Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
Essays on health and well-being
(USC Thesis Other)
Essays on health and well-being
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
ESSAYS ON HEALTH AND WELL-BEING
by
Jacqueline Smith Zweig
A Dissertation Presented to the
FACULTY OF THE USC GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(ECONOMICS)
May 2011
Copyright 2011 Jacqueline Smith Zweig
ii
ACKNOWLEDGEMENTS
I want to thank my advisors, Richard A. Easterlin and John C. Ham, who have guided
and supported me throughout graduate school and the process of writing this dissertation.
I am also thankful to the members of my guidance committee and my dissertation
committee, Tatiana Melguizo, Roger Moon, and John Strauss for their valuable advice. I
appreciate all of the administrative support and guidance provided by Young Miller and
Morgan Ponder. I am grateful to my husband, Josh Zweig, my parents, Rebecca and Jeff
Smith, my brother, Jeremy Smith, and my loyal friends who have been there for me every
step of the way.
iii
TABLE OF CONTENTS
Acknowledgements ii
List of Tables v
List of Figures vii
Abstract ix
Chapter 1: Explaining Health Satisfaction and Life Satisfaction in East and
West Germany Post-Reunification 1
1.1 Introduction 1
1.2 Background 4
1.3 Literature Review 11
1.3.1 Health 11
1.3.2 Health Care 15
1.3.3 Personal Economic and Macroeconomic Factors 17
1.3.4 Additional Factors 20
1.4 Data 21
1.5 Methods 31
1.5.1 Empirical Specification 31
1.5.2 Econometrics Issues 33
1.6 Results 41
1.6.1 Health Satisfaction 41
1.6.2 Life Satisfaction 48
1.7 Validity of Results 53
1.7.1 Selection 53
1.7.2 Comparison to Subsets of the Instrumental Variables 55
1.7.3 Additional Evidence 56
1.8 Conclusion 59
Chapter 2: Are Women Happier than Men? Evidence from the Gallup World
Poll 61
2.1 Introduction 61
2.2 Literature review 62
2.3 Data 66
2.4 Methods 69
2.4.1 Econometric Strategy 69
2.4.2 Econometric Issues 71
2.5 Results 74
2.5.1 Across-country Patterns 74
2.5.2 Within-country Results 85
2.5.3 Across-country Patterns after including control Variables 92
iv
2.6 Conclusion 96
Chapter 3: Is California Polluting Children’s Minds? The effect of Air
Pollution on Academic Performance 99
3.1 Introduction 99
3.2 Mechanisms by which Pollution Could Affect Academic
Performance 102
3.2.1 Evidence from the Economics Literature 102
3.2.2 Evidence from the Epidemiological Literature 108
3.3 Methods 112
3.4 Data 114
3.5 Results and Discussion 118
3.6 Conclusion 127
References 128
Appendices 141
Appendix 1.A: First stage results for Health Satisfaction and Life
Satisfaction 141
Appendix 1.B: Coefficient on Self-reported Health in 2SLS with
Subsets of Instrumental Variables 144
Appendix 2.A: Countries Included in the Analysis 145
Appendix 2.B: Explanatory Variables in the Analysis 146
Appendix 2.C: Percent of Variables Missing when Economic
Factors are Included in the Analysis 149
Appendix 2.D: OLS Regressions of whether a Respondent did not
Answer the Economic Questions 152
Appendix 2.E: Mean Difference in Life Satisfaction between Men
and Women from each Specification 153
Appendix 2.F: Number of Observations in Regressions 157
Appendix 3: Data Sources and Variables 161
v
LIST OF TABLES
Table 1.1: Mean Health Satisfaction in East and West Germany,
1990 and 1999 6
Table 1.2 Mean Health Satisfaction in East and West Germany by
Age Group, 1990 and 1999 8
Table 1.3 Mean Life Satisfaction in East and West Germany by Age
Group, 1990 and 1999 10
Table 1.4 Descriptive Statistics 24
Table 1.5 First stage results for Health Satisfaction and Life Satisfaction 37
Table 1.6 First Stage Regressions with Satisfaction with Housing as the
Dependent Variable 40
Table 1.7 Health Satisfaction Regression Results: First Difference and First
Difference‐2SLS 43
Table 1.8 Life Satisfaction Regression Results: First Difference and First
Difference‐2SLS 49
Table 1.9 Test for Non-random Attrition 54
Table 2.1 Descriptive Statistics: Average Female-Male Difference 69
Table 2.2 Distribution of Female-Male Differences in Life Satisfaction by
Geographic Region 76
Table 2.3 Average Female-Male Difference in Life Satisfaction by
Geographic Region 77
Table 2.4 Average Female-Male Difference in Life Satisfaction by Education
and Geographic Location 82
Table 2.5 Average Female-Male Difference in Life Satisfaction by Religion 83
Table 2.6 Comparison of the Percent with Elementary Education from
Gallup Data and UNESCO data 85
vi
Table 2.7 Average Female-Male Difference in Life Satisfaction from the Four
Specifications by the Results of the Regressions without Controls 92
Table 2.8 Average Female-Male Differences in Explanatory Variables by
Female-Male Difference in Life Satisfaction 93
Table 2.9 Average Female-Male Difference in Life Satisfaction from Four
Specifications by Geographic Region 94
Table 3.1 Descriptive Statistics across California Schools, 2002-2008 117
Table 3.2 Correlation Coefficients across California Schools, 2002-2008 118
Table 3.3 Effect of Air Pollution on the Percent of Students at Least
Proficient in Mathematics – Grade-School and Year Effects 120
Table 3.4 Effect of Air Pollution on the Percent of Students at Least
Proficient in English/Language Arts – Grade-School and Year
Effects 123
Table 3.5 Effect of Air Pollution on Academic Performance using only
Monitors Functioning throughout the Period − Grade-School and
Year Effects 126
vii
LIST OF FIGURES
Figure 1.1 Mean Health Satisfaction (0 to 10) in East and West Germany 5
Figure 1.2 Health Satisfaction in East and West Germany – 24 to 44 years old
in 1990 6
Figure 1.3 Health Satisfaction in East and West Germany – 45 to 70 years old
in 1990 7
Figure 1.4 Difference in Health Satisfaction (East Germany-West Germany)
by Age Group 7
Figure 1.5 Difference in Life Satisfaction (East Germany-West Germany)
by Age Group 9
Figure 1.6 Difference in Health Satisfaction and Ratio of Mortality Rates –
24 to 44 years old in 1990 12
Figure 1.7 Difference in Health Satisfaction and Ratio of Mortality Rates –
45 to 70 years old in 1990 12
Figure 1.8 Difference in Health Satisfaction and Ratio of Household Income
− 24 to 44 years old in 1990 18
Figure 1.9 Difference in Health Satisfaction and Ratio of Household Income
− 45 to 70 years old in 1990 18
Figure 1.10 Difference in Health Satisfaction and Self-reported Health − 24
to 44 years old in 1990 30
Figure 1.11 Difference in Health Satisfaction and Self-reported Health − 45
to 70 years old in 1990 30
Figure 1.12 Health Satisfaction and Worries about Finances by Education
Level – East Germany, 24 to 44 years old in 1990 57
Figure 2.1 Mean Life Satisfaction (0 to 10) for Men and Women 75
Figure 2.2 Female-Male Difference in Life Satisfaction and Log
GDP per Capita 78
Figure 2.3 Female-Male Difference in Life Satisfaction and Mean Life
Satisfaction 79
viii
Figure 2.4 Female-Male Difference in Life Satisfaction and Female Labor
Force Participation Rate (%) 80
Figure 2.5 Female-Male Difference in Life Satisfaction and Percent of Seats
Held by Women in National Parliament 81
Figure 2.6 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics 87
Figure 2.7 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics including Ideal Children 88
Figure 2.8 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics and Life Circumstances 89
Figure 2.9 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristic, Life Circumstances, and Economic Factors 91
Figure 2.10 Impact of the Control Variables on the Female-Male Difference
in Life Satisfaction and Log GDP per Capita 95
.
ix
ABSTRACT
This dissertation is comprised of three chapters that use microeconometric techniques to
investigate the factors that affect people’s well-being. In the first two chapters, well-
being is defined as life satisfaction or health satisfaction. The first chapter explores how
the movement from socialism to capitalism affected the life satisfaction and health
satisfaction of East Germans relative to West Germans after reunification. The second
chapter examines whether women are happier, less happy, or equally happy as men in
countries at various stages of development. The third chapter examines whether
pollution affects the academic performance of school children; their academic
performance and achievements will have important implications for their future well-
being.
1
CHAPTER 1: EXPLAINING HEALTH SATISFACTION AND LIFE
SATISFACTION IN EAST AND WEST GERMANY POST-
REUNIFICATION
1.1 INTRODUCTION
The former German Democratic Republic (GDR, also referred to as East
Germany) experienced political, economic, institutional, and social changes after
reunification with the Federal Republic of Germany (FRG, also referred to as West
Germany) in October 1990. In this paper, the German Socioeconomic Panel (SOEP) data
are used to evaluate how self-reported health and other components of the transition,
including economic circumstances, health care, and stress, affected the trends in health
and life satisfaction in East Germany relative to West Germany.
1
The first contribution of
this paper is to show the effect of self-reported health on health satisfaction and life
satisfaction. In doing so, I investigate the cause of the decline in health satisfaction for
young East Germans as compared to young West Germans. The results are then
compared to the results for the older age group to determine why there was a relative
decline (between East and West Germany) in health satisfaction for the young, but not for
the old. The final analysis in this paper aims to determine whether changes in health,
health care, or economic circumstances contributed to the trends in life satisfaction after
reunification.
Health satisfaction is one of the most important domains of life satisfaction, a
commonly used measure of well-being. It incorporates both objective health
circumstances and subjective perceptions of those circumstances. Evaluating the trends in
1
The data used here were made available by the German Socio-Economic Panel Study (2004) at the
German Institute for Economic Research (DIW), Berlin.
1
2
the perception of people’s health, rather than an “objective” measure of health, like
mortality, can provide a different policy perspective. One can gain insight into whether
there are other factors, besides objective changes to health, that contribute to people’s
satisfaction with their health.
The reunification of East and West Germany is an ideal setting to evaluate the
effect of health, and the transition from socialism to capitalism more generally, on health
satisfaction and life satisfaction. Unlike with other transition countries of Eastern
Europe, East Germany has a valid comparison group: West Germany. The East German
infrastructure was replaced by the already established West German system. The two
regions have the same currency and language. It is not obvious ex-ante how the
reunification of East and West Germany should affect trends in health satisfaction. The
East Germans benefited from reunification in several ways including having higher
incomes and access to better medical technology. On the other hand, the social safety net
to which East Germans had become accustomed was eliminated. Higher unemployment
rates, uncertainty regarding personal economic circumstances, and having to adapt to new
institutional and political systems could possibly have led to higher stress and a decline in
health satisfaction. The reunification of Germany, therefore, provides a unique setting to
determine how health and other components of the transition impacted people’s well-
being.
The second contribution in this paper is methodological. The happiness literature
has appropriately used fixed effects to control for unobserved factors that may drive both
the independent and dependent variables in a happiness equation. This literature has not
2
3
addressed two important issues that arise in the estimation of fixed effect models. First,
using such a model will accentuate bias in the estimated coefficients due to measurement
error in the explanatory variables (Griliches and Hausman, 1986). To address this issue I
use an instrumental variable (IV) approach; to the best of my knowledge this is one of the
first applications of an IV approach in an empirical investigation of the factors driving
subjective well-being with individual fixed effects.
2
Second, one can either handle the
fixed effects by taking first differences or by taking deviations from the mean; what has
not been widely recognized in this literature is that taking deviations from the means
places strong restrictions on the stochastic pattern of the explanatory variables and
instruments, i.e., strict exogeneity (Wooldridge, 2002, p. 284). Below I discuss this issue
in more detail and show that these restrictions are not realistic for my empirical model.
To the best of my knowledge this issue has not been addressed previously in the
happiness literature.
The outline of the paper is as follows. The next section includes background
information and stylized facts on health satisfaction and life satisfaction in East and West
Germany during the transition. In section 1.3 I present three possible explanations for
why health satisfaction declined in East Germany relative to the West Germany for the
younger age group: changes in health, health care, and economic circumstances. The
explanations are supported by previous research. The data are described in section 1.4
and the empirical strategy is presented in section 1.5. The results for health satisfaction
2
See, however, Ferrer-i-Carbonell and Frijters (2004) for a general discussion and examination of
econometric problems in the happiness literature.
3
4
and life satisfaction are in section 1.6. Health satisfaction declined in East Germany
relative to West Germany for the younger age group because of differences in health. I
provide evidence that stress had a direct and indirect effect on health satisfaction and that
improvements in income protected the younger age group against some of the stress
related to the transition. If the young East Germans had the same trend in health as the
older East Germans, then their health satisfaction would have been an average of 0.20
units higher than it was in 1999 and their life satisfaction would have improved by 0.08
units, ceteris paribus. If the young East Germans had the same trend in health as the
young West Germans, then their health satisfaction would have been an average of 0.25
units higher than it was in 1999 and their life satisfaction would have improved by 0.10
units. Allowing for the econometric issues discussed above does have an effect on the
estimated coefficients. The qualitative interpretations, however, are similar across
specifications. Section 1.7 contains a discussion of the robustness of my results, followed
by the conclusion in section 1.8.
1.2 BACKGROUND
Mean health satisfaction declined for both East and West Germany, but it declined more
for East Germany (see Figure 1.1). There is a statistically significant 0.36 unit decline in
health satisfaction in East Germany between 1990 and 1999. The health satisfaction
question in this study is:
“How satisfied are you with the following aspects of your life? Please give a
rating on the scale for each aspect: If you are completely dissatisfied, mark '0,' if
you are completely satisfied, mark '10.' If your feelings are mixed, give a rating
somewhere in between. How satisfied are you with your health?”
4
5
The statistics in Table 1.1 indicate that the difference between East and West Germany in
1990 is not significant, while the difference in 1999 is significant.
Figure 1.1 Mean Health Satisfaction (0 to 10) in East and West Germany
Means are computed using weights provided by SOEP. East and West Germany are determined by location
in 1989.
Since younger age groups are usually considered the “winners” of the transition, I
present the trends by age in Figures 1.2 and 1.3. Figure 1.4 contains the relative trends in
mean health satisfaction, computed as East Germany minus West Germany, for those
who were between 45 and 70 years old in 1990 (older group) compared to those between
24 and 44 years old in 1990 (younger group). The relative decline in East Germany
compared to West Germany is more dramatic for those younger than 45 than for those 45
and older. The trends for older West Germans are similar to the trends for older East
5.8
6
6.2
6.4
6.6
6.8
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Health Satisfaction (0 to 10)
East Germany West Germany
5
6
Germans. Younger East Germans, on the other hand, experienced a decline in health
satisfaction compared to younger West Germans. These observations are confirmed in
Table 1.1: Mean Health Satisfaction in East and West Germany,
1990 and 1999
East
Germany
West
Germany
East Germany-
West Germany
1990 6.63 6.59 0.04
1999 6.27 6.45 -0.18**
1999-1990 -0.36** -0.14** -0.22**
Means are computed using weights provided by SOEP. East and West
Germany are determined by location in 1989. ** significant at 1%;
* significant at 5%; + significant at 10%.
Figure 1.2 Health Satisfaction in East and West Germany – 24 to 44 years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
6.5
6.6
6.7
6.8
6.9
7
7.1
7.2
7.3
7.4
7.5
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Health Satisfaction (0 to 10)
East Germany West Germany
6
7
Figure 1.3 Health Satisfaction in East and West Germany – 45 to 70 years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
Figure 1.4 Difference in Health Satisfaction (East Germany-West Germany) by Age
Group
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
5.2
5.4
5.6
5.8
6
6.2
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Health Satisfaction (0 to 10)
East Germany West Germany
-0.6
-0.4
-0.2
0
0.2
0.4
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Health Satisfaction (East‐West)
24 to 44 years old 45 to 70 years old
7
8
Table 1.2 where the differences between the younger age group and the older age group
in 1990 and 1999 are presented. The difference between the younger group and the older
group is statistically significant in 1990, but not in 1999. By 1999, health satisfaction for
the younger East Germans had declined 0.69 units, but it only declined 0.28 units for the
older East Germans. Young East Germans’ health satisfaction declined by 0.44 units
more than young West Germans’ health satisfaction, while there was only a difference of
0.02 for the older age groups. The trends in Figures 1.2 and 1.3 and corresponding
statistics in Table 1.2 suggest that people younger than 45, were the “losers” in terms of
health satisfaction during the transition.
Table 1.2 Mean Health Satisfaction in East and West Germany by
Age Group, 1990 and 1999
East
Germany
West
Germany East-West
1990
Younger than 45 7.37 7.19 0.18**
45 and older 5.91 6.08 -0.17+
Younger - older 1.46** 1.11** 0.35**
1999
Younger than 45 6.68 6.94 -0.26**
45 and older 5.63 5.82 -0.19*
Younger - older 1.05** 1.12** -0.07
1999-1990
Younger than 45 -0.69** -0.25** -0.44**
45 and older -0.28** -0.26** -0.02+
Younger - older -0.41** 0.01 -0.42**
Means are computed using weights provided by SOEP. East and West
Germany are determined by location in 1989. Age group is based on age in
1990. ** significant at 1%; * significant at 5%; + significant at 10%.
8
9
One might expect that the trends in health satisfaction and life satisfaction would
have the same pattern. Easterlin and Plagnol (2008) show that the reunification led to an
initial decline and then slow recovery in life satisfaction for East Germany, which is
similar to the patterns found in other transition countries (Easterlin, 2009). In Figure 1.5,
I present the difference in life satisfaction by age group (see Table 1.3). The finding that
health satisfaction is worse for the younger age group is in contrast to the finding for life
satisfaction during the transition, where the younger age group and the older age group
have similar relative trends.
Figure 1.5 Difference in Life Satisfaction (East Germany-West Germany) by Age Group
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
Given that the relative decline in health satisfaction is more pronounced for the
younger age group and in contrast to what one might expect, these two groups are
analyzed separately in the rest of the paper. East Germans between 24 and 44 years old
-1.6
-1.4
-1.2
-1
-0.8
-0.6
-0.4
-0.2
0
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Life Satisfaction (East-West)
24 to 44 years old 45 to 70 years old
9
10
in 1990 were old enough to have experienced the economic, health, and labor conditions
before reunification and should have been in the labor force by the end of the study
period. The group between 45 and 70 would have been well entrenched in the pre-
unification system, and moving towards retirement during the transition period. In
addition, these age groups have been identified as having different health patterns in prior
studies on East Germany (Nolte, Shkolnikov, & McKee, 2000ab; Nolte, Britton, &
McKee, 2003).
Table 1.3 Mean Life Satisfaction in East and West Germany by Age
Group, 1990 and 1999
East
Germany
West
Germany
East Germany-
West Germany
1990
Younger than 45 6.55 7.37 -0.82**
45 and older 6.62 7.28 -0.66**
Younger - older -0.07 .09* -0.16*
1999
Younger than 45 6.37 7.01 -0.64**
45 and older 6.49 7.04 -0.55**
Younger - Older -0.12 -0.03 -0.11+
1999-1990
Younger than 45 -0.18** -0.36** 0.18
45 and older -0.13 -0.24** 0.11
Younger - older -0.05 -0.12 0.07
Means are computed using weights provided by SOEP. East and West Germany are
determined by location in 1989. Age group is based on age in 1990. ** significant at
1%; * significant at 5%; + significant at 10%.
10
11
1.3 LITERATURE REVIEW
Although the primary focus of this paper is the contribution of health to health
satisfaction and life satisfaction, the economic, social, and institutional changes in East
Germany after reunification may also have contributed to the decline in health
satisfaction (and recovery of life satisfaction) of younger East Germans relative to
younger West Germans. These factors, as well as health, could also explain the
differences in the relative health satisfaction trends across age groups if they affected the
young and the old differentially, or if the trends in these factors differed across age
groups. Health satisfaction can be thought of as a function of demographic
characteristics, health, health care, individual economic circumstances, macroeconomic
factors, aspirations, and personality. I discuss below how these factors may have
contributed to health satisfaction and life satisfaction.
1.3.1 HEALTH
There is evidence that health declined in East Germany relative to West Germany
prior to and during the transition period, which could explain why health satisfaction
declined in East Germany relative to West Germany. If relative health declined more for
the younger group than the older group, this could explain why relative health satisfaction
declined more for the younger group than the older group. In Figure 1.6, I plot the East-
West ratio in mortality rates obtained from the Federal Health Monitoring System
(2010a) for individuals between 18 and 44 against the difference in health satisfaction.
There was an increase in the mortality rates in East Germany relative to West Germany
immediately prior to and during the transition. The gap in mortality rates persisted
11
12
Figure 1.6 Difference in Health Satisfaction and Ratio of Mortality Rates – 24 to 44
years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990. Mortality rates are from the Federal Health
Monitoring System (2010a). For consistency, the mortality rates are only plotted through 1997, as the way
they were reported changed after 1997.
Figure 1.7 Difference in Health Satisfaction and Ratio of Mortality Rates – 45 to 70
years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990. Mortality rates are from the Federal Health
Monitoring System (2010a). For consistency, the mortality rates are only plotted through 1997.
1
1.2
1.4
1.6
1.8
2
-0.6
-0.4
-0.2
0
0.2
0.4
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Ratio of Mortality Rates
(East/West)
Health Satisfaction (East-West)
Health Satisfaction Ratio of Mortality Rates
1
1.2
1.4
1.6
1.8
2
-0.6
-0.4
-0.2
0
0.2
0.4
1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Ratio of Mortality Rates
(East/West)
Health Satisfaction (East-West)
Health Satisfaction Mortality Rate
12
13
through 1997. For the older age group presented in Figure 1.7, the increase in relative
mortality was more gradual.
There was also a substantial increase in mortality rates from heart disease in East
Germany between 1990 and 1993, followed by a slow decline. However, the mortality
rates from heart disease in East Germany in 1999 still had not returned to the 1990 levels.
Over the same period, West Germany experienced a steady decrease in heart disease
mortality rates (Müller-Nordhorn, Rossnagel, Mey, & Willich, 2004). One explanation
for the mortality trends is that the stress of the transition and the loss of the social safety
net caused a decline in health. The decline in health immediately prior and during the
transition may also have contributed to a slower recovery of life satisfaction for the
younger age group than if health had not declined.
There are several studies on the changes in health after the re-unification of East
and West Germany. Most studies provide evidence that health declined immediately after
reunification, especially in stress-related illnesses. Nolte et al. (2000ab) show that there
was an initial spike in mortality rates in East Germany immediately after reunification,
but that the age-standardized death rates for East Germans decreased overall between
1992 and 1997. In absolute terms, they decreased more than West Germany, which they
attribute to changes in diet caused by increased availability of fruit and imported oil.
However, the East-West gap in mortality rates still existed in 1997.
Riphahn and Zimmerman (2000) examine the mortality increase in East Germany
between 1989 and 1991. They present several theoretical models that could explain the
increase in mortality and their evidence supports a psycho-social stress theory of
13
14
mortality developments. First, the authors show that there was an increase in deaths
related to circulatory and health problems, especially for men. Second, using the SOEP
data, the authors show that worries about economic conditions had a negative and
significant effect on health satisfaction in East Germany between 1990 and 1994. Finally,
the authors show that those who were unemployed had a more negative trend in health
satisfaction that those who were employed. The analysis conducted in this paper is
similar to Riphahn and Zimmermann’s work. However, I examine both East and West
Germany rather than just East Germany, look at the differences through 1999, and use a
different empirical methodology. I also extend their findings to show how this decline in
health played a role in the recovery of life satisfaction. Health has consistently been
shown to be a determinant of life satisfaction (Dolan, Peasgood, & White, 2008).
Luschen, Niemann, and Apelt (1997) evaluate self-reported health, rather than
mortality rates, of East and West German men and women using a survey of 2,574
respondents in 1992. They conclude that the East German men have the highest self-
reported health in 1992 despite difficult working conditions and insecure jobs. West
German women have higher self-reported health status than East German women. They
also find that the elderly in East Germany have lower self-reported health status
compared to the elderly in West Germany.
Even if health changed in East Germany relative West Germany it is not
necessarily the case that a change in health would cause a change in health satisfaction or
life satisfaction. For example, Easterlin (2001) shows that more income does not make
people happier because aspirations increase as income increases. In another paper,
14
15
Easterlin (2005a) demonstrates that people’s aspirations for goods increase to the same
extent as their attainment of goods. He finds that aspirations for non-pecuniary goods,
like marriage, are less impacted by experience. Brickman, Coates, and Janoff-Bulman
(1978) claim that there is complete adaptation to changes in health. The authors examine
changes in life satisfaction from a change in disability status. However, after reviewing
the data they cite, it does seem that people with a disability are less happy than those
without a disability. In a recent article, Oswald and Powdthavee (2008), using a fixed
effect model with current disability and lagged disability included as independent
variables, estimate a 30% adaptation for a severe disability and 50% for a minor
disability. Lucas (2007), on the other hand, finds little to no adaptation to disability
status.
1.3.2 HEALTH CARE
The replacement of the socialized medical system of the former GDR by the
market-based system of the FRG on January 1, 1991 may have had an impact on health
satisfaction or life satisfaction in East Germany relative to the West Germany. Health
care could affect health satisfaction or life satisfaction indirectly through its impact on
health or directly through the benefit of knowing that quality care is available if needed.
The health care system in West Germany was considered more technologically advanced
than that in East Germany and the number of medicines offered in East Germany
increased after reunification. The chief complaints of citizens of the former GDR were
shortages of imported drugs, shortages of supplies, long wait times for elective surgery,
and physically deteriorating hospitals. One survey indicated that sixty one and a half
15
16
percent of people surveyed in Dresden in mid-1990 were willing to make out-of-pocket
expenses for better health care (Scharf, 1999).
Nolte, Scholz, Shloknikov, and McKee (2002) evaluate the contribution of
medical care to life expectancy in East and West Germany as well as Poland by
evaluating amenable mortality rates. Amenable mortality rates are the mortality rates for
a group of diseases that are responsive to medical care or health policy. Overall, the
authors find that as medical care improved in East Germany, life expectancy increased.
Between 1991/1992 and 1996/1997 life expectancy from birth to age 75 increased by
0.91 years for women and 1.41 years for men in East Germany. Over the same time
period, they estimate that reductions in death rates from amenable conditions accounted
for 25% of the increase in temporary life expectancy for women. This improvement in
the quality of medical care could have an indirect effect on health satisfaction through
improvements in health or a direct effect on health satisfaction if people are more
satisfied with their health from the knowledge that they have access to better medical
care. If the improvement in health care affected the older age group in East Germany but
not the younger age group, then that could explain why health satisfaction declined for
the younger age group, but not for the older age group.
There is also evidence to suggest that the infrastructural changes in health care in
East Germany after reunification could have led to a decrease in health satisfaction in the
East relative to the West. The West German system was characterized by competitive
sickness funds, similar to U.S. insurance companies, and autonomous doctors. The East
German system, however, was run by the government and characterized by omnipresent
16
17
ambulatory care through local clinics and occupation-based clinics.
3
The clinics were
very popular with East Germans and provided significant preventive medical services,
including vaccines and physiotherapy (Jorgen, 1997, pp. 34-37). The East German
system also boasted a streamlined system between ambulatory care and hospital care. In
contrast, in West Germany pre- and post-reunification as well as in East Germany post-
reunification, hospital services were separate from ambulatory care. Hospital staff were
not permitted to treat their patients outside of the hospital and the reverse was true for
office physicians. This led to complaints about duplication of services and discontinuity
of care between treatment in a hospital and pre- and post-hospital treatment (Stone, 1991,
pp. 401-412). The movement of physicians from polyclinics to private practices led to
additional problems. It was noted in one article that the service rate paid by the sickness
funds was 60% of that in West Germany even though cost of running it, i.e., supplies and
equipment, was nearly identical, which made it difficult for these physicians to have
adequate supplies (Swami, 2002 pp. 333-358). Given that there were benefits and
drawbacks to the new health care system, it is not entirely obvious how changes in the
health care system would have affected health satisfaction in East Germany.
1.3.3 PERSONAL ECONOMIC AND MACROECONOMIC FACTORS
East Germans experienced an increase in income after reunification which should
have allowed East Germans to purchase better health inputs or at least know that they
could buy them if necessary. This could cause an increase in health satisfaction or life
3
The outpatient clinics were called polyclinics and consisted of at least six specialties: general medicine,
obstetrics and gynecology, pediatrics, dentistry, and internal medicine.
17
18
Figure 1.8 Difference in Health Satisfaction and Ratio of Household
Income − 24 to 44 years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group based on age in 1990.
Figure 1.9 Difference in Health Satisfaction and Ratio of Household
Income − 45 to 70 years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group based on age in 1990.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-0.6
-0.4
-0.2
0
0.2
0.4
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Ration of Household Income
(East/West)
Health Satisfaction (East-West)
Health Satisfaction Household Income
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
-0.6
-0.4
-0.2
0
0.2
0.4
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Ratio of Household Income
(East/West )
Health Satisfaction (East-West)
Health Satisfaction Household Income
18
19
satisfaction. I plot the ratio of household income in East Germany relative to West
Germany for each age group in Figures 1.8 and 1.9. Although average income is lower in
East Germany than in West Germany, the relative gains during the transition for East
Germany are more dramatic for the older age group compared to the younger age group.
