Negotiations: An Exploration of Stereotype Regeneration.” Organizational Behavior and
Human Decision Processes 87(2):386–410.

Kroh, M., H. Neiss, L. Kroll & T. Lampert. 2012. “Menschen mit hohen Einkommen leben
länger.” DIW Wochenbericht 79(38):3–15.

Kroll, Lars Eric. 2011. “Konstruktion und Validierung eines allgemeinen Index für die Arbeitsbe-
lastung in beruflichen Tätigkeiten anhand von ISCO-88 und KldB-92.” Methoden, Daten,
Analysen 5(1):63–90.

Kullberg, Karin. 2012. “From Glass Escalator to Glass Travelator: On the Proportion of Men in
Managerial Positions in Social Work in Sweden.” British Journal of Social Work pp. 1–18.

Kunze, Astrid. 2008. “Gender wage gap studies: consistency and decomposition.” Empirical
Economics 35(1):63–76.

199
 Bibliography

Lawrence, P. S. 1948. “Chronic Illness and Socio-Economic Status.” Public Health Reports
63(47):1507–1533.

Lazear, Edward P. 1990. “Job Security Provisions and Employment.” The Quarterly Journal
of Economics 105(3):699 –726.

Lazear, Edward P. 2003. Firm-Specific Human Capital: A Skill-Weights Approach. Working

Paper 9679 National Bureau of Economic Research.

Lee, Lung-Fei. 1982. “Health and Wage: A Simultaneous Equation Model with Multiple
Discrete Indicators.” International Economic Review 23(1):199–221.

Leeth, John D. & John Ruser. 2003. “Compensating Wage Differentials for Fatal and Nonfatal
Injury Risk by Gender and Race.” Journal of Risk and Uncertainty 27(3):257–277.

Leino-Arjas, P., J. Liira, P. Mutanen, A. Malmivaara & E. Matikainen. 1999. “Predictors and
consequences of unemployment among construction workers: prospective cohort study.”
BMJ 319(7210):600–605.

Leuze, Kathrin & Susanne Strauß. 2009. “Lohnungleichheiten zwischen Akademikerinnen und
Akademikern: Der Einfluss von fachlicher Spezialisierung, frauendominierten Fächern und
beruflicher Segregation.” Zeitschrift für Soziologie 38(4):262–281.

Levenstein, Susan, Margot W. Smith & George A. Kaplan. 2001. “Psychosocial Predictors of
Hypertension in Men and Women.” Arch Intern Med 161(10):1341–1346.

Li, C.-Y. & F.-C. Sung. 1999. “A review of the healthy worker effect in occupational epidemi-
ology.” Occupational Medicine 49(4):225 –229.

Lichtenstein, Paul, R. Jennifer Harris, L. Nancy Pedersen & E. G. McClearn. 1992. “So-
cioeconomic Status And Physical Health, How Are They Related?: An Empirical Study
Based On Twins Reared Apart And Twins Reared Together.” Social Science & Medicine
36(4):441–450.

Liebeskind, Uta. 2004. “Arbeitsmarktsegregation und Einkommen.” KZfSS Kölner Zeitschrift

für Soziologie und Sozialpsychologie 56(4):630–652.

Lin, Nan. 1999. “Social Networks and Status Attainment.” Annual Review of Sociology
25:467–487.

Lindeboom, Maarten & Eddy van Doorslaer. 2004. “Cut-point shift and index shift in self-
reported health.” Journal of Health Economics 23(6):1083–1099.

Loughran, David S. & Julie M. Zissimopoulos. 2009. “Why Wait? The Effect of Marriage
and Childbearing on the Wages of Men and Women.” Journal of Human Resources
44(2):326–349.

200
 Bibliography

Lubke, Gitta H. & Bengt O. Muthén. 2004. “Applying Multigroup Confirmatory Factor
Models for Continuous Outcomes to Likert Scale Data Complicates Meaningful Group
Comparisons.” Structural Equation Modeling: A Multidisciplinary Journal 11(4):514–534.

Lundberg, Olle. 1991. “Childhood Living Conditions, Health Status, and Social Mobility: A
Contribution to the Health Selection Debate.” European Sociological Review 7(2):149
–162.

Lundborg, Petter, Martin Nilsson & Johan Vikström. 2011. “Socioeconomic Heterogeneity
in the Effect of Health Shocks on Earnings: Evidence from Population-Wide Data on
Swedish Workers.” IZA Discussion Paper .

Lundin, Andreas, Ingvar Lundberg, Lennart Hallsten, Jan Ottosson & Tomas Hemmingsson.
2010. “Unemployment and mortality—a longitudinal prospective study on selection and
causation in 49321 Swedish middle-aged men.” Journal of Epidemiology and Community
Health 64(01):22–28.

Lynch, John W, George Davey Smith, George A Kaplan & James S House. 2000. “Income
inequality and mortality: importance to health of individual income, psychosocial environ-
ment, or material conditions.” BMJ 320(7243):1200 –1204.

Lynch, W. J, A. G Kaplan & T. J Salonen. 1997. “Why do poor People behave poorly?:
Variation in Adult Health Behaviours and psychosocial Characteristics by Stages of the
socioeconomic Lifecourse.” Social Science & Medicine 44(6):809–819.

Maas, Victor S. & Raquel Torres-González. 2011. “Subjective Performance Evaluation and
Gender Discrimination.” Journal of Business Ethics 101(4):667–681.

Mackenbach, Johan P. 2012. “The persistence of health inequalities in modern welfare states:
The explanation of a paradox.” Social Science & Medicine 75(4):761–769.

Mackenbach, JP. 2006. “Health Inequalities: Europe in Profile: An independent, expert report
commissioned by the UK Presidency of the EU.” Rotterdam: Dept. of Public Health,
Erasmus MC 41.

MacLeod, W. Bentley. 2003. “Optimal Contracting with Subjective Evaluation.” The American
Economic Review 93(1):216–240.

Macpherson, David A & Barry T Hirsch. 1995. “Wages and Gender Composition: Why Do
Women’s Jobs Pay Less?” Journal of Labor Economics 13(3):426–71.

Madden, Janice Fanning. 2012. “Performance-Support Bias and the Gender Pay Gap among
Stockbrokers.” Gender & Society 26(3):488–518.

201
 Bibliography

Magee, William. 2004. “Effects of illness and disability on job separation.” Social Science &
Medicine 58(6):1121–1135.

Manor, Orly, Sharon Matthews & Chris Power. 2001. “Self-Rated Health and Limiting
Longstanding Illness: Inter-Relationships with Morbidity in Early Adulthood.” International
Journal of Epidemiology 30(3):600–607.

Manor, Orly, Sharon Matthews & Chris Power. 2003. “Health selection: the role of inter- and
intra-generational mobility on social inequalities in health.” Social Science & Medicine
57(11):2217–2227.

Marini, Margaret Mooney, Pi-Ling Fan, Erica Finley & Ann M. Beutel. 1996. “Gender and Job
Values.” Sociology of Education 69(1):49–65.

Marmot, Michael. 2005. “Social determinants of health inequalities.” The Lancet

365(9464):1099–1104.

Martins, Pedro. 2007. “Rent Sharing and Wages.” Reflets et Perspectives de la Vie Economique
46(2/3):23–31.

Mastekaasa, Arne. 1996. “Unemployment and Health: Selection Effects.” Journal of Community
& Applied Social Psychology 6(3):189–205.

Maume, D.J. 1999. “Glass ceilings and glass escalators occupational segregation and race and
sex differences in managerial promotions.” Work and Occupations 26(4):483–509.

McDonald, Steve. 2011. “What’s in the “old boys” network? Accessing social capital in
gendered and racialized networks.” Social Networks 33(4):317–330.

McDonald, Steve, Nan Lin & Dan Ao. 2009. “Networks of Opportunity: Gender, Race, and
Job Leads.” Social Problems 56(3):385–402.

McDonough, Peggy & Benjamin C. Amick. 2001. “The social context of health selection: a
longitudinal study of health and employment.” Social Science & Medicine 53(1):135–145.

McLeod, Jane D. & Eliza K. Pavalko. 2008. “From selection effects to reciprocal processes:
what does attention to the life course offer?” Advances in Life Course Research 13:75–104.

McMichael, J. A, R. Spirtas & L. L Kupper. 1974. “An epidemologic study of mortality within a
cohort of rubber workers, 1964-1972.” Journal of Occupational Medicine 16(7):458–464.

McMunn, Anne, Mel Bartley, Rebecca Hardy & Diana Kuh. 2006. “Life course social roles
and women’s health in mid-life: causation or selection?” Journal of Epidemiology and
Community Health 60(6):484–489.

202
 Bibliography

McPherson, Miller, Lynn Smith-Lovin & James M. Cook. 2001. “Birds of a Feather: Homophily
in Social Networks.” Annual Review of Sociology 27:415–444.

Meerding, W.J., W. IJzelenberg, M.A. Koopmanschap, J.L. Severens & A. Burdorf. 2005.
“Health problems lead to considerable productivity loss at work among workers with high
physical load jobs.” Journal of Clinical Epidemiology 58(5):517–523.

Merton, Robert K. 1968. Social theory and social structure. New York: Free Press.

Mincer, Jacob. 1974. Schooling, experience, and earnings. New York: National Bureau of
Economic Research; distributed by Columbia University Press.

Mincer, Jacob & Solomon Polachek. 1978. “An Exchange: The Theory of Human Capital
and the Earnings of Women: Women’s Earnings Reexamined.” The Journal of Human
Resources 13(1):118–134.

Montgomery, Scott M, Mel J Bartley, Derek G Cook & Michael Ej Wadsworth. 1996. “Health
and social precursors of unemployment in young men in Great Britain.” Journal of
Epidemiology and Community Health 50(4):415–422.

Mood, Carina. 2010. “Logistic Regression: Why We Cannot Do What We Think We Can Do,
and What We Can Do About It.” European Sociological Review 26(1):67 –82.

Moore, Gwen. 1990. “Structural Determinants of Men’s and Women’s Personal Networks.”
American Sociological Review 55(5):726–735.

Mount, Michael K., Marcia R. Sytsma, Joy Fisher Hazucha & Katherine E. Holt. 1997.
“Rater-ratee Race Effects In Developmental Performance Ratings Of Managers.” Personnel
Psychology 50(1):51–69.

Mulatu, Mesfin Samuel & Carmi Schooler. 2002. “Causal Connections between Socio-Economic
Status and Health: Reciprocal Effects and Mediating Mechanisms.” Journal of Health
and Social Behavior 43(1):22–41.

Mullahy, John & Jody L. Sindelar. 1991. “Gender Differences in Labor Market Effects of
Alcoholism.” The American Economic Review 81(2):161–165.

Mushkin, Selma J. 1962. “Health as an Investment.” The Journal of Political Economy

70(5):129–157.

Muthén, Linda & Bengt O. Muthén. 2012. Mplus User’s Guide. Seventh Edition. Los Angeles,
CA: Muthén & Muthén.

Nishi, Akihiro, Ichiro Kawachi, Kokoro Shirai, Hiroshi Hirai, Seungwon Jeong & Katsunori
Kondo. 2012. “Sex/Gender and Socioeconomic Differences in the Predictive Ability of
Self-Rated Health for Mortality.” PLoS ONE 7(1):e30179.

203
 Bibliography

Oakley, Judith G. 2000. “Gender-based Barriers to Senior Management Positions: Understanding

the Scarcity of Female CEOs.” Journal of Business Ethics 27(4):321–334.

O’Donnell, Aisling T., Michelle K. Ryan & Jolanda Jetten. 2013. “The hidden costs of surveil-
lance for performance and helping behaviour.” Group Processes & Intergroup Relations
pp. 246—256.

Okamoto, Dina & Paula England. 1999. “Is There a Supply Side to Occupational Sex
Segregation?” Sociological Perspectives 42(4):557–582.

Ortiz, Susan Y. & Vincent J. Roscigno. 2009. “Discrimination, Women, And Work: Processes
and Variations by Race and Class.” Sociological Quarterly 50(2):336–359.

Ostlin, P. 1988. “Negative health selection into physically light occupations.” Journal of
Epidemiology and Community Health 42(2):152 –156.

Palloni, Alberto, Carolina Milesi, Robert G White & Alyn Turner. 2009. “Early childhood
health, reproduction of economic inequalities and the persistence of health and mortality
differentials.” Social science & medicine 68(9):1574–1582.

Paringer, Lynn. 1983. “Women and Absenteeism: Health or Economics?” The American
Economic Review 73(2):123–127.

Parkin, Frank. 2003. Strategies of Social Closure in Class Formation. In The Social Analysis of
Class Structure, ed. Frank Parkin. Reprint ed. Routledge Chapman & Hall pp. 1–18.

Parsons, Talcott. 1972. Das System moderner Gesellschaften. München: Juventa.

Paul, Karsten I. & Klaus Moser. 2009. “Unemployment impairs mental health: Meta-analyses.”
Journal of Vocational Behavior 74(3):264–282.

Pavalko, Eliza K., Fang Gong & J. Scott Long. 2007. “Women’s Work, Cohort Change, and
Health.” Journal of Health and Social Behavior 48(4):352–368.

Pelkowski, Jodi Messer & Mark C Berger. 2004. “The impact of health on employment, wages,
and hours worked over the life cycle.” The Quarterly Review of Economics and Finance
44(1):102–121.

Perrott, J. & Edgar Sydenstricker. 1935. “Causal and Selective Factors in Sickness.” American
Journal of Sociology 40(6):804–812.

Phelps, S. Edmund. 1972. “The Statistical Theory of Racism and Sexism.” American Economic
Review 62(4):659–661.

204
 Bibliography

Pierenkemper, Toni & Klaus F. Zimmermann. 2009. “Zum Aufstieg und Niedergang des
Normalarbeitsverhältnisses in Deutschland 1800-2010. Ein Forschungsprojekt.” Jahrbuch
für Wirtschaftsgeschichte / Economic History Yearbook 50(2):231–242.

Piha, Kustaa, Mikko Laaksonen, Pekka Martikainen, Ossi Rahkonen & Eero Lahelma. 2010.
“Interrelationships between education, occupational class, income and sickness absence.”
The European Journal of Public Health 20(3):276–280.

Polachek, Solomon William. 1981. “Occupational Self-Selection: A Human Capital Approach

to Sex Differences in Occupational Structure.” The Review of Economics and Statistics
63(1):60–69.

Pongratz, Hans J. & G. Günter Voß. 2003. “From employee to entreployee: Towards a
self-entrepreneurial work force?” Concepts and Transformation 8(3):239–254.

Powell, G. N. & D. A. Butterfield. 1997. “Effect Of Race On Promotions To Top Management

In A Federal Department.” Academy of Management Journal 40(1):112–128.

Power, C., S. Matthews & O. Manor. 1996. “Inequalities in self rated health in the 1958 birth
cohort: lifetime social circumstances or social mobility?” BMJ : British Medical Journal
313(7055):449–453.

Prendergast, Canice & Robert H. Topel. 1996. “Favoritism in Organizations.” Journal of

Political Economy 104(5):958–78.

Quesnel–Vallée, Amélie. 2007. “Self-Rated Health: Caught in the Crossfire of the Quest for
‘true’ Health?” International Journal of Epidemiology 36(6):1161–1164.

Reskin, Barbara F. & Debra Branch McBrier. 2000. “Why Not Ascription? Organizations’
Employment of Male and Female Managers.” American Sociological Review 65(2):210–
233.

Richter, Matthias, Irene Moor & Frank J. van Lenthe. 2012. “Explaining socioeconomic
differences in adolescent self-rated health: the contribution of material, psychosocial and
behavioural factors.” Journal of Epidemiology and Community Health 66(8):691–697.

Richter, Matthias, Katharina Rathman, Saoirse Nic Gabhainn, Alessio Zambon, William Boyce
& Klaus Hurrelmann. 2012. “Welfare state regimes, health and health inequalities in ado-
lescence: a multilevel study in 32 countries.” Sociology of health & illness 34(6):858–879.

Richter, Matthias & Klaus Hurrelmann. 2009. Gesundheitliche Ungleichheit: Grundlagen,

Probleme, Perspektiven. VS Verlag für Sozialwissenschaften.

Rodin, Judith & Gail McAvay. 1992. “Determinants of Change in Perceived Health in a
Longitudinal Study of Older Adults.” Journal of Gerontology 47(6):P373–P384.

205
 Bibliography

Roelen, Corné A M, Petra C Koopmans & Johan W Groothoff. 2010. “Subjective health
complaints in relation to sickness absence.” Work (Reading, Mass.) 37(1):15–21.

Rosenbaum, James E. 1979. “Tournament Mobility: Career Patterns in a Corporation.”

Administrative Science Quarterly 24(2):220.

Roth, Philip L., Kristen L. Purvis & Philip Bobko. 2012. “A Meta-Analysis of Gender Group
Differences for Measures of Job Performance in Field Studies.” Journal of Management
38(2):719–739.

Royston, Patrick. 2006. “Explained variation for survival models.” Stata Journal 6(1):83–96.