If income has a positive effect on health satisfaction, it could be that income played a
protective role against a more dramatic decline in health satisfaction.
Frijters, Haisken-DeNew, and Shields (2005) estimate the effect of income on
health satisfaction in the context of reunification of Germany. They use a fixed effects
model using the same data that is used in this paper and find that an increase in income
did increase health satisfaction for men, although the effect is small. The authors use the
“natural experiment” of reunification based on the assumption that reunification was
unanticipated and that there was “no obvious immediate change in other health
satisfaction producing circumstances” (p. 999). As discussed in the previous section, the
East German health care system was replaced by the West German health care system.
This paper builds on their work by controlling for these changes using proxy variables for
health care obtained from the Federal Health Monitoring System. In addition, the effect
of income on health satisfaction is evaluated separately for the older and younger age
groups.
In a cross-country analysis, Deaton (2008) evaluates the effect of GDP per capita
and age on health satisfaction. Although GDP per capita is positively associated with
health satisfaction, the coefficient on the growth rate of GDP per capita between 1990
and 2005 is negative and significant, suggesting that areas of rapid growth experienced a
19
20
decline in health satisfaction. Furthermore, health satisfaction deteriorates more rapidly
over the life cycle for low- and middle-income countries compared to high-income
countries, leading one to think that higher income may protect against the effect of aging
on health.
During the transition from a socialist to a capitalist state, unemployment increased
in East Germany. Gordo (2006) estimates the effects of short- and long-term
unemployment on health satisfaction. She also uses the SOEP data set to show that long
term unemployment has a negative and significant effect for both men and women, but
that short term unemployment only negatively affects men.
Frijters, Haisken-Denew, and Shields (2004) use various econometric techniques
to investigate the determinants of life satisfaction in Germany between 1984 and 1999.
Their results suggest that income has a positive effect on life satisfaction and that
unemployment leads to lower life satisfaction. Easterlin and Plagnol (2008) show that
relative income in Germany was more closely associated with life satisfaction patterns
than absolute income. They also note than the unemployment rate was inversely related
to life satisfaction.
1.3.4 ADDITIONAL FACTORS
Demographic characteristics, aspirations, or personality may have contributed to
the relative decline in health satisfaction for the young East Germans or their recovery of
life satisfaction. Demographic characteristics may be important if they are changing
differentially for East and West Germany. Personality and aspirations may be a
contributing factor; if someone is generally optimistic or pessimistic and this differs
20
21
between East and West Germany, then it could affect the trends in health satisfaction. I
control for individual personality using individual fixed effects. There are likely
determinants of life satisfaction that are not determinants for health satisfaction. For
example, occupation and job characteristics, as well as whether or not a person has
children, could contribute to life satisfaction, but are less obvious factors of health
satisfaction. If these changed during the transition period (e.g., movement to more
entrepreneurial occupations), then they may have contributed to the initial decline and
recovery of life satisfaction during the study period.
1.4 DATA
In order to investigate the contribution of health, economic circumstances, and
health care to the trends in health satisfaction and life satisfaction, I use the SOEP survey,
which is a panel survey that started in 1984 and follows the same people over time with
refreshment groups. For a detailed discussion of the data, see Haisken-DeNew and Frick
(2005). East Germans entered the sample in 1990, but the health question was not asked
until 1992. Since this paper focuses on the transition period, the surveys from 1992
through 1999 are included in the analysis. Respondents are grouped into East and West
Germany based on where they lived in 1989. I exclude the foreigner sample and those
younger than 24 or older than 70 in 1990. One of the benefits of this data set is that it is
possible to control for unobserved time-constant individual heterogeneity, like
personality, with individual fixed effects. This is particularly important in satisfaction
equations.
21
22
Table 1.4 contains the descriptive statistics of the variables in the analysis by age
group for East and West Germany. The statistics are calculated for the study period 1992
to 1999. The independent economic and demographic variables are marital status, log
household income per capita, and employment status. Single, divorced, and widowed are
grouped together into the not-married category. Age squared controls for life cycle
changes in health satisfaction. Because the empirical specification includes individual
fixed effects, age and gender will be not be identified. Log household income per capita
is the monthly net labor income of everyone in the household including pensions,
unemployment benefits, and maternity benefits etc., divided by the number of people in
the household.
4
Household income is used rather than individual wage income so that
people who are unemployed or retired are not excluded from the analysis. Dummy
variables for unemployment and retirement are also included. Both older and younger
East Germans have significantly lower household income and are more likely to be
unemployed than their West German counterparts.
In addition to personal economic circumstances, state unemployment rates are
included in the model to control for macroeconomic conditions that may affect one’s
health or life satisfaction. The unemployment rates are significantly higher in East
Germany compared to West Germany. There are some factors in the life satisfaction
regressions that are not in the health satisfaction regressions. Specifically, I include
occupation group, whether or not the person has children, and if a family member died in
4
I use the generated version of this variable, ahinc. When no answer was provided on net household
income, the SOEP used the net household income calculated from the responses to specific income
questions of all individuals in the household (“Documentation HGEN,” 2010).
22
23
the past year. These factors might contribute to job, family, and financial satisfaction, but
not necessarily to health satisfaction.
5
Self-reported health is used as a proxy for health status. The specific question is:
“How would you describe your health at present? Very Good, Good, Satisfactory, Poor,
Very Poor.” I also include two variables as proxies for stress that indicate how worried
the respondent is about various aspects of life. The questions are: “What about the
following areas: Do they worry you... your own financial situation?... general economic
development?” The responses are very worried, slightly worried, and not worried.
Self-reported health is significantly lower in East Germany than West Germany
for the younger age group. This measure of self-reported health is widely used and
evaluated in the economics literature (for example, Adams, Hurd, McFaddeen, Merrill, &
Ribeiro, 2003; Deaton & Paxson, 1998; Ettner, 1996; Salas, 2002; Smith, 1999;
Contoyannis & Jones, 2004). Despite its widespread use, some may be concerned that
self-reported health is not a good measure of a person’s true level of health, or that is too
similar to health satisfaction. In response to the first concern, it has been demonstrated
several times that self-reported heath predicts mortality (Burström & Fredlund, 2001;
Idler & Benyamini, 1997; Idler & Kasl, 1995; Mossey & Shapiro, 1982; van Doorslaer &
Gerdtham, 2003). In particular, Idler and Benyamini (1997) review 27 studies in U.S.
and international journals that use self-reported health to predict mortality or morbidity
and conclude that self-reported health is an independent predictor of mortality even when
socioeconomic factors or other health status variables are controlled. Self
5
When included in the health satisfaction regression, these variables are all insignificant.
23
24
Table 1.4 Descriptive Statistics
1
24 to 44 years old in 1990
West Germany (N=12695) East Germany (N=8574) East-West
Mean Std dev Min Max Mean Std dev Min Max Mean t-statistic
Health satisfaction 6.905 2.05 0 10 6.683 2.02 0 10 -0.222 7.82
Life satisfaction 7.065 1.69 0 10 6.288 1.76 0 10 -0.777 32.17
Age 38.5476.39265339.260 5.9726530.7138.31
Child 0.60.50010.662 0.47010.09514.12
Spouse died 0.001 0.04 0 1 0.002 0.04 0 1 0.000 0.78
Married 0.7170.45010.794 0.40010.07612.88
Log household income per capita 7.007 0.46 3 10 6.711 0.41 4.92 8.97 -0.295 49.57
Unemployed 0.0390.19010.138 0.34010.09924.16
Retired 0.0240.15010.030 0.17010.0052.19
Unemployment rate
rt
2
0.0940.020.040.220.168 0.030.040.220.073196.97
Occupation:
No response/NIL 0.587 0.49 0 1 0.471 0.50 0 1 -0.116 16.70
Legislators, professionals 0.092 0.29 0 1 0.102 0.30 0 1 0.010 2.50
Technicians, armed forces 0.101 0.30 0 1 0.114 0.32 0 1 0.014 3.16
Clerks, sales and service workers 0.099 0.30 0 1 0.101 0.30 0 1 0.002 0.42
Agriculture, fishery, and trade
workers 0.072 0.26 0 1 0.127 0.33 0 1 0.055 12.91
Operators, elementary
occupations 0.050 0.22 0 1 0.085 0.28 0 1 0.035 9.76
No insurance 0.005 0.07 0 1 0.002 0.04 0 1 -0.003 3.98
Private insurance 0.099 0.30 0 1 0.031 0.17 0 1 -0.069 21.16
Infant Mortality Rate
rt
10.5110.678.313.411.318 1.008.3413.200.80765.38
Life expectancy at birth
rt
75.3500.5270.976.473.658 1.3570.276.4-1.691110.45
24
25
Table 1.4 (Continued) Descriptive Statistics
24 to 44 years old in 1990
West Germany (N=12695) East Germany (N=8574) East-West
Mean Std dev Min Max Mean Std dev Min Max Mean t-statistic
Not worried about finances 0.288 0.45 0 1 0.112 0.32 0 1 -0.176 33.36
Somewhat worried about finances 0.553 0.50 0 1 0.606 0.49 0 1 0.053 7.64
Very worried about finances 0.159 0.37 0 1 0.282 0.45 0 1 0.123 21.09
Not worried about country's
development 0.0640.25010.047 0.2101-0.0175.38
Somewhat worried about country's
development 0.5820.49010.552 0.5001-0.0304.29
Very worried about country's
development 0.3540.48010.400 0.49010.0476.89
Self-reported health 3.549 0.86 1 5 3.550 0.84 1 5 0.000 0.04
Doctor visits 9.150 16.68 0 368 7.394 11.93 0 200 -1.757 8.95
Traffic accidents due to alcohol
rt-1
38.520 7.83 26 94 60.018 18.37 26 105 21.498 102.26
45 to 70 years old in 1990
West Germany (N=10655) East Germany (N=6879) East-West
Health satisfaction 5.861 2.30 0 10 5.661 2.07 0 10 -0.200 5.99
Life satisfaction 7.093 1.82 0 10 6.328 1.74 0 10 -0.765 27.87
Age 62.2887.56477960.623 7.124779-1.66514.76
Child 0.0580.23010.056 0.2301-0.0020.54
Spouse died 0.009 0.09 0 1 0.008 0.09 0 1 0.000 0.14
Married 0.7420.44010.787 0.41010.0456.93
Log household income per capita 6.864 0.44 4.81 9.80 6.652 0.36 2.48 8.49 -0.212 34.95
Unemployed 0.0330.18010.097 0.30010.06416.06
Retired 0.5550.50010.573 0.49010.0172.24 25
26
Table 1.4 (Continued) Descriptive Statistics
45 to 70 years old in 1990
West Germany (N=10655) East Germany (N=6879) East-West
Mean Std dev Min Max Mean Std dev Min Max Mean t-statistic
Unemployment rate
rt
0.0940.020.040.190.171 0.020.050.220.077202.04
Occupation:
No response/NIL 0.862 0.34 0 1 0.785 0.41 0 1 -0.078 12.97
Legislators, professionals 0.025 0.16 0 1 0.047 0.21 0 1 0.022 7.37
Technicians, armed forces 0.032 0.18 0 1 0.043 0.20 0 1 0.011 3.62
Clerks, sales and service workers 0.033 0.18 0 1 0.045 0.21 0 1 0.012 3.99
Agriculture, fishery, and trade
workers 0.021 0.15 0 1 0.040 0.20 0 1 0.018 6.68
Operators, elementary
occupations 0.027 0.16 0 1 0.041 0.20 0 1 0.015 5.08
No insurance 0.003 0.05 0 1 0.001 0.04 0 1 -0.002 2.21
Private insurance 0.062 0.24 0 1 0.002 0.05 0 1 -0.060 24.83
Infant mortality rate
rt
10.5370.708.3413.3911.338 0.988.313.20.80158.82
Life expectancy at birth
rt
75.370 0.5173.976.473.607 1.3470.276.4-1.764104.27
Not worried about finances 0.412 0.49 0 1 0.208 0.41 0 1 -0.204 29.88
Somewhat worried about finances 0.456 0.50 0 1 0.562 0.50 0 1 0.106 13.74
Very worried about finances 0.131 0.34 0 1 0.230 0.42 0 1 0.099 16.33
Not worried about country's
development 0.0710.26010.047 0.2101-0.0236.57
Somewhat worried about country's
development 0.5210.50010.490 0.5001-0.0313.98
Very worried about country's
development 0.4080.49010.462 0.50010.0547.07
Self-reported health 2.978 0.93 1 5 3.029 0.86 1 5 0.051 3.72 26
27
Table 1.4 (Continued) Descriptive Statistics
45 to 70 years old in 1990
West Germany (N=10655) East Germany (N=6879) East-West
Mean Std dev Min Max Mean Std dev Min Max Mean t-statistic
Doctor visits 16.043 23.52 0 380 12.167 16.28 0 300 -3.876 12.89
Traffic accidents due to alcohol
rt-1
39.203 7.73 26 60 60.867 18.42 26 105 21.663 92.42
1. The years included are 1992 through 1999. East German and West German samples determined by location in 1989. Population weights provided in
SOEP used to calculate statistics. 2. The notation r represents the state and t represents the year. The unemployment rate, infant mortality rate, life
expectancy and traffic accidents vary by state and year.
27
28
reported health has also been shown to predict future use of health care (van Doorslaer,
Koolman & Jones, 2004). Self-reported health may also capture some elements of health
status, like pain and suffering, that would not be captured by external observation
(Graham 2009).
Recent research recommends the use of vignettes to anchor responses to self-
reported measures (King, Murray, Salomon, & Tandon., 2004; Kapteyn, Smith, & Van
Soest, 2007; Van Soest, Delaney, Harmon, Kapteyn, & Smith 2007). Much of the work
uses vignettes to make cross-country comparisons more reliable and generally in cross-
sectional settings. Kapteyn et al. (2007) find that a large part of observed differences in
reported work disability between the Netherlands and the United States can be attributed
to different response scales in answering questions on whether they have a work
disability. Van Soest et al. (2007) evaluate the use of vignettes in the context of self-
reported drinking in college students. He concludes that the use the vignettes makes the
self-reported measures more consistent with the objective measures, but also notes that
additional tests are necessary to validate the use of vignettes when there are less obvious
objective measures to use in comparison to the subjective measures. Self-reported health
would fall into the latter case with a less obvious direct objective comparison. In a first
difference model the assumption is that a one-unit change in self-reported health has the
same meaning in East and West Germany. This assumption is more innocuous than in a
cross-section where I would need to assume that a rating of “very good” is the same in
the two regions. In addition, East and West Germany after reunification have similar
28
29
cultures, and the same language and political system, so the difference should not be as
great as it might be across other regions.
In regard to the latter concern that self-reported health and health satisfaction
measure the same thing, Deaton (2008) uses cross-country data to show that health
satisfaction does not correlate well with life expectancy, infant mortality or prevalence of
HIV/AIDS. He concludes that health satisfaction should not be used as an indicator of
health. Since self-reported health does correlate with mortality and life expectancy and
health satisfaction does not, these two variables are different measures.
Easterlin (2005b) uses the General Social Survey to evaluate how changes in self-
reported health affect health satisfaction. He finds that health satisfaction changes in the
same direction as actual health, but that it changes by little more than half of what one
might expect. He attributes this finding to the fact that health standards change with
actual health. The effect of the deterioration of health on health satisfaction is therefore
mitigated by a change in standards. Easterlin’s work provides additional evidence that
health satisfaction and self-reported health are distinct measures.
Furthermore, in a simple cross-sectional ordinary least squares (OLS) regression
of self-reported health on health satisfaction for the respondents included in this study,
the R-squared is 0.59.
6
This indicates that self-reported health does not explain 40 percent
of the variation in health satisfaction. Mean self-reported health and mean health
satisfaction are plotted in Figures 1.10 and 1.11. These graphs confirm that the trends in
health and health satisfaction are similar, but that there are portions of the trends that do
6
The coefficient on self-reported health is 1.81 and statistically significant at 1%. I reject the null
hypothesis that it is equal to 2 and that it is equal to 1.
29
30
Figure 1.10 Difference in Health Satisfaction and Self-reported Health − 24 to 44 years
old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
Figure 1.11 Difference in Health Satisfaction and Self-reported Health − 45 to 70 years
old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
-0.3
-0.2
-0.1
0
0.1
0.2
-0.6
-0.4
-0.2
0
0.2
0.4
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Self-reported Health (East-West)
Health Satisfaction (East-West)
Health Satisfaction Self-reported Health
‐0.3
‐0.2
‐0.1
0
0.1
0.2
‐0.6
‐0.4
‐0.2
0
0.2
0.4
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Self‐reported Health (East‐West)
Health Satisfaction (East‐West)
Health Satisfaction Self-reported Health
30
31
not move together – between 1997 and 1999 for the younger group and between 1992
and 1995 for the older group. In the regression results in section 1.6, other variables are
significant even when self-reported health is included in the model. If self-reported
health is the same measure as health satisfaction, then no other variables would predict
health satisfaction. Thus, I conclude that self-reported health is a different measure than
health satisfaction.
Infant mortality rates and life expectancy rates for each state in each year are
included as explanatory variables. These are both from the Federal Health Monitoring
System (2010abcd). I also include lagged infant mortality rates in case people use their
previous health care quality as a reference for evaluating current health care quality.
Infant mortality is the number of deaths per 100 individuals. The specification also
includes dummy variables for whether an individual has private insurance or no
insurance, with compulsory insurance as the reference group. Life expectancy rates are
higher in West Germany than in East Germany, and vice versa for mortality rates.
1.5 METHODS
1.5.1 EMPIRICAL SPECIFICATION
Health satisfaction is a function of demographic characteristics, economic
circumstances, health, health care, macroeconomic factors, and personality. The main
empirical strategy used in this paper is a first differenced two-stage least squares model
(2SLS).
31
32
The base model is:
(1.1)
where sat
irt
represents health (or life) satisfaction for individual i, in state r, in year t; H
irt,
represents self-reported health status; X
irt
is a vector of demographic, economic, and
health care variables; is the year trend; f
i
is an individual fixed-effect which is
potentially correlated with the explanatory variables; and E
i89
is an indicator variable
equal to one if the individual lived in East Germany in 1989. The error term is
represented by
.
As mentioned in previous sections, life satisfaction is comprised of several
domains, including health satisfaction, financial satisfaction, job satisfaction, and family
satisfaction. One can also think of life satisfaction as a function of all the factors that
make up each of the domains. For example, occupation is an important factor in job
satisfaction. Thus, life satisfaction in this paper is a function of demographic
characteristics, economic circumstances, macroeconomic factors, health, health care,
occupation, family characteristics, aspirations, and personality. The empirical
specification for life satisfaction is similar to that of health satisfaction, except that there
are more components to the X
irt
vector.
To account for life-cycle changes in health satisfaction or life satisfaction, I
include age-squared as an independent variable. Age and gender are not in the
specification because they are not identified in individual fixed effects models. A time
32
33
trend and the interaction of the East Germany dummy variable with the time trend are in
the specification. A time trend is used instead of year dummy variables because year
dummy variables interacted with East Germany in a first difference model determine
whether the average change in health satisfaction in East Germany relative to West
Germany is larger or smaller, compared to the base year difference. On the other hand,
the time trend interacted with East Germany determines whether the trend for East
Germany is different than the trend for West Germany over the time period. Since the
focus of the paper is on explaining the trend, it is more useful to include the latter. When
there are no explanatory variables, the trend should be negative and significant for the
younger age group and insignificant for the older age group.
Health satisfaction (or life satisfaction) is treated as a continuous variable rather
than an ordinal variable in this model. Ferrer-i-Carbonell and Frijters (2004) show that
OLS and ordered logit regressions with life satisfaction as the dependent variable produce
similar results. They find that the inclusion of a fixed effect is more important than the
cardinality assumption. In my model, individual fixed effects are used to capture
unobserved individual time invariant traits, like personality, that may drive both health
status and health satisfaction or life satisfaction. In what follows, the standard errors are
robust to heteroskedasticity and the errors are clustered by individual.
1.5.2 ECONOMETRICS ISSUES
The previous life satisfaction literature has noted the need to use fixed effect
models to deal with correlations between the explanatory variables and the error term.
Fixed effects methods are widely used to prevent unobserved personality or other
33
34
individual specific time-constant factors from biasing the results. However, previous
research in this literature has generally overlooked two important econometric issues that
arise in fixed effects models. First, to implement the fixed effect models in deviation
from the mean form, one must assume strictly exogeneity – i.e., the error term,
, must
be uncorrelated with current, past, and future explanatory variables. This assumption is
not necessarily reasonable in this analysis because a current shock to health (or life)
satisfaction could affect one’s future health, income, or other explanatory variables.
However, if one takes the first difference of the data instead of taking deviations from the
mean, one need only assume sequential exogeneity, i.e., the error term is uncorrelated
with current and past values of the explanatory variables. I discuss below the need to use
an instrumental variable for self-reported health to account for classical measurement
errors in this model. The validity of the instruments as well as the strict assumptions
necessary for a demeaned specification suggest that a first difference model is more
appropriate than a standard fixed effects model in this context. Therefore, the main
assumption is that a shock to health satisfaction is uncorrelated with current and past
instrumental variables, as well as other explanatory variables.
7
The second issue is the fact that fixed effects models will accentuate classical
measurement error (Altonji, 1986; Griliches & Hausman, 1986). Several articles have
shown that self-reported health is measured with error (see, for example, Crossley &
7
See Imbens and Wooldridge (2008, IRP lectures 3 and 4) or Wooldridge (2002, p.284) for a discussion of
strict and sequential exogeneity.
34
35
Kennedy, 2002). Instead of observing the true level of health,
, we observe
where
is the measurement noise. Equation (1.1) becomes:
(1.2) ∆
∆
∆
∆
∆
∆
where ∆
∆
∆
. This will lead to an inconsistent estimate of because
the error term ∆
is correlated with ∆
through the measurement error ∆
. An
instrumental variable model is an appropriate way to deal with classical measurement
error.
8
The first stage equation is:
(1.3) ∆
∆
∆
∆
∆
where
represents the instrumental variables for individual i, in state r for year t. In
the second stage equation, I use the predicted values of self-reported health, ∆
to
estimate the effect of health on health satisfaction and life satisfaction.
(1.4)
∆
∆
∆
∆
∆
∆
8
By using an IV approach I am implicitly assuming that the measurement error can be modeled as classical
measurement error, which is uncorrelated with the true value of the explanatory variable. I think that this is
a reasonable approximation in my context. I realize that it is not strictly true in my application since if
health satisfaction takes on the lowest possible value the measurement error can only be positive, while if it
takes on its highest possible value the measurement error must be negative.
35
36
The instruments for self-reported health are lagged changes in doctor visits,
∆
, and lagged number of traffic accidents associated with alcohol consumption in
state r, . The latter instrumental variable is from the Federal Health Monitoring
System (2009). The definition of a traffic accident associated with alcohol is an accident
“where at least one of those involved was under the influence of alcohol.” There are a
total of sixteen states in re-unified Germany with five of them in the former GDR. The
number of accidents due to alcohol increased from 1992 to 1994 for East Germany as a
whole and then declined through 1999. In contrast, there was a steady decline in
accidents in West Germany during this time period. The number of doctor visits in year t
is the response to the question “How many times have you been to the doctor in the past 3
months?” The self-reported health question and health satisfaction question are asked
about the individual’s situation at the time of the survey, not over the past three months.
The instrumental variables are valid if they are correlated with the change in self-
reported health and if they are uncorrelated with the error term. An increase in the
number of doctor visits in the past likely means that an individual has invested more in
her health by utilizing the health care that is available. More past doctor visits indicate
that a person obtained preventive, palliative, or curative care. This would improve one’s
health in the following period. It is also possible that an increase in the number of doctor
visits could be an indication of a chronic illness. The coefficients on the instrumental
variables from the first stage regressions are in Table 1.5; the full results are presented in
Appendix 1.A. Lagged changes in doctor visits are positively and significantly associated
with self-reported health for both age groups. Changes in traffic accidents due to alcohol
36
37
Table 1.5 First stage results for Health Satisfaction and Life Satisfaction
Dependent variable=
Self-reported health
24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
Change in doctor visits
t-1
0.0035** 0.0035** 0.0019** 0.0019**
[0.001] [0.001] [0.0004] [0.0004]
Traffic accidents due to
alcohol
rt-1
0.0031**0.0030**0.0014** 0.0013**
[0.0001] [0.0001] [0.0001] [0.0001]
Observations 1777917779 14640 14640
R-squared 0.03030.03190.0192 0.0201
F-test 39.2636.0814.86 13.87
The errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in the
SOEP are used. The base year is the difference between 1992 and 1994. The traffic accidents variable is in
levels, not differences. All control variables are included in the regressions. Standard errors in parentheses.
** significant at 1%; * significant at 5%; + significant at 10%.
at the state-by-year level can be an indicator of stress and changing health behaviors.
Lagged traffic accidents due to alcohol are correlated with changes in self-reported
health. The instruments are individually and jointly significant with the expected signs.
For the younger age group, the F-statistic is 39.26 and 36.08 for the first stage of the
health satisfaction and life satisfaction equations, respectively. I reject the null
hypotheses that the first stage equations are weakly identified. The F-statistics for the
older age group are not as strong as that for the younger age group, but still reject the null
that the first stage is weakly identified. The results indicate the bias is less than 10% for
the younger age group and between 10% and 15% of the maximal IV size for the older
age group based on the Stock and Yogo (2005) critical values.
37
38
Since a shock to current health satisfaction could affect future doctor visits, strict
exogeneity is likely to be violated, motivating the use of a first difference model rather
than a fixed effects model. Of course, for the instrument to be valid, the measurement
error in lagged reported number of doctor visits needs to be orthogonal to the
measurement error in self-reported health, and uncorrelated with true health status. Since
I use a first difference model instead of a fixed effects model, the exogeneity condition
can be written as:
1.5
0 which is equivalent to
0 -1.
Current changes in doctor visits would violate the exogeneity conditions because
health satisfaction in year t-1 could predict doctor visits in year t. Therefore, lagged
changes in doctor visits, ∆
, are used instead. Using the same logic, one could argue
that ∆
could be invalid because of the simultaneity of doctor visits and shocks to
health satisfaction in t-1. I cannot completely rule out this possibility when health
satisfaction is a dependent variable in the second stage, but it is much less problematic
when life satisfaction is the dependent variable. The fact that the question on doctor visits
is related to the three months prior to the survey assuages some of the concerns about
simultaneity. In addition, I use the change in lagged doctor visits, not just doctor visits in
year t-1. Using lagged changes in doctor visits and lagged traffic as instruments in the
first difference model, I remove the endogeneity caused by measurement error, but do not
38
39
claim that this completely solves the simultaneity problem. I focus more on addressing
the measurement error problem because it is a more substantial issue in first differences.
It would be preferable to use ∆
as an instrument, but the question was not asked
until 1991 and skipped 1993, so it would not be possible to use it as an instrument and
still interpret the results in the context of the reunification of Germany. To use a two year
lag would cause the base year to be the change between 1994 and 1995, which is four
years after reunification.
The exogeneity condition would be violated if a change in past doctor visits
affects a change in satisfaction with health through an unobserved channel other than
health. The most obvious threat to this exogeneity assumption is that respondents use the
number of past doctor visits as an indicator of past availability or quality of health care,
which could be used as a reference point in determining current health satisfaction
conditional on current health care. However, I control for changes in health care using
changes in type of insurance, changes in current and lagged mortality, as well as changes
in life expectancy. Conditional on proxies for both current and lagged changes in health
care, the number of past doctor visits should only affect health satisfaction or life
satisfaction through changes in health.
The second possible concern is that unobserved time-changing mood or
personality is correlated with the instruments and the explanatory variables. In order to
investigate the exogeneity of my instruments, I conduct a placebo test. If changes in
lagged doctor visits or lagged alcohol-induced traffic accidents are correlated with
unobserved changes in personality or mood, then these instruments should predict other
39
40
self-reported measures as well. Recall that time-invariant personality is removed by the
fixed effect. Satisfaction with housing, which is on a scale of one to ten, is regressed on
the instruments and all the control variables. If the instruments are picking up changes in
mood, then these instruments should be individually and jointly significant in a
regression with satisfaction with housing, a self-reported measure, as the outcome
variable. The results in Table 1.6 show that the instruments are individually and jointly
insignificant in all of the first stage equations. This result supports the validity of these
instruments.
Table 1.6 First Stage Regressions with Satisfaction with Housing as the Dependent
Variable
24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
Change in doctor visits
t-1
0.0022 0.0022 -0.001 -0.0011
[0.002] [0.002] [0.001] [0.001]
Traffic accidents due to
alcohol
rt-1
-0.0001 0.0000 0.0009 0.0012
[0.001] [0.001] [0.001] [0.001]
Observations 17722 17722 14584 14584
R-squared 0.0107 0.0113 0.0089 0.0099
F-statistic 0.69 0.67 1.04 1.36
The errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in the
SOEP are used. All control variables are included in the regressions. The base year is the difference
between 1992 and 1994. The traffic accidents variable is in levels, not differences. Standard errors in
parentheses. ** significant at 1%; * significant at 5%; + significant at 10%.