Rubin, D. B. 1974. “Estimating causal effects of treatments in randomized and nonrandomized

studies.” Journal of Educational Psychology 66(5):688–701.

Rubin, D. B. 2005. “Causal inference using potential outcomes.” Journal of the American
Statistical Association 100(469):322–331.

Rutherford, Donald & D. Rutherford. 2005. Routledge Dictionary of Economics. Revised. ed.
Routledge Chapman & Hall.

Sackett, Paul R., Cathy L. DuBois & Ann W. Noe. 1991. “Tokenism in performance evaluation:
The effects of work group representation on male-female and White-Black differences in
performance ratings.” Journal of Applied Psychology 76(2):263–267.

Saint-Paul, Gilles. 2005. “Did European Labor Markets Become More Competitive in The
1990s? Evidence From Estimatedworker Rents.” Labor Markets Institutions pp. 281–300.

Sandanger, Inger, Jan F. Nygard, Sören Brage & Gunnar Tellnes. 2000. “Relation between
health problems and sickness absence: gender and age differences.” Scandinavian Journal
of Public Health 28(4):244 –252.

Schmitz, Hendrik. 2011. “Why are the unemployed in worse health? The causal effect of
unemployment on health.” Labour Economics 18(1):71–78.

Schneiderman, N., G. Ironson & S. D Siegel. 2005. “Stress and health: Psychological, behavioral,
and biological determinants.” Annual Review of Clinical Psychology 1:607–628.

Schnitzlein, Daniel D. 2011. “Umfang und Folgen der Nichtinanspruchnahme von Urlaub in
Deutschland.” DIW Wochenbericht 78(51/52):14–20.

Schurer, Stefanie. 2008. “Discrete heterogeneity in the impact of health shocks on labour
market outcomes.” Ruhr Economic Paper (71).

206
 Bibliography

Schuring, Merel, Lex Burdorf, Anton Kunst & Johan Mackenbach. 2007. “The effects of ill
health on entering and maintaining paid employment: evidence in European countries.”
Journal of epidemiology and community health 61(7):597–604.

Schwartz, J E, H S Friedman, J S Tucker, C Tomlinson-Keasey, D L Wingard & M H Criqui.

1995. “Sociodemographic and psychosocial factors in childhood as predictors of adult
mortality.” American Journal of Public Health 85(9):1237–1245.

Schwartz, S.H. & T. Rubel. 2005. “Sex differences in value priorities: cross-cultural and
multimethod studies.” Journal of personality and social psychology 89(6):1010.

Shapiro, Carl & Joseph E Stiglitz. 1984. “Equilibrium Unemployment as a Worker Discipline
Device.” American Economic Review 74(3):433–44.

Shmueli, Amir. 2003. “Socio-economic and demographic variation in health and in its measures:
the issue of reporting heterogeneity.” Social Science & Medicine 57(1):125–134.

Siegrist, Johannes & Michael Marmot. 2004. “Health inequalities and the psychosocial
environment–two scientific challenges.” Social Science & Medicine 58(8):1463–1473.

Singer, Judith D. & John B. Willett. 1993. “It’s About Time: Using Discrete-Time Survival
Analysis to Study Duration and the Timing of Events.” Journal of Educational and
Behavioral Statistics 18(2):155–195.

Singh-Manoux, Archana, Alice Gueguen, Pekka Martikainen, Jane Ferrie, Michael Marmot
& Martin Shipley. 2007. “Self-Rated Health and Mortality: Short- and Long-Term
Associations in the Whitehall II Study.” Psychosom Med 69(2):138–143.

Singh-Manoux, Archana, Aline Dugravot, Martin J Shipley, Jane E Ferrie, Pekka Martikainen,
Marcel Goldberg & Marie Zins. 2007. “The association between self-rated health and
mortality in different socioeconomic groups in the GAZEL cohort study.” International
Journal of Epidemiology 36(6):1222–1228.

Sluiter, J K. 2003. “Need for recovery from work related fatigue and its role in the development
and prediction of subjective health complaints.” Occupational and Environmental Medicine
60(>90001):62i–70.

Sluiter, J K, A J van der Beek & M H Frings-Dresen. 1999. “The influence of work characteristics
on the need for recovery and experienced health: a study on coach drivers.” Ergonomics
42(4):573–583.

Sluiter, J K, M H Frings-Dresen, A J van der Beek & T F Meijman. 2001. “The relation
between work-induced neuroendocrine reactivity and recovery, subjective need for recovery,
and health status.” Journal of Psychosomatic Research 50(1):29–37.

207
 Bibliography

Smith, G. D. 2000. “Inequalities in health continue to grow despite government’s pledges.”

BMJ 320(7234):582–582.

Smith, G D, M Bartley & D Blane. 1990. “The Black report on socioeconomic inequalities in
health 10 years on.” BMJ : British Medical Journal 301(6748):373–377.

Smith, George Davey. 2003. Health Inequalities: Lifecourse Approaches. The Policy Press.

Smith, James P. 1999. “Healthy Bodies and Thick Wallets: The Dual Relation between Health
and Economic Status.” The Journal of Economic Perspectives 13(2):145–166.

Snyder, Karrie Ann & Adam Isaiah Green. 2008. “Revisiting the Glass Escalator: The Case of
Gender Segregation in a Female Dominated Occupation.” Social Problems 55(2):271–299.

Sobel, Michael E. 1987. “Direct and Indirect Effects in Linear Structural Equation Models.”
Sociological Methods & Research 16(1):155–176.

Socialstyrelsen. 2010. Social rapport 2010. Stockholm: Socialstyrelsen.

Staff, Jeremy & Jeylan T. Mortimer. 2012. “Explaining the Motherhood Wage Penalty During
the Early Occupational Career.” Demography 49(1):1–21.

Stansfeld, Stephen A., Charlotte Clark, Bryan Rodgers, Tanya Caldwell & Chris Power. 2011.
“Repeated exposure to socioeconomic disadvantage and health selection as life course
pathways to mid-life depressive and anxiety disorders.” Social Psychiatry and Psychiatric
Epidemiology 46(7):549–558.

Stefansdottir, Ingibjorg & Runar Vilhjalmsson. 2007. “Dimensions of health-related lifestyle in

young adult-hood: results from a national population survey.” Scandinavian Journal of
Caring Sciences 21:321–328.

Stern, Jon. 1983. “Social Mobility and the Interpretation of Social Class Mortality Differentials.”
Journal of Social Policy 12(01):27–49.

Stevens, Cynthia K., Anna G. Bavetta & Marilyn E. Gist. 1993. “Gender differences in the
acquisition of salary negotiation skills: The role of goals, self-efficacy, and perceived
control.” Journal of Applied Psychology 78(5):723–735.

Stewart, Jennifer M. 2001. “The impact of health status on the duration of unemployment
spells and the implications for studies of the impact of unemployment on health status.”
Journal of Health Economics 20(5):781–796.

Stiglitz, Joseph E, Amartya Sen & Jean-Paul Fitoussi. 2009. “Report by the commission on
the measurement of economic performance and social progress.”.

208
 Bibliography

Stone, Pamela & Meg Lovejoy. 2004. “Fast-Track Women and the “Choice” to Stay Home.”
The ANNALS of the American Academy of Political and Social Science 596(1):62–83.

Stowasser, Till, Florian Heiss, Daniel McFadden & Joachim Winter. 2011. “’Healthy, Wealthy
and Wise?’ Revisited: An Analysis of the Causal Pathways from Socio-economic Status
to Health.” NBER Working Paper Series (17273).

Stronks, K., H. van de Mheen, J. van den Bos & J. P. Mackenbach. 1997. “The interrelationship
between income, health and employment status.” International Journal of Epidemiology
26(3):592–600.

Stuhlmacher, Alice F. & Amy E. Walters. 1999. “Gender Differences in Negotiation Outcome:
A Meta-Analysis.” Personnel Psychology 52(3):653–677.

Subramanian, S. V. & Karen Ertel. 2008. “Is the Use of Self-Rated Health Measures to Assess
Health Inequalities Misleading?” International Journal of Epidemiology 37(6):1436–1440.

Sørensen, A. B. 1983. “Processes of allocation to open and closed positions in social structure.”
Zeitschrift für Soziologie 12(3):203–224.

Sørensen, Aage B. 1996. “The Structural Basis of Social Inequality.” American Journal of
Sociology 101(5):1333–1365.

Tavory, Iddo & Nina Eliasoph. 2013. “Coordinating Futures: Toward a Theory of Anticipation.”
American Journal of Sociology 118(4):908–942.

Thrane, Christer. 2006. “Explaining educational-related inequalities in health: Mediation and

moderator models.” Social Science & Medicine 62(2):467–478.

Thurow, Lester. 1975. Generating inequality: mechanisms of distribution in the U.S. economy.
New York: Basic Books.

Tiemann, Michael. 2010. “Wissensintensive Berufe.” Wissenschaftliche Diskussionspapiere des

Bundesinstituts für Berufsbildung, Heft (114).

Timms, Duncan. 1998. “Gender, social mobility and psychiatric diagnoses.” Social Science &
Medicine 46(9):1235–1247.

Townsend, Peter & Nick Davidson. 1982. Inequalities in Health: The Black Report - The
most important critique of the health service and general health standards written in this
country since the war. Middlessex: Penguin Books Ltd.

Tunistra, Jolanda, W. Johan Groothoff, J. A. Wim van den Heuvel & Doeke Post. 1998.
“Socio-Economic Differences in Health Risk Behavior in Adolescence: Do they exist?”
Social Science & Medicine 47(1):67–74.

209
 Bibliography

Unger, Rainer. 2008. Gesundheit im Lebenslauf Zur relativen Bedeutung von Selek-
tions gegenüber Kausaleffekten am Beispiel des Familienstands. In Health Inequalities:
Determinanten und Mechanismen gesundheitlicher Ungleichheit (Gesundheit und
Gesellschaft), ed. Ullrich Bauer, Uwe H. Bittlingmayer & Matthias Richter. 2008 ed.
VS Verlag für Sozialwissenschaften pp. 430–454.

Vaillant, Nicolas & François-Charles Wolff. 2012. “On the reliability of self-reported health:
Evidence from Albanian data.” Journal of Epidemiology and Global Health 2(2):83–98.

van De Mheen, H. Dike, Karien Stronks & Johann P. Mackenbach. 1998. “A Lifecourse Perspec-
tive on Socio-economic Inequalities in Health: The Influence of Childhood Socio-economic
Conditions and Selection Processes.” Sociology of Health & Illness 20(5):754–777.

van de Mheen, H., K. Stronks, C. W. N Looman & P. J Mackenbach. 1998. “Does childhood
sodoeconomic status influence adult health through behavioural factors?” International
Journal of Epidemiology 27:431–437.

Virtanen, P., J. Vahtera, M. Kivimäki, J. Pentti & J. Ferrie. 2002. “Employment security and
health.” Journal of epidemiology and community health 56(8):569.

Virtanen, P., U. Janlert & A. Hammarström. 2013. “Health status and health behaviour as
predictors of the occurrence of unemployment and prolonged unemployment.” Public
Health 127(1):46–52.

Virtanen, Pekka, Urban Janlert & Anne Hammarström. 2012. “Suboptimal health as a
predictor of non-permanent employment in middle age: a 12-year follow-up study of
the Northern Swedish Cohort.” International Archives of Occupational and Environmental
Health pp. 1–7.

Viscusi, W. Kip. 1978. “Wealth Effects and Earnings Premiums for Job Hazards.” The Review
of Economics and Statistics 60(3):408–416.

Viscusi, W. Kip & Michael J. Moore. 1987. “Workers’ Compensation: Wage Effects, Benefit
Inadequacies, and the Value of Health Losses.” The Review of Economics and Statistics
69(2):249–261.

Vonneilich, Nico, Karl-Heinz Jöckel, Raimund Erbel, Jens Klein, Nico Dragano, Johannes
Siegrist & Olaf von dem Knesebeck. 2012. “The mediating effect of social relationships
on the association between socioeconomic status and subjective health – results from the
Heinz Nixdorf Recall cohort study.” BMC Public Health 12(1):1–11.

Voß, G.G. & H.J. Pongratz. 1998. “Der Arbeitskraftunternehmer.” Kölner Zeitschrift für
Soziologie und Sozialpsychologie 50(1):131–158.

210
 Bibliography

Wagenaar, Alfred F., Michiel A. J. Kompier, Irene L. D. Houtman, Seth N. J. van den Bossche
& Toon W. Taris. 2012. “Employment Contracts and Health Selection.” Journal of
Occupational and Environmental Medicine 54(10):1192–1200.

Wagner, G. Gert, R. Joachim Frick & Jürgen Schupp. 2007. “The German Socio-Economic
Panel Study (SOEP) - Scope, Evolutions and Enhancements.” Schmollers Jahrbuch
127(1):139–169.

Wahrendorf, Morten, David Blane, Mel Bartley, Nico Dragano & Johannes Siegrist. 2013.
“Working conditions in mid-life and mental health in older ages.” Advances in Life Course
Research 18(1):16–25.

Ward, Andrew. 2009. “Causal criteria and the problem of complex causation.” Medicine,
Health Care and Philosophy 12(3):333–343.

Warren, John Robert. 2009. “Socioeconomic Status and Health across the Life Course: A Test
of the Social Causation and Health Selection Hypotheses.” Social Forces 87(4):2125–2153.

Warren, John Robert, Peter Hoonakker, Pascale Carayon & Jennie Brand. 2004. “Job
characteristics as mediators in SES–health relationships.” Social Science & Medicine
59(7):1367–1378.

Watts, Jacqueline H. 2009. “Leaders of men: women ‘managing’ in construction.” Work,

Employment & Society 23(3):512–530.

Weber, Max. 1976. Wirtschaft und Gesellschaft: Grundriss der verstehenden Soziologie. Tübin-
gen: J.C.B. Mohr (Paul Siebeck).

Weeden, Kim A. 2002. “Why Do Some Occupations Pay More than Others? Social Closure and
Earnings Inequality in the United States.” The American Journal of Sociology 108(1):55–
101.

Weichselbaumer, Doris & Rudolf Winter-Ebmer. 2005. “A Meta-Analysis of the International

Gender Wage Gap.” Journal of Economic Surveys 19(3):479–511.

Wendt, Claus & Christof Wolf, eds. 2006. Soziologie der Gesundheit. VS Verlag für Sozialwis-
senschaften.

West, Patrick. 1991. “Rethinking the health selection explanation for health inequalities.”
Social Science & Medicine 32(4):373–384.

White, Harrison C. 1970. Chains of opportunity: system models of mobility in organizations.

Harvard University Press.

211
 Bibliography

WHO. 1946. Preamble to the Constitution of the World Health Organization as adopted by
the International Health Conference. World Health Organization.

Wickrama, K. A. S., Rand D. Conger, Lora Ebert Wallace & Glen H. Elder. 1999. “The
Intergenerational Transmission of Health-Risk Behaviors: Adolescent Lifestyles and Gender
Moderating Effects.” Journal of Health and Social Behavior 40(3):258–272.

Wilkinson, Richard & Kate Pickett. 2009. The spirit level: Why more equal societies almost
always do better. London: Allen Lane.

Williams, Christine L. 1992. “The Glass Escalator: Hidden Advantages for Men in the ”Female”
Professions.” Social Problems 39(3):253–267.

Williamson, Oliver E. 1973. “Markets and Hierarchies: Some Elementary Considerations.” The
American Economic Review 63(2):316–325.

Williamson, Oliver E. 1981. “The Economics of Organization: The Transaction Cost Approach.”
American Journal of Sociology 87(3):548–577.

Williamson, Oliver E., Michael L. Wachter & Jeffrey E. Harris. 1975. “Understanding the
Employment Relation: The Analysis of Idiosyncratic Exchange.” The Bell Journal of
Economics 6(1):250–278.

Wilz, Sylvia M. 2002. Organisation und Geschlecht: Strukturelle Bindungen und kontingente
Kopplungen. 2002 ed. VS Verlag für Sozialwissenschaften.

Windolf, Paul. 2008. “Eigentümer ohne Risiko - Die Dienstklasse des Finanzmarkt-Kapitalismus.”
Zeitschrift für Soziologie 37(6):516–535.

Winship, Christopher & Michael Sobel. 2001. “Causal Inference in Sociological Studies.”
Havard University Working Paper .

Wooldridge, Jeffrey M. 2009. Introductory Econometrics: A Modern Approach. Cengage

Learning.

Wooldridge, M. Jeffrey. 2002. Econometric Analysis fo Cross-Section and Panel Data. Cam-
bridge: MIT Press.

Wright, Sewall. 1934. “The method of path coefficients.” The Annals of Mathematical
Statistics 5(3):161–215.

Zambon, Alessio, Will Boyce, Ester Cois, Candace Currie, Patrizia Lemma, Paola Dalmasso,
Alberto Borraccino & Franco Cavallo. 2006. “Do welfare regimes mediate the effect of
socioeconomic position on health in adolescence? A cross-national comparison in Europe,
North America, and Israel.” International Journal of Health Services 36(2):309–329.