40
41
1.6 RESULTS
1.6.1 HEALTH SATISFACTION
There is a statistically significant downward trend for young East Germans
compared to young West Germans, while there is no significant difference for the older
age groups. The coefficient on the East German trend in the regressions for the younger
age group approaches zero and insignificance when self-reported health is included. The
variables that determine health satisfaction differ between the older and younger age
group. Household income appears to be a determinant of health satisfaction for the
younger age group, but not for the older group. The negative and significant coefficients
on concerns about personal finances show that the stress of the transition contributed to
the decline in health satisfaction. The coefficients on concerns about finances are larger
for the older age group than for the younger group. Since concerns in the East were
declining relative to the West more rapidly for the older group, this could help explain
why their health satisfaction did not decline as much.
The baseline regression (i.e., first difference without instrumental variables)
results presented in Table 1.7 columns (1) and (5) include self-reported health, a time
trend, and an indicator for East Germany interacted with the time trend.
9
These baseline
regressions show that there is no difference between East and West Germany for the
younger age group when self-reported health is included in the model. For the older age
group, there is a slight positive relative trend. I add in demographic and economic
characteristics in columns (2) and (6) and all explanatory variables in columns (3) and
(7). The coefficient on self-reported health remains stable and is similar across age
9
The coefficient on the time trend is identified because the base difference is the difference between 1994
and 1992, while the subsequent differences are one year differences.
41
42
groups when all explanatory variables are included in the model. In columns (4) and (8), I
present the 2SLS results; the coefficients on self-reported health remain significant and
increase in magnitude.
In the first difference regressions, self-reported health has a positive and
statistically significant effect on health satisfaction, and the magnitude of the coefficient
is similar across age groups. People do not completely adapt to changes in health. If there
were complete adaptation, then the coefficient on self-reported health would be zero.
Concerns about finances have a negative and significant effect on health satisfaction for
both age groups. Concerns about the country’s development are not significant in the
regressions for either age group. This finding suggests that stress about one’s personal
economic situation affects health satisfaction, but concerns about the country as a whole
do not. Income is significant for the younger age group and older age group. However,
labor force participation and the unemployment rate are only significant for the older age
group. The second stage of the 2SLS model for the younger age group is presented in
column (4) and shows that the decline in self-reported health contributed to their decline
in health satisfaction. The coefficient on health is positive, significant, and larger than in
the first-difference specification in column (3). The larger coefficient in the 2SLS
equation is what one would expect if there was classical measurement error in self-
reported health. Income still has a positive and significant effect on health satisfaction.
Concerns about finances have both a direct and indirect effect on health satisfaction. By
indirect, I mean that it affects self-reported health (i.e. significant in the first stage) which
42
43
Table 1.7 Health Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
Age squared 0.000 0.000 -0.0002 0.000 0.000 -0.0004
[0.001][0.001][0.001] [0.001][0.001][0.001]
Married -0.180+-0.190+-0.1883+ 0.0870.0830.0704
[0.105][0.105][0.106] [0.140][0.138][0.133]
Log household income per capita 0.272* 0.265* 0.2601* 0.164+ 0.147+ 0.1084
[0.119][0.117][0.119] [0.087][0.088][0.105]
Unemployed 0.0640.0840.0765 0.241*0.261**0.2849**
[0.083][0.083][0.085] [0.098][0.098][0.109]
Retired 0.2330.2340.2314 0.1040.1040.057
[0.295][0.289][0.296] [0.115][0.113][0.128]
Unemployment rate
rt
0.0070.6650.9361 -2.968+-2.674+-2.8304+
[1.430][1.497][1.591] [1.516][1.559][1.600]
No insurance -0.175 -0.192 -0.095 -0.0455
[0.292][0.304] [0.461][0.454]
Private insurance 0.015 0.0269 0.312* 0.3337*
[0.134][0.138] [0.141][0.141]
Infant mortality Rate
rt
-0.012-0.0075 0.0620.0777
[0.103][0.104] [0.102][0.109]
Infant mortality Rate
rt-1
-0.001-0.0137 -0.112-0.1125
[0.071][0.077] [0.090][0.091]
Life expectancy
rt
0.0890.0915 0.1170.0903
[0.089][0.090] [0.110][0.113]
43
44
Table 1.7 (Continued) Health Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
Somewhat worried about finances
t
-0.099*-0.0915+ -0.117*-0.1017+
[0.049][0.054] [0.047][0.052]
Very worried about finances
t
-0.159*-0.1417 -0.249**-0.2088*
[0.072][0.088] [0.080][0.101]
Somewhat worried about finances
t-1
-0.081+-0.0832+ -0.0050.0015
[0.044][0.045] [0.047][0.048]
Very worried about finances
t-1
-0.078-0.0784 -0.174*-0.1515*
[0.063][0.063] [0.071][0.077]
Somewhat worried about country's
development
t
0.0650.0658 0.0340.0666
[0.070][0.070] [0.073][0.085]
Very worried about country's
development
t
0.0180.0163 0.0850.1226
[0.081][0.081] [0.082][0.095]
Year -0.030*-0.041-0.053-0.0392 -0.069**-0.037-0.075-0.0161
[0.013][0.077][0.078] [0.085] [0.013][0.088] [0.092] [0.119]
East Germany*year -0.029 -0.045* -0.060* -0.0570+ 0.036+ 0.03 0.003 0.0187
[0.020][0.021][0.029] [0.030] [0.019][0.020] [0.027] [0.035]
Self-reported health 1.164** 1.162** 1.160** 1.2873** 1.141**1.140** 1.136** 1.5294**
[0.031][0.031][0.031] [0.331] [0.034][0.034] [0.034] [0.548]
Observations 17924177791777917779 14713146401464014640
44
45
Table 1.7 (Continued) Health Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
R-squared 0.2340.2360.2370.2347 0.2 0.201 0.204 0.1815
The notation r represents the state and t represents the year. The unemployment rate, infant mortality rate, life expectancy and traffic accidents vary by
state and year. The errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in the SOEP are used. The base year
is the difference between 1992 and 1994. Standard errors in parentheses. ** significant at 1%; * significant at 5%; + significant at 10%.
45
46
then affects health satisfaction. Lagged concerns about finances also have a negative
effect on health satisfaction.
The 2SLS results for the older age group show that health has a positive and
significant effect on health satisfaction and that the effect is underestimated in the first
difference specification. Both current and lagged increases in worries about finances
decrease one’s health satisfaction. Income does not have a significant effect on health
satisfaction for the older age group, though being unemployed and the unemployment
rate do have significant effects. The only health care variable that is significant in the
second stage is private insurance for the older age group. Few people elect to have
private insurance, so this group may have different characteristics than the rest of the
population. The other proxies for health care do not appear to be determinants of health
satisfaction. This could be because health care is not important conditional on health, or
because the benefits negate the drawbacks of the new health care system in East
Germany.
Taken together, these results indicate that health is an important determinant of
health satisfaction and that stress about personal economic circumstances contributes
directly and indirectly to one’s health satisfaction. Recall from Figures 1.6 an 1.7 that
health declined for both age groups, but the decline was more substantial for the younger
age group. A one-unit increase in self-reported health would increase health satisfaction
by 1.28 units and 1.53 units for the younger group and older group, respectively. The
significant coefficient on health indicates that there is less than complete short-term
adaption to changes in health. If the young East Germans had the same trend in self-
46
47
reported health as the young West Germans did during this study period, then their health
satisfaction would have been 0.25 units higher than it was by 1999. This would have
decreased the East-West gap by 95%.
Improvements in economic circumstances have a positive effect on health
satisfaction, which suggests that income may have protected against a more severe
decline in health satisfaction for the younger age group. Current and lagged concerns
about finances were important for both age groups. East Germans became less concerned
about personal economic circumstances relative to West Germans, but the improvements
were more dramatic for the older group. In 1999, average concerns about finances for the
young were still 0.27 units higher in East Germany than West Germany. Much of the
decline in relative concerns was due to an increase in concerns in the West, rather than a
decline in the East. In contrast, the relative decline for the older group was due to a
decrease in concerns in the East. This supports the hypothesis that stress from the
transition contributed to the decline in health satisfaction for the younger age group and
that a reduction in stress for the older age group contributed to their relatively stable
trends in health satisfaction. If the young East Germans moved from being somewhat
concerned to not concerned about finances, then their health satisfaction would have been
0.09 units higher than it was in 1999. This would have decreased the East-West gap by
34%. Taking into account the indirect effect of concerns on health satisfaction through
health, health satisfaction would have been 0.16 units higher, which would have
decreased the East-West gap by 60%.
47
48
1.6.2 LIFE SATISFACTION
As previously discussed, there was a recovery in life satisfaction in East Germany
relative to West Germany starting in 1992. The life satisfaction regressions are similar to
those in Table 1.7 for health satisfaction except that the model includes additional
covariates (i.e., occupation and family characteristics). The results are in Table 1.8.
Although income and labor force participation are important determinants of life
satisfaction, the decline in health for the younger age group was also a factor in the
recovery of life satisfaction.
The results of the first difference model are presented in columns (1) and (5). The trend
for East Germany is positive and significant, confirming that life satisfaction in East
Germany relative to West Germany recovered during the transition period. Self-reported
health has a positive and significant effect on life satisfaction. It remains significant
when economic and demographic characteristics are included in columns (2) and (6) and
when all explanatory variables are included in columns (3) and (7). The results for the
younger and older groups show that economic circumstances are important determinants
of the recovery of life satisfaction, which is consistent with previous studies (see
Easterlin and Plagnol, 2008; Frijters, Haisken-DeNew, & Shields, 2005). For both age
groups, income and unemployment are significant with the expected signs.
The 2SLS results for the younger age group and the older age group in columns
(4) and (8), respectively, show that health is a determinant of life satisfaction. Changes in
health during this transition period contributed to the recovery of life satisfaction. The
coefficients on health are larger in the 2SLS models than in the first difference models.
Once again this suggests that there is classical measurement error in self-reported health.
48
49
Table 1.8 Life Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
Age squared 0.001 0.001 0.0006 -0.001 -0.001 -0.0009
[0.001][0.001][0.001] [0.001][0.001][0.001]
Married 0.467**0.418**0.4198** 0.1710.160.1656
[0.107][0.099][0.100] [0.210][0.204][0.194]
Child 0.0590.0610.0619 -0.21-0.202-0.1854
[0.078][0.074][0.075] [0.267][0.259][0.251]
Spouse died 0.024 -0.03 -0.0676 -0.298 -0.318 -0.2357
[0.245][0.224][0.220] [0.243][0.241][0.256]
Log household income per capita 0.384** 0.319** 0.3121** 0.493** 0.448** 0.3994**
[0.082][0.083][0.083] [0.078][0.075][0.086]
Unemployed -0.601**-0.485**-0.4950** -0.291*-0.228+-0.1981
[0.097][0.088][0.090] [0.125][0.120][0.124]
Retired -0.078-0.055-0.0573 0.0890.0990.0315
[0.219][0.206][0.202] [0.089][0.087][0.111]
Unemployment rate
rt
-4.027**-2.452-2.109 -2.961*-1.635-1.8564
[1.441][1.492][1.615] [1.473][1.537][1.590]
Legislators, professionals -0.058 -0.035 -0.0418 -0.038 -0.021 -0.0525
[0.056][0.053][0.055] [0.076][0.079][0.097]
Technicians, armed forces 0.101* 0.097* 0.0887* -0.047 -0.027 -0.0181
[0.044][0.042][0.044] [0.070][0.064][0.070]
Clerks, sales and service workers 0.013 0.007 0.0087 -0.053 -0.046 -0.0424
[0.048][0.048][0.048] [0.071][0.070][0.075]
49
50
Table 1.8 (Continued)) Life Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
Agriculture, fishery, and trade
workers 0.005 0.046 0.0443 -0.189* -0.181* -0.1918*
[0.050][0.049][0.049] [0.083][0.082][0.092]
Operators, elementary
occupations -0.117+-0.115+-0.1057 -0.027-0.038-0.0579
[0.063][0.063][0.066] [0.125][0.124][0.129]
No insurance -0.197 -0.2185 0.478 0.545
[0.334][0.335] [0.483][0.535]
Private insurance 0.022 0.0373 0.081 0.1088
[0.122][0.126] [0.110][0.118]
Infant mortality rate
rt
0.0640.0683 0.241*0.2609*
[0.093][0.095] [0.107][0.111]
Infant mortality rate
rt-1
0.0210.0053 -0.117-0.1165
[0.062][0.069] [0.084][0.088]
Life expectancy
rt
0.1080.1117 0.02-0.0135
[0.100][0.101] [0.125][0.132]
Somewhat worried about finances
t
-0.342**-0.3325** -0.233**-0.2114**
[0.040][0.043] [0.049][0.051]
Very worried about finances
t
-0.944**-0.9216** -0.623**-0.5679**
[0.068][0.077] [0.076][0.085]
Somewhat worried about
finances
t-1
-0.065-0.0675 -0.02-0.0116
[0.042][0.043] [0.044][0.045]
50
51
Table 1.8 (Continued) Life Satisfaction Regression Results: First Difference and First Difference‐2SLS
24 to 44 years old in 1990 45 to 70 years old in 1990
(1) (2) (3)
(4)
2SLS (5) (6) (7)
(8)
2SLS
Very worried about finances
t-1
-0.132+-0.1322+ -0.141*-0.1084
[0.073][0.073] [0.070][0.075]
Somewhat worried about country's
development
t
0.0160.0167 0.1030.1481+
[0.067][0.068] [0.074][0.079]
Very worried about country's
development
t
-0.047-0.0494 0.0610.1125
[0.076][0.077] [0.081][0.092]
Year -0.054**-0.104-0.094-0.0765 -0.0480.0310.020.1007
[0.013] [0.081] [0.082] [0.085] [0.012] [0.091] [0.096] [0.116]
East Germany*year 0.101** 0.081** 0.061** 0.0640** 0.119** 0.087** 0.083** 0.1033**
[0.019] [0.019] [0.023] [0.023] [0.019] [0.020] [0.025] [0.030]
Self-reported health 0.362** 0.361** 0.342** 0.4992* 0.504** 0.494** 0.485** 1.0298*
[0.030] [0.030] [0.029] [0.243] [0.032] [0.032] [0.031] [0.449]
Observations 17924177791777917779 14713146401464014640
R-squared 0.0340.060.0960.0905 0.056 0.068 0.083 0.0204
The notation r represents the state and t represents the year. The unemployment rate, infant mortality rate, life expectancy and traffic accidents vary by
state and year. The errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in the SOEP are used. The base year
is the difference between 1992 and 1994. Standard errors in parentheses. Significance levels: +10%, *5%, **1%.
51
52
Since concerns about finances are significant in both the first and second stages, it seems
that stress from economic circumstances has both a direct effect on life satisfaction and
an indirect effect through health. The determinants of life satisfaction are generally
similar across age groups. Two exceptions are marital status and being unemployed.
They are significant for the younger age group, but not for the older age group. As
expected, the magnitude of the coefficient on health in these regressions is smaller than it
is in the health satisfaction regressions.
Based on the 2SLS results, if health were to increase by one unit, then life
satisfaction would increase by approximately 0.5 units for the younger age group and by
one unit for the older age group. Since changes in health probably play a larger role in the
ability to perform day-to-day activities in the older age group than in the younger age
group, it is not surprising that the coefficient on health is larger for the older age group. If
the young East Germans had the same trend in self-reported health as the young West
Germans during this study period, then their life satisfaction would have been 0.10 units
higher than it was by 1999. If young East Germans moved from being very concerned to
not being concerned about finances, then their life satisfaction would increase by 0.92.
For the older age group, it would increase by 0.57. If one were to take into account the
indirect effect of these concerns about finances through health, then the improvement in
life satisfaction would have been 0.99 and 0.66 units for the younger and older groups,
respectively. Therefore, the effect of stress from the transition appears to have a larger
overall role in the life satisfaction of the younger age group. Based on these findings, if
health had not declined so dramatically during the transition period, life satisfaction
52
53
likely would have recovered sooner. This does not suggest that it was the main factor in
the recovery of life satisfaction, just that it was a contributing factor.
1.7 VALIDITY OF RESULTS
In order to examine the validity of my results, I first discuss the possibility of
selection biasing my results. Then I investigate the stability of the results by doing the
2SLS analysis using each of the instruments separately. Finally, I present evidence from
the SOEP and other sources to support the finding that stress contributed directly and
indirectly to the decline in health satisfaction for the younger age group during the
transition period.
1.7.1 SELECTION
One possible concern with these data is that there is non-random attrition. If
young East Germans with high health satisfaction are more likely to leave the sample
than those with low health satisfaction and the same pattern did not exist for the West
Germans, then the decline in health satisfaction in East Germany relative to West
Germany may be a statistical artifact. Recall that East Germany is defined by where
people lived in 1989, so these attriters are likely to be people who left the country or died,
not people who moved to West Germany; the respondents who moved to West Germany
should have remained in the East German sample.
Table 1.9 column (1) contains the results of a simple OLS regression of whether a
person attrits anytime during the study period on their health satisfaction in 1992,
whether they lived in East Germany in 1989, and the interaction of the two variables.
53
54
Column (2) contains the coefficients from an OLS regression of whether or not a person
leaves the sample in the next period on their health satisfaction in the current period,
whether or not they lived in East Germany in 1989, and the interaction.
Table 1.9 Test for Non-random Attrition
24 to 44 years old in 1990 45 to 70 years old in 1990
Attrit ever
Attrit next
period Attrit ever
Attrit next
period
(1) (2) (3) (4)
East Germany 0.127+ 0.008 -0.017 -0.003
[0.068] [0.014] [0.057] [0.012]
Health satisfaction 0.008 0.000 -0.014* -0.005**
[0.006] [0.002] [0.006] [0.001]
East Germany *
health
satisfaction -0.017+ -0.001 -0.003 -0.001
[0.009] [0.002] [0.009] [0.002]
Constant 0.201** 0.036** 0.370** 0.071**
[0.046] [0.012] [0.037] [0.008]
Observations 3814 24038 3234 19630
R-squared 0.002 0.0001 0.006 0.004
The errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in the
SOEP are used. Standard errors in parentheses. ** significant at 1%; * significant at 5%; + significant at
10%.
For the younger age group, the significant coefficient on the interaction term
between health satisfaction and East Germany indicates that people who lived in East
Germany in 1989 and have higher health satisfaction are less likely to leave the sample.
Since the people with better health satisfaction are more likely to stay, then the decline in
health satisfaction may have been even more severe if those who left the sample had
54
55
actually stayed. I would be more concerned that selection was driving my results if the
coefficient on the interaction term was positive. In column (2), neither the coefficient on
heath satisfaction nor the interaction is significant. For the older age group in column
(3), I find that health satisfaction does affect whether or a not a person ever leaves the
sample, but that the difference is the same in East and West Germany. Since the
likelihood of leaving the sample is not different for East and West Germany, I am less
concerned that the conclusions regarding their relative trends are affected by attrition. In
column (4), the coefficient on health satisfaction is significant again, but the interaction
term is not. Overall the coefficients suggest that the relative trends in health satisfaction
are not driven by attrition.
1.7.2 COMPARISON TO SUBSETS OF THE INSTRUMENTAL VARIABLES
In order to investigate the stability of the coefficients on self-reported health in
the 2SLS model, I re-run these regressions using each of the instruments separately. The
results are presented in Appendix 1.B. The coefficients in the health satisfaction and life
satisfaction regressions for the younger group remain fairly stable, ranging from 1.21 to
1.33 and 0.411 to 0.676, respectively. For the older age group, the results are less stable,
which is consistent with the fact that the F-statistic is smaller in the main first-stage
results. For health satisfaction, the self-reported health coefficient ranges from 1.25 to
2.9; for life satisfaction, it ranges from 0.97 to 1.36. The larger coefficients occur when
lagged traffic accidents is the only instrumental variable. The effect of health is revealed
for the subpopulations whose health is predicted by alcohol related traffic accidents. One
would expect that the coefficient for these subpopulations would be greater than for the
55
56
subpopulations captured by lagged doctor visits, because these are a more select group of
people. Therefore, the effect of health on their life satisfaction or health satisfaction may
be greater. Overall, the stability of the coefficients, especially for the young group,
supports my empirical methodology.
1.7.3 ADDITIONAL EVIDENCE
This paper has shown that the decline in health was a major contributor to the
decline in health satisfaction and slow recovery of life satisfaction for the younger age
group of East Germans after reunification. The regression results support the hypothesis
that stress from uncertain economic conditions contributed to both directly and indirectly
to the trends in health satisfaction and life satisfaction. People between 24 and 44 years
old are at the point in the life cycle where financial obligations are at their peak, with
spouses and children to support. People between 45 and 70, on the other hand, have
fewer financial obligations and dependents. As they move into retirement, they become
the beneficiaries of a generous pension system. In the following paragraphs, I present
evidence that people who would be most vulnerable to stress, the young and especially
the less-educated young, report higher stress and exhibit more stress-related behaviors.
If stress contributed to the decline in health and health satisfaction, then young
East Germans with lower education should have reported more concerns about their
finances compared to those with higher education, and their concerns should have
increased during the transition period. In addition, health satisfaction should have
declined more for those with less education than those with more education. In Figure
56
57
1.12, I plot mean health satisfaction and mean concerns about finances for those with
secondary education compared to those who received upper secondary education.
Figure 1.12 Health Satisfaction and Worries about Finances by Education Level –
East Germany, 24 to 44 years old in 1990
Means are computed using weights provided by SOEP. East and West Germany are determined by
location in 1989. Age group is based on age in 1990.
Those in the latter group likely continued on to university; whereas the former group
likely did not. Figure 1.12 is consistent with the stress hypothesis; concerns about
finances increased more for those with less education and health satisfaction declined
more for that same group.
Data from other sources on stress-related health behaviors, such as alcohol use
and smoking, support the notion that health declined during the transition, and impacted
the younger age group more than the older age group. Nolte et al. (2003) estimate that in
1.5
1.7
1.9
2.1
2.3
2.5
2.7
2.5
3
3.5
4
4.5
5
5.5
6
6.5
7
7.5
8
1990 1991 1992 1993 1994 1995 1996 1997 1998 1999
Worries about Finances (1 to 3)
Health Satisfaction (0 to 10)
Upper Secondary Health Satisfaction Lower Secondary Health Satisfaction
Upper Secondary Worries Lower Secondary Worries
57
58
1992 the net death rate from alcohol disease in East Germany was five times higher than
in West Germany. The net death rate takes into account both the positive and negative
effects of alcohol on health. The overall net death rate declined in East Germany and
increased in West Germany, narrowing the gap to 2.5 times higher in East Germany by
1998. However, there was almost no change in relative death rates for younger people,
while the net death rates decreased in the East compared to the West for older people.
This again indicates that young East Germans experienced a high level of stress
immediately before and through the years immediately following reunification, while
there was an improvement for older East Germans.
The smoking patterns in East and West Germany also suggest that stress
contributed to the decline in health satisfaction. Junge and Nagel (1999) studied smoking
patterns in East and West Germany using a representative sample of 7,124 men and
women between 18 to 79 years old conducted in 1998. The authors conclude that
compared to a previous survey in 1990/92, the proportion of male smokers dropped by
3% in West Germany, but remained the same in East Germany. The proportion of female
smokers rose by 1% in West Germany, but by 8% in East Germany. Junge and Nagel
(1999) also show that the percentage of West German men between 25 and 49 that
smoked decreased between 1990/1992 and 1998, but it increased for East Germans. On
the other hand, the percentage of men between 50 and 69 that smoked decreased between
1990/1992 and 1998 for both East Germany and West Germany. Again, these data point
to the younger age groups in East Germany exhibiting more poor health behaviors
58
59
associated with stress than in West Germany, without a commensurate finding for the
older age groups.
1.8 CONCLUSION
In this paper, I have shown that health satisfaction in East Germany declined
relative to West Germany for people between 24 and 44 in 1990 because of a decline in
health. The first-stage results suggest that stress from economic uncertainty contributed
to the decline in health. The younger age group was at the middle to the peak of their
careers when the transition from socialism to capitalism occurred, and therefore, they
were the most susceptible to higher unemployment rates and economic uncertainty. The
older age group did not experience the same decline in health satisfaction because they
did not experience a decline in health and they had less relative concerns about their
personal economic circumstances. The people in the older age group were the
beneficiaries of a generous pension system, which could explain why they did not
experience as much economic uncertainty as the younger group. The older East Germans
had less severe worries about their personal economic circumstances compared to
younger East Germans. The decline in health for the younger and older age groups
contributed to the slow recovery in life satisfaction. If health had not declined for the
younger age group, life satisfaction would have recovered to nearly pre-reunification
levels sooner.
In addition to contributing to the literature on health and well-being, this paper
also provides insight into the methodology of the well-being literature. Classical
measurement error attenuates the effect of health on both health satisfaction and life
59
60
satisfaction in a fixed effects model. The qualitative interpretation of the coefficients on
health in the 2SLS model and the first difference model are the same. This finding
suggests that not correcting for classical measurement error in satisfaction regressions
with fixed effects yields similar qualitative results, but that the size of the effect of health
is somewhat underestimated without an instrumental variable strategy.
60
61
CHAPTER 2: ARE WOMEN HAPPIER THAN MEN? EVIDENCE
FROM THE GALLUP WORLD POLL
2.1 INTRODUCTION
Women in nearly all countries of the world have lower incomes, are less
educated, are more likely to be widowed or divorced, and report lower levels of health
than men. Despite these inequalities, this paper provides evidence that women are
equally happy or happier than men in nearly all countries examined. This is contrary to
feminist expectations and is true in all regions of the world and in countries at all stages
of economic development.
Are women with the same life circumstances as men also happier than men? In
addition to presenting the mean differences in happiness between women and men, this
paper also examines the differences after controlling for various individual circumstances
to come closer to the “pure effect” of being female. In other words, mean differences are
presented after controlling for, among other things, income, education, marital status, and
health.
These analyses are conducted using the Gallup World Poll Survey (2009a) that
covers over 100 countries around the world and includes four waves between 2005 and
2008. One of the benefits of these data is that the same questions are asked in all
countries. This allows for comparisons to be made both within countries and across
countries, which is why these data can be used to address the research questions
discussed above. It has been shown that time series and cross-section data yield different
results in the happiness literature (Easterlin and Angelescu, 2011). Consequently, one
cannot infer the time-series from this paper. The focus of this paper is, therefore, on
61
62
establishing the facts about the female-male happiness gap in countries of different
geographic locations and stages of development at a point in time. In the following
section, I discuss the previous literature on female-male differences in well-being. The
existing economics of happiness literature has largely focused on female-male differences
in happiness in developed countries and typically reports the female-male differences in
happiness after controlling for various life circumstances. This paper contributes to the
literature by presenting the female-male differences in life satisfaction for numerous
countries. In addition, the analysis starts with female-male differences in life satisfaction
without controlling for any individual circumstances. Additional details about the data
are presented in section 2.3, and the empirical strategy is discussed in section 2.4. The
results are discussed in section 2.5 and the conclusion is in section 2.6. As in much of the
happiness literature, I use the terms life satisfaction, happiness, and subjective well-being
interchangeably.
2.2 LITERATURE REVIEW
The consensus based on reviews of the economics of happiness literature is that
women are generally slightly happier than men after controlling for individual
circumstances (Dolan et. al, 2008; Frey & Stutzer, 2002), but this evidence is largely
based on analyses from developed countries. Easterlin (2001) uses the U.S. General
Social Survey to explore life-cycle changes in happiness. In doing so, he finds that
women were happier relative to men in the 1970s, but that the reverse was true by the
1990s. Blanchflower and Oswald (2004) present evidence that women in the U.S. and
Britain reported being happier than men after controlling for life circumstances, but the
62
63
gap has declined for white women in the United States. Stevenson and Wolfers (2009)
confirm the findings of Easterlin (2001) that female happiness in the United States has
declined relative to men in recent decades. Stevenson and Wolfers then conduct a similar
analysis for 11 European countries using the Eurobarometer and find a similar decline in
the female-male happiness gap. Easterlin and Marcelli (2010) use the General Social
Survey to show that American women are happier than men earlier in the life cycle, but
that the reverse is true later in the life cycle. They attribute this life-cycle reversal to
changes in financial security and marital status, which favor women earlier in the life-
cycle and favor men later in the life-cycle.
As previously mentioned, there are few studies on differences in subjective well-
being between men and women in developing countries. Knight, Song, and Gunatilaka
(2009) show that rural women in China are happier than rural men. This result is found
when they include both objective and subjective explanatory variables in their
specification. When looking at the happiness equations separately for men and women,
the authors find that income and material comparisons are more important for men than
women. Graham and Pettinato (2001) investigate the determinants of happiness for 15
Latin American countries using the Latinobarometer. They report, after controlling for
individual circumstances, that there is no happiness difference between men and women.
They contrast this result to that of Russia, where men are significantly happier than
women. Arku, Filson, and Shute (2008) organize focus groups to discuss well-being in
three communities in rural Ghana. Instead of asking about their life circumstances and
overall life satisfaction, the men and women are asked about the factors that are
63
64
“promoters of well-being.” The top five indicators for women are engaging in petty
trading, having a trustworthy pastor in their local churches, large family size, ability to go
to church regularly, and social interaction. The top 5 indicators for men are card playing,
being a member of a church committee, drinking palm wine together, listening to radios,
and farming. The authors conclude that different factors are important for well-being for
women and men. Although this paper does not take the standard approach, it does suggest
that men and women report different specific factors as important for their well-being,
but the themes of leisure and income-generating activities are important for both sexes.
There are also studies that investigate the differences in the impact of specific life
circumstances on life satisfaction of men and women. Using data for 28 OECD countries,
Stavrova, Schlösser, and Fetchenhauer (2011) confirm previous findings that women tend
to be less negatively affected by unemployment. The effect of marriage on happiness for
men and women is still an area of debate, but most studies have found that the benefits of
marriage are the same for women and men (Frey & Stutzer, 2002). Kroll (2010)
investigates the effects of different types of social capital on life satisfaction. He finds
that in the United Kingdom, informal socializing is more important for women than for
men, but that civic engagement is only associated with higher life satisfaction for men
and childless women; it has no impact on women who have children. Schoon, Hansson,
and Salmela-Aro (2005) investigate the effects of work and parenthood on women and
men in Estonia, Finland, and the UK. First, they find that women are happier than men in
all three countries. Second, married women in the UK and Finland who are employed
and have children are more satisfied with their lives than those who are employed but do
64
65
not have children. This is also true for men in the UK and Estonia, but the reverse is true
for women in Estonia and men in Finland. These results suggests that there is some
interaction effect between home production and employment on life satisfaction.
The psychology literature is in agreement with the economics of happiness
literature that women report being happier than men (Nolen-Hoeksema & Rusting 1999).
They also conclude that women are more likely to have anxiety or be depressed. The
authors discuss the evidence for three possible explanations for gender differences in
mood: personality, social norms, and biology. There does not seem to be enough
convincing evidence that biology or personality can explain gender differences in
happiness or depression. Nolen-Hoeksema and Rusting conclude that social norms are the
likely cause of differences in happiness, but that more research needs to be done.
According to the happiness literature, there are also gender differences in specific
domains of life, including job satisfaction, financial satisfaction, and satisfaction with
leisure. Women in Denmark, Australia, and Russia are found to be more satisfied with
their financial situations than men (Bonke, Deding, & Lausten, 2009; Marks & Fleming,
1999; Schyns, 2001). They have equal satisfaction with leisure in Denmark (Bonke et al.,
2009). In the United States and United Kingdom, women report higher job satisfaction
than men despite the fact that they are paid less (Clark, 1997; Clark & Oswald, 1996;
Sloane & Williams, 2000; Sousa-Poza & Sousa-Poza, 2000).
My contribution to this literature is to show the patterns of female-male
differences in life satisfaction across countries at various stages of development,
including developed, developing, and transition countries. First, I present the mean
65
66
differences between women and men. Then I control for individual circumstances to get
closer to the “pure effect” of being female. Finally, I will determine whether individual
circumstances have the same effect on the magnitude of the female-male happiness gap
in all countries.
2.3. DATA
The Gallup World Poll Survey (2009a) data are the primary data used in this
paper. The first three waves of the surveys, conducted between 2005 and 2008, are
pooled to form a cross-sectional dataset. The analysis starts with the 92 countries used in
Easterlin, Angelescu, and Zweig (2011). Due to geographic coverage and variable
availability, these 92 countries are reduced to 73 countries. The sample contains 20
developed countries, 12 transition countries in Eastern and Central Europe, 16 Asian
countries, 17 Latin American countries, and 8 African countries. (See Appendix 2.A for
a list of the countries in each group.)
10
The surveys in these countries are nationally
representative. Telephone surveys are typically used in developed countries while face-
to-face interviews are used in developing countries. For additional information on the
sampling procedure and quality of the data, see Gallup World Poll (2009b).
10
The nineteen countries and reasons for elimination are: Saudi Arabia survey confined to Saudis; Darfur
area was not surveyed in Sudan; in Laos, survey coverage was confined largely to urban areas; seventy per
cent of the population of Madagascar was not surveyed in the wave for which several other variables were
available; marital status was not available for China; the education variable was not available for Paraguay;
the education variable was not available in the same year as the employment status variable for Ethiopia,
Nigeria, and Tunisia; the location variable was not available for Yemen, Algeria, Angola, Cuba, Malawi,
Pakistan, South Africa, and Zambia; the location variable was not available in the same years as the
education variable in Egypt; the education, location, and employment status variables were not available in
the same year for Morocco.
66
67
The life satisfaction question is:
Please imagine a ladder with steps numbered from zero at the bottom to ten at the
top. Suppose we say that the top of the ladder represents the best possible life for
you, and the bottom of the ladder represents the worst possible life for you. On
which step of the ladder would you say you personally feel you stand at this time,
assuming that the higher the step the better you feel about your life, and the lower
the step the worse you feel about it? Which step comes closest to the way you
feel?
The respondents answer on a scale from 0 to 10. The life satisfaction question used in this
analysis is different from the typical life satisfaction question. The more typical life
satisfaction question is asked in some countries and waves, but only 46% of individuals
who responded to the best/worst possible life satisfaction question responded to the
typical question. Thus, I use the best/worst question in this paper.
The main independent variable is whether an individual is male or female. In the
regressions where I control for individual circumstances I examine objective factors that
are typically included in happiness regressions.
11
The objective factors include
demographic characteristics and life circumstances. The specific questions for these
variables and response categories are given in Appendix 2.B. The demographic and life
circumstance variables are age, marital status, education, employment status, attendance
at a religious ceremony in the previous week, residential location, and health. Marital
status is divided into single, married, and previously married. The latter includes
divorced, separated, and widowed. The categories for education level are elementary,
secondary, and tertiary. Residential location is divided into rural, small town, and large
city. For the purpose of this analysis, the first two categories are grouped together. The
11
See Dolan et al. (2008) for a summary of the main determinants of happiness based on the economics of
happiness literature.
67
68
health variable is whether or not the individual has any health problems. There is not a
variable on the number of children in the household. However, for some countries and
waves, there is a question on the ideal number of children. This latter variable is used in
one specification as a proxy for the number of children.
Income and occupation are also included in some specifications. These two
variables are available in a limited number of surveys and countries. The 12 occupation
groups in the survey are reduced to six categories: white collar, business owner, service
worker, non-farm manual worker, farmer, and other. Other includes all individuals who
chose “other” in the list of possible responses or reported having a job, but did not answer
the occupation question. This category is only constructed for waves and countries where
some of the respondents answered the occupation question. The income variable is a
continuous measure that only includes cash income.
Table 2.1 shows the descriptive statistics for the average differences between
women and men in the outcome and explanatory variables, where the unit of observation
is a country. These averages give equal weight to each country regardless of the number
of observations in the survey. The differences for each country are computed by
subtracting the average for men from the average for women. If the difference is
positive, then the average value for women is higher than for men, and vice versa if it is
negative. Women are more likely to be previously married, unemployed, only have an
elementary education, have health problems, and have lower income. In addition, they are
less likely to be white collar workers. The only factors that tend to favor women are and
being less likely to be farmers or non-farm manual workers.
68
69
Table 2.1 Descriptive Statistics: Average Female-Male Difference
N Mean
Standard
deviation Minimum Maximum
Life satisfaction 73 0.04 0.18 -0.32 0.62
Single (%) 73 -0.09 0.04 -0.18 0.03
Married (%) 73 0.01 0.06 -0.15 0.13
Previously married (%) 73 0.09 0.04 -0.01 0.19
Elementary education (%) 73 0.04 0.06 -0.10 0.16
Secondary education (%) 73 -0.03 0.06 -0.15 0.20
Tertiary education (%) 73 -0.01 0.04 -0.19 0.10
Without health problems (%) 73 -0.03 0.04 -0.15 0.09
Employed (%) 73 -0.19 0.11 -0.64 -0.02
Live in a large city (%) 73 0.00 0.04 -0.10 0.14
Attend a religious ceremony (%) 73 0.06 0.10 -0.26 0.21
Income (*10^-13) 66 -3.09 3.61 -15.82 0.48
White collar (%) 71 -0.01 0.04 -0.10 0.08
Business owner (%) 71 0.00 0.06 -0.21 0.09
Service worker (%) 71 0.01 0.03 -0.04 0.10
Non-farm manual worker (%) 71 -0.14 0.06 -0.26 -0.03
Farmer (%) 71 -0.04 0.05 -0.29 0.00
Other (%) 71 -0.01 0.04 -0.17 0.04
Means are computed using weights provided in the Gallup World Poll Survey data (2009a).
2.4 METHODS
2.4.1 ECONOMETRIC STRATEGY
The first part of the analysis focuses on the average differences in life satisfaction
allowing the respondents’ individual circumstances to vary. The average differences will
then be compared to several country-level factors. Satisfaction with life for individual i,
S
i
, is regressed on a dummy variable, F
i,
which is equal to one if the individual is female
and zero if the individual is male.
69
70
The baseline specification is:
2.1
.
This regression is run separately for each of the 73 countries in the analysis, resulting in
73 estimates of , where indicates the size of the female-male happiness gap in a
country. Recall that if is positive and significant, then women are on average happier
than men.
In the second step of the analysis, demographic characteristics, economic factors,
and life circumstances are included as explanatory variables. If differences in these
variables account for the female-male happiness gap, then the coefficients on female
should approach zero and should not be significant. I add each set of control variables to
the regressions separately. First, demographic characteristics represented by X
i
are added
to each country’s regression:
2.2
.
The demographic variables are age, age squared, marital status, and level of
education. To see whether children are important in explaining the female-male
happiness gap, the number of ideal children is included as an explanatory variable in a
separate regression. It is added separately because the number of observations is reduced
considerably when it is included in the model.
70
71
Life circumstances, represented by L
i
, are then added to the specification:
2.3
.
Life circumstances include whether or not the individual has health problems,
employment status, whether she attended a religious ceremony in the previous week, and
whether she lives in a large city (with rural or small town as the reference group). In the
fourth step, economic factors, i.e., occupation and income, represented by E
i
. are added
to the specification:
2.4
.
Again, if economic circumstances explain the female-male happiness gap then the
coefficients on female should approach zero. These regressions provide insight into
whether women with the same circumstances as men are as happy as men.
2.4.2 ECONOMETRIC ISSUES
In all specifications, ordinary least squares (OLS) is used rather than an ordered
probit, which requires the assumption that the responses to the life satisfaction question
are cardinal even though they are actually ordinal. The happiness literature suggests the
results of ordinal and cardinal methods tend to be very similar in terms of levels of
significance, especially for responses on a scale of zero to ten (see Frey & Stutzer, 2002,
p. 43; Powdthavee, 2010, pp. 27-29; and van Pragg, 2005, pp. 205-206). In order to
71
72
verify that the results are consistent, I run the baseline regressions using ordered probit
and OLS. All of the female-male differences in life satisfaction that are significant in
OLS regressions are also significant in the order probit regressions. Greece and
Lithuania are the only countries where the differences are significant in the ordered probit
regression and not in the OLS regression.
Because the number of observations changes depending on the variables included
in the analysis, the coefficients on female from the regressions with controls are always
compared to the coefficients on female when there are no controls, after restricting the
number of observations to that in the controls case. For example, when I present the
results for equation (2.2) against equation (2.1), I plot the coefficients from (2.2) against
the coefficients from a regression of life satisfaction on female without controls for
observations where age, marital status, and education level are also available.
It is possible that the observations that are dropped in each specification are not
random and could bias the results. Therefore, I do a Wald test of the equality of the
female-male differences in life satisfaction in the baseline regressions with all
observations and with only the observations available with controls. For each
specification, I cannot reject the null hypothesis that the no-controls coefficients are equal
to the no-controls coefficients limited to the observations available when the control
variables are included in the specification.
Concern that the missing observations are non-random is particularly important
when the economic variables are included in the model. It is possible that the income
variable would be missing more for women than for men. If that is true, and the women
72
73
who have missing income data are the least happy, then the results would be biased
upward. The percent of women and men in each country who are missing the economic
variables, as well as the differences in these percentages, are presented in Appendix 2.C.
The percent of the observations that are dropped in each country is similar across men
and women. There are only three countries for which the difference in the percent
missing for men and women is greater than 9 percent: Nepal, Norway, and the
Netherlands. On average, there is less than a 1% difference in the number of observations
missing for men and women. Unlike in the other regions, more male observations are
dropped than female observations in Asia and the transition counties.
As mentioned above, the coefficients on female could be overestimated in the
regressions if (1) women are less likely to respond to the income question than men and
(2) it is the women with lower life satisfaction who are less likely to respond. This is
tested more formally using an OLS regression where the dependent variable is equal to
one if the respondent did not answer the income question and zero otherwise. The
independent variables are female, life satisfaction, and the interaction of female with life
satisfaction. The female-male differences may be overestimated if the coefficient on the
interaction term is negative and significant and the coefficient on female is positive and
significant. This only occurs in eight out of 66 countries, which are shown in Appendix
2.D. The eight countries are Norway, Sweden, Australia, Columbia, Mexico, Dominican
Republic, Indonesia, and Tajikistan. In general, this suggests that the magnitudes of the
coefficients on female in the regressions that include income are not driven by this type
73
74
of selection. However, the results for the eight countries listed above should be
interpreted with caution.
The regressions are weighted by the weights provided in the Gallup data. The
weights adjust for gender, age, and, where reliable comparative population data are
available, education or socioeconomic status (Gallup World Poll, 2009b). I include
dummy variables for waves to control for any factors that affect all observations in a
specific wave. Although some of the variables can be thought of as exogenous, such as
age, health, and to some extent, marital status and education, others are the result of
individual choices. Because of this and the fact that I use cross-section data, I do not
claim that the results that include all of the control variables can be interpreted as causal.
The aim of this paper is, therefore, to present evidence on the patterns of the female-male
happiness gap. In the results section, the phrase equally happy refers to there being no
significant difference in happiness.
2.5 RESULTS
2.5.1 ACROSS-COUNTRY PATTERNS
Are women happier than men? Figure 2.1 plots mean life satisfaction for men in
the 73 countries examined in this analysis against mean life satisfaction for women. The
dashed line is the 45‐degree line where the data points would be if men and women had
the same levels of life satisfaction. Mean life satisfaction for women is above the 45-
degree line for 40 of the 73 countries. In Table 2.2, the female-male differences in life
satisfaction are grouped by statistical significance and geographic region. The average
differences in life satisfaction are presented in Table 2.3.
74
75
Figure 2.1 Mean Life Satisfaction (0 to 10) for Men and Women
Means are computed using weights provided in the Gallup World Poll Survey data (2009a).
Using a two-tailed t-test, the coefficient on the dummy variable for female is
statistically significant in 18 countries at a 90% significance level; twelve countries are
significant at a 95% significance level. The coefficient is positive in 14 of the 18
countries, which indicates that the women are more satisfied with life than the men.
There are four countries where the men are more satisfied and there is no significant
difference between men and women in the remaining 55 countries. Women are generally
at least as happy as men in all geographic regions. The only notable geographic pattern to
the significance of the female-male happiness gap is that women and men are equally
happy in all of the transition countries. The average difference in happiness is negative in
ARG
AUS
AUT
BEL
BFA
BGD
BGR
BLR
BOL
BRA
CAN
CHE
CHL
CMR
COL
CRI
CZE
DEU
DNK
DOM
ECU
ESP
EST
FIN
FRA
GBR
GRC
GTM
HND
HUN
IDN
IND
IRL
IRN
ITA
JPN
KAZ
KEN
KGZ
KHM
KOR
LTU
LVA
MEX
MLI
MNG
MOZ
MYS
NIC
NLD
NOR
NPL
PAN
PER
PHL
POL
PRT
ROM RUS
SEN
SLV
SVK
SVN
SWE
THA
TJK
TUR
TZA
UGA
URY
USA
VEN
VNM
4 5 6 7 8
Mean Life Satisfaction for Women
4 5 6 7 8
Mean Life Satisfaction for Men
Mean Life Satisfaction 45-degree Line
75
76
Table 2.2 Distribution of Female-Male Differences in Life Satisfaction by
Geographic Region
No significant
difference
Women
happier
Men
happier Total
Developed countries 16 2 2 20
Transition countries 12 0 0 12
Asia 10 6 0 16
Latin America 12 3 2 17
Africa 5 3 0 8
Total 55 14 4 73
The categories no significant difference, women happier, and men happier are based on whether
the coefficient on the dummy variable for female is signifcantly different from zero in a two-tailed
t-test.
the transition countries, while Asia and Africa have the greatest overall positive
differences. In the countries where women are significantly happier, the happiness gap is
similar across regions. Without controlling for any individual factors, women are at least
as happy as men in nearly all countries and there is not a systematic pattern by region.
The coefficients and number of observations for each specification are presented in
Appendix 2.E and 2.F, respectively.
Is the female-male happiness gap larger at more advanced stages of development?
One might expect that well-being in less developed countries would favor men because
women do not have the same opportunities and status as men. In Figure 2.2, the female-
male happiness gap from the regression with no controls ( β from equation 2.1) is plotted
against log GDP per capita of the country obtained from the World Bank (2010). Recall
that a positive difference indicates that women are happier and a negative difference
76
77
Table 2.3 Average Female-Male Difference in Life Satisfaction by Geographic
Region
No significant
difference
Women
happier
Men
happier Total
Developed countries 0.01 0.33 -0.29 0.01
[0.02] [0.13] [0.04] [0.04]
Transition countries -0.07 n/a n/a -0.07
[0.02] [0.02]
Asia 0.02 0.36 n/a 0.14
[0.03] [0.07] [0.05]
Latin America 0.001 0.23 -0.20 0.02
[0.04] [0.02] [0.03] [0.04]
Africa 0.04 0.28 n/a 0.13
[0.05] [0.03] [0.05]
Total -0.005 0.31 -0.24 0.04
[0.01] [0.03] [0.03] [0.02]
The categories no significant difference, women happier, and men happier are based on whether
the coefficient on the dummy variable for female is signifcantly different from zero using a two-
tailed t-test. Standard errors are in parentheses.
indicates that men are happier. The female-male happiness gap is not a reflection of a
country’s stage of development. In an OLS regression of the happiness gap on log GDP
per capita, I cannot reject the null hypothesis at a 95% significance level (but not at a
90% significance level) that the coefficient on log GDP per capita is equal to zero. There
is a slightly negative slope, which suggests, if anything, that women are happier relative
to men at lower levels of development and that at advanced stages of development, the
happiness gap disappears. The countries where women are the happiest relative to men
(Iran, Japan, South Korea, Tanzania, and Turkey) are not what one might expect.
77
78
Figure 2.2 Female-Male Difference in Life Satisfaction and Log GDP per Capita
GDP per capita data are for 2006 in PPP constant 2005 international dollars. GDP data are from World
Bank (2010) World Development Indicators Online (WDI) database. Data retrieved March 1, 2011. The
OLS regression is SatLife
f
-SatLife
m
= 0.31 - 0.03lnGDP. R
2
= 0.04 N=73.
[1.98] [1.73]
If the female-male happiness gap is not a reflection of a country’s stage of
development, then perhaps it is associated with overall well-being. One might speculate
that the female-male happiness gap would be larger in countries with higher levels of
overall well-being if increases in well-being are a result of economic opportunities,
freedom, and/or education. One could also speculate that the opposite would be true. If
reported happiness takes into account aspirations and if women’s aspirations are lower
relative to men at lower levels of overall well-being then the happiness gap could
FRA
JPN
AUT
PRT
NOR
ESP
BEL
NLD
IRL
SWE
CAN
GBR
FIN
DEU
AUS
CHE
USA
DNK
ITA
GRC
SVK
ROM
BGR
EST
RUS
LVA
BLR
LTU
POL
CZE
HUN
SVN
IDN
MYS
TJK
MNG
PHL
KAZ
IRN
NPL
BGD
KOR
KHM
VNM
TUR
KGZ
IND
THA
PER
BOL
HND
CHL
MEX
NIC
ECU
SLV
VEN
GTM
ARG
COL
PAN
URY
CRI
DOM
BRA
CMR
MLI
UGA
KEN
TZA
MOZ
SEN
BFA
-.4 -.2 0 .2 .4 .6
Female-Male Difference in Life Satisfaction
7 8 9 10 11
Log GDP per Capita
78
79
decrease at lower levels of overall life satisfaction. In Figure 2.3, the female-male life
satisfaction gap is plotted against mean life satisfaction for each country. There is no
association between overall well-being and the female-male happiness gap.
Figure 2.3 Female-Male Difference in Life Satisfaction and Mean Life Satisfaction
The OLS regression is SatLife
f
-SatLife
m
= 0.18 - 0.02MeanSatLife. R
2
= 0.02 N=73.
[1.60] [1.24]
Does the size of the female-male happiness gap depend on women’s rights or
education? Figure 2.4 plots the female-male happiness gap and the labor participation
rate of women. Figure 2.5 shows the female-male happiness gap and the percent of seats
FRA
JPN
AUT
PRT
NOR
ESP
BEL
NLD
IRL
SWE
CAN
GBR
FIN
DEU
AUS
CHE
USA
DNK
ITA
GRC
SVK
ROM
BGR
EST
RUS
LVA
BLR
LTU
POL
CZE
HUN
SVN
IDN
MYS
TJK
MNG
PHL
KAZ
IRN
NPL
BGD
KOR
KHM
VNM
TUR
KGZ
IND
THA
PER
BOL
HND
CHL
MEX
NIC
ECU
SLV
VEN
GTM
ARG
COL
PAN
URY
CRI
DOM
BRA
CMR
MLI
UGA
KEN
TZA
MOZ
SEN
BFA
-.4 -.2 0 .2 .4 .6
Female-Male Difference in Life Satisfaction
4 5 6 7 8
Mean Life Satisfaction
79
80
held by women in the national parliament for each country. Both variables are from the
World Bank Development Indicators (2010). The figures show that the size of the
happiness gap is not associated with either measure of women’s rights.
Figure 2.4 Female-Male Difference in Life Satisfaction and Female Labor Force
Participation Rate (%)
Female labor force participation rates are for 2005 or the closest year available from 2006 through 2008.
Data are from World Bank (2010) World Development Indicators Online (WDI) database. Data retrieved
March 1, 2011. The OLS regression is SatLife
f
-SatLife
m
= 0.11 + 0.57Labor
f
. R
2
= 0.001 N=73.
[0.12] [0.32]
It could be that women with higher education have higher aspirations, and
therefore, their happiness should be closer to that of men. On the other hand, it could be
FRA
JPN
AUT
PRT
NOR
ESP
BEL
NLD
IRL
SWE
CAN
GBR
FIN
DEU
AUS
CHE
USA
DNK
ITA
GRC
SVK
ROM
BGR
EST
RUS
LVA
BLR
LTU
POL
CZE
HUN
SVN
IDN
MYS
TJK
MNG
PHL
KAZ
IRN
NPL
BGD
KOR
KHM
VNM
TUR
KGZ
IND
THA
PER
BOL
HND
CHL
MEX
NIC
ECU
SLV
VEN
GTM
ARG
COL
PAN
URY
CRI
DOM
BRA
CMR
MLI
UGA
KEN
TZA
MOZ
SEN
BFA
-.4 -.2 0 .2 .4 .6
Female-Male Difference in Life Satisfaction
.2 .4 .6 .8 1
Female Labor Force Participation Rate (%)
80
81
that women with higher education are happier relative to men if education increases
women’s opportunities relative to men. Table 2.4 contains the average female-male
difference in happiness by geographic region and reported education level – elementary,
secondary, and tertiary. The reported numbers are averages of each country’s mean
female-male difference in life satisfaction, giving equal weight to each country. Women
Figure 2.5 Female-Male Difference in Life Satisfaction and Percent of Seats Held by
Women in National Parliament
The percent of seats held by women in national parliament are for 2005 or the closest year available from
2006 through 2008. Data are from World Bank (2010) World Development Indicators Online (WDI)
database. Data retrieved March 1, 2011. The OLS regression is SatLife
f
-SatLife
m
= 0.09 - 0.29Parliament
f
.
R
2
= 0.001 N=73. [2.12] [1.24]
who have an elementary education are happier relative to men in Asia, Latin America,
and Africa. However, men with an elementary education are happier than women in the
FRA
JPN
AUT
PRT
NOR
ESP
BEL
NLD
IRL
SWE
CAN
GBR
DEU
AUS
CHE
USA
DNK
ITA
GRC
SVK
ROM
BGR
EST
RUS
LVA
BLR
LTU
POL
CZE
HUN
SVN
IDN
MYS
TJK
MNG
PHL
KAZ
IRN
NPL
BGD
KOR
KHM
VNM
TUR
KGZ
IND
THA
PER
BOL
HND
CHL
MEX
NIC
ECU
SLV
VEN
GTM
ARG
COL
PAN
URY
CRI
DOM
BRA
CMR
MLI
UGA
KEN
TZA
MOZ
SEN
BFA
-.4 -.2 0 .2 .4 .6
Female-Male Difference in Life Satisfaction
0 .1 .2 .3 .4 .5
Percent of Seats Held by Women in National Parliament
81
82
transition countries and there is almost no difference between men and women in the
developed countries. Looking at respondents with a secondary education, women are
Table 2.4 Average Female-Male Difference in Life
Satisfaction by Education and Geographic Location
Elementary Secondary Tertiary
Developed countries -0.02 0.10 0.10
[0.15] [0.04] [0.05]
Transition countries -0.16 -0.03 -0.08
[0.05] [0.05] [0.04]
Asia 0.10 0.23 0.19
[0.07] [0.06] [0.08]
Latin America 0.10 0.05 0.04
[0.06] [0.05] [0.05]
Africa 0.23 0.04 n/a
[0.07] [0.05]
Total 0.05 0.09 0.07
[0.04] [0.02] [0.03]
Countries where there were less than 75 observations for a particular
education level were excluded from the average for that education
level. Standard errors are in parenthesis.
at least as happy in all regions. Other than in the transition countries, women of all
education levels are at least as satisfied with their lives as men. In Latin America and
Africa, the size of the female-male happiness gap is larger for respondents with a
secondary education compared to respondents with a primary education. In Asia, the
happiness gap is larger for the secondary education group compared to the elementary
education group. These findings suggest that the happiness gaps may depend on
82
83
education. In some regions, there is a larger happiness gap for more educated women and
men, and in other regions, the reverse is true.
What about religion? The average differences in female-male happiness by a
country’s primary religion are presented below in Table 2.5. A religion is designated as
primary if 50% or more of the population practiced a specific religion based on the
Central Intelligence Agency’s (CIA’s) World Fact Book (2010). The “other” category
Table 2.5 Average Female-Male Difference in
Life Satisfaction by Religion
Primary religion Mean difference N
Buddhism 0.15 5
[0.08]
Catholicism -0.02 33
[0.03]
Christianity 0.02 18
[0.03]
Islam 0.11 11
[0.06]
Other 0.21 6
[0.11]
Total 0.04 73
[0.02]
The primary religion is determined by whether the CIA
World Fact Book (2010) stated that at least 50% of the
population was a specific religion. Other includes
religions that did not fall into one of the four categories
and countries without a primary religion. Standard
errors in parentheses.
includes countries without a primary religion or with a primary religion other than
Catholicism, Christianity, Buddhism, or Islam. Women are the happiest relative to men
83
84
in Islamic and Buddhist countries, which are primarily in the Middle East and Asia.
Since this finding may not be what one would expect, it is important to verify that the
female populations studied here are in fact representative of the populations as a whole.
For example, if the Gallup Poll is only surveying educated women in the Middle East and
people with higher education are happier, then this difference could be due to data
limitations.
Table 2.6 compares the percent of the female population with only an elementary
education in the Gallup data to the United Nations Educational, Scientific, and Cultural
Organization’s (UNESCO’s) data (2011). There are large differences between Gallup and
UNESCO data in Burkina Faso, Kazakhstan, and Cambodia. The Gallup data contain a
larger percentage of the uneducated in Burkina Faso and Kazakhstan, which should drive
down happiness, rather than increase it. After excluding Burkina Faso and Kazakhstan,
the average difference increases to 0.17 for the Islamic countries. The patterns by
education presented in Table 2.4 remain the same.
So far, the evidence suggests that women are at least as happy as men in nearly all
countries studied, and the size of the female-male happiness gap is not correlated with a
country’s stage of development, geographic location, overall well-being, or even two
measures of women’s rights. The female-male happiness gap is larger in Islamic
countries and Buddhist countries and larger for more educated respondents compared to
less educated respondents in Asia, the transition countries, and the developed countries.
The reverse is true for Latin America and Africa. However, in all the developing regions,
women in all education categories are at least as happy as men. The cross-country
84
85
analysis suggests that women are happier or equally happy as men in nearly all countries
in spite of the fact that most objective factors tend to favor men.
Table 2.6 Comparison of the Percent with Elementary Education from
Gallup Data and UNESCO data
Gallup UNESCO Difference
Islamic Countries
Bangladesh 50 446
Burkina Faso 94 42 52
Indonesia 72 657
Iran 30 35-5
Kazakhstan 19 118
Kyrgyzstan 21 138
Malaysia 48 57-9
Mali 87 89-2
Senegal 87 95-8
Tajikistan 29 263
Turkey 77 689
Buddhist/Daoist Countries
Cambodia 28 90-62
Japan 18 28-10
Mongolia 19 22-3
Thailand 66 70-4
Vietnam 47 n/a n/a
The reported percent with an elementary education is the sum of the following educational
categories: no schooling, incomplete primary, and primary (ISCED 1). For Iran,Tajikistan,
and Japan, elementary education also includes lower secondary (ISCED 2). Data are from
United Nations Educational, Scientific, and Cultural Organization (2011). Educational
Attainment of the Population Aged 25 years and Older/ Latest Year Available. Data
retrieved on March 1, 2011.