212
 Bibliography

Zok, K. 2008. Krank zur Arbeit: Einstellungen und Verhalten von Frauen und Männern beim
Umgang mit Krankheit am Arbeitsplatz. In Fehlzeiten-Report 2007, ed. Bernhard Badura,
Helmut Schröder & Christian Vetter. Berlin, Heidelberg: Springer Berlin Heidelberg
pp. 121–144.

213
A. Appendix

A.1. Results Tables

Table A.1.: The Effects of Subjective Health on Job Status

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public

Subjective Health 0.198 -0.005 0.005 0.100

(0.099) (0.118) (0.071) (0.170)
Age 0.440 0.555 0.371 0.685
(0.107) (0.146) (0.107) (0.219)
Age2 -0.006 -0.006 -0.005 -0.005
(0.001) (0.001) (0.001) (0.002)
East Germany 0.591 0.340 -1.113 0.968
(0.754) (0.906) (0.498) (0.807)
Years of Education 0.025 0.019 0.178 0.254
(0.111) (0.118) (0.078) (0.168)
Years of FT experience 0.031 -0.058 0.004 -0.256
(0.043) (0.046) (0.087) (0.121)
log. Wage per Hour 0.139 0.256 0.481 0.259
(0.181) (0.256) (0.148) (0.294)
Job tenure 0.085 0.007 0.031 -0.071
(0.021) (0.029) (0.010) (0.070)
Hours of work 0.030 0.034 0.015 0.012
(0.009) (0.012) (0.008) (0.020)
Hours of overtime 0.046 0.039 0.010 -0.025
(0.016) (0.022) (0.009) (0.025)
Size of HH -0.136 0.171 -0.133 -0.174
(0.129) (0.164) (0.088) (0.215)
Number of children in HH -0.320 -0.141 -0.001 0.116
(0.166) (0.190) (0.084) (0.175)
Marital status: Single -0.053 1.573 -0.211 -0.019
(0.298) (0.606) (0.224) (0.455)
Marital status: Other -0.178 -0.366 -0.315 -0.306
(0.284) (0.410) (0.223) (0.554)
Amount of housework+childcare -0.081 -0.017 -0.059 0.022
(0.028) (0.034) (0.027) (0.055)
Age of youngest HH-member -0.009 0.016 -0.017 -0.026
(0.011) (0.012) (0.007) (0.016)
Firmsize:< 5 0.241 -21.940 0.352 -0.269
(0.296) (22389.620) (0.245) (1.192)
Firmsize: 20-199 0.202 0.214 -0.104 -0.101
(0.223) (0.317) (0.158) (0.564)
Firmsize: 200-2.000 0.296 0.570 -0.310 -0.220
(0.264) (0.352) (0.183) (0.582)
Firmsize:> 2.000 0.570 0.860 -0.356 -0.391
(0.287) (0.359) (0.200) (0.623)
Branch: Agriculture and fishing 2.103 -1.229 -1.197
(1.390) (0.809) (2.361)
Branch: Energy and water supply -0.603 0.161 0.300
(0.969) (0.455) (2.062)
Branch: Construction 0.581 0.198 0.255
(0.744) (0.230) (2.278)
Branch: Wholesale, hotel and restaurant industry -0.035 22.969 0.193 1.333
(0.309) (.) (0.171) (1.466)
Branch: Transportation and information -0.666 21.447 0.008 0.910
(0.577) (54643.324) (0.314) (2.225)
Branch: Financial intermediate -0.135 1.374 -0.566 -0.565
(0.579) (34259.349) (0.424) (2.214)
Branch: Real estate, law counseling 0.288 -0.502 -0.267 0.508
(0.357) (34259.349) (0.219) (2.174)
Branch: Public Administration 0.552 0.444 0.169 0.050

i
 A. Appendix

Table A.1 – Continued from previous page

(0.393) (34259.349) (0.424) (2.065)

Branch: Public and private Services 0.057 -0.021 0.181 0.536
(0.391) (34259.349) (0.258) (2.048)
Psychological strain of occupation 0.053 -0.191 0.042 -0.102
(0.036) (0.062) (0.024) (0.065)
Physical strain of occupation -0.107 0.174 -0.085 0.046
(0.056) (0.100) (0.026) (0.071)
Great Worries about job security -0.312 -0.145 -0.214 -0.194
(0.174) (0.220) (0.115) (0.321)
Great Worries about own economic situation 0.394 0.169 0.052 0.106
(0.154) (0.181) (0.109) (0.263)
Satisfaction with work 0.079 -0.047 0.076 0.100
(0.031) (0.042) (0.022) (0.054)
Observations 3291 2098 6564 1351
Individuals 511 319 933 208
Note: Standard errors in parentheses.

Table A.2.: The Effects of log. Days of Sickness Absence on Job Status

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public

log. Days of sickness absence -0.034 0.024 -0.069 0.009

(0.044) (0.053) (0.022) (0.054)
Age 0.411 0.298 0.398 0.522
(0.109) (0.146) (0.065) (0.144)
Age2 -0.005 -0.004 -0.005 -0.004
(0.001) (0.001) (0.000) (0.001)
East Germany 0.436 0.232 -0.406 0.372
(0.798) (0.896) (0.372) (0.699)
Years of Education 0.010 0.050 0.136 0.185
(0.114) (0.117) (0.048) (0.089)
Years of FT experience 0.044 -0.053 0.029 -0.129
(0.043) (0.046) (0.049) (0.082)
log. Wage per Hour 0.083 0.312 0.532 0.193
(0.185) (0.263) (0.102) (0.253)
Job tenure 0.082 0.006 0.034 -0.009
(0.022) (0.030) (0.007) (0.037)
Hours of work 0.030 0.035 0.021 0.015
(0.010) (0.012) (0.006) (0.017)
Hours of overtime 0.045 0.045 0.022 0.002
(0.016) (0.023) (0.007) (0.019)
Size of HH -0.188 0.236 -0.131 -0.087
(0.133) (0.166) (0.059) (0.141)
Number of children in HH -0.328 -0.158 0.112 0.009
(0.169) (0.191) (0.056) (0.130)
Marital status: Single -0.043 1.351 -0.119 -0.064
(0.316) (0.597) (0.153) (0.340)
Marital status: Other -0.436 -0.299 -0.469 -0.494
(0.300) (0.418) (0.153) (0.354)
Amount of housework+childcare -0.083 -0.005 -0.021 -0.004
(0.029) (0.035) (0.022) (0.048)
Age of youngest HH-member -0.012 0.016 -0.011 -0.022
(0.011) (0.012) (0.005) (0.010)
Firmsize:< 5 0.200 -33.911 0.220 -0.930
(0.310) (9109915.841) (0.188) (0.890)
Firmsize: 20-199 0.115 0.292 -0.198 -0.080
(0.229) (0.330) (0.114) (0.421)
Firmsize: 200-2.000 0.265 0.689 -0.392 -0.162
(0.270) (0.362) (0.135) (0.436)
Firmsize:> 2.000 0.489 0.927 -0.479 -0.161
(0.291) (0.367) (0.145) (0.456)
Branch: Agriculture and fishing 2.640 -0.244 -1.231
(1.401) (0.573) (1.420)
Branch: Energy and water supply -0.986 0.285 0.120
(1.087) (0.352) (0.883)
Branch: Construction 1.133 0.377 -1.184
(0.815) (0.160) (1.157)
Branch: Wholesale, hotel and restaurant industry -0.128 34.614 0.230 1.255
(0.318) (13144367.903) (0.122) (1.310)
Branch: Transportation and information -0.817 32.799 -0.187 0.588
(0.598) (17770789.131) (0.252) (0.897)
Branch: Financial intermediate -0.105 1.285 -0.354 -0.125

ii
 A. Appendix

Table A.2 – Continued from previous page

(0.590) (.) (0.312) (1.058)

Branch: Real estate, law counseling 0.281 -0.448 -0.323 -0.083
(0.370) (1.288) (0.164) (0.962)
Branch: Public Administration 0.476 0.388 -0.302 0.153
(0.402) (1.178) (0.317) (0.819)
Branch: Public and private Services -0.047 -0.128 0.429 0.441
(0.413) (1.274) (0.210) (0.837)
Psychological strain of occupation 0.042 -0.161 0.055 -0.074
(0.038) (0.063) (0.018) (0.045)
Physical strain of occupation -0.125 0.162 -0.078 -0.015
(0.058) (0.101) (0.020) (0.056)
Great Worries about job security -0.390 -0.099 -0.038 0.185
(0.181) (0.224) (0.091) (0.244)
Great Worries about own economic situation 0.379 0.115 -0.132 0.041
(0.157) (0.187) (0.088) (0.215)
Satisfaction with work 0.091 -0.067 0.097 0.047
(0.031) (0.043) (0.017) (0.042)
Observations 3129 1989 11671 2294
Individuals 493 306 1365 298
Note: Standard errors in parentheses.

Table A.3.: The Effects of Subjective Health on Job Status - Mediated by Knowledge Intensity

(1) (2)
W - Private M - Private

Subjective Health -0.197 -0.040

(0.199) (0.134)
log. Knowledge intensity 0.100 0.437
(0.195) (0.143)
Interaction effect -0.373 -0.057
(0.163) (0.128)
Age 0.443 0.381
(0.107) (0.107)
Age2 -0.006 -0.005
(0.001) (0.001)
East Germany 0.572 -1.093
(0.757) (0.500)
Years of Education 0.020 0.185
(0.111) (0.078)
Years of FT experience 0.031 -0.007
(0.043) (0.088)
log. Wage per Hour 0.143 0.437
(0.182) (0.149)
Job tenure 0.083 0.030
(0.021) (0.010)
Hours of work 0.029 0.014
(0.009) (0.008)
Hours of overtime 0.045 0.010
(0.016) (0.009)
Size of HH -0.144 -0.126
(0.129) (0.088)
Number of children in HH -0.330 -0.031
(0.166) (0.086)
Marital status: Single -0.054 -0.220
(0.298) (0.224)
Marital status: Other -0.123 -0.327
(0.283) (0.223)
Amount of housework+childcare -0.086 -0.058
(0.028) (0.027)
Age of youngest HH-member -0.010 -0.018
(0.011) (0.007)
Firmsize:< 5 0.244 0.370
(0.298) (0.245)
Firmsize: 20-199 0.212 -0.070
(0.225) (0.159)
Firmsize: 200-2.000 0.339 -0.270
(0.264) (0.184)
Firmsize:> 2.000 0.632 -0.306
(0.289) (0.201)
Branch: Agriculture and fishing 14.821 -1.549
(721.234) (0.997)
Branch: Energy and water supply -0.532 0.195

iii
 A. Appendix

Table A.3 – Continued from previous page

(0.967) (0.455)
Branch: Construction 0.629 0.205
(0.754) (0.230)
Branch: Wholesale, hotel and restaurant industry -0.022 0.209
(0.312) (0.171)
Branch: Transportation and information -0.626 0.041
(0.578) (0.315)
Branch: Financial intermediate -0.093 -0.582
(0.583) (0.424)
Branch: Real estate, law counseling 0.325 -0.264
(0.360) (0.220)
Branch: Public Administration 0.615 0.180
(0.401) (0.422)
Branch: Public and private Services 0.113 0.171
(0.394) (0.258)
Psychological strain of occupation 0.052 0.025
(0.038) (0.025)
Physical strain of occupation -0.089 -0.042
(0.061) (0.030)
Great Worries about job security -0.318 -0.234
(0.174) (0.116)
Great Worries about own economic situation 0.394 0.056
(0.155) (0.110)
Satisfaction with work 0.079 0.078
(0.031) (0.023)
Observations 3288 6537
Individuals 510 931
Note: Standard errors in parentheses.

Table A.4.: The Effects of log. Days of Sickness Absence on Job Status - Mediated by Knowledge Intensity

(1) (2)
W - Private M - Private

log. Days of sickness absence -0.129 -0.052

(0.079) (0.049)
log. Knowledge intensity 0.503 0.430
(0.172) (0.114)
Interaction effect -0.098 0.016
(0.064) (0.045)
Age 0.203 0.405
(0.072) (0.066)
Age2 -0.003 -0.005
(0.001) (0.000)
East Germany 1.587 -0.397
(0.619) (0.372)
Years of Education 0.106 0.141
(0.070) (0.048)
Years of FT experience 0.074 0.028
(0.026) (0.050)
log. Wage per Hour 0.469 0.507
(0.147) (0.103)
Job tenure 0.051 0.032
(0.014) (0.007)
Hours of work 0.043 0.020
(0.007) (0.006)
Hours of overtime 0.055 0.022
(0.012) (0.007)
Size of HH -0.200 -0.117
(0.098) (0.060)
Number of children in HH -0.009 0.078
(0.114) (0.057)
Marital status: Single 0.125 -0.140
(0.224) (0.154)
Marital status: Other -0.306 -0.467
(0.232) (0.154)
Amount of housework+childcare -0.067 -0.023
(0.023) (0.022)
Age of youngest HH-member -0.013 -0.011
(0.008) (0.005)
Firmsize:< 5 -0.000 0.232
(0.225) (0.190)
Firmsize: 20-199 0.046 -0.140

iv
 A. Appendix

Table A.4 – Continued from previous page

(0.168) (0.116)
Firmsize: 200-2.000 -0.010 -0.334
(0.202) (0.136)
Firmsize:> 2.000 0.285 -0.411
(0.216) (0.147)
Branch: Agriculture and fishing 2.716 -0.647
(1.389) (0.840)
Branch: Energy and water supply -0.646 0.309
(0.902) (0.353)
Branch: Construction -0.066 0.373
(0.400) (0.161)
Branch: Wholesale, hotel and restaurant industry 0.280 0.258
(0.221) (0.124)
Branch: Transportation and information -0.510 -0.143
(0.451) (0.253)
Branch: Financial intermediate 0.679 -0.342
(0.428) (0.313)
Branch: Real estate, law counseling 0.441 -0.329
(0.264) (0.165)
Branch: Public Administration 0.503 -0.309
(0.299) (0.317)
Branch: Public and private Services 0.446 0.411
(0.297) (0.210)
Psychological strain of occupation 0.076 0.037
(0.030) (0.019)
Physical strain of occupation -0.143 -0.035
(0.049) (0.022)
Great Worries about job security -0.245 -0.036
(0.151) (0.092)
Great Worries about own economic situation 0.215 -0.133
(0.129) (0.089)
Satisfaction with work 0.100 0.097
(0.025) (0.017)
Observations 5188 11563
Individuals 661 1358
Note: Standard errors in parentheses.

Table A.5.: The Effects of Subjective Health on Job Status in Different Occupations - Women

(1) (2) (3)

Male Dominated Mixed Female Dominated

Subjective Health -0.060 0.047 0.484

(0.349) (0.178) (0.167)
Age 0.444 0.350 0.530
(0.400) (0.204) (0.178)
Age2 -0.011 -0.006 -0.004
(0.005) (0.002) (0.002)
East Germany 12.280 0.734 -14.164
(1780.158) (1.707) (991.360)
Years of Education 0.739 0.106 -0.557
(0.784) (0.193) (0.238)
Years of FT experience 0.066 0.115 -0.004
(0.179) (0.086) (0.073)
log. Wage per Hour 0.121 0.302 0.025
(0.554) (0.392) (0.271)
Job tenure 0.088 0.073 0.073
(0.071) (0.058) (0.035)
Hours of work 0.043 0.037 0.021
(0.033) (0.021) (0.015)
Hours of overtime -0.015 0.086 0.034
(0.061) (0.026) (0.029)
Size of HH 1.206 -0.349 -0.315
(0.635) (0.270) (0.220)
Number of children in HH -0.316 -0.452 -0.150
(0.720) (0.282) (0.293)
Marital status: Single -0.455 -0.462 -0.175
(1.019) (0.488) (0.582)
Marital status: Other -0.035 -0.724 0.710
(1.013) (0.529) (0.564)
Amount of housework+childcare -0.032 -0.085 -0.084
(0.108) (0.049) (0.046)
Age of youngest HH-member 0.163 -0.024 -0.025

v
 A. Appendix

Table A.5 – Continued from previous page

(0.056) (0.017) (0.020)

Firmsize:< 5 4.122 -0.713 0.149
(1.751) (1.135) (0.441)
Firmsize: 20-199 2.648 -0.203 0.369
(1.307) (0.465) (0.373)
Firmsize: 200-2.000 1.488 0.351 1.116
(1.786) (0.527) (0.531)
Firmsize:> 2.000 0.859 0.811 1.116
(1.966) (0.557) (0.588)
Branch: Construction 2.507 14.811
(1.900) (2342.390)
Branch: Wholesale, hotel and restaurant industry -2.486 -0.132 0.754
(1.263) (0.611) (0.771)
Branch: Transportation and information 2.565 0.520 0.409
(5.706) (1.647) (1.547)
Branch: Financial intermediate 10.415 -1.869
(1209.297) (1.342)
Branch: Real estate, law counseling 0.554 -0.038 1.306
(1.611) (0.809) (0.855)
Branch: Public Administration -0.193 0.721 2.180
(1.617) (0.898) (0.990)
Branch: Public and private Services 3.792 -1.573 1.451
(2.027) (0.871) (0.897)
Psychological strain of occupation -0.304 0.004 0.051
(0.234) (0.093) (0.121)
Physical strain of occupation 0.054 -0.018 -0.018
(0.232) (0.152) (0.166)
Great Worries about job security -0.655 -0.348 0.180
(0.561) (0.290) (0.324)
Great Worries about own economic situation 0.597 0.079 0.458
(0.504) (0.281) (0.241)
Satisfaction with work 0.293 0.022 0.097
(0.117) (0.052) (0.055)
Observations 412 1102 1213
Individuals 96 204 226
Note: Standard errors in parentheses.