2.5.2 WITHIN-COUNTRY RESULTS
The next step of the analysis is to determine whether accounting for individual
circumstances reduces or increases the size of the female-male happiness gap. This
85
86
analysis will determine whether women with the same circumstances as men are equally
happy, happier, or less happy than men. Demographic characteristics, including age, age
squared, education, and marital status, are the first set of explanatory variables added to
the regression. The happiness gap would decrease if the factors that make people happy
are the same factors that are greater for women. For example, Easterlin and Zimmerman
(2006) show that marriage increases happiness and if women are on average more likely
to be married, then controlling for marriage should reduce the size of the female-male
happiness gap. In Figure 2.6, I plot the female-male happiness gap from the no-controls
case ( β from equation 1) on the x-axis and the female-male happiness gap after
controlling for demographic characteristics ( β
1
from equation 2.2) on the y-axis. The
dashed line is the 45-degree line; if the coefficients from regressions (2.1) and (2.2) are
the same, the coefficients would be on the 45-degree line. If these factors reduce the size
of the female-male life satisfaction gap then the coefficients should be clustered around
zero on the y-axis. Neither of these situations occurs. The female-male happiness gap
increases when the controls are included in the regressions and the coefficients are nearly
all above the 45-degree line. The positive coefficient on female is now significant in 18
of the countries compared to 14 in the no-controls case and there are no countries in
which men are more satisfied than women.
Figure 2.6 includes the predicted values from an OLS regression of the
coefficients from equation (2.2) against the coefficients for equation (2.1). The fact that
the fitted line is below the 45-degree line indicates that including the independent
variables increases the size of the female-male happiness gap in nearly all countries, and
86
87
the slope of the line suggests that the controls have a larger effect on countries where the
gap between women and men is smaller (or reversed). The coefficient after controlling
for demographic characteristics is statistically significantly larger than the coefficient
from the regression without controls in 26 of 73 countries.
Figure 2.6 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics
The regressions with controls include the following explanatory variables: age, age squared, marital status,
education level, and wave effects. South Korea is out of the range of the graph. The OLS regression is
(SatLife
f
-SatLife
m
)
controls
= 0.07 + 0.90(SatLife
f
-SatLife
m
)
nocontrols
. R
2 =
0.88 N=72.
[10.15] [23.07]
ARG
AUS
AUT
BEL
BFA
BGD
BGR
BLR
BOL
BRA
CAN
CHE
CHL
CMR
COL
CRI
CZE
DEU
DNK
DOM
ECU
ESP
EST
FIN
FRA
GBR
GRC
GTM
HND
HUN
IDN
IND
IRL
IR
ITA
JPN
KAZ
KEN
KGZ
KHM
LTU
LVA
MEX
MLI
MNG
MOZ
MYS
NIC
NLD
NOR
NPL
PAN
PER
PHL
POL
PRT
ROM
RUS
SEN
SLV
SVK
SVN
SWE
THA
TJK
TUR
TZA
UGA
URY
USA
VEN
VNM
-.4 -.2 0 .2 .4 .6
Controls for Demographic Charactersitics
-.4 -.2 0 .2 .4
No Controls
Coefficient on Female Predicted Values
45-degree Line
87
88
The results when the variable on ideal children is added to the regression are in
Figure 2.7. This variable is only available in 58 countries, but the results are very similar
to the previous findings. The coefficients are mostly above the 45-degree line and the
slope of the fitted regression line is nearly identical to that in Figure 2.6. These figures
suggest women with the same age, marital status, reported ideal number of children, and
education level are equally happy or happier than men in nearly all countries.
Figure 2.7 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics including Ideal Children
The regressions with controls include the following explanatory variables: age, age squared, marital
status, health problems, employment status, education, attendance at a religious ceremony, residential
location, and wave effects. Colombia is beyond the range of the graph. The OLS regression is
(SatLife
f
-SatLife
m
)controls= 0.07 + 0.90(SatLife
f
-SatLife
m
)nocontrols. R
2
= 0.88 N=58.
[8.36] [20.14]
ARG
AUS
AUT
BEL
BFA
BGD
BLR
BOL
CAN
CHL
CMR
DEU
DOM
ECU
ESP
EST
FIN
FRA
GBR
GTM
HND
IDN
IND
IRL
IRN
ITA
JPN
KAZ
KEN
KGZ
KHM
LTU
LVA
MEX
MLI MNG
MOZ
MYS
NIC
NLD
NOR
NPL
PER
PHL
PRT RUS
SEN
SLV
SWE
THA
TJK
TUR
TZA
UGA
USA
VEN
VNM
-.2 0 .2 .4 .6
Controls for Demographic Charactersitics
including Ideal Children
-.2 0 .2 .4 .6
No Controls
Coefficient on Female Predicted Values
45-degree Line
88
89
Figure 2.8 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristics and Life Circumstances
The regressions with controls include the following explanatory variables: age, age squared, marital status,
health problems, employment status, education, attendance at a religious ceremony, residential location,
and wave effects. South Korea is beyond the range of the graph. The OLS regression is
(SatLife
f
-SatLife
m
)
controls
= 0.11 + 0.85(SatLife
f
-SatLife
m
)
nocontrols
. R
2
= 0.81 N=73.
[12.48] [17.46]
The next set of regressions includes life circumstances as explanatory variables –
residential location, employment status, attendance at a religious ceremony, and health
problems. In Figure 2.8, the coefficients from the no-controls case are plotted against the
coefficients after controlling for demographic characteristics and life circumstances.
Again, the coefficients are nearly all above the 45-degree line and are greater than zero.
This indicates that between men and women of the same demographic characteristics and
life circumstances, women are happier. In fact, controlling for these life circumstances
ARG
AUS
AUT
BEL
BFA
BGD
BGR BLR
BOL
BRA
CAN
CHE
CHL
CMR
COL
CRI
CZE
DEU
DNK
DOM
ECU
ESP
EST
FIN
FRA GBR
GRC
GTM
HND
HUN
IDN
IND
IRL
IRN
ITA
JPN
KAZ
KEN
KGZ
KHM
LTU
LVA
MEX
MLI
MNG
MOZ
MYS
NIC
NLD
NOR
NPL
PAN
PER
PHL
POL
PRT
ROM
RUS
SEN
SLV
SVK
SVN
SWE
THA
TJK
TUR
TZA
UGA
URY
USA
VEN
VNM
-.4 -.2 0 .2 .4 .6
Controls for Demographic Charactersitics
and Life Circumstances
-.4 -.2 0 .2 .4 .6
No Controls
Coefficient on Female Predicted Values
45-degree Line
89
90
makes the coefficients on female significant in 22 countries, compared to 18 countries
when only controlling for demographic characteristics. The coefficients are statistically
significantly larger in 28 of the countries; the average female-male difference increases to
0.15 from 0.11.
Does accounting for economic circumstances increase the size of the female-male
happiness gap? In Figure 2.9, the coefficients with no controls are plotted against the
coefficient after controlling for demographic characteristics, life circumstances, and
economic factors. The economic factors are income and occupation. Because of data
limitations, seven countries are dropped from the analysis and 25% of the observations
from the remaining countries are missing.
12
The results are consistent with the previous
figures – nearly all coefficients are above the 45-degree line and positive. The
coefficients are significantly larger than the no-controls case in 19 of 66 countries. If I
remove the countries where the coefficients may be artificially high due to selection (as
discussed in the data section), then the coefficients are significantly larger than in the no-
controls case in 16 countries.
13
Women in the same occupations, with the same income,
demographic characteristics, and life circumstances as men are significantly happier than
men in about a third of the countries studied, and there is only one country, Costa Rica,
where women are less happy than men.
12
The occupation variable is not available for Brazil and Hungary. Income data are not available for the
Czech Republic, Denmark, Mali, Mozambique, and Panama.
13
The three countries that may have significant coefficients dues to selection are Mexico, the Dominican
Republic, and Tajikistan.
90
91
Figure 2.9 Coefficients on Female: No Controls vs. Controls for Demographic
Characteristic, Life Circumstances, and Economic Factors
The regressions with controls include the following explanatory variables: age, age squared, marital status,
health problems, employment status, education, residential location, attendance at a religious ceremony,
income, occupation, and wave effects. South Korea is beyond the range of the graph. The OLS regression
is (SatLife
f
-SatLife
m
)
controls
=0.11 + 0.85(SatLife
f
-SatLife
m
)
nocontrols
. R
2
= 0.7 N=66.
[9.54] [15.05]
From these figures it is clear that objective factors increase the size of the male-
female happiness gap. This means that women with the same demographic
characteristics, life circumstances, and economic circumstances are at least as happy as
their male counterparts.
ARG
AUS
AUT
BEL
BFA
BGD
BGR
BLR
BOL
CAN
CHE
CHL
CMR
COL
CRI
DEU
DOM
ECU
ESP
EST
FIN
FRA
GBR
GRC
GTM
HND
IDN
IND
IRL
IRN
ITA
JPN
KAZ
KEN
KGZ
KHM
LTU
LVA
MEX
MNG
MYS
NIC
NLD
NOR
NPL
PER
PHL
POL
PRT
ROM
RUS
SEN
SLV
SVK
SVN
SWE
THA
TJK
TUR
TZA
UGA
URY
USA
VEN
VNM
-.4 -.2 0 .2 .4 .6
Controls for Demographic, Economic,
and Life Circumstances
-.4 -.2 0 .2 .4
No Controls
Coefficient on Female Predicted Values
45-degree Line
91
92
2.5.3 ACROSS-COUNTRY PATTERNS AFTER INCLUDING CONTROL
VARIABLES
This paper has shown that women are at least as happy as men in nearly all
countries and the size and significance of the happiness gap increases when controlling
for individual circumstances and perceptions. Looking at the slopes of the fitted lines in
Figures 2.6-2.9, there is some evidence to suggest that the happiness gap increases more
for countries where the initial female-male difference is smaller or even negative. This
Table 2.7 Average Female-Male Difference in Life Satisfaction from the Four
Specifications by the Results of the Regressions without Controls
Specification
No significant
difference
Women
happier
Men
happier All N
No controls -0.005 0.31 -0.24 0.04 73
[0.01] [0.03] [0.03] [0.02]
Controls:
Demographic controls 0.06 0.36 -0.10 0.11 73
[0.01] [0.04] [0.04] [0.02]
Demographic characteristics
and life circumstances 0.10 0.38 -0.03 0.15 73
[0.01] [0.04] [0.08] [0.02]
Demographic characteristics,
life circumstances, and
economic factors 0.10 0.37 0.02 0.16 66
[0.02] [0.05] [0.09] [0.02]
See the notes to Figures 2.6-2.9 for lists of the explanatory variables used in the regressions. The categories
no significant difference, women happier, and men happier are based on whether the coefficient on the
dummy variable for female is signifcantly different from zero in the no-controls specification using a two-
tailed t-test. Standard errors are in parentheses.
is confirmed in Table 2.7, which contains the average of the coefficients in each of the
specifications grouped by whether women are equally happy, happier, or less happy than
men in the no-controls case. Adding controls has a larger effect on countries where men
92
93
are happier or equally happy. The largest overall increase in coefficients occurs between
the no-controls specification and the specification that includes demographic controls.
Table 2.8 Average Female-Male Differences in Explanatory Variables by Female-
Male Difference in Life Satisfaction
No significant
difference
Women happier
Men happier
N Mean N Mean N Mean
Life satisfaction 55 0.00 14 0.31 4 -0.24
Single (%) 55 -0.09 14 -0.10 4 -0.12
Married (%) 55 -0.01 14 0.04 4 0.04
Previously married (%) 55 0.09 14 0.06 4 0.08
Elementary education (%) 55 0.04 14 0.04 4 0.09
Secondary education (%) 55 -0.03 14 -0.03 4 -0.07
Tertiary education (%) 55 -0.01 14 -0.02 4 -0.03
Without health problems (%) 55 -0.04 14 -0.01 4 -0.06
Employed (%) 55 -0.17 14 -0.26 4 -0.23
Live in a large city (%) 55 0.00 14 0.02 4 -0.04
Attend a religious ceremony (%) 55 0.07 14 0.02 4 0.16
Income (*10^-13) 49 -3.49 14 -1.40 4 -4.23
White collar (%) 53 0.00 14 -0.03 4 -0.05
Business owner (%) 53 0.01 14 -0.04 4 -0.03
Service worker (%) 53 0.02 14 0.00 4 0.00
Non-farm manual worker (%) 53 -0.15 14 -0.11 4 -0.12
Farmer (%) 53 -0.04 14 -0.08 4 -0.02
Other (%) 53 -0.01 14 -0.01 4 -0.01
The categories no significant difference, women happier, and men happier are based on statistical
significance.
In Table 2.8, the mean differences for the explanatory variables are presented by whether
the difference in the no-controls case is statistically significant. In the countries where
men are happier, the differences in education, percent married, and income are greater
93
94
than in countries where women are equally happy or happier than men. This can explain
why controlling for these factors matters most in the countries where men are happier.
If the countries where there are large female-male differences in the explanatory
variables are the same countries where the control variables have the largest effect, then
some of the female-male differences in life satisfaction can be attributed to differences in
the endowments of the sexes. Thus, it is possible that the controls have differential
Table 2.9: Average Female-Male Difference in Life Satisfaction from Four
Specifications by Geographic Region
Developed
countries
Transition
countries Asia
Latin
America Africa Total
No controls 0.01 -0.07 0.14 0.02 0.13 0.04
[0.04] [0.02][0.05] [0.04] [0.05][0.02]
Controls:
Demographic characteristics 0.08 0.04 0.21 0.09 0.13 0.11
[0.04] [0.02] [0.05] [0.04] [0.05] [0.02]
Demographic characteristics
and life circumstances 0.10 0.08 0.25 0.16 0.14 0.15
[0.04] [0.03] [0.05] [0.03] [0.05] [0.02]
Demographic characteristics,
life circumstances, and
economic factors 0.13 0.08 0.23 0.16 0.14 0.15
[0.05] [0.04] [0.06] [0.05] [0.08] [0.02]
See notes to Figures 2.6-2.9 for lists of the explanatory variables used in the regressions.
impacts based on geographic location or economic growth. Table 2.9 shows the average
coefficients from each specification by geographic region. The controls have similar
effects across regions, except Africa where the controls have, on average, no effect on the
female-male happiness gap. In developed countries, Asia, and Latin America, including
94
95
demographic characteristics and life circumstances as controls in the regressions,
increases the size of the coefficients by 0.07 on average. These controls have a larger
effect on the transition countries, where the objective factors favor men even more than
in the other countries. Including economic circumstances causes a small increase in the
Figure 2.10 Impact of the Control Variables on the Female-Male Difference in Life
Satisfaction and Log GDP per Capita
See notes to Figures 2.2 and 2.9. The OLS regression is Coeff
4
-Coeff
1
=- 0.05 + 0.02lnGDP. R
2
= 0.03
N=58. [0.44] [1.38]
FRA
JPN
AUT
PRT
NOR
ESP
BEL
NLD
IRL
SWE
CAN
GBR
FIN
DEU
AUS
CHE
USA
ITA
GRC
SVK
ROM
BGR
EST RUS
LVA
BLR
LTU
POL
SVN
IDN
MYS
TJK
MNG
PHL
KAZ
IRN
NPL
BGD
KOR
KHM VNM
TUR
KGZ
IND
THA
PER
BOL
HND
CHL
MEX
NIC
ECU
SLV
VEN
GTM
ARG
COL
URY
CRI
DOM
CMR
UGA
KEN
TZA
SEN BFA
-.2 0 .2 .4 .6
Difference in Coefficients
from Equation 1 and Equation 4
7 8 9 10 11
Log GDP per Capita
95
96
mean of the developed countries’ coefficients, while there is small decrease for the Asian
countries. However, the average change in female-male differences in life satisfaction is
always less than 0.11 when controls are included, confirming that female-male
differences in life satisfaction do not depend on geography.
Lastly, I examine whether the impact of the control variables on the coefficients
depends on a country’s stage of economic development. For each country, I compute the
difference in coefficients from the no-controls case and the coefficients after controlling
for economic factors, life circumstances, and demographic characteristics. The
differences in those coefficients are plotted against log GDP per capita in Figure 2.10.
There is no systematic relationship between stage of development and the impact of the
control variables on female-male differences in life satisfaction.
2.6 CONCLUSION
This paper has shown that women are at least as satisfied as men in nearly all
countries studied, regardless of a country’s stage of development, overall well-being, or
geographic location. This is in spite of the fact that women are on average less educated,
have lower income, and are more likely to be widowed or divorced. The results after
controlling for individual circumstances that usually favor men show that the “pure
effect” of being female is larger than the average effect. Women are statistically
significantly more satisfied with life than men in about a third of the countries. Women
of the same age, education level, occupational status, etc. are equally happy or happier
than their male counterparts. The increase in the happiness gap is larger in countries
with smaller initial differences in life satisfaction; however there is no consistent pattern
96
97
in the size of the differences across geographic regions or stage of economic
development.
These findings have possibly led to more questions than answers. If women’s
objective circumstances cannot explain why women are happier, then what can explain
it? Why are women in countries with low levels of women’s rights happier than men?
There are a several possible explanations that are beyond the scope of the paper, but
worth mentioning. Two possibilities, biology and personality, have largely been rejected
by the psychology literature (Nolen-Hoeksema & Rusting 1999). The third and most
likely explanation is that aspirations formed from culture and social norms play an
important role in well-being. It is possible that women have lower aspirations than men
so when they evaluate their circumstances in terms of life satisfaction, they report higher
well-being.
Lalive and Stutzer (2010) investigate whether social norms can explain why
women in Switzerland are happier relative to men despite being paid less. They use as a
proxy for social norms the percent of the community that voted for legislation for equal
rights for men and women. Surprisingly, they find that employed women have lower life
satisfaction if they live in communities where a larger percentage of the population voted
for equal rights. Easterlin and Plagnol (2008) show that early in adult life women have
higher happiness because they are more likely than men to fulfill their material goods and
family life aspirations. Later in life, the reverse is true leading men to be more
satisfaction with life. These articles suggest that aspirations and social norms play an
important role in people’s evaluation of their circumstances.
97
98
Thus, this paper, which is the first to examine female-male differences in
countries of all stages of development, is a starting point for much additional research on
why women are at least as happy as men despite the fact that so many factors favor men.
98
99
CHAPTER 3: IS CALIFORNIA POLLUTING CHILDREN’S MINDS?
THE EFFECT OF AIR POLLUTION ON ACADEMIC
PERFORMANCE
14
3.1 INTRODUCTION
The effects of air pollution on child and adult health have been widely studied.
Air pollution is associated with asthma, lower lung function, hay fever, infant mortality,
and emergency room visits (Chay & Greenstone, 2003ab; Currie & Neidell, 2005,
Gauderman et al., 2000; McConnell et al., 2002; McConnell et al., 2003; Neidell, 2004;
and Rabinovitch, Strand, & Gelfand, 2006). Recently, economists and epidemiologists
have noted that increased air pollution also increases school absenteeism (Currie,
Hanushek, Kahn, Neidell, & Rivkin, 2009; Gilliland et al., 2001; Ransom & Pope, 1992),
and that asthma may reduce school performance (Currie, Stabile, Manivong, & Roos,
2010).
If air pollution negatively affects children’s health and increases school
absenteeism, it is plausible that the educational attainment of students would also be
affected. We contribute to the literature by determining i) whether recent changes in air
pollution affected the academic performance of school children in California and ii) the
size of the effect. The central difficulty in identifying the effects of air pollution on
academic performance is that air pollution is likely to be correlated with socioeconomic
status; higher income families are likely to sort into lower-pollution neighborhoods.
Since children from lower income families tend to have lower test scores than those from
14
This chapter is based ongoing work with John Ham and Edward Avol. Any opinions, findings,
conclusions, or recommendations in this material are my own. I am responsible for any errors.
99
100
higher income families, a finding of a negative effect of pollution on test scores may
simply reflect selection. Of course this problem is not unique to our paper, and all of the
papers in the economics literature attempt to eliminate the confounding factor of
socioeconomic status when estimating the effect of pollution on an outcome variable. The
articles use appropriate conditioning variables and fixed effects and/or use variation
across time intervals that presumably are too short to reflect location behavior, albeit at
the cost of making the strong separability assumption discussed below. We follow the
first strategy in our work.
Specifically, we use the results of the California Standards Tests in mathematics
and English/language arts as measures of academic performance (California Department
of Education, 2002-2008c). Since test scores are reported for each grade in each school,
we calculate the pollution measures for each school in California from all monitors
(weighted by distance from the school) within a twenty mile radius of the school. The
pollution measures used in this study are the percent of days above the standard for
carbon monoxide (CO), nitrogen dioxide (NO
2
), ozone (O
3
), coarse particulate matter
(PM
10
), and fine particulate matter (PM
2.5
). Our sample is limited to the years 2002 to
2008 because that is the period during which the content of the tests remained constant.
The tests are taken within a 10 day window of when 85% of the instructional year is
completed (California Department of Education Laws ch. 2, § 855, 2007). California
public schools have 180 days of instruction, so tests should take place near the 153
rd
day.
The school year typically begins in August or September and ends in late May or June,
which would mean that testing occurs in April or May. Therefore, we use an average of
100
101
the daily pollution levels for September through May. To avoid the issue of confounding
factors or selection biasing our results, we include grade-school fixed effects, year
effects, as well as a host of time-varying school quality, demographic, and community
characteristics in the regressions. We also consider a number of robustness checks. Thus
this paper makes an important contribution to the pollution literature by being the first to
estimate the causal effect of pollution on test scores.
The paper proceeds as follows. We review the relevant literature in the following
section. We first discuss the economics literature to date on the related topics of the effect
of pollution on health, the effect of pollution on school absenteeism, and the effect of
asthma, which is exacerbated by and possibly caused by pollution, on academic
performance. Next, we review work from the public health literature concerning the
above issues as well as examining correlations between pollution and test scores. We
conclude that since the epidemiological literature is based on cross-section data, it is
unlikely to have sufficient controls to consider the estimated effects as causal. In section
3.3 we discuss our empirical (and identification) strategy, while we describe the data in
section 3.4. We present our results in section 3.5, where we measure (separately) the
effects of five pollutants on two measures of academic performance: i) the percent of the
students considered at least proficient in mathematics, and ii) the percent of students in a
grade considered at least proficient in English/language arts. We find that air pollution,
specifically ozone and particulate matter, has a significantly negative, but small, effect on
the academic performance of school children. The average percent of days above the
standard for ozone is 1.9. If this increased to 100%, the percent of a grade scoring at
101
102
least proficient in mathematics would decrease by an average of 9.3 percentage points. If
the percent of days above the standard for PM
10
increased to 100% from the mean of
11.71%, then the percent scoring at least proficient would decrease by 2.1. Increasing the
percent of days above the standard to 100% for PM
10
and ozone would increase the
percent at least proficient in language arts by 1.9 and 16.6, respectively.
3.2 MECHANISMS BY WHICH POLLUTION COULD AFFECT ACADEMIC
PERFORMANCE
There are three mechanisms by which pollution could affect academic
performance: (i) school absenteeism due to illness caused by pollution; (ii) attention
problems in school due to illness caused by pollution; and (iii) fatigue when doing
homework due to illness caused by pollution.
3.2.1 EVIDENCE FROM THE ECONOMICS LITERATURE
Mechanisms (i)-(iii) rely on pollution having a negative effect on health, and then
health impacting students’ academic performance, and here we highlight some of the
economics articles investigating these mechanisms.
15
Chay and Greenstone (2003ab)
examine the effect of air pollution on infant mortality rates in United States counties
between 1980 to 1982. Their initial identification strategy is based on assuming that
county fixed effects, state trends, year effects and socioeconomic controls are sufficient
to eliminate most spurious correlations between pollution and infant mortality. Their
15
There are several articles outside the economics literature that establish this link as well (see McConnell
et al., 2002, Gauderman et al., 2000, and McConnell et al., 2003); however, for brevity, we focus on the
economics papers.
102
103
socioeconomic controls include mother-specific characteristics aggregated to the county-
level, including education, ethnicity, income, prenatal care, and age. They chose this
period in the hope that much of the remaining variation in pollution after controlling for
these variables comes from the differential impacts of the 1980 recession on pollution
levels. Therefore, changes in pollution are transitory and less likely to affect location
choice. One caveat to this identification strategy is that one must ignore the fact that the
recession will also directly affect location decisions as adults move from hard-hit labor
markets to more prosperous labor markets, i.e. there may still be selection at work.
They weaken their identifying assumptions in two ways. First they treat current
pollution as endogenous, instrumenting for the change in pollution with lagged pollution
levels; the latter will be a valid instrumental variable (IV) if there is no autocorrelation in
pollution. Second, they find all counties with low levels of manufacturing employment in
1980, and then look at neighboring counties with and without high levels of
manufacturing employment. If a neighboring county experiences a substantial decrease in
manufacturing employment, it is likely to experience a reduction in Total Suspended
Particles (TSPs). Because of wind and other weather components, a reduction in TSPs in
a neighboring county should affect a county’s own pollution levels. The authors then
compare the effect of changes in pollution on changes in infant mortality in counties that
had neighbors with a large reduction in employment to those in counties with neighbors
who did not experience a large reduction in employment. Thus their new identifying
assumption is that demand shocks in a neighboring county will not have spillover effects
and induce migration from the county under consideration.
103
104
Neidell (2004) evaluates how seasonal changes in pollution change asthma
hospital admission rates for different age groups by month, conditional on zip code-year
fixed effects and year-month fixed effects. For each zip code, he constructs a weekly
measure of pollution by taking the average of pollution levels recorded at monitors within
20 miles of the centroid of the zip code weighted by the inverse distance to the monitor.
The outcome variable is the number of asthma-related emergency room visits in each zip
code-month observation, where a visit is classified as asthma-related based on the
principle diagnosis from the California Hospital Discharge Data. The control variables
include the sex, race, and age of the patient, expected source of payment to the hospital,
weather, and housing prices. Neidell finds that of the pollutants considered, carbon
monoxide has a significant effect on hospitalizations for asthma among children ages 1–
18, while none of the pollutants he considered has a clear impact on hospitalizations for
infants. Using his estimated coefficients and the expected number of asthma admissions
from 1992 and 1998 pollution levels, Neidell calculates that the decline in pollution
during this time period led the asthma admission rates to go from 13.5% to 4.6%. He
also tests whether families display avoidance behavior by including the number of smog
alerts as a control variable. He concludes that a smog alert decreases asthma
hospitalizations by roughly 1%, while including these announcements raises the effect of
O
3
on admissions, although O
3
does not appear to significantly affect hospitalization
rates. Since he uses a large number of fixed effects, the only caveat to his results is that
he must assume that pollution in previous months does not affect admissions in
subsequent months; this assumption would be violated if previous pollution caused
104
105
individuals to develop asthma which made them more sensitive to current pollution.
While of course this criticism could also be leveled at studies using annual data, as in the
consumer demand literature, it is likely that separability over time periods is less credible
as the size of the period decreases.
Currie and Neidell (2005) evaluate the effect of increased air pollution on infant
mortality during the period 1989 to 2000. The authors construct a weekly pollution
measure similar to that in Neidell (2004), by taking the average of pollution levels
recorded at monitors within 20 miles from of centroid of the zip code weighted by the
inverse distance to the monitor. The authors include zip code month fixed effects and zip
code year fixed effects. The authors also include various mother-specific factors,
including mother’s age, race, ethnicity, education, marital status, zip code of maternal
residence, use of prenatal care, and private/public insurance. Other covariates include
weekly county-level averages for weather, date of birth, birth weight and gestation
period. They use a flexible discreet hazard model where the outcome variable is equal to
one if the child died within the week. They find that in periods of higher pollution, infant
mortality rates are higher, but that prenatal exposure to pollution does not affect infant
mortality. They often find that ozone has the incorrect sign, but attribute that to the fact
that there is a negative correlation between ozone and other pollutants. One possible
criticism is that they must assume there is no unobserved heterogeneity at the individual
or zip code level, since either form of heterogeneity will cause their parameter estimates
to be biased.
105
106
Currie et al. (2009) investigate the effect of pollution on school absences using
data from the Texas Schools Project, a longitudinal administrative data set on student
absenteeism in Texas. They aggregate pollution data from the Texas Commission on
Environmental Quality into six week time blocks, and merge these data with the
administrative absenteeism data. Their identification comes from the variation in
pollution across six-week attendance periods within a year or within an attendance period
across years. In the former case, they include school by attendance period fixed effects
and in the latter case, they include school by year fixed effects. They measure pollution
by determining whether each day is 0-25%, 25-50%, 50-75%, 75-100% and greater than
100% of the relevant Environmental Protection Agency (EPA) threshold for the pollutant.
They then calculate the shares of days in each category for the six-week attendance
period. Their main finding is that CO between 75-100% of the air quality standards
threshold and above the threshold has a positive and significant effect on school
absences. Ozone is not statistically significant in most specifications, but they did find a
statistically significant increase in absences associated with PM
10
levels between 50-75%
of the EPA threshold. As the authors acknowledge, this latter result must be viewed with
caution since one significant result among many coefficients can occur by chance. Again
the identifying assumption in their work is that past pollution levels do not affect current
absences; as noted above this would be violated if previous pollution levels increased
the incidence of asthma and the effects of pollution in the current period.
Of course, air pollution will affect academic performance through health only if
health problems affect performance. Currie et al. (2010) evaluate the effect of various
106
107
childhood diseases, including asthma, on (i) performance on a literacy exam, (ii) whether
the students enrolled in a college preparatory math class, (iii) whether they were in the
twelfth grade by age 17, and (iv) whether they used social assistance. They match school
administrative data, social assistance records, and health records for young adults in
Manitoba, Canada born between 1979 and 1987. With a mother fixed effect (which
controls for time constant family characteristics), they investigate whether having been
treated for asthma at various ages (0-3, 4-8, 9-13, 14-18) affects these young adult
outcomes by using the variation across siblings in the incidence of asthma. They find (at
the 10% level) that (a) asthma at ages 9 to 13 had a significant negative effect on taking a
college preparatory math class and (b) asthma at ages 14 to 18 sometimes had a negative
effect on the literacy score in the 12
th
grade. They find no effect of earlier asthma,
conditional on current asthma. As the authors acknowledge, their results must be viewed
with caution since two significant coefficients could happen by chance in this framework.
Their identifying assumption is that there are no time varying family characteristics, i.e.,
socioeconomic status, that would be correlated with both asthma and these outcome
variables.
We next discuss below a number of related studies from the epidemiological
literature, all of which are based on cross-section data and a limited number of controls.
As a result, these researchers are much more limited in their ability to deal with selection
and endogeneity issues; this latter problem is accentuated by the fact that none consider
instrumental variable estimation.
107
108
3.2.2 EVIDENCE FROM THE EPIDEMIOLOGICAL LITERATURE
Gilliland et al. (2001) use the Children’s Health Study data to evaluate the effect
of pollution on absenteeism. They study a cohort of 2,081 4
th
grade students who reside
in 12 southern California communities. They track the students’ absences for the first 6
months of 1996 and followed up with the students’ parents to determine if the absence is
illness-related or not, and if so, whether it is an upper-respiratory, lower-respiratory, or
gastro-intestinal illness. The type of illness is determined by the symptoms described
during phone interviews. Using daily pollution levels from monitors located near the
schools and a community fixed effect model, the authors use within-community variation
in pollution across the six month period to determine its effect on average daily absences
due to respiratory illness. They find that ozone has a statistically significant relationship
(partial correlation) with reported absences from upper respiratory and lower respiratory
illness rates. In order to obtain a causal effect they need to assume that within a
community, families do not sort themselves based on permanent differences in pollution
across the community; which is a strong, but necessary assumption given their data.
We now consider a number of other studies which use cross-section data and no
fixed effects, rendering them less credible in terms of estimating causal effects. Fowler,
Davenport, and Garg (1992) analyze the effect of asthma on different outcomes for the
United States. They use data for 10,362 children in first through twelfth grade from the
1988 United States National Health Interview Survey. They find that children with
asthma are more likely to have a learning disability than children who do not have
asthma. In addition, among households with incomes below $20,000, asthmatic children
are twice as likely to fail a grade as those without asthma, but among higher income
108
109
families, asthmatic children have only a slightly higher failure rate than non-asthmatic
children.
16
With a sample of 1,058 kindergarten-age children from Rochester, New York
in 1998, Halterman et al. (2001) compare the parent-reported development skills of
asthmatic children to non-asthmatic children. After controlling for type of health
insurance, education of the care-giver, gender, and pre-kindergarten education, the
authors find that asthmatic kindergarten-aged children scored lower in school readiness
skills (one category of reported development skills), than their non-asthmatic peers. Butz
et al. (1995) obtain demographic, asthma symptoms and psychosocial information for
392 children in kindergarten through eighth grade in 42 schools in Baltimore, Maryland.
Asthma symptoms are divided into low, medium, and high levels. A child is considered
to be exhibiting behavior problems if her score on a questionnaire containing
standardized psychosocial questions is higher than a given threshold. Using logistic
regressions, the authors conclude that parents who report that their children have higher
levels of asthma symptoms are twice as likely to report a behavioral problem compared to
parents who report lower levels of asthma symptoms.
Bussing, Halfon, Benjamin, and Wells (1995) first use responses to the 1988
National Health Interview Survey on Child Health to categorize children into those who
suffer from asthma alone, those who suffer from asthma combined with other chronic
conditions, those who suffer from other chronic conditions alone or those who have no
chronic (including asthmatic) conditions. They then combine this information with a
Behavior Problem Index constructed from psychosocial questions in the survey. Using
16
This suggests the possibility of heterogeneous asthma effects by socioeconomic status, but we felt we did
not have sufficient data to explore this possibility in our analysis.
109
110
logistic regressions, the authors find that children with severe asthma alone are nearly
three times as likely to have severe behavioral problems as children without a chronic
condition. Halterman et al. (2006) investigate the relationship between behavioral
problems and asthma symptoms for a cohort of 1,619 inner-city students in Rochester,
New York. The parents of these kindergarten-age children were surveyed about their
children’s health and behavior. The authors find that children with persistent asthma
score worse on peer interactions and task orientation, and are more likely to exhibit shy
and anxious behaviors compared to non-asthmatic children.
17
Finally, Pastor, Sadd, and Morello-Frosch (2004) evaluate the relationship
between academic performance and environmental hazards in the Los Angeles Unified
School District in 1999. They combine data on schools’ Academic Progress Index (API)
with information on their proximity to Toxic Release Inventory (TRI) emissions and
census tract-level estimated respiratory risks associated with concentrations of 148
ambient air toxins. This latter measure of exposure at the tract-level is the sum of hazard
ratios for each pollutant, where the hazard ratio is calculated by dividing the EPA’s tract-
level exposure estimate for a particular pollutant by the amount of toxicant below which
there should be no adverse health effects. According to the California Department of
Education (2010, p. 6), the API “is calculated by converting a student’s performance on
statewide assessments across multiple content areas into points on the API scale. These
points are then averaged across all students and all tests.” Each school receives one API
17
According to the National Heart, Blood and Lung Institute of the National Institutes of Health (2007, p.
72), asthma is considered persistent if the patient experiences symptoms more than two days per week,
limitation in activities, some nighttime awakenings or use of short acting beta
2
agonists combined with
either more than two exacerbations requiring oral steroids or more than four wheezing episodes longer
lasting than a day per year.
110
111
score. In an OLS regression, the authors regress the API score on the respiratory risk
index, a dummy equal to one if a facility releasing substances covered by the TRI and in
the 33/50 program is within one mile of the school.
18
The authors find that having a
33/50 facility within a 1 mile radius has a negative and significant effect on academic
performance even after controlling for some socioeconomic status variables such as
parents’ education, percent minority, and percent who are English learners.
Pastor, Morello-Frosch, and Sadd (2006) expand their previous analysis to all
schools in California. They again use the API score as their measure of academic
performance and construct similar respiratory risk indices. To examine whether the
mechanism by which exposure to air pollution affects academic performance is through
asthma, the authors first run a tobit regression of the three-year averaged, age-adjusted,
asthma hospitalization rates by Zip Code Tabulation Area on their measure of exposure
controlling for socioeconomic status. They find that areas with higher respiratory risk
have higher hospitalization rates. They then turn to academic performance, and find
again that schools located in higher pollution areas have lower API scores. They estimate
that moving from the seventy-fifth percentile to the median level of the respiratory hazard
ratio would improve test scores by about 1.2%. However, the assumptions necessary to
interpret their estimates of the effect of pollution on school performance as causal are
identical to the epidemiological studies discussed above and hence likely to be much too
strong. Below we aim to use an econometric approach very similar to those used in the
18
The 33/50 program was a voluntary program by the EPA set to reduce the release of 17 targeted priority
chemicals. It was enacted in 1991 and its goals was to reduce the release and transfer of chemical by 50%
by 1995, measured against a 1988 baseline (EPA, 1999).
111
112
economics papers discussed above, so that our estimates of the effect of pollution on
school performance can be credibly viewed as causal.
3.3 METHODS
Our data, described in much more detail below, consists of approximately 24,000
grade-school units observed for up to seven years. Given that we have panel data, our
first empirical specification is:
where
gst
S represents a measure of performance on a given standardized test for grade g
in school s (located in county c) in year t;
st
P represents pollution at school s at time t;
gst
X represents racial composition in grade g at school s at time t;
st
W represents school
specific characteristics for school s at time t;
ct
Z represent time-changing county level
factors;
gs
f represents school-grade fixed effects;
t
D represents a full set of time
dummies; and
gst
is the error term.
As noted above, being able to account for confounding factors is crucial to the
credibility of our analysis (or any such analysis). As control variables, we use yearly data
for each grade of each school on the students’ ethnicity from the California Basic
Educational System Data (California Department of Education, 2002-2008b). Our other
educational controls are from the Academic Performance Index (API) data files
,
12 3 4
(3.1)
gst gst st gst st ct gs t
SP X W Zf D
112
113
(California Department of Education, 2002-2008a). We first condition on average class
size, which is measured separately for grades 4 through 6 and kindergarten through 3
rd
grade. We also control for the following variables at the school-year level: the percent of
students receiving free or reduced-price lunches; the educational make up of parents; the
percent of students who are native English speakers; the percent of teachers who are fully
certified; and total enrollment. We also control for annual expenditure per student at the
district level from the National Center for Education Statistics’ Common Core of Data
(2002-2008). Finally, we control for a number of business cycle variables: the
unemployment rate and taxable transactions at the county level (the lowest level of
geographical aggregation available). We adjust taxable transactions and expenditures per
student for inflation.
Identification comes from assuming that all the variation in pollution over time at
a specific school, after controlling for grade-school time changing characteristics,
, , and
gst st ct
X WZ , and year,
t
D , is uncorrelated with any remaining unobservables
driving school performance. We would argue that our rich set of fixed effects and control
variables renders our identifying assumptions on par with those made in the economic
studies discussed above. Finally, we make the standard GLS (heteroskedasticity)
adjustment of weighting observations by the respective square root of the number of
students in the grade-school-year observation. However, to allow for autocorrelation
over time and any other sources of heteroskedasticity, the standard errors are still
clustered at the school level.
113
114
3.4 DATA
The summary statistics for the variables used in this study: test performance,
pollution, school characteristics, and district/county characteristics are shown in Table
3.1. Panel A contains summary statistics for the percent in each grade that were at least
proficient on the California Standards Tests (CSTs). The average percent at least
proficient in mathematics is 50.8% and the percent at least proficient in English/language
arts is 43.5%. The goal in California is for all students to score at least proficient. A
scaled score above 350 out of 600 is considered proficient, where a score above 400 is
considered advanced. We constructed the percent at least proficient by adding together
the percent advanced and the percent proficient. We include years 2002 through 2008
because this is a period during which the format of the exams remained the same. Our
analysis includes grades 2 through 6 because the same tests are administered to all
students within each grade. We avoid data from grade 7 on, since from grade 7 on,
students may take different mathematics courses based on ability – for example, algebra,
geometry, or basic mathematics – which would raise difficult selection issues for our
analysis.
Our main explanatory variables of interest are measures of pollution. As noted
above, we consider five pollutants in this paper: coarse particulate matter (PM
10
); fine
particulate matter (PM
2.5
); nitrogen dioxide (NO
2
); carbon monoxide (CO); and Ozone
(O
3
). We use these specific pollutants in our analysis because they have been studied in
the previous literature (Currie et al., 2009, Gilliland et al., 2001) and are correlated with
various diseases (Gauderman et al., 2005; Grahame & Schlesinger, 2007; Kurt,
Mogielnicki, Chandler, 1978; Linn, Szlachcic, Gong, Kinney, & Berhane, 2000;
114
115
McConnell et al., 2002, Pope & Dockery, 2006; Russell & Brunekreef, 2009; and Yu,
Sheppard, Lumley, Koenig, & Shapiro, 2000). A major source of PM
10
, PM
2.5
, NO
2
, and
CO is vehicle exhaust. Other sources of particulate matter include dust from the earth's
surface, pollen, forest fires, power plants, and factories. CO is also formed through the
improper burning of various fuels, but the greatest exposure comes from smoking
cigarettes. NO
2
is emitted from coal-burning power plants and the burning of fossil fuels.
O
3
is formed through a chemical reaction between nitrogen oxides, sunlight, and various
gaseous pollutants, which are often emitted from vehicles.
19
The pollution data are from the Air Resources Board of California, (Daily Data,
2010). The only feasible way of measuring pollution is at the school-year level, since
there is no way to obtain addresses for the students. (Thus there is no variation in
pollution across grades for a given school in a specific year.) The pollution measure used
in this study is the percent of days that exceed the California Standard for that pollutant.
The California one-hour standards are 20 parts per million (ppm) for CO, 0.18 ppm for
NO
2
, and 0.09 ppm for O
3
. The standards for PM
10
and PM
2.5
are based on a 24-hour
measure rather than a one-hour measure. The 24-hour standard is 50 micrograms per
cubic meter ( μg/m3) for PM
10
and 35 μg/m3 for PM
2.5
(California Environmental
Protection Agency, 2009). The California standards are stricter than the federal standards
for all pollutants except for PM
2.5
, which is the same as the federal standard.
To obtain our pollution measure, we first use the longitude and latitude for each
school and for each pollution monitor in California to find all monitors within a 20-mile
19
For additional information on these pollutants, see Environmental Protection Agency (2011).
115
116
radius of each school. For a given pollutant and monitor, we calculate the total number
of days that exceed the standards for that pollutant and then divide by the total number of
days that are tested. Since students usually take the California Standards Tests in April or
May, we use pollution data from September through May as an approximation of the
pollution experienced during the school year. Then for a given pollutant at a given
school in a given year, we take the weighted average of the percent of days exceeding the
standard at each monitor, where the weighting is based on the inverse distance to the
school. Thus we give monitors that are closer to the school more weight relative to ones
that are further away. We have placed the summary statistics for our pollution variables
in Panel B of Table 3.1, while in Table 3.2 we show the correlation matrix for the
pollution variables. For all schools and years, an average of 0.0004%, 0.003%, 1.94%,
11.78%, and 28.32% days of the school year are above the standards for CO, NO
2
, O
3
,
PM
10
, and PM
2.5
, respectively. The correlation matrix for the pollutants in Table 3.2
suggests that simultaneously using the different pollution measures is likely to cause a
severe multicollinearity problem, and thus we follow the literature and enter them one at
a time. PM
10
and PM
2.5
are particularly highly correlated; O
3
is uncorrelated with the
other measures.
The summary statistics for the control variables are in Panel C of Table 3.1. In
terms of the ethnic composition of the students, on average, 35% of the students are
White, 11.1% Black, 42.6% Hispanic, 7.6% Asian, and 3.8% Other. The average class
size is 26 and nearly all teachers are fully certified at an average of 95%. Further, on
average, 50% of students receive a free or reduced-price lunch. The average percent of
116
117
Table 3.1 Descriptive Statistics Across California Schools, 2002-2008
Mean
Standard
deviation
Minimum Maximum
Panel A: Test score variables
Percent at least proficient
Mathematics 50.7421.360.00 100.00
English/language arts 43.35 21.25 0.00 100.00
Panel B: Pollution variables
Percent of days that exceeded the pollution standard
Carbon monoxide (CO) 0.0004 0.11 0.00 7.40
Nitrogen dioxide (NO
2
) 0.0030.020.00 0.63
Ozone (O
3
) 1.942.500.00 24.91
Coarse particulate matter (PM
10
) 11.7812.48 0.00 79.80
Fine particulate matter (PM
2.5
) 28.3219.030.00 94.10
Panel C: Control variables
Time-varying grade-school characteristics
White (%) 35.00 28.36 0.00 100.00
Asian (%) 11.04 15.05 0.00 100.00
Hispanic (%) 42.56 29.75 0.00 100.00
African American (%) 7.58 12.51 0.00 100.00
Other (%) 3.81 5.98 0.00 98.31
Time-varying school characteristics
Reduced or free meals (%) 51.39 30.65 0.00 100.00
Parent is a high school graduate (%) 19.38 18.50 0.00 100.00
Parent has some college education (%) 24.98 13.20 0.00 100.00
Parent is a college graduate (%) 19.53 13.28 0.00 100.00
Parent attended graduate school (%) 11.82 13.91 0.00 100.00
Fully certified teachers (%) 94.72 9.31 0.00 100.00
Non-native English speakers (%) 26.24 21.91 0.00 100.00
Total school enrollment 406.89 214.76 13.00 3,110.00
Average class size 26.54 4.97 5.00 50.00
Time-varying district characteristics
Expenditure per student (*10^-3) 8.83 2.56 1.87 70.76
Time-varying county characteristics
County taxable transactions (*10^-5) 384.52 417.05 0.17 1,215.06
County unemployment rate 6.38 2.10 3.40 22.40
Average class size is measured separately for kindergarten through third grade and fourth through sixth
grade. The means for the outcome variables and the control variables are based on 143,041 observations.
The means for CO, NO
2,
O
3,
PM10, and PM2.5 are based on 143,041 observations, 131,446 observations,
138,259 observations, 136,476 observations, and 136,214 observations, respectively.
117
118
students who are non-native English speakers is 26. The average enrollment and
expenditure per student are 407 and $8,828, respectively. The average county
unemployment rate is 6.3% and the average value of taxable transactions is
approximately $384 million.
Table 3.2 Correlation Matrix of Pollution Variables, 2002-2008
Percent of days that exceed the standard:
CO NO
2
O
3
PM
10
PM
2.5
Percent of days that exceeded the standard:
CO 1
NO
2
0.221
O
3
0.01 -0.01 1
PM
10
0.150.130.55 1
PM
2.5
0.120.160.520.92 1
3.5 RESULTS AND DISCUSSION
In Table 3.3 we show the effect of pollution on the percent at least proficient in
mathematics. The main explanatory variable in each of the columns is the percent of days
where each pollutant is above the one-hour standard. In column (1), we include the
percent of days above the standard for carbon monoxide and then in columns (2)-(5), we
include the percent of days above the standard for NO
2
, O
3
, PM
10
, and PM
2.5
,
respectively. If air pollution affects academic performance, then the coefficients on the
pollution variables should be negative and significant. In every regression, we include
controls for school quality, school composition, and community characteristics. We
expect that some of these control variables would be significant with coefficients that are
larger in magnitude than those on pollution; one would think that background
118
119
characteristics and school quality are more important for academic performance than
pollution. Although the coefficients are not presented, we also include year effects to
control for any factors that change over time for all schools in California. The
regressions are weighted by the square root of the number of students in each grade-
school-year cell. The errors are clustered at the school level and are robust to
heteroskedasticity.
The coefficient on the percent of days above the standard for CO is negative, but
insignificant in column (1), indicating that it does not affect the percent at least proficient
in mathematics. An increase in the average class size, decrease in the percent of staff
who are full-time equivalent, increase in the percent of students who receives free or
reduced price lunches, or increase in the percent of the student body who are non-native
English speakers, would decrease the percent of the grade scoring at least proficient in
mathematics. Expenditures per student have no effect on academic performance. The
amount of taxable transactions in the county, a measure of economic activity, has a
negative effect on the percent scoring at least proficient. The unemployment rate has a
positive effect on the percent proficient. It could be that this is reflecting migration from
lower performing counties to higher performing counties and subsequent lowering of test
scores in the high performing counties and improvements of scores in lower performing
counties. The coefficients on the control variables are generally consistent in terms of
significance and magnitude across columns.
119
120
Table 3.3 Effect of Air Pollution on the Percent of Students at Least Proficient
in Mathematics – Grade-School and Year Effects
(1) (2) (3) (4) (5)
Percent of days that exceed the standard for:
CO -0.123
[0.277]
NO
2
-3.323+
[1.791]
O
3
-0.089**
[0.032]
PM
10
-0.024**
[0.007]
PM
2.5
0.002
[0.005]
Asian (%) 0.013 0.008 0.013 0.011 0.011
[0.012] [0.012] [0.012] [0.012] [0.012]
Hispanic (%) -0.190** -0.197** -0.194** -0.196** -0.196**
[0.008] [0.009] [0.008] [0.008] [0.008]
African American (%) -0.292** -0.299** -0.296** -0.298** -0.299**
[0.013] [0.014] [0.013] [0.013] [0.014]
Other (%) -0.133** -0.141** -0.131** -0.137** -0.137**
[0.011] [0.012] [0.011] [0.012] [0.012]
Average class size -0.123** -0.132** -0.126** -0.128** -0.129**
[0.017] [0.018] [0.017] [0.017] [0.017]
Reduced or free meals (%) -0.038** -0.039** -0.039** -0.039** -0.040**
[0.007] [0.008] [0.007] [0.008] [0.008]
County unemployment rate 0.235* 0.258* 0.302** 0.346** 0.332**
[0.103] [0.107] [0.102] [0.103] [0.104]
County taxable transactions -0.006** -0.007** -0.006** -0.006** -0.006**
[0.002] [0.002] [0.002] [0.002] [0.002]
Parent is a:
High school graduate (%) 0.029** 0.030** 0.031** 0.031** 0.033**
[0.010] [0.011] [0.010] [0.011] [0.010]
Some college (%) 0.032** 0.036** 0.033** 0.035** 0.036**
[0.010] [0.011] [0.011] [0.011] [0.011]
College graduate (%) 0.042** 0.038** 0.042** 0.041** 0.042**
[0.011] [0.011] [0.011] [0.011] [0.011]
Graduate school (%) 0.026* 0.026* 0.026* 0.025* 0.026*
[0.011] [0.012] [0.011] [0.012] [0.012]
120
121
Table 3.3 (Continued) Effect of Air Pollution on the Percent of Students at Least
Proficient in Mathematics – Grade-School and Year Effects
(1) (2) (3) (4) (5)
Fully certified teachers (%) 0.065** 0.064** 0.066** 0.064** 0.065**
[0.008] [0.009] [0.009] [0.009] [0.009]
Expenditure per student 0.019 0.006 0.025 -0.002 0.002
[0.029] [0.034] [0.030] [0.030] [0.030]
Non-native English speakers (%) -0.063** -0.063** -0.061** -0.061** -0.061**
[0.010] [0.010] [0.010] [0.010] [0.010]
School enrollment -0.003** -0.003** -0.003** -0.003** -0.003**
[0.001] [0.001] [0.001] [0.001] [0.001]
Observations 142320130832137600135815 135571
R-squared 0.860.8650.8620.862 0.862
Standard errors in parentheses are clustered by school and robust to heteroskedasticity. ** significant at
1%; * significant at 5%; + significant at 10%.
In column (2), the coefficient on the percent of days where NO
2
is above the
standard is negative and significant. If the percent of days decreased by one then the
percent of the grade that scored at least proficient would increase by 3.3. At first glance,
this seems like a large increase in performance for a small increase in pollution.
However, on average 0.003% of days are above the standard, so a 1% increase in the
days above the standard is a huge increase in pollution.
In column (3), the coefficient on O
3
is negative and significant. A one percent
decrease in the days above the standard for O
3
would increase the percent of the grade
that was proficient in mathematics by 0.09. The average percent of days above the
standard for O
3
is 1.9. If this increased to 100%, then the percent of a grade scoring at
least proficient would decrease by 8.8 percentage points. The coefficient on the percent
121
122
of days on which PM
10
is above the California standard is negative and significant in
column (4). This indicates that after controlling for school quality, community and
school characteristics, time-invariant grade-school factors, and time-varying
macroeconomic factors, a decrease in the percent of days where PM
10
is above the
standard would increase the percent at least proficient in mathematics. A one percent
increase in the days of above the standard for PM
10
would increase the percent proficient
by 0.026. If the percent of days above the standard for PM
10
increased to 100% from the
mean of 11.71%, then the percent at least proficient would decrease by 2.1 percentage
points. The coefficient in column (4) for fine particulate matter is positive, but not
significant.
Table 3.4 contains the results for the percent proficient in English/language arts,
the second measure of academic performance in this study. Based on columns (3) and
(4), both PM
10
and O
3
are negative and statistically significant. Increasing the percent of
days above the standard to 100% for PM
10
and O
3
would decrease the percent at least
proficient in language arts by 1.67 and 16.28, respectively. Taken together, these results
indicate that exceptionally high levels of PM
10
and O
3
have negative effects on children’s
performance on standardized exams. NO
2,
PM
2.5
, and CO are not significant in these
regressions.
To put these results in perspective, we do a back-of-the-envelope calculation of
the benefits of a decrease in pollution to low income neighborhoods. Using the median
of free or reduced-price lunches as the threshold to determine high- and low-income
122
123
Table 3.4 Effect of Air Pollution on the Percent of Students at Least Proficient
in English/Language Arts – Grade-School and Year Effects
(1) (2) (3) (4) (5)
Percent of days that exceed the standard for:
CO 0.177
[0.176]
NO
2
-0.876
[1.380]
O
3
-0.166**
[0.022]
PM
10
-0.019**
[0.005]
PM
2.5
0.003
[0.004]
Asian (%) 0.007 0.004 0.006 0.006 0.006
[0.010] [0.010] [0.010] [0.010] [0.010]
Hispanic (%) -0.197** -0.202** -0.201** -0.200** -0.200**
[0.007] [0.007] [0.007] [0.007] [0.007]
African American (%) -0.230** -0.234** -0.233** -0.235** -0.235**
[0.011] [0.011] [0.011] [0.011] [0.011]
Other (%) -0.106** -0.107** -0.104** -0.109** -0.108**
[0.009] [0.010] [0.009] [0.009] [0.009]
Average class size -0.005 -0.006 -0.001 -0.004 -0.004
[0.012] [0.012] [0.012] [0.012] [0.012]
Reduced or free meals (%) -0.044** -0.042** -0.045** -0.043** -0.045**
[0.005] [0.006] [0.006] [0.006] [0.006]
County unemployment rate 0.358** 0.414** 0.425** 0.442** 0.414**
[0.071] [0.075] [0.070] [0.072] [0.072]
County taxable transactions 0.001 0.001 0.002 0.001 0.0001
[0.001] [0.001] [0.001] [0.001] [0.001]
Parent is a:
High school graduate (%) 0.026** 0.028** 0.028** 0.028** 0.030**
[0.007] [0.008] [0.008] [0.008] [0.008]
Some college (%) 0.025** 0.031** 0.026** 0.028** 0.028**
[0.008] [0.008] [0.008] [0.008] [0.008]
College graduate (%) 0.052** 0.052** 0.053** 0.054** 0.054**
[0.008] [0.008] [0.008] [0.008] [0.008]
Graduate school (%) 0.038** 0.040** 0.039** 0.039** 0.039**
[0.008] [0.009] [0.008] [0.009] [0.008]
123
124
Table 3.4 (Continued) Effect of Air Pollution on the Percent of Students at Least
Proficient in English/Language Arts – Grade-School and Year Effects
(1) (2) (3) (4) (5)
Fully certified teachers (%) 0.048** 0.046** 0.049** 0.046** 0.047**
[0.006] [0.007] [0.006] [0.006] [0.006]
Expenditure per student 0.012 0.032 0.016 0.003 0.007
[0.027] [0.025] [0.028] [0.028] [0.028]
Non-native English speakers (%) -0.110** -0.113** -0.111** -0.112** -0.114**
[0.007] [0.008] [0.008] [0.008] [0.008]
School enrollment -0.002** -0.002** -0.002** -0.002** -0.002**
[0.001] [0.001] [0.001] [0.001] [0.001]
Observations 142320130832137600135815 135571
R-squared 0.9040.9080.9050.905 0.905
Standard errors in parentheses are clustered by school and robust to heteroskedasticity. ** significant at
1%; * significant at 5%; + significant at 10%.
schools, the percent at least proficient in mathematics is 22.5 percentage points higher in
high-income schools (61.8%) compared to low-income schools (39.3%). The percent of
days above the standard for PM
10
in low-income schools is 14.3 and it is 9.3 for high-
income schools − a gap of 5 percentage points. If these low-income schools had the
pollution levels of the high income schools then the percent at least proficient would
increase by 0.12. This would reduce the gap between low- and high- income schools by
half a percentage point.
Another way to calculate the impact of the change in pollution is to examine the
trends over time. Using all grade-school observations, the average percent of days that
O
3
was above the standard was 1.78 in 2002 and 1.06 in 2008. This is a 0.72 percentage
point decrease in the percent of days above the O
3
standard. During this same time
124
125
period, the percent at least proficient in mathematics increased by approximately 17
percentage points. Based on our estimated coefficients, this 0.72 percentage point
decrease in the percent of days above the O
3
standard was responsible for 0.37% of the
improvement in the percent scoring proficient in mathematics during this time period,
ceteris paribus. Doing this same calculation for English/language arts, we can attribute
1% of the gain in English/language arts scores to the reductions in O
3
from 2002 to 2008.
Using either outcome variable, the contribution of the decrease in pollution to the
improvement in the percent at least proficient is small.