Table A.6.: The Effects of Subjective Health on Job Status in Different Occupations - Men

(1) (2) (3)

Male Dominated Mixed Female Dominated

Subjective Health 0.074 0.093 -0.133

(0.096) (0.150) (0.312)
Age 0.375 0.529 1.542
(0.160) (0.219) (0.682)
Age2 -0.004 -0.007 -0.006
(0.001) (0.001) (0.003)
East Germany -0.556 -1.719 -14.729
(0.766) (1.168) (2638.151)
Years of Education 0.151 0.166 0.337
(0.120) (0.155) (0.390)
Years of FT experience -0.017 -0.007 -0.900
(0.137) (0.182) (0.577)
log. Wage per Hour 0.148 0.831 0.479
(0.214) (0.299) (0.651)
Job tenure 0.042 0.030 -0.049
(0.016) (0.022) (0.041)
Hours of work 0.009 0.019 0.013
(0.011) (0.015) (0.035)
Hours of overtime -0.005 0.006 -0.055
(0.012) (0.018) (0.040)
Size of HH 0.049 -0.370 0.040
(0.122) (0.186) (0.459)
Number of children in HH 0.038 -0.625 0.992
(0.113) (0.196) (0.470)
Marital status: Single -0.391 -0.511 1.108
(0.334) (0.404) (0.820)
Marital status: Other -0.733 0.497 -0.095
(0.300) (0.553) (0.878)
Amount of housework+childcare -0.067 -0.052 0.279
(0.036) (0.056) (0.117)
Age of youngest HH-member -0.014 -0.038 -0.028

vi
 A. Appendix

Table A.6 – Continued from previous page

(0.009) (0.014) (0.029)

Firmsize:< 5 0.612 -0.125 0.009
(0.320) (0.671) (1.138)
Firmsize: 20-199 -0.208 0.020 0.277
(0.219) (0.376) (0.738)
Firmsize: 200-2.000 -0.335 -0.394 -0.806
(0.254) (0.422) (0.988)
Firmsize:> 2.000 -0.195 -0.479 -0.152
(0.290) (0.454) (1.260)
Branch: Energy and water supply -0.159 0.702 17.194
(0.528) (1.620) (955.512)
Branch: Construction 0.353 -1.388
(0.275) (0.925)
Branch: Wholesale, hotel and restaurant industry 0.298 0.088 3.270
(0.240) (0.434) (1.235)
Branch: Transportation and information -0.436 0.867 16.887
(0.467) (0.703) (1522.600)
Branch: Financial intermediate -0.650 -1.055 17.351
(0.836) (0.819) (1188.875)
Branch: Real estate, law counseling -0.112 -0.171 3.372
(0.285) (0.705) (1.613)
Branch: Public Administration 0.739 -0.780
(0.566) (0.924)
Branch: Public and private Services 0.694 -0.555 0.815
(0.369) (0.656) (1.646)
Psychological strain of occupation -0.026 0.168 -0.211
(0.038) (0.069) (0.241)
Physical strain of occupation -0.076 -0.299 0.320
(0.033) (0.096) (0.337)
Great Worries about job security -0.200 -0.206 -0.783
(0.156) (0.237) (0.435)
Great Worries about own economic situation 0.082 0.211 -0.230
(0.147) (0.224) (0.444)
Satisfaction with work 0.055 0.108 0.210
(0.030) (0.047) (0.105)
Observations 3561 1686 465
Individuals 551 291 93
Note: Standard errors in parentheses.

Table A.7.: The Effects of log. Days of Sickness Absence on Job Status in Different Occupations - Women

(1) (2) (3)

Male Dominated Mixed Female Dominated

log. Days of sickness absence -0.260 0.056 -0.123

(0.154) (0.078) (0.071)
Age 0.389 0.373 0.456
(0.444) (0.210) (0.180)
Age2 -0.009 -0.006 -0.004
(0.005) (0.002) (0.002)
East Germany 14.532 0.637 -12.963
(2950.242) (1.710) (673.092)
Years of Education 0.613 0.082 -0.668
(0.801) (0.194) (0.259)
Years of FT experience 0.008 0.099 0.006
(0.182) (0.088) (0.074)
log. Wage per Hour 0.096 0.260 0.048
(0.585) (0.400) (0.278)
Job tenure 0.072 0.078 0.061
(0.080) (0.061) (0.039)
Hours of work 0.045 0.037 0.028
(0.036) (0.021) (0.016)
Hours of overtime 0.002 0.091 0.030
(0.063) (0.027) (0.030)
Size of HH 1.181 -0.449 -0.367
(0.650) (0.286) (0.223)
Number of children in HH 0.043 -0.466 -0.136
(0.738) (0.285) (0.300)
Marital status: Single -0.104 -0.373 -0.638
(1.139) (0.511) (0.644)
Marital status: Other -0.087 -1.091 0.056
(1.083) (0.568) (0.611)
Amount of housework+childcare -0.092 -0.057 -0.108

vii
 A. Appendix

Table A.7 – Continued from previous page

(0.128) (0.052) (0.048)

Age of youngest HH-member 0.165 -0.028 -0.025
(0.057) (0.018) (0.021)
Firmsize:< 5 4.342 -0.581 0.112
(2.173) (1.145) (0.454)
Firmsize: 20-199 2.364 -0.284 0.243
(1.317) (0.471) (0.384)
Firmsize: 200-2.000 1.335 0.388 0.991
(1.715) (0.541) (0.536)
Firmsize:> 2.000 0.644 0.877 1.013
(1.854) (0.568) (0.593)
Branch: Construction 15.704 13.620
(893.896) (1299.414)
Branch: Wholesale, hotel and restaurant industry -2.427 -0.193 0.384
(1.232) (0.622) (0.797)
Branch: Transportation and information 2.186 0.007 -0.167
(5.182) (1.681) (1.625)
Branch: Financial intermediate 11.729 -2.076
(2356.387) (1.332)
Branch: Real estate, law counseling 0.673 -0.274 1.206
(1.655) (0.843) (0.898)
Branch: Public Administration -0.163 0.718 1.746
(1.695) (0.918) (1.000)
Branch: Public and private Services 2.109 -1.489 1.052
(2.183) (0.880) (0.912)
Psychological strain of occupation -0.349 0.053 0.052
(0.260) (0.096) (0.126)
Physical strain of occupation 0.164 -0.049 -0.079
(0.239) (0.156) (0.174)
Great Worries about job security -0.519 -0.579 0.187
(0.589) (0.311) (0.331)
Great Worries about own economic situation 0.575 0.117 0.382
(0.521) (0.289) (0.245)
Satisfaction with work 0.253 0.024 0.113
(0.115) (0.051) (0.055)
Observations 387 1042 1163
Individuals 91 193 218
Note: Standard errors in parentheses.

Table A.8.: The Effects of log. Days of Sickness Absence on Job Status in Different Occupations - Men

(1) (2) (3)

Male Dominated Mixed Female Dominated

log. Days of sickness absence -0.091 -0.063 -0.174

(0.033) (0.058) (0.118)
Age 0.479 0.533 0.236
(0.129) (0.176) (0.390)
Age2 -0.004 -0.007 -0.003
(0.001) (0.001) (0.003)
East Germany -0.054 -1.072 -15.479
(0.602) (0.873) (2343.506)
Years of Education 0.110 0.144 0.102
(0.082) (0.109) (0.232)
Years of FT experience -0.103 -0.002 0.072
(0.110) (0.143) (0.297)
log. Wage per Hour 0.260 0.920 0.150
(0.183) (0.274) (0.554)
Job tenure 0.047 0.018 -0.039
(0.013) (0.020) (0.035)
Hours of work 0.015 0.014 -0.012
(0.010) (0.014) (0.027)
Hours of overtime 0.005 0.014 -0.002
(0.010) (0.016) (0.033)
Size of HH 0.067 -0.348 0.526
(0.101) (0.153) (0.382)
Number of children in HH 0.066 -0.362 0.257
(0.092) (0.165) (0.358)
Marital status: Single -0.267 -0.310 0.970
(0.262) (0.362) (0.667)
Marital status: Other -0.393 -0.275 -0.525
(0.253) (0.427) (0.843)
Amount of housework+childcare -0.041 -0.071 0.245

viii
 A. Appendix

Table A.8 – Continued from previous page

(0.033) (0.055) (0.118)

Age of youngest HH-member -0.011 -0.033 0.031
(0.008) (0.012) (0.024)
Firmsize:< 5 0.304 0.028 2.205
(0.269) (0.613) (1.148)
Firmsize: 20-199 -0.119 -0.158 -0.320
(0.180) (0.349) (0.682)
Firmsize: 200-2.000 -0.528 -0.390 -0.492
(0.214) (0.392) (0.908)
Firmsize:> 2.000 -0.412 -0.495 0.225
(0.242) (0.416) (1.113)
Branch: Agriculture and fishing 1.739
(1.515)
Branch: Energy and water supply -0.054 0.784 18.712
(0.469) (1.611) (1156.273)
Branch: Construction 0.467 -0.531 16.527
(0.228) (0.736) (648.139)
Branch: Wholesale, hotel and restaurant industry 0.328 0.099 3.811
(0.207) (0.358) (1.047)
Branch: Transportation and information -0.387 0.841 16.902
(0.426) (0.605) (1465.035)
Branch: Financial intermediate -0.085 -0.948 16.736
(0.713) (0.675) (939.074)
Branch: Real estate, law counseling -0.018 -0.637 2.969
(0.244) (0.563) (1.264)
Branch: Public Administration 0.838 -0.669
(0.587) (0.720)
Branch: Public and private Services 0.662 -0.283 3.918
(0.331) (0.560) (1.363)
Psychological strain of occupation -0.014 0.170 -0.159
(0.033) (0.061) (0.183)
Physical strain of occupation -0.058 -0.289 0.150
(0.029) (0.088) (0.269)
Great Worries about job security -0.157 -0.137 -0.752
(0.137) (0.222) (0.416)
Great Worries about own economic situation -0.014 0.136 -0.134
(0.130) (0.205) (0.414)
Satisfaction with work 0.076 0.106 0.146
(0.026) (0.042) (0.091)
Observations 4701 2039 564
Individuals 664 330 115
Note: Standard errors in parentheses.

Table A.9.: The Effects of Long-Term Poor Health on Job Status

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public

1 year of poor health -0.031 -0.020 -0.055 0.035

(0.113) (0.137) (0.077) (0.183)
2-3 years of poor health -0.243 -0.070 0.045 0.054
(0.164) (0.191) (0.107) (0.249)
4+ years poor health -0.453 0.050 -0.244 -0.267
(0.163) (0.188) (0.110) (0.251)
Age 0.313 0.256 0.457 0.566
(0.079) (0.103) (0.074) (0.157)
Age2 -0.004 -0.003 -0.005 -0.005
(0.001) (0.001) (0.000) (0.001)
East Germany 1.271 0.374 -0.481 0.498
(0.575) (0.746) (0.358) (0.717)
Years of Education 0.131 0.163 0.112 0.151
(0.075) (0.089) (0.051) (0.091)
Years of FT experience 0.045 -0.021 -0.028 -0.146
(0.031) (0.032) (0.056) (0.092)
log. Wage per Hour 0.303 0.307 0.617 0.116
(0.153) (0.214) (0.118) (0.254)
Job tenure 0.047 -0.007 0.037 -0.036
(0.015) (0.024) (0.008) (0.042)
Hours of work 0.043 0.035 0.019 0.017
(0.008) (0.010) (0.006) (0.017)
Hours of overtime 0.056 0.049 0.021 -0.013
(0.013) (0.019) (0.007) (0.020)
Size of HH -0.105 -0.056 -0.120 -0.303

ix
 A. Appendix

Table A.9 – Continued from previous page

(0.103) (0.128) (0.067) (0.161)

Number of children in HH -0.056 -0.201 0.108 -0.043
(0.120) (0.126) (0.062) (0.140)
Marital status: Single -0.010 0.230 0.077 -0.265
(0.236) (0.407) (0.168) (0.376)
Marital status: Other -0.251 -0.281 -0.445 -0.999
(0.237) (0.317) (0.171) (0.390)
Amount of housework+childcare -0.062 0.020 -0.035 0.005
(0.023) (0.029) (0.023) (0.050)
Age of youngest HH-member -0.006 0.004 -0.009 -0.036
(0.008) (0.009) (0.005) (0.011)
Firmsize:< 5 0.146 -1.339 0.252 -0.220
(0.235) (0.796) (0.196) (0.903)
Firmsize: 20-199 0.135 0.114 -0.193 -0.168
(0.176) (0.243) (0.122) (0.438)
Firmsize: 200-2.000 0.063 0.022 -0.407 0.078
(0.210) (0.265) (0.144) (0.457)
Firmsize:> 2.000 0.375 0.416 -0.495 -0.231
(0.227) (0.268) (0.157) (0.483)
Branch: Agriculture and fishing 2.219 -15.131 -0.416 -1.200
(1.291) (841.470) (0.647) (1.512)
Branch: Energy and water supply -0.292 -14.956 0.187 0.103
(0.844) (841.471) (0.369) (1.056)
Branch: Construction 0.129 -3.619 0.400 -0.790
(0.436) (1130.986) (0.174) (1.296)
Branch: Wholesale, hotel and restaurant industry 0.343 -3.439 0.316 1.070
(0.239) (1119.301) (0.130) (1.316)
Branch: Transportation and information -0.509 -2.848 -0.032 0.394
(0.467) (1139.743) (0.260) (1.169)
Branch: Financial intermediate 0.762 -16.173 -0.177 -0.242
(0.465) (841.470) (0.332) (1.225)
Branch: Real estate, law counseling 0.338 -17.859 -0.106 0.255
(0.277) (841.470) (0.173) (1.117)
Branch: Public Administration 0.714 -16.704 -0.098 0.071
(0.309) (841.470) (0.327) (0.985)
Branch: Public and private Services 0.480 -16.811 0.486 0.315
(0.303) (841.470) (0.209) (0.992)
Psychological strain of occupation 0.110 -0.099 0.045 -0.119
(0.030) (0.050) (0.019) (0.048)
Physical strain of occupation -0.180 0.102 -0.077 0.016
(0.046) (0.077) (0.021) (0.058)
Great Worries about job security -0.293 -0.120 -0.083 0.117
(0.154) (0.180) (0.095) (0.254)
Great Worries about own economic situation 0.291 0.104 -0.040 0.077
(0.132) (0.153) (0.090) (0.216)
Satisfaction with work 0.091 0.001 0.092 0.073
(0.026) (0.034) (0.018) (0.044)
Observations 4708 3150 9725 2118
Individuals 625 411 1208 281
Note: Standard errors in parentheses.