In the main specification we estimate pollution at the school from all monitors
during this time period. As a robustness check, we limit the monitors to only those that
are functioning throughout the time period, which eliminates the possibility that the
variation in pollution reflects the introduction of new monitors rather than changes in
pollution. By removing some of the noise in the pollution data caused by the variation in
the monitors used to calculate the averages, the average changes in pollution may be
more accurate. On the other hand, this reduces the number of pollution monitors available
and could therefore produce less accurate measures of pollution for schools where the
new monitors are closer. We present these results in Table 3.5. We only show the
coefficients on the pollution monitors because the coefficients on the control variables are
very similar to those in the previous tables.
Again, the effect of each pollutant is estimated in separate regressions. PM
10
and
O
3
are negative and significant, which is consistent with Tables 3.3 and 3.4. However,
PM
2.5
is now negative in both specifications, but significant in only one of them. The
125
126
significance and magnitude of the other coefficients are similar to those reported in the
main tables. It is not surprising that the largest impact of excluding new monitors was on
PM
2.5
. Monitoring of PM
2.5
did not start until 1998 and approximately 35% of PM
2.5
monitors used in the main calculations either started after 2002 or terminated before
2008. By restricting the monitors to those available between 2002 and 2008, we
eliminated the fluctuations in PM
2.5
caused by adding new monitors rather than changes
in pollution. In this case, a one percentage point decrease in the percent above the
standards for PM
2.5
would increase the percent at least proficient in language by 0.008
percentage points.
Table 3.5 Effect of Air Pollution on Academic Performance
using only Monitors Functioning Throughout the Period −
Grade-School and Year Effects
Mathematics
English/
language arts
(1) (2)
Percent of days that exceed the standard for:
CO -0.192 0.249
[0.402] [0.261]
NO
2
-2.821+ -0.202
[1.638] [1.246]
O
3
-0.091** -0.175**
[0.032] [0.022]
PM
10
-0.026** -0.024**
[0.007] [0.005]
PM
2.5
-0.005 -0.008*
[0.005] [0.004]
Standard errors in parentheses are clustered by school and robust to
heteroskedasticity. ** significant at 1%; * significant at 5%; + significant at
10%.
126
127
3.6 CONCLUSION
We have shown in this paper that a reduction in air pollution increases
performance on standardized tests of school children. The effects are small, but
significant. There are a few possible reasons for the small effects. First, we are only
estimating the effects of extreme pollution on test scores. From the descriptive statistics,
it is clear that the percent of days above the standard for all these pollutants is small
during the school year in this time period. It could be that less extreme pollution also
impacts students’ performance and could have a larger overall impact. Second, by
looking at grade-school observations, we are essentially examining the average impact of
pollution on the tests scores of all children in a particular grade and school. Since
asthmatic children are more susceptible to the negative health effects of pollution, these
children would likely suffer more in terms of their academic performance than their
peers. Although it is beyond the scope of this paper, it would be beneficial to tease out
the mechanisms by which pollution affects test scores and whether or not certain
subgroups benefit more from reductions in pollution. Despite these limitations, using
California grade-school data on academic performance and estimates for pollution at each
school, we have shown here that increases in pollution negatively affect academic
performance. This finding extends the work of Pastor et al. (2004, 2006) and highlights
another benefit of reducing air pollution levels in California.
127
128
REFERENCES
Adams, P., Hurd, M.D., McFadden, D., Merrill, A., & Ribeiro, T. (2003). Healthy,
wealthy and wise? Tests for direct causal paths between health and socioeconomic status.
Journal of Econometrics, 112(1), 3-56.
Altonji, J. (1986). Intertemporal substitution in labor supply: Evidence from micro data.
Journal of Political Economy, 94(3), S176-S215.
Arku, F.S., Filson G.C., & Shute, J. (2008). An empirical approach to the study of well-
being among rural men and women in Ghana. Social Indicators Research, 88(2), 365-
387.
Blanchflower, D., & Oswald, A. (2004). Well-being over time in Britain and the USA.
Journal of Public Economics, 88(7-8), 1359-1386.
Bonke, J., Deding, M., & Lausten, M. (2009). Time and money: A simultaneous analysis
of men’s and women’s domain satisfactions. Journal of Happiness Studies, 10(2), 113-
131.
Brickman, P., Coates, D., & Janoff-Bulman, R. (1978). Lottery winners and accident
victims – is happiness relative? Journal of Personality and Social Psychology, 36(8),
917-927.
Burström, B., & Fredlund, P. (2001). Self rated health: Is it as good a predictor of
subsequent mortality among adults in lower as well as in higher social classes? Journal of
Epidemiology and Community Health, 55(11), 836-840.
Bussing, R., Halfon N., Benjamin, B., & Wells, K.B. (1995). Prevalence of behavior
problems in US children with asthma. Archives of Pediatric Adolescent Medicine, 149(5),
565-572.
Butz, A.M., Malveaux F.J., Eggleston, P., Thompson, L., Huss, K., Kolodner, K., et al.
(1995). Social factors associated with behavioral problems in children with asthma.
Clinical Pediatrics, 34(11), 581-590.
128
129
California Board of Equalization (2002-2008). Taxable Sales in California. Retrieved
from California Board of Equalization website: http://www.boe.ca.gov/news/
tsalescont.htm.
California Department of Education (2002-2008a). Academic Performance Index Data
Files. Retrieved from California Department of Education website: http://www.cde.ca.
gov/ta/ac/ap/apidatafiles.asp#updates.
California Department of Education (2002-2008b). California Basic Educational Data
System (CBEDS) School Enrollment and Staffing Data Files. Retrieved from California
Department of Education website: http://www.cde.ca.gov/ds/sd/ cb/studentdatafiles.asp.
California Department of Education (2002-2008c). California Standards Tests Research
Files. Standardized Testing and Reporting (STAR) Program. Retrieved from California
Department of Education website: http://star.cde.ca.gov/.
California Department of Education Laws ch. 2, § 855, 2007.
California Department of Education (2010). 2009-10 Academic Performance Index
Reports: Information Guide. Retrieved from California Department of Education
website: http://www.cde.ca.gov/ta/ac/ap/.
California Employment Development Department (2009). Sub-County Areas Labor
Force and Unemployment Data. Retrieved from California Employment Development
Department website: http://www.labormarketinfo.edd.ca.gov/cgi/dataanalysis.
California Environmental Protection Agency (2009). The California Almanac of
Emissions and Air Quality - 2009 Edition. Retrieved from California Environmental
Protection Agency website: http://www.arb.ca.gov/aqd/almanac/almanac09/almanac09.
htm.
California Environmental Protection Agency (2010). Daily Data. 2010 Air Quality Data
DVD. Air Resources Board.
129
130
Chay, K., & Greenstone, M. (2003a). Air quality, infant mortality, and the Clean Air
Act of 1970. NBER Working Paper #10053.
Chay, K., & Greenstone, M. (2003b). The impact of air pollution on infant mortality:
Evidence from geographic variation in pollution shocks induced by a recession.
Quarterly Journal of Economics, 118(3), 1121-1167.
Clark, A. (1997). Job satisfaction and gender: Why are women so happy at work? Labour
Economics, 4(3), 341-372.
Clark, A., & Oswald, A. (1996). Satisfaction and comparison income. Journal of Public
Economics, 61(3), 359-381.
Contoyannis, P., & Jones, A. (2004). Socio-economic status, health and lifestyle. Journal
of Health Economics, 23(5), 965-995.
Crossley T.F., & Kennedy, S. (2002). The reliability of self-assessed health status.
Journal of Health Economics, 21(4), 643-658.
Currie, J., & Neidell, M. (2005). Air pollution and infant health: What can we learn from
California’s recent experience? Quarterly Journal of Economics, 120(2), 1003-1030.
Currie, J., Hanushek, E., Kahn, M., Neidell, M., & Rivkin, S. (2009). Does pollution
increase school absences? Review of Economics and Statistics, 91(4), 682–694.
Currie, J., Stabile, M., Manivong, P., & Roos, L. (2010). Child health and young adult
outcomes. Journal of Human Resources, 45(3), 517-548.
Deaton, A. (2008). Income, aging, health and wellbeing around the world: Evidence from
the Gallup World Poll. Journal of Economic Perspectives, 22(2), 53-72.
Deaton, A.S., & Paxson, C.H. (1998). Aging and inequality in income and health.
American Economic Review, Papers and Proceedings, 88(2), 248-253.
130
131
“Documentation HGEN” (2010): Household related Household-related status variables
and generated variables. The German Socio Economic Panel Study. Retrieved from the
German Socio Economic Panel Study website: http://www.diw.de/en/diw_02.c.
238110.en/gene rated_variables.html.
Dolan, P., Peasgood, T., & White, M. (2008). Do we really know what makes us happy?
A review of the economic literature on the factors associated with subjective well-being.
Journal of Economic Psychology, 29(1), 94-122.
Easterlin, R.A. (2001). Income and happiness: Towards a unified theory. The Economic
Journal, 111(473), 465-484.
Easterlin, R.A. (2005). A puzzle for adaptive theory. Journal of Economic Behavior &
Organization, 56(4), 513-521.
Easterlin, R.A. (2005). Health and health satisfaction over the life cycle. Unpublished
Manuscript.
Easterlin, R.A. (2009). Lost in Transition. Journal of Economic Behavior and
Organization, 71(2), 130-145.
Easterlin, R.A., & Angelescu, L. (2011). Modern economic growth and quality of life:
cross sectional and time series evidence. Forthcoming in K.C. Land (Ed.), Handbook of
Social Indicators and Quality-of-Life Research, New York and London: Springer.
Easterlin, R.A., Angelescu, L., & Zweig, J. (forthcoming 2010). The impact of modern
economic growth on urban-rural differences in life satisfaction. World Development.
Easterlin, R.A., & Marcelli, E. (2010). The cross-over in the life cycle happiness of
American women and men. In H. Hinte & K. Zimmermann (Eds.). Happiness Growth
and the Life Cycle. New York: Oxford University Press.
Easterlin, R.A., & Plagnol, A.C. (2008). Life satisfaction and economic outcomes in
Germany pre- and post-unification. Journal of Economic Behavior and Organization,
68(3-4), 433-444.
131
132
Easterlin, R.A., & Zimmerman, A.C. (2006). Happily ever after? Cohabitation, marriage,
divorce, and happiness in Germany. Population and Development Review, 32(3), 511-
528.
Environmental Protection Agency (1999). 33/50 Program The Final Record EPA-745-R-
99-004. Available at http://www.epa.gov/opptintr/3350/. Accessed on January 21, 2011.
Environmental Protection Agency (2011). Six Common Air Pollutants. Retrieved from
the Environmental Protection Agency website: http://www.epa.gov/air/urbanair/.
Ettner, S. (1996). New evidence on the relationship between income and health. Journal
of Health Economics, 15(1), 67-85.
Federal Health Monitoring System (2009). Number of people injured or killed in road
traffic accidents under the influence of alcohol per 100,000 inhabitants. Retrieved from
the Federal Health Monitoring System website: http://www.gbe-bund.de/gbe10/.
Federal Health Monitoring System (2010a). Deaths per 100,000 inhabitants (1980-1997).
Retrieved from the Federal Health Monitoring System website: http://www.gbe-
bund.de/gbe10/.
Federal Health Monitoring System (2010b). Deaths per 100,000 inhabitants (from 1998).
Retrieved from the Federal Health Monitoring System website: http://www.gbe-
bund.de/gbe10/.
Federal Health Monitoring System (2010c). Average Life Expectancy at Birth (1980-
1997). Retrieved from the Federal Health Monitoring System website: http://www.gbe-
bund.de/gbe10/.
Federal Health Monitoring System (2010d). Average Life Expectancy at Birth (from
1997). Retrieved from the Federal Health Monitoring System website: http://www.gbe-
bund.de/gbe10/.
Ferrer-i-Carbonell, A., & Frijters, P. (2004). How important is methodology for the
estimates of the determinants of happiness? The Economic Journal, 114(497), 641-659.
132
133
Fowler, M.G., Davenport, M.G., & Garg, R. (1992). School functioning of US children
with asthma. Pediatrics, 90(6), 939-944.
Frey, B.S., & Stutzer, A. (2002). Happiness and Economics. Princeton and Oxford:
Princeton University Press.
Frijters, P., Haisken-DeNew, J., & Shields, M. (2004). Investigating the patterns and
determinants of life satisfaction in Germany following reunification. Journal of Human
Resources, 39(3), 649-674.
Frijters, P., Haisken-DeNew, J., & Shields, M. (2005). The causal effect of income on
health: Evidence from German reunification. Journal of Health Economics, 24(5), 997-
1017.
Gallup World Poll (2009a). World Poll Survey Data, Washington DC: Gallup, Inc.
Gallup World Poll (2009b). World Poll Methodology, Washington DC: Gallup, Inc.
Gauderman, J., McConnell, R., Gilliland, F., London, S., Thomas, D., Avol, E. et al.
(2000). Association between air pollution and lung function growth in southern
California. American Journal of Respiratory Critical Care Medicine, 162(4), 1383-1390.
Gauderman, J., Avol, E., Lurmann, F., Kuenzli, N., Gilliland, F., Peters, J. &. McConnell,
R. (2005). Childhood asthma and exposure to traffic and nitrogen dioxide. Epidemiology,
16(6), 737-743.
German Institute for Economic Research (DIW) (2004). German Socioeconomic Panel
Study, Berlin.
Gilliland, F., Berhane, K., Rappaport, E., Thomas, D., Avol, E., Gauderman, J. et al.
(2001). The effects of ambient air pollution on school absenteeism due to respiratory
illness. Epidemiology, 12(1), 43-54.
133
134
Gordo, L.R. (2006). Effects of short- and long-term unemployment on health satisfaction:
evidence from German data. Applied Economics, 38(20), 2335-2350.
Graham, C. (2009). Happiness around the world: the paradox of happy peasants and
miserable millionaires. New York: Oxford University Press.
Graham, C., & Pettinato, S., (2001). Happiness, markets, and democracy: Latin America
in comparative perspective. Journal of Happiness Studies, 2(3), 237–268.
Grahame, T.J., & Schlesinger, R.B. (2007). Health effects of airborne particulate matter:
Do we know enough to consider regulating specific particle types or sources? Inhalation
Toxicology, 19(6-7), 457-481.
Griliches, Z., & Hausman, J. (1986). Errors in variables in panel data. Journal of
Econometrics, 31(1), 93-118.
Haisken-DeNew, J.P., & Frick, J.R. (Eds.) (2005). DTC Desktop Companion to the
German Socio-Economic Panel (SOEP). Available at
www.diw.de/deutsch/sop/service/dtc/dtc.pdf.
Halterman, J., Montes, G., Aligne, A., Kaczorowski, J., Hightower, A., & Szilagyi, P.
(2001). School readiness among urban children with asthma. Ambulatory Pediatrics,
1(4), 201–05.
Halterman, J., Conn, K., Forbes-Jones, E., Fagnano, M., Hightower, A., & Szilagyi, P.
(2006). Behavior problems among inner-city children with asthma: Findings from a
community-based sample. Pediatrics, 117(2), e192-e199.
Idler, E. L., & Benyamini, Y. (1997). Self-rated health and mortality: a review of twenty-
seven community studies. Journal of Health and Social Behavior, 38(1), 21-37.
Idler, E.L., & Kasl, S.V. (1995). Self-ratings of health: Do they also predict change in
functional ability? Journal of Gerontology, 50B, S344-S353.
134
135
Imbens, G., & Wooldridge, J. (2008) Institute for Research on Poverty Applied
Microeconomics Workshop: Lecture Notes 3 and 4. Retrieved from the Institute for
Research on Poverty website: http://www.irp.wisc.edu/newsevents/workshops/
appliedmicroeconometrics/schedule1.htm.
Jorgen, M., & Groenewegen, P. (1997). Back to Bismarck: Eastern European health care
systems in transition. Aldershot, U.K.: Avebury.
Junge B., & Nagel, M. (1999). Smoking behavior in Germany. Gesundheitswesen, 61,
S121-S125.
Kapteyn, A., Smith, J., & Van Soest, A. (2007). Vignettes and self-reports of work
disability in the United States and the Netherlands. American Economic Review, 97(1),
461-473.
King , G., Murray, C. J.L., Salomon, J.A., & Tandon, A. (2004). Enhancing the validity
and cross-cultural comparability of measurement in survey research. American
Political Science Review, 98(1), 567- 583.
Knight, J., Song, L., & Gunatilaka, R., (2009). Subjective well-being and its
determinants in rural China. China Economic Review, 20(4), 635-649.
Kroll, C. (2010). Different things make different people happy: Examining social capital
and subjective well-being by gender and parental status. Social Indicators Research. 1-
21.
Kurt, T.L., Mogielnicki, R.P., & Chandler, J.E. (1978). Association of the frequency of
acute cardiorespiratory complaints with ambient levels of carbon monoxide. Chest, 74(1),
10-14.
Linn, W.S., Szlachcic, Y., Gong, H., Kinney, P.L., & Berhane, T. (2000). Air pollution
and daily hospital admissions in metropolitan Los Angeles. Environmental Health
Perspectives, 108(5), 427-434.
135
136
Lucas, R. (2007). Long-term disability is associated with lasting changes in subjective
well-being: Evidence from two nationally representative longitudinal studies. Journal of
Personality and Social Psychology, 92(4), 717-730.
Luschen, G., Niemann, S., & Apelt, P. (1997). The integration of two health care
systems: Social stratification, work and health in East and West Germany. Social Science
and Medicine, 44(6), 883-899.
Marks, G. N., & Fleming, N. (1999). Influences and consequences of well-being among
Australian young people: 1980–1995. Social Indicators Research, 46(3), 301-323.
McConnell, R., Berhane, K., Gilliland, F., Islam T., Gauderman, W.J., Avol, E., et al.
(2002). Asthma in exercising children exposed to ozone: A cohort study. Lancet,
359(9304), 386-391.
McConnell, R., Berhane, K., Gilliland, F., Molitor, J., Thomas, D., Lurmann, F., Avol,
E., et al. (2003). Prospective study of air pollution and bronchitic symptoms in children
with asthma. American Journal of Respiratory and Critical Care Medicine, 168(7), 790-
797.
Mossey, J.M., & Shapiro, E. (1982). Self-related health: A predictor of mortality among
the elderly. American Journal of Public Health, 72(8), 800-808.
Müller-Nordhorn, J., Rossnagel, K., Mey, W., & Willich, S. (2004). Regional variation
and time trends in mortality from ischaemic heart disease: East and West Germany 10
years after reunification. Journal of Epidemiology Community Health, 58(6), 481-485.
National Center for Education Statistics (2002-2008). Common Core of Data. Retrieved
from National Center for Education Statistics website: http://nces.ed.gov/ccd/.
National Heart, Blood and Lung Institute of the National Institutes of Health (2007).
Expert panel report 3 (EPR3): Guidelines for the diagnosis and management of asthma.
Available at http://www.nhlbi.nih.gov/guidelines/asthma/04_sec3_comp.pdf.
136
137
Neidell, M. (2004). Air Pollution, health, and socio-economic status: the effect of outdoor
air quality on childhood asthma. Journal of Health Economics, 23(6), 1209–1236.
Nolen-Hoeksema, S. & Rusting, C. (1999). Gender differences in well-being. In D.
Kahneman, E. Diener, & N. Schwarz (Eds.), Foundations of hedonic psychology:
Scientific perspectives on enjoyment and suffering. (pp. 330-352). New York:
Russell Sage Foundation.
Nolte, E., Britton, A., & McKee, M. (2003). Trends in mortality attributable to current
alcohol consumption in East and West Germany. Social Science and Medicine, 56(7),
1385-1395.
Nolte E., Scholz R., Shkolnikov V., & McKee, M. (2002). The contribution of medical
care to changing life expectancy in Germany and Poland. Social Science and Medicine,
55(7), 1905-1921.
Nolte, E., Shkolnikov, V., & McKee, M. (2000a). Changing mortality patterns in East
and West Germany and Poland. I: long term trends (1960-1997). Journal of
Epidemiology Community Health, 54(12), 884-898.
Nolte, E., Shkolnikov, V., & McKee, M. (2000b). Changing mortality patterns in East
and West Germany and Poland. II: Short-term trends during transition and in the 1990s,
Journal of Epidemiology Community Health, 54(12), 899-906.
Oswald, A., & Powdthavee, N. (2008). Does happiness adapt? A longitudinal study of
disability with implications for economists and judges. Journal of Public Economics,
92(5-6), 1061-1077.
Pastor, M., Morello-Frosch, R., Sadd, J. (2006). Breathless: Air quality, schools, and
environmental justice in California. Policy Studies Journal, 34(3), 337-362.
Pastor, M., Sadd, J.L., & Morrello-Frosch, R. (2004). Reading, writing, and toxics:
children's health, academic performance, and environmental justice in Los Angeles.
Environment and Planning C: Government and Policy, 22(2), 271-290.
137
138
Plagnol, A.C., & Easterlin, R. (2008). Aspirations, attainments, and satisfaction: Life
cycle differences between American women and men. Journal of Happiness Studies,
9(4), 601-619.
Pope, C.A., & Dockery, D.W. (2006). Health effects of fine particulate air pollution:
Lines that connect. Journal of the Air and Waste Management Association, 54, 709-742.
Powdthavee, N. (2010). The Happiness equation: The surprising economics of our most
valuable asset. London: Icon Books.
Rabinovitch, N., Strand, M., & Gelfand, E. (2006). Particulate levels are associated with
early asthma qorsening in children with persistent disease. American Journal of
Respiratory Critical Care Medicine, 173(10), 1098-1105.
Ransom, M. & Pope, C. (1992). Elementary school absences and PM
10
pollution in Utah
Valley. Environmental Research, 58(2), 204-219.
Central Intelligence Agency (2010). Religions The World Fact Book. Retrieved from the
Central Intelligence Agency website: https://www.cia.gov/library/publications/the-world-
factbook/fields/2122.html.
Riphahn, R., & Zimmermann, K. (2000). The mortality crisis in east Germany. In G.A.
Cornia and R. Paniccià (Eds.), The Mortality Crisis in Transitional Economies (pp. 227-
252). Oxford: Oxford University Press.
Russell, A.G. & B. Brunekreef (2009). A focus on particulate matter and health.
Environmental Science and Technology, 43(13), 4620-4625.
Salas, C. (2002). On the empirical association between poor health and low
socioeconomic status at old age. Health Economics, 11(3), 207-220.
Scharf, B. (1999). German unity and health care reform. In F.D. Powell & A.F. Wessen
(Eds.), Health care systems in transition: An international perspective (pp. 101-112).
California: Sage.
138
139
Schyns, P. (2001). Income and satisfaction in Russia. Journal of Happiness Studies, 2(2),
173-204.
Sloane, P.J., & Williams, H. (2000). Job satisfaction, comparison earnings and gender.
Labour, 14(3), 473-502.
Smith, J.P. (1999). Healthy bodies and thick wallets: the dual relationship between health
and economic status. Journal of Economic Perspectives, 13(2), 145-166.
Sousa-Poza, A., & Sousa-Poza, A.A. (2000). Taking another look at the gender/job
satisfaction paradox, KYKLOS, 53(2), 135-152.
Stavrova,O., Schlösser, T., & Fetchenhauer, D. (2011). Are the unemployed equally
unhappy all around the world? The role of the social norms to work and welfare state
provision in 28 OECD countries. Journal of Economic Psychology, 32(1), 159-171.
Stevenson, B., & Wolfers, J. (2009). The Paradox of Declining Female Happiness.
American Economic Journal: Economic Policy, 1(2), 190-225.
Stock, J., & Yogo, M. (2005). Testing for weak instruments in linear IV regression. In
D.W.K. Andrews & J.H. Stock (Eds.), Identification and Inference for Econometric
Models: Essays in Honor of Thomas Rothenberg (pp. 80-108), Cambridge: Cambridge
University Press.
Stone, D. (1991). German unification: East meets West in the doctor’s office. Journal of
Health Politics, Policy and Law, 16(2), 401-412.
Swami, B. (2002). The German health care system. In . K.V. Thai, E.T Wimberley, &
S.M. McManus (Eds.), Handbook of International Health Care Systems (pp. 333-358),
New York: Marcel Dekker.
United Nations Educational, Scientific, and Cultural Organization (UNESCO) (2005-
2008). Educational Attainment of the Population Aged 25 years and Older/ Latest Year
Available. Retrieved from UNESCO website: http://stats.uis.unesco.org/unesco.
139
140
van Doorslaer, E., & Gerdtham, U-G.(2003). Does inequality in self-assessed health
predict inequality in survival by income? Evidence from Swedish data. Social Science
and Medicine, 57(9), 1621-1629.
van Doorslaer, E., Koolman, X., & Jones, A.M. (2004). Explaining income-related
inequalities in doctor utilisation in Europe. Health Economics, 13(7), 629-647.
Van Soest, A., Delaney, L., Harmon, C., Kapteyn, A., & J. Smith (2007). Validating the
use of vignettes for subjective threshold scales. IZA Discussion Paper 2860. Retrieved
from IZA website: http://www.iza.org/en/webcontent/publications/papers.
van Pragg, B.M.S. (2005). The connection between old and new approaches to financial
satisfaction. In L. Bruni & P.L. Porta (Eds.), Economics and Happiness: Framing the
Analysis (pp. 196-222). New York: Oxford University Press.
Wooldridge, J. (2002). Econometric Analysis of Cross Section and Panel Data.
Massachusetts: MIT Press.
World Bank (2010). World Development Indicators Online (WDI) database. Data
retrieved from World Bank website: http://data.worldbank.org/data-catalog/world-
development indicators.
Yu, O., Sheppard, L., Lumley, T., Koenig, J., & Shapiro, G. (2000). Effects of ambient
air pollution on symptoms of asthma in Seattle-area children enrolled in the CAMP study.
Environmental Health Perspectives, 108(12), 1209-1214.
140
141
APPENDIX 1.A: FIRST STAGE RESULTS FOR HEALTH
SATISFACTION AND LIFE SATISFACTION
Dependent variable = self-
reported health
24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
Age squared 0.0005 0.0005 0.0006+ 0.0006+
[0.000] [0.000] [0.000] [0.000]
Married -0.019-0.01540.0329 -0.0076
[0.058] [0.058] [0.087] [0.088]
Child -0.0024-0.0257
[0.038] [0.081]
Spouse died 0.2281 -0.1430*
[0.188] [0.063]
Log household income per
capita 0.04270.04380.1039** 0.0960*
[0.033] [0.033] [0.040] [0.040]
Unemployed 0.0550.0551-0.0612 -0.0569
[0.037] [0.038] [0.053] [0.053]
Retired 0.02750.01520.1108** 0.1145**
[0.103] [0.102] [0.040] [0.040]
Unemployment rate
rt
-1.4924*-1.5372*0.7587 0.7435
[0.634] [0.638] [0.645] [0.644]
Legislators, professionals 0.0379 0.0533
[0.024] [0.044]
Technicians, armed forces 0.0436+ -0.0201
[0.025] [0.047]
Clerks, sales and service
workers -0.0155 -0.0099
[0.024] [0.041]
Agriculture, fishery, and trade
workers -0.003 0.0194
[0.022] [0.050]
Operators, elementary
occupations -0.0613 0.0339
[0.039] [0.046]
No insurance 0.1356 0.1394 -0.1158 -0.1128
[0.103] [0.104] [0.151] [0.149]
Private insurance -0.0938 -0.0933 -0.0515 -0.049
[0.070] [0.070] [0.068] [0.068]
141
142
APPENDIX 1.A (CONTINUED): FIRST STAGE RESULTS FOR
HEALTH SATISFACTION AND LIFE SATISFACTION
Dependent variable = self-
reported health
24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
Infant mortality rate
rt
-0.0112 -0.007-0.0255 -0.0226
[0.039] [0.039] [0.044] [0.043]
Infant mortality rate
rt -1
0.0727* 0.0725*-0.0064 -0.0066
[0.028] [0.028] [0.031] [0.031]
Life expectancy
rt
-0.0388-0.04230.0487 0.0444
[0.047] [0.047] [0.052] [0.051]
Somewhat worried about
finances
t
-0.0561*-0.0566**-0.0376+ -0.0387+
[0.022] [0.022] [0.022] [0.023]
Very worried about finances
t
-0.1367** -0.1371** -0.1016** -0.1023**
[0.031] [0.031] [0.036] [0.036]
Somewhat worried about
finances
t-1
0.02060.0185-0.0148 -0.0148
[0.020] [0.021] [0.021] [0.021]
Very worried about finances
t-1
0.0036 0.0013 -0.0594* -0.0602*
[0.029] [0.030] [0.029] [0.029]
Somewhat worried about
country's development
t
-0.0045 -0.0045-0.0844* -0.0840*
[0.040] [0.039] [0.037] [0.037]
Very worried about country's
development
t
0.01410.0138-0.0976* -0.0967*
[0.041] [0.041] [0.040] [0.040]
Year -0.1707**-0.1698**-0.1847** -0.1813**
[0.040] [0.041] [0.047] [0.048]
East Germany*year -0.0772** -0.0750** -0.0647** -0.0609**
[0.014] [0.015] [0.015] [0.015]
Change in doctor visits
t-1
0.0035** 0.0035**0.0019** 0.0019**
[0.001] [0.001] [0.0004] [0.0004]
Traffic accidents due to
alcohol
rt-1
0.0031**0.0030**0.0014** 0.0013**
[0.0001] [0.0001] [0.0001] [0.0001]
142
143
APPENDIX 1.A (CONTINUED): FIRST STAGE RESULTS FOR
HEALTH SATISFACTION AND LIFE SATISFACTION
Dependent variable = self-
reported health 24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
Observations 177791777914640 14640
R-squared 0.03030.03190.0192 0.0201
F-test 39.2636.0814.86 13.87
The notation r represents the state and t represents the year. The unemployment rate, infant mortality, life
expectancy and traffic accidents vary by state and year. The errors are clustered by individual and are
robust to heteroskedasticity. Survey weights provided in the SOEP are used. The base year is the
difference between 1992 and 1994. The notation r represents the state and t represents the year. The
unemployment rate, mortality rate, life expectancy and traffic accidents vary by state and year. Standard
errors in parentheses. ** significant at 1%; * significant at 5%; + significant at 10%.