Table A.10.: The Effects of Subjective Health on Job Status - Applicants

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
Subjective Health 0.233 0.068 -0.065 0.169
(0.103) (0.054) (0.081) (0.088)
Age 0.002 -0.130 -0.040 -0.089
(0.057) (0.029) (0.041) (0.042)
Age2 -0.000 0.001 0.000 0.001
(0.001) (0.000) (0.000) (0.001)
East Germany -0.084 -0.059 -0.057 0.137
(0.158) (0.088) (0.131) (0.132)
Years of Education 0.254 0.056 0.242 0.040
(0.030) (0.021) (0.026) (0.029)
Years of FT experience 0.013 -0.004 0.033 -0.023
(0.013) (0.005) (0.015) (0.013)
log. Wage per Hour 0.923 -0.248 0.806 0.258
(0.160) (0.110) (0.156) (0.152)
Job tenure 0.005 0.003 0.004 -0.006
(0.010) (0.005) (0.007) (0.007)
Hours of work 0.035 0.005 0.021 -0.021
(0.009) (0.005) (0.008) (0.008)

x
 A. Appendix

Table A.10 – Continued from previous page

Hours of overtime 0.069 0.031 0.032 0.030

(0.015) (0.013) (0.011) (0.014)
Size of HH -0.174 -0.111 -0.034 -0.045
(0.074) (0.050) (0.067) (0.067)
Number of children in HH 0.042 0.015 0.122 0.115
(0.153) (0.074) (0.086) (0.102)
Marital status: Single 0.034 -0.194 0.031 -0.153
(0.161) (0.119) (0.145) (0.168)
Marital status: Other -0.166 -0.184 -0.168 0.238
(0.229) (0.142) (0.255) (0.195)
Amount of housework+childcare -0.056 0.000 -0.115 0.005
(0.035) (0.014) (0.039) (0.031)
Age of youngest HH-member -0.017 -0.002 -0.006 0.005
(0.008) (0.004) (0.005) (0.006)
Firmsize:< 5 0.075 0.211 0.248 -0.052
(0.230) (0.200) (0.211) (0.289)
Firmsize: 20-199 -0.155 -0.169 -0.146 -0.275
(0.193) (0.105) (0.137) (0.181)
Firmsize: 200-2.000 0.110 -0.344 -0.154 -0.343
(0.202) (0.117) (0.153) (0.185)
Firmsize:> 2.000 0.207 -0.351 -0.340 -0.328
(0.220) (0.123) (0.162) (0.188)
Branch: Agriculture and fishing -23.518 -0.699 -0.668 -2.530
(0.301) (0.465) (0.679) (1.015)
Branch: Energy and water supply -0.564 -0.620 0.521 -0.813
(0.652) (0.329) (0.330) (0.253)
Branch: Construction -0.177 -0.055 0.242 -0.711
(0.512) (0.404) (0.181) (0.289)
Branch: Wholesale, hotel and restaurant industry 0.233 -0.640 0.149 -0.016
(0.189) (0.197) (0.142) (0.252)
Branch: Transportation and information 0.120 -0.783 -1.051 -0.968
(0.275) (0.220) (0.266) (0.208)
Branch: Financial intermediate -0.226 -1.169 -0.425 -1.168
(0.249) (0.227) (0.200) (0.317)
Branch: Real estate, law counseling 0.080 -1.899 -0.044 -0.856
(0.235) (0.400) (0.165) (0.282)
Branch: Public Administration 0.426 -1.212 0.275 -1.247
(0.196) (0.153) (0.228) (0.165)
Branch: Public and private Services 0.339 -1.472 -0.349 -1.110
(0.334) (0.229) (0.283) (0.219)
Psychological strain of occupation 0.080 0.038 0.024 0.015
(0.032) (0.018) (0.021) (0.020)
Physical strain of occupation -0.133 0.012 -0.231 -0.017
(0.045) (0.025) (0.024) (0.023)
Great Worries about job security -0.356 0.058 -0.344 0.062
(0.200) (0.103) (0.150) (0.169)
Great Worries about own economic situation 0.451 -0.021 0.186 -0.077
(0.151) (0.091) (0.130) (0.146)
Satisfaction with work 0.122 -0.049 0.044 -0.032
(0.038) (0.019) (0.028) (0.031)
Observations 16842 5376 17198 2707
Individuals
Note: Standard errors in parentheses.

Table A.11.: The Effects of Subjective Health on Job Status - Incumbents

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
Subjective Health -0.107 -0.309 0.033 -0.190
(0.086) (0.246) (0.083) (0.374)
Age -0.055 -0.194 -0.072 0.002
(0.055) (0.155) (0.048) (0.303)
Age2 0.000 0.002 0.001 -0.001
(0.001) (0.002) (0.001) (0.003)
East Germany 0.109 0.828 -0.346 -2.084
(0.154) (0.654) (0.143) (1.156)
Years of Education -0.104 -0.054 -0.115 -0.249
(0.030) (0.102) (0.023) (0.144)
Years of FT experience -0.009 -0.014 0.010 0.090
(0.011) (0.042) (0.014) (0.129)
log. Wage per Hour -0.308 0.225 -0.933 -3.230
(0.143) (0.625) (0.148) (1.533)
Job tenure 0.003 -0.015 0.006 -0.093

xi
 A. Appendix

Table A.11 – Continued from previous page

(0.009) (0.035) (0.006) (0.036)

Hours of work -0.011 -0.019 -0.024 -0.054
(0.008) (0.039) (0.008) (0.038)
Hours of overtime -0.055 -0.249 -0.055 -0.120
(0.023) (0.068) (0.014) (0.056)
Size of HH 0.115 -0.185 0.102 0.569
(0.110) (0.343) (0.087) (0.301)
Number of children in HH -0.251 0.302 -0.061 -0.191
(0.159) (0.494) (0.088) (0.464)
Marital status: Single -0.473 0.303 0.182 1.094
(0.176) (0.644) (0.158) (0.693)
Marital status: Other -0.216 0.702 -0.216 0.725
(0.250) (0.836) (0.269) (0.768)
Amount of housework+childcare 0.068 -0.159 0.029 -0.294
(0.023) (0.212) (0.037) (0.252)
Age of youngest HH-member 0.018 -0.022 0.010 -0.004
(0.008) (0.026) (0.006) (0.028)
Firmsize:< 5 0.073 -24.532 -1.005 -24.014
(0.209) (2.304) (0.329) (1.672)
Firmsize: 20-199 0.205 -0.769 -0.192 -0.587
(0.164) (0.797) (0.169) (1.083)
Firmsize: 200-2.000 0.094 -1.379 0.087 -1.695
(0.201) (0.898) (0.177) (1.269)
Firmsize:> 2.000 -0.064 -1.440 -0.230 -0.325
(0.205) (1.210) (0.193) (1.333)
Branch: Agriculture and fishing -0.592 0.549
(0.815) (0.554)
Branch: Energy and water supply 0.685 24.284 0.469 0.116
(1.061) (1.119) (0.338) (1.023)
Branch: Construction -0.529 -0.217 -6.478
(0.502) (0.229) (2.266)
Branch: Wholesale, hotel and restaurant industry 0.411 21.947 0.024 -22.056
(0.233) (1.607) (0.182) (1.823)
Branch: Transportation and information 0.630 -0.950 0.304 -25.952
(0.314) (1.395) (0.253) (1.417)
Branch: Financial intermediate 0.732 27.386 0.475 1.077
(0.255) (.) (0.213) (1.223)
Branch: Real estate, law counseling -0.101 27.347 0.269 -0.761
(0.272) (1.399) (0.160) (1.465)
Branch: Public Administration 0.420 23.697 0.315 -1.366
(0.235) (1.140) (0.213) (0.936)
Branch: Public and private Services 0.521 23.826 0.657 -0.153
(0.305) (1.236) (0.200) (0.922)
Psychological strain of occupation 0.034 0.019 -0.003 -0.017
(0.028) (0.111) (0.024) (0.114)
Physical strain of occupation 0.029 0.021 0.083 -0.028
(0.043) (0.171) (0.025) (0.102)
Great Worries about job security 0.279 1.201 0.132 -0.228
(0.201) (0.818) (0.188) (1.008)
Great Worries about own economic situation 0.087 1.860 -0.005 -0.452
(0.179) (0.498) (0.196) (0.909)
Satisfaction with work -0.075 -0.144 -0.046 0.182
(0.037) (0.118) (0.030) (0.131)
Observations 1013 209 4924 301
Individuals
Note: Standard errors in parentheses.

Table A.12.: The Effects of log. Days of Sickness Absence on Job Status - Applicants

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
log. Days of sickness absence 0.037 0.011 -0.062 0.012
(0.041) (0.023) (0.031) (0.032)
Age 0.012 -0.145 -0.000 -0.076
(0.046) (0.025) (0.034) (0.035)
Age2 -0.000 0.001 -0.000 0.001
(0.001) (0.000) (0.000) (0.000)
East Germany -0.076 -0.146 -0.027 0.132
(0.136) (0.078) (0.108) (0.109)
Years of Education 0.247 0.061 0.221 0.078
(0.024) (0.017) (0.020) (0.025)
Years of FT experience 0.008 0.004 0.017 -0.019
(0.011) (0.005) (0.012) (0.011)

xii
 A. Appendix

Table A.12 – Continued from previous page

log. Wage per Hour 0.722 -0.224 0.887 0.147

(0.119) (0.097) (0.083) (0.119)
Job tenure 0.013 0.003 0.004 0.000
(0.008) (0.005) (0.006) (0.006)
Hours of work 0.027 -0.001 0.017 -0.014
(0.008) (0.004) (0.006) (0.007)
Hours of overtime 0.049 0.021 0.040 0.028
(0.015) (0.012) (0.009) (0.011)
Size of HH -0.101 -0.109 -0.018 -0.015
(0.064) (0.044) (0.046) (0.053)
Number of children in HH 0.077 0.086 0.082 0.094
(0.113) (0.060) (0.066) (0.068)
Marital status: Single 0.059 -0.415 0.065 -0.106
(0.133) (0.097) (0.113) (0.133)
Marital status: Other -0.031 -0.154 -0.077 0.021
(0.189) (0.120) (0.195) (0.170)
Amount of housework+childcare -0.055 0.002 -0.051 -0.007
(0.027) (0.012) (0.028) (0.027)
Age of youngest HH-member -0.011 0.003 -0.004 0.005
(0.007) (0.004) (0.004) (0.004)
Firmsize:< 5 -0.020 -0.010 0.230 -0.053
(0.195) (0.159) (0.172) (0.260)
Firmsize: 20-199 -0.148 -0.236 -0.259 -0.287
(0.155) (0.088) (0.109) (0.147)
Firmsize: 200-2.000 0.129 -0.482 -0.317 -0.472
(0.162) (0.097) (0.124) (0.153)
Firmsize:> 2.000 0.163 -0.447 -0.435 -0.416
(0.179) (0.101) (0.128) (0.153)
Branch: Agriculture and fishing -0.937 -0.675 -0.191 -2.829
(0.982) (0.378) (0.443) (1.019)
Branch: Energy and water supply -0.344 -0.554 0.486 -1.052
(0.674) (0.271) (0.255) (0.204)
Branch: Construction -0.065 -0.101 0.374 -0.893
(0.377) (0.478) (0.136) (0.228)
Branch: Wholesale, hotel and restaurant industry 0.454 -0.511 0.189 -0.085
(0.151) (0.184) (0.116) (0.210)
Branch: Transportation and information 0.211 -0.607 -0.698 -0.909
(0.251) (0.184) (0.201) (0.141)
Branch: Financial intermediate -0.099 -1.049 -0.269 -1.209
(0.219) (0.201) (0.162) (0.236)
Branch: Real estate, law counseling 0.185 -1.352 -0.137 -0.955
(0.197) (0.320) (0.146) (0.256)
Branch: Public Administration 0.537 -1.152 0.069 -1.327
(0.163) (0.145) (0.206) (0.129)
Branch: Public and private Services 0.108 -1.210 -0.098 -1.326
(0.300) (0.193) (0.236) (0.184)
Psychological strain of occupation 0.053 0.026 0.018 0.007
(0.026) (0.014) (0.017) (0.015)
Physical strain of occupation -0.135 0.009 -0.225 -0.018
(0.036) (0.020) (0.018) (0.018)
Great Worries about job security -0.329 -0.034 -0.215 -0.016
(0.169) (0.091) (0.124) (0.138)
Great Worries about own economic situation 0.202 -0.001 0.037 -0.019
(0.132) (0.079) (0.109) (0.117)
Satisfaction with work 0.131 -0.050 0.062 -0.027
(0.031) (0.015) (0.021) (0.021)
Observations 24834 8630 30153 4567
Individuals
Note: Standard errors in parentheses.

Table A.13.: The Effects of log. Days of Sickness Absence on Job Status - Incumbents

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
log. Days of sickness absence 0.065 0.107 0.024 -0.025
(0.044) (0.126) (0.040) (0.197)
Age -0.026 -0.120 -0.050 -0.013
(0.051) (0.138) (0.044) (0.257)
Age2 0.000 0.001 0.001 -0.001
(0.001) (0.002) (0.000) (0.003)
East Germany -0.026 0.437 -0.300 -1.634
(0.153) (0.634) (0.124) (0.722)
Years of Education -0.095 -0.092 -0.085 -0.125

xiii
 A. Appendix

Table A.13 – Continued from previous page

(0.026) (0.075) (0.020) (0.121)

Years of FT experience -0.008 0.010 0.013 0.129
(0.009) (0.028) (0.012) (0.097)
log. Wage per Hour -0.453 0.222 -0.948 -2.505
(0.148) (0.689) (0.124) (0.905)
Job tenure 0.008 -0.001 0.010 -0.103
(0.008) (0.025) (0.005) (0.029)
Hours of work -0.016 -0.030 -0.024 -0.056
(0.007) (0.029) (0.007) (0.033)
Hours of overtime -0.045 -0.166 -0.042 -0.114
(0.019) (0.060) (0.011) (0.052)
Size of HH 0.050 -0.039 0.059 0.322
(0.096) (0.231) (0.077) (0.317)
Number of children in HH -0.250 -0.115 -0.028 -0.053
(0.141) (0.321) (0.075) (0.413)
Marital status: Single -0.276 0.305 0.126 1.191
(0.165) (0.547) (0.139) (0.711)
Marital status: Other -0.201 0.282 -0.049 0.961
(0.253) (0.752) (0.225) (0.668)
Amount of housework+childcare 0.015 -0.103 0.031 -0.196
(0.029) (0.117) (0.034) (0.179)
Age of youngest HH-member 0.008 -0.026 0.007 -0.000
(0.007) (0.022) (0.005) (0.023)
Firmsize:< 5 0.109 -1.903 -0.940 -2.881
(0.204) (1.058) (0.292) (1.240)
Firmsize: 20-199 0.168 -0.763 -0.149 -0.552
(0.148) (0.647) (0.136) (0.849)
Firmsize: 200-2.000 0.001 -1.707 0.012 -1.759
(0.186) (0.714) (0.149) (1.138)
Firmsize:> 2.000 -0.084 -1.864 -0.262 -0.383
(0.184) (0.886) (0.159) (1.167)
Branch: Agriculture and fishing -1.231 0.236 -24.830
(0.537) (0.394) (1.672)
Branch: Energy and water supply 0.086 23.272 0.397 -0.553
(1.015) (1.360) (0.317) (1.074)
Branch: Construction -0.834 -0.187 -4.194
(0.508) (0.186) (1.086)
Branch: Wholesale, hotel and restaurant industry 0.272 21.430 0.204 1.619
(0.210) (1.395) (0.148) (1.092)
Branch: Transportation and information 0.664 -1.773 0.427 -0.227
(0.288) (1.313) (0.224) (0.972)
Branch: Financial intermediate 0.744 26.156 0.585 0.596
(0.223) (1.259) (0.174) (1.033)
Branch: Real estate, law counseling -0.153 25.779 0.217 -0.367
(0.245) (.) (0.146) (1.182)
Branch: Public Administration 0.402 23.015 0.404 -1.645
(0.204) (0.987) (0.194) (0.884)
Branch: Public and private Services 0.476 22.917 0.781 -0.705
(0.261) (1.242) (0.169) (0.994)
Psychological strain of occupation 0.030 0.079 -0.009 0.055
(0.027) (0.098) (0.019) (0.083)
Physical strain of occupation 0.016 -0.070 0.074 0.130
(0.040) (0.143) (0.021) (0.100)
Great Worries about job security 0.284 0.744 0.056 -0.849
(0.182) (0.560) (0.156) (0.904)
Great Worries about own economic situation 0.112 2.018 -0.032 0.092
(0.180) (0.396) (0.161) (0.774)
Satisfaction with work -0.038 -0.259 -0.052 0.157
(0.032) (0.101) (0.024) (0.103)
Observations 1269 290 6569 430
Individuals
Note: Standard errors in parentheses.

Table A.14.: Cross-Lagged Panel Model - Cross-Sectional

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
Subjective Health at t+1
High Status Job at t 0.014 0.006 0.006 0.002
(0.014) (0.018) (0.010) (0.019)
Subjective Health at t 0.580 0.603 0.604 0.610
(0.008) (0.012) (0.007) (0.015)
Years of Education 0.008 0.008 0.009 0.010

xiv
 A. Appendix

Table A.14 – Continued from previous page

(0.002) (0.003) (0.002) (0.004)

Years of FT experience -0.000 0.001 0.002 0.001
(0.001) (0.001) (0.001) (0.002)
log. Wage per Hour 0.020 0.047 0.024 0.046
(0.010) (0.017) (0.009) (0.020)
Job tenure 0.001 -0.000 0.001 0.001
(0.001) (0.001) (0.000) (0.001)
Size of HH 0.017 -0.002 0.010 0.017
(0.005) (0.008) (0.004) (0.009)
Number of children in HH 0.005 -0.001 0.009 -0.010
(0.007) (0.010) (0.005) (0.011)
Marital status: Single -0.002 0.009 0.025 -0.008
(0.013) (0.021) (0.010) (0.023)
Marital status: Other 0.032 -0.036 -0.002 -0.014
(0.011) (0.018) (0.012) (0.025)
Amount of housework+childcare -0.004 -0.005 0.004 0.007
(0.001) (0.002) (0.002) (0.005)
Age of youngest HH-member 0.000 -0.001 0.001 0.002
(0.000) (0.001) (0.000) (0.001)
Firmsize:< 5 -0.003 -0.032 0.026 0.005
(0.012) (0.042) (0.016) (0.062)
Firmsize: 20-199 -0.014 -0.009 0.012 0.013
(0.010) (0.018) (0.010) (0.030)
Firmsize: 200-2.000 -0.041 -0.007 0.013 0.024
(0.012) (0.020) (0.011) (0.031)
Firmsize:> 2.000 -0.019 -0.025 0.008 0.006
(0.013) (0.020) (0.011) (0.031)
Branch: Agriculture and fishing -0.001 0.081 -0.004 -0.018
(0.042) (0.114) (0.028) (0.069)
Branch: Energy and water supply 0.059 -0.148 0.015 -0.027
(0.035) (0.112) (0.025) (0.053)
Branch: Construction 0.034 0.084 0.020 0.037
(0.023) (0.104) (0.011) (0.056)
Branch: Wholesale, hotel and restaurant industry -0.011 -0.043 0.011 0.063
(0.011) (0.092) (0.010) (0.075)
Branch: Transportation and information 0.003 -0.006 0.018 0.031
(0.022) (0.087) (0.013) (0.043)
Branch: Financial intermediate -0.021 -0.015 0.022 0.021
(0.020) (0.087) (0.017) (0.049)
Branch: Real estate, law counseling -0.025 -0.028 0.029 0.015
(0.014) (0.088) (0.013) (0.049)
Branch: Public Administration -0.026 -0.037 -0.008 0.017
(0.013) (0.082) (0.017) (0.040)
Branch: Public and private Services -0.024 -0.032 0.018 0.043
(0.017) (0.085) (0.020) (0.043)
Psychological strain of occupation -0.005 -0.000 0.003 -0.002
(0.002) (0.003) (0.001) (0.003)
Physical strain of occupation -0.000 -0.002 -0.002 0.001
(0.002) (0.004) (0.002) (0.003)
Hours of work -0.001 -0.001 -0.000 0.000
(0.000) (0.001) (0.001) (0.001)
Hours of overtime -0.004 -0.001 -0.002 -0.002
(0.001) (0.002) (0.001) (0.002)
Great Worries about job security -0.040 -0.048 -0.020 -0.018
(0.011) (0.018) (0.009) (0.026)
Great Worries about own economic situation -0.063 -0.061 -0.037 -0.085
(0.010) (0.016) (0.009) (0.021)
Satisfaction with work 0.016 0.017 0.018 0.021
(0.002) (0.003) (0.002) (0.004)
Age -0.008 -0.004 -0.015 -0.012
(0.003) (0.005) (0.003) (0.006)
Age2 0.000 -0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
East Germany 0.009 -0.023 -0.003 0.015
(0.010) (0.014) (0.008) (0.016)
Constant 0.080 0.013 0.086 -0.121
(0.078) (0.150) (0.066) (0.161)
High Status Job at t+1
High Status Job at t 0.656 0.630 0.641 0.662
(0.013) (0.017) (0.010) (0.020)
Subjective Health at t 0.005 -0.001 -0.002 0.002
(0.002) (0.004) (0.003) (0.006)
Years of Education 0.018 0.026 0.020 0.026
(0.001) (0.002) (0.001) (0.003)
Years of FT experience 0.000 -0.000 -0.002 -0.002
(0.000) (0.000) (0.001) (0.001)

xv
 A. Appendix

Table A.14 – Continued from previous page

log. Wage per Hour 0.034 0.048 0.091 0.053

(0.004) (0.008) (0.005) (0.012)
Job tenure 0.000 0.000 -0.000 0.001
(0.000) (0.000) (0.000) (0.000)
Size of HH -0.003 0.005 0.001 -0.000
(0.002) (0.004) (0.002) (0.005)
Number of children in HH 0.006 0.008 0.003 0.011
(0.002) (0.005) (0.002) (0.006)
Marital status: Single -0.002 0.021 -0.003 -0.007
(0.005) (0.010) (0.005) (0.012)
Marital status: Other -0.005 0.003 -0.011 0.010
(0.004) (0.008) (0.005) (0.011)
Amount of housework+childcare -0.001 -0.002 -0.003 -0.005
(0.000) (0.001) (0.001) (0.002)
Age of youngest HH-member -0.000 0.001 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
Firmsize:< 5 0.005 -0.010 0.014 -0.018
(0.004) (0.017) (0.007) (0.031)
Firmsize: 20-199 -0.000 0.000 -0.003 0.011
(0.004) (0.009) (0.004) (0.015)
Firmsize: 200-2.000 -0.002 -0.009 -0.015 0.003
(0.005) (0.009) (0.005) (0.015)
Firmsize:> 2.000 0.003 -0.001 -0.011 -0.002
(0.005) (0.010) (0.005) (0.015)
Branch: Agriculture and fishing 0.010 -0.036 0.031 -0.022
(0.011) (0.030) (0.012) (0.022)
Branch: Energy and water supply 0.020 -0.059 0.019 -0.014
(0.032) (0.033) (0.015) (0.022)
Branch: Construction 0.022 -0.006 0.028 -0.007
(0.011) (0.067) (0.005) (0.022)
Branch: Wholesale, hotel and restaurant industry 0.007 -0.032 0.004 0.064
(0.004) (0.027) (0.005) (0.047)
Branch: Transportation and information -0.007 -0.034 -0.030 -0.024
(0.008) (0.029) (0.006) (0.019)
Branch: Financial intermediate -0.006 -0.070 -0.031 -0.049
(0.008) (0.024) (0.010) (0.026)
Branch: Real estate, law counseling 0.006 -0.061 -0.005 0.024
(0.005) (0.029) (0.007) (0.025)
Branch: Public Administration 0.007 -0.039 0.022 -0.005
(0.005) (0.022) (0.010) (0.018)
Branch: Public and private Services 0.027 -0.031 -0.005 -0.007
(0.007) (0.025) (0.009) (0.020)
Psychological strain of occupation 0.005 0.005 0.002 0.000
(0.001) (0.001) (0.001) (0.001)
Physical strain of occupation -0.003 -0.005 -0.014 -0.007
(0.001) (0.002) (0.001) (0.002)
Hours of work 0.001 0.001 0.002 -0.000
(0.000) (0.000) (0.000) (0.001)
Hours of overtime 0.004 0.005 0.004 0.003
(0.001) (0.001) (0.000) (0.001)
Great Worries about job security -0.006 0.003 -0.009 -0.015
(0.004) (0.008) (0.004) (0.011)
Great Worries about own economic situation 0.004 0.004 -0.002 -0.000
(0.003) (0.006) (0.004) (0.010)
Satisfaction with work 0.002 0.003 0.002 0.003
(0.001) (0.001) (0.001) (0.002)
Age 0.000 0.005 -0.001 -0.007
(0.001) (0.002) (0.001) (0.003)
Age2 -0.000 -0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
East Germany -0.009 -0.003 0.012 0.008
(0.004) (0.007) (0.004) (0.009)
Constant -0.332 -0.542 -0.381 -0.211
(0.031) (0.064) (0.033) (0.083)
var(e.fh)
Constant 0.292 0.297 0.259 0.263
(0.004) (0.006) (0.003) (0.008)
var(e.fhq)
Constant 0.041 0.060 0.061 0.067
(0.002) (0.003) (0.002) (0.004)
cov(e.fh,e.fhq)
Constant 0.000 -0.002 0.003 0.000
(0.001) (0.001) (0.001) (0.002)
Observations 25316 11111 32351 6335
Individuals
Note: Standard errors in parentheses.

xvi
 A. Appendix

Table A.15.: Cross-Lagged Panel Model with Fixed-Effects

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
Subjective Health at t+1
High Status Job at t -0.008 0.058 -0.016 -0.014
(0.021) (0.025) (0.013) (0.027)
Subjective Health at t 0.030 0.030 0.063 0.073
(0.010) (0.017) (0.009) (0.025)
Age -0.009 -0.011 -0.025 -0.052
(0.008) (0.015) (0.010) (0.020)
Age2 -0.000 -0.000 -0.000 0.000
(0.000) (0.000) (0.000) (0.000)
East Germany -0.117 -0.104 -0.022 0.208
(0.077) (0.070) (0.046) (0.159)
Years of Education 0.003 -0.032 0.009 0.020
(0.011) (0.013) (0.009) (0.017)
Years of FT experience -0.002 0.001 0.010 0.016
(0.003) (0.005) (0.009) (0.013)
log. Wage per Hour -0.002 0.030 0.021 0.033
(0.013) (0.022) (0.015) (0.035)
Job tenure -0.003 0.001 -0.002 -0.006
(0.002) (0.004) (0.001) (0.004)
Hours of overtime -0.000 0.002 0.000 -0.003
(0.002) (0.002) (0.001) (0.003)
Hours of work -0.003 -0.001 -0.001 0.001
(0.001) (0.001) (0.001) (0.002)
Size of HH 0.009 -0.008 0.009 -0.003
(0.010) (0.017) (0.009) (0.020)
Number of children in HH -0.031 -0.027 -0.015 0.020
(0.012) (0.016) (0.009) (0.019)
Marital status: Single 0.051 0.018 0.047 -0.044
(0.032) (0.056) (0.024) (0.056)
Marital status: Other 0.057 0.006 0.022 -0.041
(0.027) (0.042) (0.026) (0.056)
Amount of housework+childcare -0.001 -0.003 0.003 0.001
(0.002) (0.003) (0.002) (0.006)
Age of youngest HH-member 0.001 0.000 -0.000 0.002
(0.001) (0.001) (0.001) (0.002)
Firmsize:< 5 0.022 0.038 -0.008 0.078
(0.020) (0.073) (0.024) (0.098)
Firmsize: 20-199 -0.015 0.008 0.025 -0.003
(0.018) (0.033) (0.016) (0.056)
Firmsize: 200-2.000 -0.024 0.013 0.043 0.021
(0.024) (0.036) (0.020) (0.064)
Firmsize:> 2.000 -0.009 0.009 0.047 0.014
(0.025) (0.038) (0.022) (0.065)
Branch: Agriculture and fishing 0.112 0.431 -0.094 0.218
(0.065) (0.556) (0.050) (0.150)
Branch: Energy and water supply 0.030 0.131 -0.051 0.066
(0.104) (0.558) (0.060) (0.120)
Branch: Construction -0.038 0.404 0.060 0.193
(0.038) (0.552) (0.021) (0.147)
Branch: Wholesale, hotel and restaurant industry -0.010 0.347 0.023 0.197
(0.021) (0.548) (0.020) (0.236)
Branch: Transportation and information 0.066 0.323 0.020 0.108
(0.048) (0.563) (0.028) (0.147)
Branch: Financial intermediate 0.082 0.420 0.074 -0.049
(0.055) (0.550) (0.069) (0.136)
Branch: Real estate, law counseling 0.003 0.355 0.049 0.190
(0.027) (0.544) (0.022) (0.121)
Branch: Public Administration -0.056 0.329 0.030 0.085
(0.033) (0.546) (0.042) (0.107)
Branch: Public and private Services -0.011 0.378 0.020 0.119
(0.030) (0.547) (0.028) (0.114)
Psychological strain of occupation 0.001 -0.001 0.001 -0.005
(0.003) (0.006) (0.003) (0.007)
Physical strain of occupation -0.005 -0.004 -0.002 0.006
(0.005) (0.009) (0.003) (0.010)
Great Worries about job security -0.013 -0.006 -0.008 0.060
(0.013) (0.019) (0.009) (0.030)
Great Worries about own economic situation -0.015 -0.011 -0.011 -0.033
(0.011) (0.018) (0.009) (0.024)
Satisfaction with work 0.003 0.001 0.005 0.008
(0.003) (0.004) (0.002) (0.006)
Constant -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
High Status Job at t+1
High Status Job at t 0.030 0.005 0.075 0.024

xvii
 A. Appendix

Table A.15 – Continued from previous page

(0.021) (0.027) (0.015) (0.033)

Subjective Health at t 0.006 0.001 -0.001 0.007
(0.003) (0.005) (0.003) (0.009)
Age 0.009 0.022 0.024 0.041
(0.003) (0.007) (0.005) (0.014)
Age2 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
East Germany 0.010 0.028 -0.054 0.132
(0.021) (0.028) (0.025) (0.055)
Years of Education 0.007 0.010 0.012 0.018
(0.006) (0.010) (0.006) (0.012)
Years of FT experience 0.002 -0.002 -0.003 -0.020
(0.001) (0.002) (0.005) (0.011)
log. Wage per Hour 0.007 0.011 0.029 0.007
(0.005) (0.010) (0.008) (0.018)
Job tenure 0.002 0.000 0.002 -0.001
(0.001) (0.002) (0.001) (0.002)
Hours of overtime 0.002 0.003 0.001 -0.000
(0.001) (0.001) (0.000) (0.001)
Hours of work 0.001 0.002 0.001 0.000
(0.000) (0.001) (0.000) (0.001)
Size of HH -0.004 0.007 -0.006 -0.006
(0.004) (0.007) (0.005) (0.010)
Number of children in HH -0.004 0.000 0.000 0.004
(0.004) (0.006) (0.004) (0.011)
Marital status: Single 0.000 0.044 -0.012 -0.025
(0.013) (0.029) (0.012) (0.034)
Marital status: Other -0.010 -0.014 -0.013 0.007
(0.011) (0.017) (0.013) (0.020)
Amount of housework+childcare -0.002 -0.000 -0.002 0.002
(0.001) (0.001) (0.001) (0.002)
Age of youngest HH-member 0.000 0.001 -0.001 -0.001
(0.000) (0.001) (0.000) (0.001)
Firmsize:< 5 0.006 -0.034 0.012 -0.020
(0.006) (0.030) (0.011) (0.071)
Firmsize: 20-199 0.001 0.016 -0.008 -0.006
(0.006) (0.017) (0.008) (0.025)
Firmsize: 200-2.000 0.002 0.027 -0.019 -0.009
(0.008) (0.018) (0.010) (0.027)
Firmsize:> 2.000 0.014 0.042 -0.019 -0.027
(0.010) (0.019) (0.011) (0.028)
Branch: Agriculture and fishing 0.023 -0.005 -0.016 -0.033
(0.018) (0.025) (0.025) (0.069)
Branch: Energy and water supply -0.001 0.005 0.023 0.008
(0.053) (0.019) (0.043) (0.050)
Branch: Construction 0.017 0.003 0.014 0.024
(0.011) (0.026) (0.010) (0.052)
Branch: Wholesale, hotel and restaurant industry -0.007 0.064 0.011 0.085
(0.006) (0.032) (0.011) (0.116)
Branch: Transportation and information -0.022 0.011 -0.005 0.028
(0.015) (0.039) (0.012) (0.062)
Branch: Financial intermediate -0.011 0.052 -0.058 -0.097
(0.019) (0.046) (0.038) (0.114)
Branch: Real estate, law counseling 0.014 -0.065 -0.016 0.025
(0.011) (0.047) (0.013) (0.062)
Branch: Public Administration 0.008 0.009 0.009 0.017
(0.013) (0.016) (0.027) (0.051)
Branch: Public and private Services -0.003 0.007 0.011 0.046
(0.011) (0.024) (0.015) (0.054)
Psychological strain of occupation 0.002 -0.007 0.001 -0.005
(0.001) (0.003) (0.001) (0.003)
Physical strain of occupation -0.002 0.005 -0.006 0.001
(0.002) (0.004) (0.002) (0.003)
Great Worries about job security -0.006 -0.002 -0.006 -0.006
(0.004) (0.008) (0.004) (0.012)
Great Worries about own economic situation 0.009 0.010 0.000 0.003
(0.004) (0.007) (0.004) (0.011)
Satisfaction with work 0.002 -0.002 0.003 0.004
(0.001) (0.002) (0.001) (0.003)
Constant -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
var(e.d fh)
Constant 0.163 0.165 0.150 0.148
(0.003) (0.004) (0.002) (0.006)
var(e.d fhq)
Constant 0.022 0.032 0.035 0.035

xviii
 A. Appendix

Table A.15 – Continued from previous page

(0.001) (0.002) (0.001) (0.002)

cov(e.d fh,e.d fhq)
Constant 0.000 -0.000 0.001 -0.000
(0.000) (0.001) (0.000) (0.001)
Observations 25316 11111 32351 6335
Individuals
Note: Standard errors in parentheses.