143
144
APPENDIX 1.B: COEFFICIENT ON SELF-REPORTED HEALTH
IN 2SLS WITH SUBSETS OF INSTRUMENTAL VARIABLES
24 to 44 years old in 1990 45 to 70 years old in 1990
Sathealth Satlife Sathealth Satlife
(1) (2) (3) (4)
IV=change in doctor visits
t-1
and
traffic accidents due to alcohol
rt-1
1.287** 0.499* 1.529** 1.03*
[0.331] [0.243] [0.548] [0.449]
IV=change doctor visits
t-1
1.328**0.4111.248+ 0.970+
[0.489] [0.318] [0.641] [0.512]
IV=traffic accidents due to alcohol
rt-1
1.209** 0.676* 2.904** 1.356+
[0.309] [0.292] [0.854] [0.716]
Standard errors are clustered by individual and are robust to heteroskedasticity. Survey weights provided in
the SOEP are used. The base year is the difference between 1992 and 1994. The notation r represents the
state and t represents the year. The unemployment rate, infant mortality, life expectancy, and traffic
accidents vary by state and year. ** significant at 1%; * significant at 5%; + significant at 10%.
144
145
APPENDIX 2.A: COUNTRIES INCLUDED IN THE ANALYSIS
1. Developed countries (20) 3. Less developed countries (41)
Australia (AUS) 3.1 Asia (16) 3.3 Africa (8)
Austria (AUT) Bangladesh (BGD) Burkina Faso (BFA)
Belgium (BEL) Cambodia (KHM) Cameroon (CMR)
Canada (CAN) India (IND) Kenya (KEN)
Denmark (DNK) Indonesia (IDN) Mali (MLI)
Finland (FIN) Iran (IRN) Mozambique (MOZ)
France (FRA) Kazakhstan (KAZ) Senegal (SEN)
Germany (DEU) Kyrgyzstan (KGZ) Tanzania (TZA)
Greece (GRC) Malaysia (MYS) Uganda (UGA)
Ireland (IRL) Mongolia (MNG)
Italy (ITA) Nepal (NPL)
Japan (JPN) Philippines (PHL)
Netherlands (NLD) South Korea (KOR)
Norway (NOR) Tajikistan (TJK)
Portugal (PRT) Thailand (THA)
Spain (ESP) Turkey (TUR)
United Kingdom (GBR) Vietnam (VNM)
United States (USA)
Sweden (SWE) 3.2. Latin America (17)
Switzerland (CHE) Argentina (ARG)
Bolivia (BOL)
2. Transition countries (12) Brazil (BRA)
Belarus (BLR) Chile (CHL)
Bulgaria (BGR) Colombia (COL)
Czech Republic (CZE) Costa Rica (CRI)
Estonia (EST) Dominican Republic (DOM)
Hungary (HUN) Ecuador (ECU)
Latvia (LVA) El Salvador (SLV)
Lithuania (LTU) Guatemala (GTM)
Poland (POL) Honduras (HND)
Romania (ROM) Mexico (MEX)
Russia (RUS) Nicaragua (NIC)
Slovakia (SVK) Panama (PAN)
Slovenia (SVN) Peru (PER)
Uruguay (URY)
Venezuela (VEN)
145
146
APPENDIX 2.B: EXPLANATORY VARIABLES IN THE ANALYSIS
Life satisfaction
Question Please imagine a ladder with steps numbered from zero at the bottom
to ten at the top. Suppose we say that the top of the ladder represents
the best possible life for you, and the bottom of the ladder represents
the worst possible life for you. On which step of the ladder would you
say you personally feel you stand at this time, assuming that the
higher the step the better you feel about your life, and the lower the
step the worse you feel about it? Which step comes closest to the way
you feel?
answer
Worst possible*01*02*03*04*05*06*07*08*09*Best possible
Don't know
Refused
Marital status
question
What is your current marital status?
answer Single/never been married
Married
Separated
Divorced
Widowed
Domestic partner
Don't know
Refused
Ideal children
question What do you think is the ideal number of children for a family to
have?
answer Continuous number
Don't know
Refused
Location
question
Respondent lives in:
answer A rural area or on a farm
In a small town or village
In a large city
In the suburb of a large city
Don't know
Refused
Employment Status
question
Do you currently have a job or work (either paid or unpaid work)?
answer Yes
No
Don't know
Refused
146
147
APPENDIX 2.B (CONTINUED): EXPLANATORY VARIABLES
IN THE ANALYSIS
Education
question
EDUCATION_CAT
answer Elementary - Completed elementary education or less (up to 8 years
of basic education)
Secondary - Completed some education beyond elementary education
(9 to 15 years of education)
Tertiary - Completed four years of education beyond high school
and/or received a 4-year college degree.
Don't know
Refused
Health problems
question Do you have any health problems that prevent you from doing any of
the things people your age normally can do?
answer Yes
No
Don't know
Refused
Income
question What is your total monthly household income in [local currency],
before taxes? Please include income from wages and salaries,
remittances from family members living elsewhere, farming and all
other sources.
answer Continuous number
Don't know
Refused
Attend a religious ceremony
question Have you attended a place of worship or religious service within the
last seven days?
answer Yes
No
Don’t know
Refused
Occupation
question Could you tell me the general category of work you do in your
primary job?
answer Professional worker--lawyer, doctor, scientist, teacher, engineer,
nurse, accountant, computer programmer, architect, investment
banker, stock broker, marketing, musician, artist
Manager, Executive or Official--in a business, government agency, or
other organization
Business Owner--such as a store, factory, plumbing contractor, etc.
(self employed)
147
148
APPENDIX 2.B (CONTINUED): EXPLANATORY VARIABLES
IN THE ANALYSIS
answer Clerical or Office Worker--in business, government agency, or other
type of organization--such as a typist, secretary, postal clerk,
telephone operator, computer operator, data entry, bank clerk, etc.
Sales worker - clerk in a store, door-to-door salesperson, sales
associate, manufacturer's representative, outside sales person
Service worker - policeman/woman, fireman, waiter or waitress,
maid, nurse’s aide, attendant, barber or beautician, fast-food,
landscaping, janitorial, personal care worker
Construction or Mining worker - construction manager, plumber,
carpenter electrician, other construction trades, miner, or other
extraction worker
Manufacturing or Production worker - operates a machine in a
factory, is an assembly line worker in a factory, includes non-
restaurant food preparation (baker), printer, print shop worker,
garment, furniture and all other manufacturing
Transportation worker - drives a truck, taxi cab, bus or etc., works
with or on aircraft (including pilots and flight attendants), trains,
boats, teamster, longshoreman, delivery company worker or driver,
moving company worker
Installation or Repair worker - garage mechanic, linesman, other
installation, maintenance or repair worker
Farming, Fishing or Forestry worker - Farmer, farm worker,
aquaculture or hatchery worker, fisherman, deck hand on fishing
boat, lumberjack, forest management worker
Other (list)____________
Don't know
Refused
148
149
APPENDIX 2.C: PERCENT OF VARIABLES MISSING WHEN
ECONOMIC FACTORS ARE INCLUDED IN THE ANALYSIS
Women Men Average
Difference
(women-men)
Developed countries:
Australia 30 29 30 0.40
Austria 36 37 36 -1.11
Belgium 74 65 71 8.09
Canada 23 18 21 5.15
Finland 25 18 22 7.32
France 20 14 18 5.44
Germany 31 27 29 3.84
Greece 57 57 57 -0.17
Ireland 29 24 27 5.05
Italy 47 47 47 0.37
Japan 31 25 28 6.85
Netherlands 20 10 15 9.91
Norway 20 10 15 9.77
Portugal 30 30 30 -0.52
Spain 45 38 43 7.15
Sweden 15 14 15 0.45
Switzerland 19 14 17 4.06
United Kingdom 30 24 28 6.01
United States 60 61 60 -0.79
Transition countries:
Belarus 46 48 47 -1.59
Bulgaria 16 24 19 -8.13
Estonia 20 26 23 -5.67
Hungary 8 10 9 -2.16
Latvia 20 27 23 -7.29
Lithuania 18 23 20 -5.61
Poland 72 77 74 -5.08
Romania 12 15 13 -3.34
Russia 34 37 35 -3.91
Slovakia 17 20 18 -2.29
Slovenia 14 14 14 -0.22
Asia:
Bangladesh 0 1 1 -0.05
Cambodia 52 57 54 -4.84
149
150
APPENDIX 2.C (CONTINUED): PERCENT OF VARIABLES
MISSING WHEN ECONOMIC FACTORS ARE INCLUDED IN
THE ANALYSIS
Women Men Average
Difference
(women-men)
India 13 10 11 3.52
Indonesia 4 4 4 -0.24
Iran 69 70 70 -0.81
Kazakhstan 18 24 20 -6.56
Kyrgyzstan 8 9 8 -1.08
Malaysia 25 25 25 0.18
Mongolia 8 7 8 1.24
Nepal 48 60 53 -12.22
Philippines 48 48 48 -0.77
South Korea 57 56 56 1.64
Tajikistan 16 12 14 3.05
Thailand 3 3 3 -0.50
Turkey 10 16 13 -5.94
Vietnam 12 13 13 -1.34
Latin America:
Argentina 22 20 21 2.04
Bolivia 46 40 43 6.02
Chile 4 6 5 -1.78
Colombia 13 13 13 0.43
Costa Rica 32 32 32 -0.26
Dominican Republic 44 44 44 -0.07
Ecuador 5 4 4 1.01
El Salvador 15 16 16 -0.19
Guatemala 33 28 31 4.71
Honduras 20 20 20 -0.22
Mexico 15 13 14 2.38
Nicaragua 6 6 6 0.51
Peru 12 10 11 1.19
Uruguay 69 71 70 -2.11
Venezuela 23 17 21 6.19
Africa:
Burkina Faso 29 25 26 4.43
Cameroon 23 18 20 5.46
Kenya 37 30 33 6.32
150
151
APPENDIX 2.C (CONTINUED): PERCENT OF VARIABLES
MISSING WHEN ECONOMIC FACTORS ARE INCLUDED IN
THE ANALYSIS
Women Men Average
Difference
(women-men)
Senegal 12 8 10 3.94
Tanzania 47 40 44 6.76
Uganda 10 6 8 3.53
151
152
APPENDIX 2.D: OLS REGRESIONS OF WHETHER A RESPONDENT DID NOT
ANSWER THE ECONOMIC QUESTIONS
Dependent variable=1 if
income or occupation is
missing; 0 otherwise Norway Sweden Australia Tajikistan Indonesia Colombia Mexico
Dominican
Republic
Female 0.59**0.30*0.16+0.17**0.06*0.13**0.15+0.05*
[3.54] [2.27] [1.90] [2.84] [2.15] [3.83] [1.94] [2.19]
Life satisfaction 0.04** 0.03* 0.02** 0.04** -0.004 0.02** 0.01 0.01**
[2.96] [2.35] [2.96] [4.40] [-0.39] [4.82] [1.42] [3.64]
Female * life satisfaction -0.06** -0.04* -0.02+ -0.03* -0.01* -0.02** -0.02+ -0.01*
[-2.73] [-2.21] [-1.67] [-2.29] [-2.23] [-3.70] [-1.73] [-1.99]
Constant -0.22*-0.09-0.07-0.030.01-0.08**0.060.08**
[-2.18] [-0.84] [-1.31] [-0.74] [0.43] [-2.63] [1.02] [4.31]
Observations 969972314519483085292819592885
R-squared 0.040.010.140.020.030.030.010.64
T-statistics in parentheses. ** significant at 1%; * significant at 5%; + significant at 10%.
152
153
APPENDIX 2.E: MEAN DIFFERENCE IN LIFE SATISFACTION BETWEEN MEN
AND WOMEN FROM EACH SPECIFICATION
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Developed Countries:
Australia 7.417.217.310.20**0.24** 0.160.23**0.33**
Austria 7.147.297.21-0.15-0.11 -0.09-0.14-0.12
Belgium 7.217.167.190.060.13 0.070.16+0.09
Canada 7.587.437.500.150.21+ 0.24+0.190.20
Denmark 7.907.827.860.08 0.06 0.10
Finland 7.727.657.690.080.11 0.120.120.09
France 7.096.967.030.130.16 0.130.190.22
Germany 6.486.506.49-0.02 0.12 0.09 0.12 0.10
Greece 6.436.266.350.170.30** 0.33**0.63**
Ireland 7.507.647.56-0.14-0.11 -0.13-0.09-0.17
Italy 6.656.976.80-0.32*-0.21 -0.17-0.24-0.23
Japan 6.245.786.010.46**0.56** 0.50**0.56**0.48**
Netherlands 7.597.69 7.64-0.1 -0.03 -0.002 0.06 0.06
Norway 7.637.677.65-0.04-0.04 0.010.030.12
Portugal 5.445.685.55-0.25+-0.05 0.05-0.030.01
Spain 7.307.357.32-0.050.02 -0.020.02-0.002
Sweden 7.467.597.52-0.13-0.13 -0.13-0.060.09
Switzerland 7.537.42 7.470.11 0.18 0.17 0.06
United Kingdom 6.90 6.85 6.87 0.04 0.14 0.08 0.19* 0.26*
United States 7.40 7.41 7.40 -0.02 0.04 0.07 0.07 0.18
153
154
APPENDIX 2.E (CONTINUED): MEAN DIFFERENCE IN LIFE SATISFACTION
BETWEEN MEN AND WOMEN FROM EACH SPECIFICATION
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Transition Countries:
Belarus 5.575.605.58-0.030.02 0.030.030.03
Bulgaria 3.793.933.86-0.14-0.04 0.030.003
Czech Republic 6.45 6.54 6.49 -0.09 0.004 -0.08
Estonia 5.355.435.38-0.080.04 0.070.060.03
Hungary 5.175.245.20-0.070.06 0.16
Latvia 4.754.834.79-0.08-0.002 -0.020.080.02
Lithuania 5.775.895.83-0.12-0.04 -0.050.009-0.01
Poland 5.785.705.740.080.13 0.160.34
Romania 5.265.215.230.050.21+ 0.31*0.30*
Russia 5.255.295.27-0.050.07 0.050.090.06
Slovakia 5.255.325.28-0.070.09 0.13-0.07
Slovenia 5.705.925.81-0.22-0.04 0.010.05
Asia
Bangladesh 4.964.67 4.810.29**0.42** 0.57**0.36** 0.30*
Cambodia 4.324.354.33-0.02 0.10 0.06 0.10 0.02
India 5.085.065.070.020.13 0.080.25*0.29**
Indonesia 4.984.914.950.070.16* 0.16+ 0.08 0.09
Iran 5.505.035.260.47**0.56** 0.53**0.67**0.60**
Kazakhstan 5.755.88 5.81-0.13 -0.07 -0.02 0.03 -0.02
Kyrgyzstan 4.654.72 4.68-0.07 -0.01 -0.01 0.08 0.06
Malaysia 6.155.996.070.15+0.21** 0.190.23*0.21+
154
155
APPENDIX 2.E (CONTINUED): MEAN DIFFERENCE IN LIFE SATISFACTION
BETWEEN MEN AND WOMEN FROM EACH SPECIFICATION
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Mongolia 4.544.434.480.10 0.09 0.10 0.12 0.13
Nepal 4.734.444.590.30**0.51** 0.36**0.48**0.35*
Philippines 4.824.76 4.790.07 0.14 0.11 0.17 0.13
South Korea 5.85 5.23 5.55 0.62** 0.63** 0.71** 0.87**
Tajikistan 4.734.794.76-0.07-0.06 -0.03 0.090.16+
Thailand 5.725.555.640.170.16 0.190.150.15
Turkey 5.274.975.120.30+0.36* 0.32+0.45*0.23
Vietnam 5.415.385.390.030.10 0.080.070.07
Latin America:
Argentina 6.045.835.940.200.36* 0.37*0.29+ 0.27
Bolivia 5.385.495.44-0.120.03 -0.080.050.09
Brazil 6.736.546.640.190.22 0.36*
Chile 5.635.825.72-0.19-0.07 -0.10.080.14
Colombia 6.026.256.13-0.22*-0.02 -0.36+ 0.09 0.18
Costa Rica 7.39 7.48 7.44 -0.1 -0.08 -0.10 -0.31+
Dominican Republic 5.03 4.98 5.01 0.05 0.02 0.07 0.14 0.46**
Ecuador 5.065.235.14-0.17+-0.12 -0.180.080.12
El Salvador 5.50 5.28 5.39 0.22* 0.28** 0.31* 0.30** 0.24*
Guatemala 6.306.45 6.37-0.15 -0.05 -0.16 -0.01 -0.06
Honduras 5.315.175.240.14 0.15 0.45*0.25*0.32*
Mexico 6.766.566.670.20+0.23* 0.30+0.26*0.33*
Nicaragua 5.154.895.020.26*0.31* 0.49**0.34**0.33*
Panama 6.916.886.890.040.05 0.21
155
156
APPENDIX 2.E (CONTINUED): MEAN DIFFERENCE IN LIFE SATISFACTION
BETWEEN MEN AND WOMEN FROM EACH SPECIFICATION
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Peru 5.025.115.06-0.090.09 0.210.110.16
Uruguay 5.725.785.75-0.05-0.04 0.03-0.04
Venezuela 6.456.37 6.410.09 0.12 0.12 0.20 0.20
Africa:
Burkina Faso 3.79 3.92 3.85 -0.13 -0.12 -0.09 -0.17 -0.14
Cameroon 4.344.284.310.06 0.03 0.05 0.15 0.01
Kenya 4.034.004.020.030.10 0.100.120.09
Mali 4.174.054.110.120.11 0.100.09
Mozambique 4.704.59 4.650.11 0.16 0.16 0.19+
Senegal 4.784.544.660.24*0.29* 0.30*0.28*0.23
Tanzania 4.504.154.330.34*0.28+ 0.25+0.26+ 0.28
Uganda 4.704.454.570.25+ 0.2 0.20.230.35*
1. Equation 2 includes the following explanatory variables: age, age squared, marital status, education, and wave effects. 2. Equation 2a
includes the following explanatory variables: age, age squared, marital status, education, ideal children, and wave effects. 3. Equation 3
includes the following explanatory variables: age, age squared, marital status, health problems, employment status, education, residential
location, attendance at a religious ceremony, and wave effects. 4. Equation 4 includes the following explanatory variables: age, age
squared, marital status, health problems, employment status, education, residential location, attendance at a religious ceremony, income,
occupation, and wave effects. ** significant at 1%; * significant at 5%; + significant at 10%
156
157
APPENDIX 2.F: NUMBER OF OBSERVATIONS IN REGRESSIONS
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Developed Countries:
Australia 1,5721,573314531453145 94827502211
Austria 618358976976976 924975623
Belgium 1,193671186418641864 9101854 548
Canada 570407977977977 918977770
Denmark 585381966966966 966
Finland 561414975975975 917972762
France 584402986986986 948984812
Germany 1,074868194219421942 184019401375
Greece 1,116796191219121912 1903821
Ireland 601374975975975 940975710
Italy 646317963963963 894959511
Japan 2,0492,002405140514051 290940482915
Netherlands 550434984984 984 903 984 832
Norway 507462969969969 866969823
Portugal 1,199662186118611861 87618561305
Spain 624356980980980 907978561
Sweden 550422972972972 884972829
Switzerland 600381981981 981 980 815
United Kingdom 1,222 891 2113 2113 2113 903 2113 1525
United States 1,186 1,004 2190 2190 2190 938 2190 871
Transition Countries:
157
158
APPENDIX 2.F (CONTINUED): NUMBER OF OBSERVATIONS IN REGRESSIONS
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Belarus 1,7711,248301930193019 284429801603
Bulgaria 580340920920920 904748
Czech Republic 597 417 1014 1014 1014 1003 2915
Estonia 1,3871,096248324832483 234924721916
Hungary 571433100410041004 1002
Latvia 1,3651,051241624162416 234923611865
Lithuania 1,1941,181237523752375 223623531893
Poland 1,181726190719071907 1887496
Romania 1,205706191119111911 19041667
Russia 3,5243,025654965496549 624751324234
Slovakia 585404989989989 981809
Slovenia 637353990990990 988854
Asia
Bangladesh 1,0521,14121932193 2193 1000 2189 2182
Cambodia 1,207689189618961896 9251887 873
India 2,0462,937498349834983 165149384424
Indonesia 1,7481,337308530853085 209130642960
Iran 1,5571,673323032303230 10152215978
Kazakhstan 1,06180218631863 1863 1767 1840 1482
Kyrgyzstan 1,7501,19429442944 2944 2835 2937 2704
Malaysia 1,055983203820382038 70620211524
Mongolia 505436941941941 899923870
Nepal 1,115872198719871987 9941987927
Philippines 1,8601,27431343134 3134 2128 3129 1633
158
159
APPENDIX 2.F (CONTINUED): NUMBER OF OBSERVATIONS IN REGRESSIONS
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
South Korea 1,015 1,027 2042 2042 2042 2040 890
Tajikistan 1,218730194819481948 187619371668
Thailand 634372100610061006 9931004974
Turkey 464518982982982 910974856
Vietnam 1,4821,272275427542754 170426922408
Latin America:
Argentina 605378983983983 915980773
Bolivia 1,5291,317284628462846 86228341626
Brazil 586433101910191019 1019
Chile 1,141915205620562056 104220511955
Colombia 1,933995292829282928 96529272549
Costa Rica 480 501 981 981 981 981 667
Dominican Republic 1,650 1,235 2885 2885 2885 950 2884 1613
Ecuador 1,240799203920392039 97020381955
El Salvador 1,504 1,422 2926 2926 2926 920 2915 2470
Guatemala 955927188218821882 94518701301
Honduras 974958193219321932 95719171551
Mexico 1,031928195919591959 95419511679
Nicaragua 958992195019501950 94519481837
Panama 489488977977977 967
Peru 1,6321,316294829482948 96529432624
Uruguay 1,146794194019401940 1916 591
Venezuela 1,140679181918191819 93918121435
159
160
APPENDIX 2.F (CONTINUED): NUMBER OF OBSERVATIONS IN REGRESSIONS
Life satisfaction Coefficient on female
Women Men Average
No
controls Equation 2
1
Equation 2a
2
Equation 3
3
Equation 4
4
Africa:
Burkina Faso 414 558 972 972 972 903 966 715
Cameroon 459521980980 980 956 976 780
Kenya 1,0011,154215521552155 203721541440
Mali 462530992992992 973989
Mozambique 499482981981 981 970 981
Senegal 443543986986986 953982885
Tanzania 432524956956956 896950540
Uganda 454533987987987 986985908
1. Equation 2 includes the following explanatory variables: age, age squared, marital status, education, and wave effects. 2. Equation 2a
includes the following explanatory variables: age, age squared, marital status, education, ideal children, and wave effects. 3. Equation 3
includes the following explanatory variables: age, age squared, marital status, health problems, employment status, education, residential
location, attendance at a religious ceremony, and wave effects. 4. Equation 4 includes the following explanatory variables: age, age
squared, marital status, health problems, employment status, education, residential location, attendance at a religious ceremony, income,
occupation, and wave effects.
.
160
161
APPENDIX 3: DATA SOURCES AND VARIABLES
Data sources and variables
varies by
(g=grade, s=school,
d=district, c=county, t=time)
California Department of Education (2002-2008). California Standards Tests Research
Files. Standardized Testing and Reporting (STAR) Program. Retrieved from California
Department of Education website: http://star.cde.ca.gov/.
Percent at least proficient in mathematics gst
Percent at least proficient in English/language arts gst
California Environmental Protection Agency (2010).Daily Data. 2010 Air Quality Data
DVD. Air Resources Board.
Percent of days that exceeded the pollution standard:
Carbon monoxide (CO ppm) st
Nitrogen dioxide (NO
2
ppm) st
Ozone (O
3
ppm) st
Coarse particulate matter (PM
10
μg/m3) st
Fine particulate matter (PM
2.5
μg/m3) st
California Department of Education (2002-2008). California Basic Educational Data
System (CBEDS) School Enrollment and Staffing Data Files. Retrieved from California
Department of Education website: http://www.cde.ca.gov/ds/sd/ cb/studentdatafiles.asp.
White (%) gst
Asian (%) gst
Hispanic (%) gst
African American (%) gst
Other (%) gst
California Department of Education (2002-2008). Academic Performance Index Data Files.
Retrieved from California Department of Education website: http://www.cde.ca.
gov/ta/ac/ap/apidatafiles.asp#updates.
Reduced or free meals (%) st
Parent is a high school graduate (%) st
Parent has some college education (%) st
Parent is a college graduate(%) st
Parent attended graduate school (%) st
Fully certified teachers (%) st
Non-native English speakers (%) st
National Center for Education Statistics (2002-2008). Common Core of Data.
Retrieved from National Center for Education Statistics website:
http://nces.ed.gov/ccd/.
Expenditure per student (in thousands) dt
161
162
APPENDIX 3 (CONTINUED): DATA SOURCES AND
VARIABLES
Data sources and variables
varies by
(g=grade, s=school,
d=district, c=county, t=time)
California Employment Development Department (2009). Sub-County Areas Labor Force
and Unemployment Data. Retrieved from California Employment Development
Department website: http://www.labormarketinfo.edd.ca.gov/cgi/dataanalysis.
County unemployment rate ct
California Board of Equalization (2002-2008). Taxable Sales in California. Retrieved from
California Board of Equalization website: http://www.boe.ca.gov/news/tsalescont.htm.
County taxable transactions ct
162
Abstract (if available)
Abstract
This dissertation is comprised of three chapters that use microeconometric techniques to investigate the factors that affect people’s well-being. In the first two chapters, well-being is defined as life satisfaction or health satisfaction. The first chapter explores how the movement from socialism to capitalism affected the life satisfaction and health satisfaction of East Germans relative to West Germans after reunification. The second chapter examines whether women are happier, less happy, or equally happy as men in countries at various stages of development. The third chapter examines whether pollution affects the academic performance of school children
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
Essays on the economics of subjective well-being in transition countries
PDF
Explaining well-being: essays on the socio-economic factors accompanying life satisfaction
PDF
Four essays on how policy, the labor market, and age relate to subjective well-being
PDF
Essays in empirical health economics
PDF
Essays on human capital accumulation -- health and education
PDF
Three essays on social policy: institutional development, and subjective well-being as a cause and consequence of labor market outcomes
PDF
What leads to a happy life? Subjective well-being in Alaska, China, and Australia
PDF
Adaptation, assets, and aspiration. Three essays on the economics of subjective well-being
PDF
Essays on financial markets
PDF
Effectiveness of a heart plus brain health-focused nutrition intervention among Latinas: a randomized controlled trial of Buenos Hábitos Alimenticios Para Una Buena Salud (Good Eating Habits for ...
PDF
Essays on bundling and discounts
PDF
Preparing students for a twenty-first century global workplace in an era of accountability
PDF
Essays on labor and development economics
PDF
Essays in international economics
PDF
Essays on health economics
PDF
Three essays on health & aging
PDF
Reporting differences among sexually assaulted college women: a cultural exploration
PDF
Clinical research in reproductive health
PDF
Three essays on distance: examing the role of institutional distance on foreign firm entry, local isomorphism strategy and subsidiary performance
PDF
Essays in behavioral and financial economics
Asset Metadata
Creator
Zweig, Jacqueline Smith
(author)
Core Title
Essays on health and well-being
School
College of Letters, Arts and Sciences
Degree
Doctor of Philosophy
Degree Program
Economics
Publication Date
04/26/2012
Defense Date
03/23/2011
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
Happiness,OAI-PMH Harvest,well-being
Place Name
California
(states),
Germany
(geographic subject),
USA
(countries)
Language
English
Contributor
Electronically uploaded by the author
(provenance)
Advisor
Easterlin, Richard A. (
committee chair
), Ham, John C. (
committee chair
), Melguizo, Tatiana (
committee member
)
Creator Email
jackiesmith04@yahoo.com,smith2@usc.edu
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-m3782
Unique identifier
UC1151710
Identifier
etd-Zweig-4500 (filename),usctheses-m40 (legacy collection record id),usctheses-c127-466953 (legacy record id),usctheses-m3782 (legacy record id)
Legacy Identifier
etd-Zweig-4500.pdf
Dmrecord
466953
Document Type
Dissertation
Rights
Zweig, Jacqueline Smith
Type
texts
Source
University of Southern California
(contributing entity),
University of Southern California Dissertations and Theses
(collection)
Repository Name
Libraries, University of Southern California
Repository Location
Los Angeles, California
Repository Email
cisadmin@lib.usc.edu
Tags
well-being