Table A.16.: Cross-Lagged Panel Model for log. Days of Sickness Absence - Cross-Sectional

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
log. Days of sickness absence at t+1
High Status Job at t -0.041 -0.058 -0.105 -0.045
(0.029) (0.038) (0.020) (0.044)
log. Days of sickness absence at t 0.375 0.397 0.345 0.401
(0.008) (0.011) (0.006) (0.013)
Years of Education -0.022 -0.025 -0.031 -0.027
(0.004) (0.007) (0.004) (0.009)
Years of FT experience -0.002 -0.003 -0.006 -0.006
(0.001) (0.002) (0.002) (0.004)
log. Wage per Hour 0.111 0.119 0.028 -0.004
(0.018) (0.035) (0.018) (0.038)
Job tenure -0.002 -0.002 -0.002 0.000
(0.001) (0.002) (0.001) (0.002)
Size of HH -0.040 -0.038 0.006 -0.037
(0.010) (0.016) (0.008) (0.023)
Number of children in HH 0.020 0.023 0.012 0.032
(0.014) (0.020) (0.011) (0.023)
Marital status: Single -0.047 -0.069 -0.031 -0.037
(0.026) (0.042) (0.022) (0.049)
Marital status: Other 0.002 0.104 0.033 0.076
(0.024) (0.039) (0.025) (0.058)
Amount of housework+childcare 0.006 0.007 0.006 0.018
(0.003) (0.005) (0.004) (0.010)
Age of youngest HH-member -0.001 0.000 0.000 -0.000
(0.001) (0.001) (0.001) (0.002)
Firmsize:< 5 -0.091 -0.200 -0.060 -0.056
(0.023) (0.068) (0.033) (0.135)
Firmsize: 20-199 0.132 0.033 0.053 0.022
(0.021) (0.040) (0.020) (0.063)
Firmsize: 200-2.000 0.229 0.122 0.136 0.133
(0.026) (0.041) (0.022) (0.064)
Firmsize:> 2.000 0.253 0.189 0.122 0.112
(0.027) (0.042) (0.023) (0.065)
Branch: Agriculture and fishing 0.043 -0.043 -0.085 -0.020
(0.088) (0.203) (0.060) (0.182)
Branch: Energy and water supply -0.213 -0.077 0.068 -0.014
(0.104) (0.165) (0.059) (0.103)
Branch: Construction -0.036 -0.093 0.043 0.313
(0.046) (0.250) (0.023) (0.129)
Branch: Wholesale, hotel and restaurant industry -0.041 0.214 -0.059 0.206
(0.023) (0.165) (0.022) (0.159)
Branch: Transportation and information 0.060 -0.043 -0.034 0.061
(0.048) (0.142) (0.032) (0.096)
Branch: Financial intermediate 0.051 -0.162 0.038 -0.012
(0.041) (0.138) (0.037) (0.102)
Branch: Real estate, law counseling 0.013 -0.023 0.051 0.040
(0.029) (0.157) (0.026) (0.111)
Branch: Public Administration -0.019 0.018 -0.012 -0.019
(0.027) (0.129) (0.043) (0.087)
Branch: Public and private Services -0.023 -0.018 -0.013 0.065
(0.034) (0.138) (0.038) (0.096)
Psychological strain of occupation 0.002 0.004 0.001 0.003
(0.004) (0.006) (0.003) (0.006)
Physical strain of occupation 0.021 0.000 0.023 0.001
(0.005) (0.009) (0.003) (0.008)
Hours of work 0.012 0.012 -0.000 0.004
(0.001) (0.002) (0.001) (0.003)
Hours of overtime -0.007 -0.012 -0.007 -0.017
(0.003) (0.004) (0.002) (0.004)
Great Worries about job security -0.009 0.035 -0.051 0.002
(0.023) (0.038) (0.019) (0.054)

xix
 A. Appendix

Table A.16 – Continued from previous page

Great Worries about own economic situation 0.066 0.069 0.022 0.152
(0.020) (0.032) (0.019) (0.045)
Satisfaction with work -0.043 -0.062 -0.052 -0.059
(0.004) (0.007) (0.004) (0.008)
Age -0.018 -0.015 -0.027 -0.019
(0.006) (0.010) (0.006) (0.013)
Age2 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
East Germany 0.035 0.018 -0.005 -0.025
(0.023) (0.032) (0.019) (0.040)
Constant 0.935 1.164 1.704 1.684
(0.152) (0.274) (0.140) (0.332)
High Status Job at t+1
High Status Job at t 0.658 0.624 0.648 0.664
(0.012) (0.016) (0.008) (0.017)
log. Days of sickness absence at t -0.001 -0.003 -0.004 -0.003
(0.001) (0.002) (0.001) (0.002)
Years of Education 0.016 0.026 0.019 0.025
(0.001) (0.002) (0.001) (0.002)
Years of FT experience 0.000 -0.000 -0.001 -0.002
(0.000) (0.000) (0.000) (0.001)
log. Wage per Hour 0.030 0.042 0.072 0.051
(0.003) (0.006) (0.004) (0.009)
Job tenure 0.000 0.000 -0.000 0.001
(0.000) (0.000) (0.000) (0.000)
Size of HH -0.002 0.001 0.001 -0.004
(0.001) (0.003) (0.001) (0.004)
Number of children in HH 0.005 0.004 0.002 0.010
(0.002) (0.004) (0.002) (0.005)
Marital status: Single -0.002 0.011 -0.006 -0.008
(0.005) (0.009) (0.004) (0.010)
Marital status: Other -0.004 0.006 -0.010 0.004
(0.004) (0.007) (0.004) (0.009)
Amount of housework+childcare -0.001 -0.001 -0.003 -0.005
(0.000) (0.001) (0.001) (0.002)
Age of youngest HH-member -0.000 0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
Firmsize:< 5 0.002 -0.018 0.010 0.016
(0.004) (0.012) (0.006) (0.034)
Firmsize: 20-199 0.000 -0.002 -0.004 0.016
(0.003) (0.008) (0.004) (0.012)
Firmsize: 200-2.000 0.000 -0.009 -0.008 0.004
(0.004) (0.008) (0.004) (0.013)
Firmsize:> 2.000 0.004 0.004 -0.007 0.004
(0.004) (0.009) (0.004) (0.013)
Branch: Agriculture and fishing 0.015 0.018 0.022 0.008
(0.009) (0.030) (0.010) (0.032)
Branch: Energy and water supply 0.008 -0.026 0.010 0.002
(0.028) (0.029) (0.013) (0.018)
Branch: Construction 0.021 0.039 0.023 0.018
(0.011) (0.046) (0.004) (0.023)
Branch: Wholesale, hotel and restaurant industry 0.005 -0.007 0.000 0.044
(0.003) (0.022) (0.005) (0.043)
Branch: Transportation and information -0.012 -0.011 -0.035 -0.017
(0.008) (0.019) (0.005) (0.016)
Branch: Financial intermediate -0.006 -0.037 -0.029 -0.022
(0.007) (0.017) (0.010) (0.023)
Branch: Real estate, law counseling 0.005 -0.037 -0.006 0.025
(0.005) (0.022) (0.007) (0.023)
Branch: Public Administration 0.003 -0.014 0.015 0.006
(0.004) (0.015) (0.010) (0.016)
Branch: Public and private Services 0.023 -0.008 -0.003 -0.005
(0.006) (0.017) (0.008) (0.018)
Psychological strain of occupation 0.005 0.005 0.001 0.001
(0.001) (0.001) (0.001) (0.001)
Physical strain of occupation -0.003 -0.004 -0.013 -0.008
(0.001) (0.002) (0.001) (0.001)
Hours of work 0.001 0.001 0.002 -0.000
(0.000) (0.000) (0.000) (0.001)
Hours of overtime 0.004 0.004 0.003 0.003
(0.001) (0.001) (0.000) (0.001)
Great Worries about job security -0.007 -0.003 -0.007 -0.013
(0.003) (0.007) (0.003) (0.009)
Great Worries about own economic situation -0.000 -0.001 -0.005 0.000
(0.003) (0.006) (0.003) (0.009)
Satisfaction with work 0.002 0.003 0.002 0.001

xx
 A. Appendix

Table A.16 – Continued from previous page

(0.001) (0.001) (0.001) (0.002)

Age 0.000 0.004 -0.001 -0.005
(0.001) (0.002) (0.001) (0.002)
Age2 0.000 -0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
East Germany -0.013 0.002 0.003 0.013
(0.004) (0.006) (0.004) (0.008)
Constant -0.295 -0.501 -0.307 -0.235
(0.026) (0.053) (0.026) (0.069)
var(e.fh)
Constant 1.409 1.527 1.490 1.421
(0.015) (0.023) (0.014) (0.030)
var(e.fhq)
Constant 0.038 0.056 0.058 0.063
(0.001) (0.002) (0.001) (0.003)
cov(e.fh,e.fhq)
Constant -0.004 -0.005 -0.004 -0.001
(0.001) (0.003) (0.001) (0.003)
Observations 31733 14400 45520 8485
Individuals
Note: Standard errors in parentheses.

Table A.17.: Cross-Lagged Panel Model with Fixed-Effects for log. Days of Sickness Absence

(1) (2) (3) (4)

W - Private W - Public M - Private M - Public
log. Days of sickness absence at t+1
High Status Job at t 0.029 -0.024 -0.013 0.008
(0.040) (0.049) (0.027) (0.064)
log. Days of sickness absence at t 0.028 0.002 0.032 0.037
(0.008) (0.012) (0.007) (0.015)
Age -0.062 -0.114 -0.125 -0.042
(0.013) (0.021) (0.020) (0.033)
Age2 0.001 0.001 0.001 0.001
(0.000) (0.000) (0.000) (0.000)
East Germany 0.131 -0.109 0.100 0.172
(0.163) (0.175) (0.092) (0.213)
Years of Education 0.031 0.003 0.004 0.023
(0.019) (0.028) (0.015) (0.028)
Years of FT experience 0.016 0.007 0.033 0.002
(0.005) (0.008) (0.019) (0.020)
log. Wage per Hour 0.107 0.179 0.056 -0.137
(0.029) (0.052) (0.031) (0.069)
Job tenure 0.011 0.015 0.012 0.011
(0.003) (0.007) (0.002) (0.008)
Hours of overtime -0.007 -0.009 -0.007 -0.013
(0.003) (0.006) (0.002) (0.005)
Hours of work 0.013 0.013 0.001 0.006
(0.002) (0.003) (0.002) (0.005)
Size of HH -0.062 -0.045 -0.002 -0.047
(0.019) (0.028) (0.014) (0.036)
Number of children in HH 0.031 0.032 0.018 0.065
(0.021) (0.029) (0.016) (0.038)
Marital status: Single -0.017 -0.018 -0.113 -0.001
(0.053) (0.097) (0.042) (0.085)
Marital status: Other -0.014 0.037 -0.063 0.164
(0.045) (0.074) (0.046) (0.103)
Amount of housework+childcare -0.001 0.010 -0.001 0.014
(0.004) (0.007) (0.005) (0.011)
Age of youngest HH-member -0.003 -0.002 -0.001 -0.005
(0.002) (0.002) (0.001) (0.003)
Firmsize:< 5 0.013 0.156 -0.049 0.188
(0.037) (0.123) (0.049) (0.299)
Firmsize: 20-199 0.091 0.078 0.058 -0.218
(0.034) (0.065) (0.031) (0.102)
Firmsize: 200-2.000 0.153 0.114 0.114 -0.107
(0.042) (0.070) (0.038) (0.108)
Firmsize:> 2.000 0.155 0.135 0.095 -0.096
(0.048) (0.072) (0.042) (0.107)
Branch: Agriculture and fishing -0.018 -0.197 -0.185 0.163
(0.152) (0.233) (0.109) (0.391)
Branch: Energy and water supply -0.413 -0.111 0.033 -0.036
(0.170) (0.329) (0.133) (0.327)

xxi
 A. Appendix

Table A.17 – Continued from previous page

Branch: Construction 0.067 -0.089 -0.028 0.333

(0.067) (0.333) (0.041) (0.344)
Branch: Wholesale, hotel and restaurant industry 0.053 0.035 -0.077 0.055
(0.039) (0.276) (0.037) (0.520)
Branch: Transportation and information -0.004 0.187 -0.079 0.096
(0.087) (0.264) (0.060) (0.330)
Branch: Financial intermediate 0.136 -0.105 -0.128 0.300
(0.099) (0.288) (0.095) (0.403)
Branch: Real estate, law counseling -0.019 -0.111 -0.097 0.101
(0.051) (0.213) (0.047) (0.334)
Branch: Public Administration 0.005 -0.123 -0.024 0.142
(0.066) (0.161) (0.079) (0.307)
Branch: Public and private Services -0.167 -0.082 -0.106 0.147
(0.061) (0.188) (0.058) (0.318)
Psychological strain of occupation -0.005 -0.001 -0.006 0.004
(0.007) (0.013) (0.005) (0.012)
Physical strain of occupation 0.000 0.003 0.008 0.008
(0.010) (0.021) (0.006) (0.017)
Great Worries about job security -0.044 -0.017 -0.060 -0.050
(0.026) (0.045) (0.022) (0.070)
Great Worries about own economic situation -0.007 0.028 -0.023 0.055
(0.023) (0.036) (0.021) (0.051)
Satisfaction with work -0.029 -0.052 -0.035 -0.034
(0.005) (0.008) (0.005) (0.011)
Constant 0.000 0.000 0.000 0.000
(0.000) (0.000) (0.000) (0.000)
High Status Job at t+1
High Status Job at t 0.086 0.059 0.151 0.098
(0.019) (0.025) (0.014) (0.028)
log. Days of sickness absence at t -0.000 -0.000 -0.003 0.001
(0.001) (0.002) (0.001) (0.002)
Age 0.006 0.011 0.017 0.031
(0.002) (0.004) (0.004) (0.011)
Age2 -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
East Germany 0.058 0.027 -0.018 0.063
(0.038) (0.044) (0.022) (0.030)
Years of Education 0.007 0.019 0.009 0.019
(0.004) (0.007) (0.004) (0.009)
Years of FT experience 0.002 0.000 -0.002 -0.015
(0.001) (0.002) (0.004) (0.009)
log. Wage per Hour 0.016 0.018 0.035 0.012
(0.005) (0.009) (0.006) (0.015)
Job tenure 0.001 -0.002 0.001 -0.002
(0.001) (0.002) (0.001) (0.002)
Hours of overtime 0.003 0.004 0.002 -0.000
(0.001) (0.001) (0.000) (0.001)
Hours of work 0.001 0.002 0.001 0.001
(0.000) (0.001) (0.000) (0.001)
Size of HH -0.005 -0.001 -0.001 -0.008
(0.003) (0.006) (0.002) (0.008)
Number of children in HH 0.003 -0.012 0.003 0.008
(0.003) (0.008) (0.003) (0.009)
Marital status: Single 0.002 -0.014 -0.014 -0.009
(0.011) (0.021) (0.009) (0.023)
Marital status: Other -0.004 -0.004 -0.015 -0.000
(0.007) (0.012) (0.009) (0.022)
Amount of housework+childcare -0.001 0.000 -0.002 -0.001
(0.001) (0.001) (0.001) (0.002)
Age of youngest HH-member -0.000 0.000 -0.000 -0.001
(0.000) (0.000) (0.000) (0.001)
Firmsize:< 5 0.003 -0.023 0.012 -0.043
(0.005) (0.026) (0.009) (0.062)
Firmsize: 20-199 0.002 0.015 -0.005 -0.008
(0.005) (0.015) (0.006) (0.021)
Firmsize: 200-2.000 -0.001 0.015 -0.015 -0.017
(0.006) (0.015) (0.007) (0.024)
Firmsize:> 2.000 0.006 0.024 -0.021 -0.021
(0.008) (0.016) (0.009) (0.024)
Branch: Agriculture and fishing 0.027 0.068 -0.006 -0.028
(0.011) (0.109) (0.022) (0.062)
Branch: Energy and water supply -0.012 0.090 0.020 0.015
(0.040) (0.076) (0.031) (0.048)
Branch: Construction -0.003 0.033 0.013 -0.012
(0.012) (0.031) (0.008) (0.055)
Branch: Wholesale, hotel and restaurant industry 0.002 0.008 0.012 0.063

xxii
 A. Appendix

Table A.17 – Continued from previous page

(0.006) (0.011) (0.008) (0.094)

Branch: Transportation and information -0.016 0.063 -0.007 0.005
(0.015) (0.035) (0.009) (0.058)
Branch: Financial intermediate 0.030 0.071 -0.000 0.006
(0.014) (0.031) (0.028) (0.063)
Branch: Real estate, law counseling 0.021 -0.037 -0.017 0.008
(0.010) (0.038) (0.012) (0.058)
Branch: Public Administration 0.016 0.011 -0.022 0.028
(0.012) (0.010) (0.020) (0.050)
Branch: Public and private Services 0.013 0.011 0.021 0.026
(0.010) (0.017) (0.013) (0.052)
Psychological strain of occupation 0.003 -0.002 0.002 -0.004
(0.001) (0.003) (0.001) (0.003)
Physical strain of occupation -0.006 0.005 -0.004 -0.001
(0.002) (0.004) (0.002) (0.003)
Great Worries about job security -0.006 -0.003 -0.002 0.008
(0.003) (0.008) (0.003) (0.010)
Great Worries about own economic situation 0.004 0.004 -0.004 0.001
(0.003) (0.006) (0.003) (0.010)
Satisfaction with work 0.003 -0.001 0.003 0.001
(0.001) (0.001) (0.001) (0.002)
Constant -0.000 -0.000 -0.000 -0.000
(0.000) (0.000) (0.000) (0.000)
var(e.d fh)
Constant 0.896 0.948 1.015 0.875
(0.013) (0.019) (0.012) (0.024)
var(e.d fhq)
Constant 0.022 0.032 0.036 0.035
(0.001) (0.002) (0.001) (0.002)
cov(e.d fh,e.d fhq)
Constant -0.000 -0.002 -0.002 0.004
(0.001) (0.002) (0.001) (0.002)
Observations 31733 14400 45520 8485
Individuals
Note: Standard errors in parentheses.

A.2. Strategies of Closure on the Labor Market

A.2.1. Protection Against Dismissal
Permanent contracts can be seen as a form of closure based on legal regulations (Giesecke
2006, Giesecke & Groß 2007). It is easier for employers not to continue a temporary contract
than to fire a worker who holds a permanent contract. Therefore permanent workers face less
competition and their position is (at least partly) closed. They need not fear job loss and they
can extract a rent in form of higher wages and better working conditions.
Some scholars argue that permanent contracts are actually a reward in themselves for closure
based on individual skill. The argument is as follows: An open-ended contract comes with
much higher layoff costs due to possible severance pays and administrative costs (Aparicio 2011,
13). At the same time permanently employed workers are more productive than temporarily
employed ones (Emerson 1988, Lazear 1990). Therefore it pays to sign permanent contracts
for both employer and employee in some cases. They share a composite rent. In the end
permanent contracts have both the characteristics of a closure mechanism and of a reward for
closure.
Another way how skill can lead to informal closure of a labor market position is through the
acquisition of firm-specific human capital (Parsons 1972). Firm specific human capital consists
of skills and training a worker usually gets by on-the-job-training. These skills are also called

xxiii
 A. Appendix

idiosyncratic skills (Williamson, Wachter & Harris 1975). This means that the human capital
of employees is bound to a certain company and cannot or only to certain degree be transferred
to other employers. Therefore workers who have acquired a lot of skills via on-the-job training
strongly bound their employer towards them and vice versa. As the employer has already
invested much into the specific human capital of the employee and other workers outside the
firm cannot compete with the incumbent because they lack these skills, it would cost the
employer dearly to replace the initial already trained worker with an untrained one (for an
example of the calculation of such costs, see Hamermesh 1987).

A.2.2. Unionization
With regard to the effects of unions on closure a special focus should lie on the effect of
collective bargaining agreements. They can on the one hand reduce the power of the employer,
increase wages and improve working conditions as they shield their members from competition
by non-union workers (Jacobsen, Skillman & Jacobsen 2004, 130). In this way they function
as instruments of social closure as they are only valid for a specific group of workers. Wage
differentials should increase between occupation/industries and between union members and
non-members if collective bargaining agreements are unequally distributed (which they are). If
collective bargaining is covered by law or state regulations unions can significantly increase
wages for their members (and non-members through spill-over effects) (Weeden 2002, 63). It
can be claimed that unions have an effect on industry wages rather than occupation wages as
they are often organized around industries and cut across occupations (Weeden 2002, 64).
It can also be shown that labor mobility and turnover are reduced by unionization, which in
turn reduces competition. This induces a feedback mechanism where longer tenure under
unionization leads to increased company specific human capital (Elias 1994, 563).

A.2.3. Licensing and Registration

Licensing is done or demanded by the state or state empowered private organizations and often
introduced with the argument that it protects consumers from incompetence or malfeasance
in areas in which it is hard to assess such things and where incompetence or malfeasance
may have dire consequences for consumers (e.g. their health). An official authority has to
give permission to individuals to use certain (occupational) titles. With the introduction of
such a mechanism several costs arise due to possible fees, the adaption of minimum standards
required by the responsible authority, and the time consumed by the bureaucratic process. This
restricts the supply of labor in this group or occupation as not all who are able to work in a
certain occupation can cover these costs (Weeden 2002, 62). Licenses can also be voluntary.
As voluntary licenses their most important contribution to closure is that they channel demand
to a certain group by claiming only holders of the license or certificate can provide the product.

xxiv
 A. Appendix

At the very least the license is an organized claim that a service or product bought from holders
of such licenses are of quality (Weeden 2002, 66-67).

A.2.4. Monopoly and Ownership

This strategy of closure focuses on employers as actors and not as workers. This includes a
certain change of perspective from worker to employer and it necessitates a link between the
closure strategy of the employer and the rent generation of the employee.
Based on ownership of certain means of production like technological rent (Rutherford &
Rutherford 2005, 407) and/or monopoly/oligopoly in certain product or service markets, firms
are able to generate high profits which are far above those profits they would generate under
perfect competition. These profits thus classify as a rent and the strategy is generally classified
as rent-seeking (Rutherford & Rutherford 2005, 348). When employers share part of their
profit with their employees in form of higher wages or other non-pecuniary forms, this is called
rent-sharing (for a short review on the literature, see Martins 2007). Paying wages above
competitive wages is called efficiency wages (see e.g. Akerlof & Yellen 1986) and is theorized
to be used to avoid (among other things) shirking (Shapiro & Stiglitz 1984).
Rent or profit sharing reduces turnover, because wages are flexible. In hard times wages fall with
profits and in good times they rise with profits. Therefore there is less incentive for employer
and employee to terminate the labor contract. This results in reduced turnover rates (Azfar &
Danninger 2000) and should increase employment stability and job security for workers.
Furthermore, successful companies can offer better terms for employees, therefore they get
better workers. Sick workers drop out of good firms. This constitutes a selection effect which
results in company specific health inequalities.
If a company produces in a closed position it can offer better working employment conditions
to its employees in forms of efficiency wages. In branches or industries where there is heavy
product competition or where companies are largely owned by investors who invest on a
short term basis and who expect maximum share-holder-value the pressure put on the firms
(Windolf 2008) will be redirected to the employees. Another important characteristic is the
firm size and the total revenue which constrain companies in offering good working conditions.
Product market competition can increase respective labor market competition when product
markets are deregulated or increasingly penetrated by international trade (Saint-Paul 2005,
282). Generally speaking a lower price-elasticity of demand on the product-market-sector
means a lower wage-elasticity of labor demand in the respective sector which increases wages
(Saint-Paul 2005, 284). Other labor market outcomes beyond wages like wage inequality,
returns to education or training, managers’ incentives, or discrimination vary with product
market competition (Aparicio 2011, 2).
It can be concluded that processes of closure are not only important among workers, but also

xxv
 A. Appendix

among companies. Closure on the employer level can also induce closure on the employee level.

A.3. Open and Closed Positions - Segmented Labor

Market Theory
A second big influence on the theory of open and closed position should be mentioned. It helps
understanding the theoretical contribution Sørensen made to the field. The works of theorists
of segmented labor market (SLM) theory are important points of inspiration for the theory of
open and closed positions. SLM theorists often started off with empirical facts of the (US)
labor market. Certain phenomena which could not be fully explained by neoclassical theory
were pointed out. Among others these phenomena are (Cain 1976, 1217-1220):

1. persistence of poverty

2. persistence of income inequality

3. education and training failed to eradicate poverty

4. education and training as criteria for discrimination

5. discrimination

6. level and form of unemployment

7. monopolies, unions, protected labor markets

Driven by these problems dual-labor-market or segmented labor market theories emerged

in the 1970s. They saw a division of the labor market into two parts with little mobility
between them. One was characterized by stable well paid jobs with benefits and union coverage,
the other by low paying, hazardous work, with little job security or benefits (Kalleberg &
Sorensen 1979, 356). In the following I will give a brief description of some of the most
important works from that strand of research.

Thurow’s (1975) theory states that the number of jobs is technologically determined and
that workers’ reservation wage does not influence the number of job positions actually filled.
Wages are seen as fixed, workers queuing for jobs at a given wage make up the supply side
(Cain 1976, 1221-1222). Employers use screening devices to select workers according to their
potential for further training (Thurow 1975, 87). The criterion of potential for further is
training is important, because most skills are acquired on the job, after the end of the formal
education process. This pays tribute to the fact that direct worker-to-worker training on-the-job
is the cheapest way of training workers (Thurow 1975, 76,79). Thurow points out that job

xxvi
 A. Appendix

competition and wage competition are not mutually exclusive and exist at the same time
in reality. Based on the position in the queue workers are allocated to jobs which are paid
based on characteristics of the job, not the worker (Thurow 1975, 76). In classical economic
terms he formulates that “[...], the marginal product resides in the job and not in the man.”
(Thurow 1975, 77).
When employers do not have perfect information about their potential employees’ productivity
they resort to signals of productivity. About certain groups like women or blacks only less
reliable information is available. Employers transfer their information disadvantage from the
individual to the group level (Cain 1976, 1232-1233). Education is used as a screening device
(Cain 1976, 1219).
Doeringer & Piore (1985) make one of the major contributions to the theory of open and
closed positions. They use a different terminology and speak of internal labor markets. They
try to explain differences in the determination of wages across different sectors in the industry.
Doeringer and Piore state that the primary labor market is dominated by jobs in large firms
with a high degree of unionization which are paying higher wages, have better career prospects,
better working conditions and provide more job security. The secondary labor market (external)
contains low paying jobs of discriminated groups with very unstable employment relations
(Cain 1976, 1222). They argue that in internal labor markets hiring, promotion and layoff rules
are regulated in collective bargaining agreements or management manuals. These regulations
follow a generalized bureaucratic or administrative logic and less an individualized profit
maximizing logic. The rigidity of these regulations is the defining moment of internal labor
markets in contrast to external labor markets. They suggest that the existence of internal
labor markets should be tested against a theoretically chosen set of variables which determines
wages and demonstrate that the actual data indicates a different or modified set of variables as
wage determining (Doeringer & Piore 1985, 5). Doeringer and Piore consider the neoclassical
theory of wage determination to be the best choice for a reference theory, as it was common in
segmented labor market theories:

Nevertheless, since the time of Adam Smith, the classical, and then the neo-
classical, school of labor economics has been a common target, representing the
orthodoxy to be challenged. (Cain 1976, 1215)

Due to a lack of available data Doeringer an Piore propose some indicators for the rigidity of
internal labor markets which are more heuristic of nature (Doeringer & Piore 1985, 6-7):

1. The time an internal labor market already exists

2. Comments from labor and management about certain rules

3. Investments in company specific human capital and on-the-job-training

4. Customs and traditions regarding wage structure, promotions, etc.

xxvii
 A. Appendix

Internal labor markets are mainly generated by three factors. These are (Doeringer &
Piore 1985, 13):

1. skill specificity

2. on-the-job training

3. customary law

The skill and job specificity increases the costs of employers for screening and hiring employees.
The specificity is often not a choice made by the management, but more a tribute to certain
job or technology demands (Doeringer & Piore 1985, 15). On-the-job training provides a major
portion of skills used by both blue and white collar workers. For the latter the formal education
is often just a screening device for the management (Doeringer & Piore 1985, 18). On-the-job
training is usually organized very informally (Doeringer & Piore 1985, 19-20).
Labor turnover is reduced in internal labor markets, because workers value these internal labor
markets, and benefits are often awarded according to length of service. In addition management
faces costs of termination and of replacement which are considerably higher in internal labor
markets than in external labor markets1 . Replacement costs consist of costs for recruitment,
screening, and training. Termination costs can be explicit (unemployment insurance or severance
pay) or implicit (difference in benefit-payments) (Doeringer & Piore 1985, 29-30). Doeringer
& Piore (1985, 43) differentiate between open and closed internal labor markets. Positions
in closed internal labor markets are only filled with persons from within the same internal
labor market. This can lead to a process of so called vacancy-chains through promotion
ladders, where a chain of workers is sequentially promoted (see e.g. White 1970, Chase 1991).
Most internal labor markets are between completely open and completely closed (Doeringer &
Piore 1985, 43-45). The degree to which workers can involuntarily lose their job in an internal
labor market is also an important characteristic. In contrast to external labor markets, in
internal labor markets the options for layoffs for employers are reduced (Doeringer & Piore 1985,
49).
Doeringer & Piore (1985, 74-77) point out that the neoclassical view on labor market and
wage determination fails to take certain facts into account which often appear in internal labor
markets. These are permanency of employment, meaning contracts are unlimited and cannot
be canceled at any time. Labor costs are fixed and wages cannot be reduced freely as suggested
by the neoclassical theory. And wage decisions usually apply to groups and not to individuals.

1
Where according to neoclassical theory they should be virtually non-existent.

xxviii
 A. Appendix

A.4. Review of the Literature on Reporting

Heterogeneity in Self-Rated Health
Crossley & Kennedy’s (2002) study is widely cited, because they address the reliability of the
subjective health measure in an innovative fashion. They use a kind of randomized questionnaire
experiment with two measures of subjective health. In one group the question is asked before a
block of specific health questions, in the control group the subjective health question is asked
before and after the block with slightly different wording. They find significant differences in the
distributions between the questions asked before and after the health block in the questionnaire.
They argue that these differences can be attributed to measurement error, different interview
forms (face-to-face vs. written questionnaire), and “learning” about (or becoming more aware
of) health issues on part of the respondents by answering the specific health questions (Crossley
& Kennedy 2002, 651). Bowling & Windsor (2008) also come to the conclusion that wording
and order of the self-rated health item has an impact on its distribution for example by reducing
ceiling and increasing optimistic bias effects by changing the highest category from “very good”
to “excellent”.

Lindeboom & van Doorslaer (2004) use the health utility index III (HUI3) as an objective
index of health which is supposed to represent “true” health. They want to address the problem
if certain groups use different thresholds in the subjective health measure given the same “true”
health status. They define two possible problems, the index-shift and the cut-point-shift. The
former means that a group (e.g. women) rate their health given same “true” health on average
higher or lower than the other group. The relative distance between the thresholds remains
the same for the groups (Lindeboom & van Doorslaer 2004, 1084). This could mean that
women tend to report lower overall health than men, but the relation of a reported “good” to
a reported “poor” means the same for men and women. A cut-point-shift on the other hand
is defined as giving thresholds different meaning. They cannot be compared across groups
(Lindeboom & van Doorslaer 2004, 1084). In this case, relation between the answers to the
subjective health question are not the same for men and women. The difference between
“good” and “poor” might be greater for men than for women given a measure of “true” health.
Lindeboom & van Doorslaer (2004) come to the conclusion that in their dataset women and
older people report better health and that therefore a direct comparison of the self-rated health
across gender and age is not feasible.

Eriksson, Undén & Elofsson (2001) compare 3 self-rated health measures. The standard
5-point scale, a 7-point scale without specific answers to each category and a 5-point scale
with a comparison to the respondent’s age group. They conclude that the differences in a
cross-sectional setting are small and that the measures can be seen as very similar (Eriksson,
Undén & Elofsson 2001, 332).

xxix
 A. Appendix

Gunasekara, Carter & Blakely (2012) use longitudinal data to assess whether computed
change in SRH or an explicit self-reported change in actual health is a better measure of change
in health. They try to validate both measures using an objective health indicator, hospital visit
within the last year. They conclude that self-reported change in health is a better indicator than
SRH, because it is stronger associated with a hospital visit between two points of observation.
A problem they do not address is that a hospital visit can influence the self-reported change in
health (SRCH) item. A visit to the hospital is usually seen as an indicator that not all is/was
well with one’s health. So two persons who have the same change in “true” health, but one
has been to the hospital and the other has not, could report different changes in health. If
then SRCH is associated stronger with hospital visits than SRH this could also mean that
the hospital visit influences the reporting behavior of SRCH, but not of SRH. The stronger
association would then not be a measure of external reliability, but rather of measurement error
caused by an unusual health treatment which signals worsening health to the respondent even
if true health change is the same.
Benı́tez-Silva & Ni (2008) also compare SRCH and SRH. They agree with the results of
Gunasekara, Carter & Blakely (2012) that SRCH is more consistent than computed changes in
SRH. Vaillant & Wolff (2012) point in the comparison of computed changes of SRH and SRCH
to the strong state-dependence of SRH. This could be one explanation of the better performance
of SRCH which can capture smaller changes and does not suffer from a ceiling-effect. The
state-dependence and inferiority of computed changes in SRH (CCSRH) to SRCH is one of the
problems I need to address in my methodological approach.
Idler & Kasl (1995) show in a very important study that SRH is not only predictive of
mortality, but also functional ability, meaning the way the body works in every-day life. They
do not find gender differences in the predictive power (Idler & Kasl 1995, S349). Similarly
Manor, Matthews & Power (2001) find that SRH is associated with longstanding (non-mortal)
illnesses, and other specific health problems. A change in SRH has also been associated with
new illnesses, more visits to a physician, and psychosocial factors of well-being (Rodin &
McAvay 1992).
A different approach at estimating “true” health is used by Shmueli (2003). In this study
confirmatory factor analysis, in particular a multiple indicator, multiple causes model (MIMIC)
is used to estimate a latent health variable and to differentiate between systematic and random
measurement error. The author uses the short-form-36 health questionnaire (SF-36), an index
of chronic illnesses and a subjective visual health reporting variable. Deviant reporting behavior
is found for age (older people rate their health as better) and gender with regard to the SF-36.
In this case women understate their health. The author argues that his method is superior to
most studies so far, because those studies just assume that they are estimating latent health
when in fact they are using just a (supposedly) more reliable (objective) health indicator. In all
these cases it is a theoretical assumption which is not tested.

xxx
View publication stats