Essays On Gender and Health

University of Pennsylvania
ScholarlyCommons
Publicly Accessible Penn Dissertations
1-1-2012
Essays on Gender and Health

Carla Medalia
University of Pennsylvania, medalia@sas.upenn.edu
Follow this and additional works at: http://repository.upenn.edu/edissertations

Part of the Demography, Population, and Ecology Commons
Recommended Citation
Medalia, Carla, "Essays on Gender and Health" (2012). Publicly Accessible Penn Dissertations. Paper 546.
This paper is posted at ScholarlyCommons. http://repository.upenn.edu/edissertations/546

For more information, please contact repository@pobox.upenn.edu.
Essays on Gender and Health

Abstract
The relationship between gender and health is complex. Although women live longer than men in almost
every country throughout the world, women also tend to be sicker than men. While biological sex differences
likely contribute to sex gaps in health, cross-national, historical, and life course variation suggest that social
factors also play a role. This dissertation is composed of three chapters which examine social explanations for
gender gaps in mortality and morbidity. The first chapter looks at the relationship between gender equality in
the public sphere, and sex gaps in life expectancy throughout the world. I find that influence of gender equality
on the sex gap in life expectancy depends on the level of economic development. The second chapter takes an
historical perspective to examine the trend in the sex gap in depression in the United States between 1971 and
2008. In examining this trend, I find that the sex gap in depression has decreased over the past forty years, due
to a decrease in depression among women that is primarily attributable to an increase in women's labor force
participation and attachment. In the third chapter, I examine the relationship between gender, aging, and
depression using longitudinal data for the population over age fifty in the United States. In doing so, I find that
age does not increase depression until age 75, after which point depression increases for both sexes, but
particularly for men, leading to a reversal in the sex gap in depression at the end of the lifespan. Furthermore,
while the majority of the age effect on depression is explained by social and health changes, I conclude that
there is a net effect of age per se on depression after age 75.
Degree Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)

Graduate Group
Sociology
First Advisor
Jerry A. Jacobs
Keywords
depression, gender equality, life expectancy

Subject Categories
Demography, Population, and Ecology | Sociology
This dissertation is available at ScholarlyCommons: http://repository.upenn.edu/edissertations/546
ESSAYS ON GENDER AND HEALTH

Carla Medalia
A DISSERTATION
in
Sociology and Demography
Presented to the Faculties of the University of Pennsylvania
in
Partial Fulfillment of the Requirements for the
Degree of Doctor of Philosophy
2012
_______________________________
Jerry A. Jacobs, Professor of Sociology
Supervisor of Dissertation
_______________________________
Emily Hannum, Associate Professor of
Sociology and Education.
Graduate Group Chairperson (Sociology)
_______________________________
Hans-Peter Kohler, Professor of Sociology.
Graduate Group Chairperson (Demography)
Dissertation Committee:
Jerry A. Jacobs (Chair), Professor of Sociology
Samuel H. Preston, Professor of Sociology and Demography
Jason Schnittker, Associate Professor of Sociology
This dissertation is dedicated to my Nana,

Judith Klubock Medalia
ii
ACKNOWLEDGEMENTS
This dissertation is the culmination of six years of graduate education and
research in the fields of sociology and demography at the University of Pennsylvania.
Although this project is my own, the success of my work would not have been possible
without the advice from several faculty members and students at the University, as well
as the never-ending support of my family. First and foremost, I want to acknowledge
Jerry A. Jacobs, my advisor and dissertation committee chair. Jerry has been an
exceptionally supportive mentor, and his feedback and advice throughout the process of
researching and writing my dissertation has been indispensable. Sam Preston and Jason
Schnittker, the other members of my dissertation committee, have also provided helpful
theoretical, methodological, and editorial advice. My entire dissertation committee was
wonderful, and I feel fortunate to have worked with such knowledgeable and supportive
scholars in my field.
The first chapter of my dissertation was published in the International Journal of
Comparative Sociology. I would like to thank my coauthor, Virginia W. Chang, for her
hard work and dedication in getting this chapter published. In addition, the editor of IJCS,
David A. Smith, as well as the reviewers, provided insightful comments that strengthened
the paper.
Other faculty members at the University of Pennsylvania whose advice
contributed to my research include Paul Allison, Herb Smith, Kristen Harknett, HansPeter Kohler, Janis Madden, and Robin Leidner. I am also very appreciative of the
iii
advice and support I received from my fellow sociology and demography students,
including Ruth Burke, Andy Fenelon, Vitor Miranda, Allison Sullivan, Sabrina
Danielsen, Junhow Wei, Whitney Schott, and Arielle Kuperberg. I would especially like
to thank Julie Szymczak, a dear friend and indispensable asset to me throughout my
tenure in graduate school. In addition to the faculty and students, I would like to
acknowledge the incredibly helpful staff in sociology, demography, and in the Population
Studies Center library, including Audra Rogers, Carolanne Saunders, Yuni Thorton, John
McCabe, Nancy Bolinski, Aline Rowens, Karen Cooke, Shellie Copp, Addie Mtivier,
and Nykia Perez Kibler.
I would not have been able to write this dissertation without the unrelenting
support from my family. Special thanks to my mom, Deborah Medalia, who has
encouraged and supported me throughout my life, and particularly in the past six years. I
am also tremendously grateful to have such a loving partner and fianc, Drew Dallas,
who has always has my back and helps me keep my cool. Thanks also to my dad,
Anthony Phillips, and his wife, Maureen Smith, and everyone else in my family for their
love and encouragement. I would also like to acknowledge the inspiration of my Papa,
Dr. Avrom Medalia, who passed away in 2002; ever since I was eight years old I wanted
to get a PhD like him.
Finally, I dedicate this dissertation to my Nana, Judy Medalia, who motivated me
to finish this dissertation even when it felt impossible to do so. Although she passed
away in May 2011, her encouragement has stayed with me and continued to help me
iv
through the process of completing my PhD. One message she reiterated when I felt
overwhelmed with school was that If it were easy to get a PhD, everyone would do it.
This message may sound simple, but has reminded me that it was OK that writing a
dissertation was hard. It certainly was difficult to write this dissertation, but I am so glad
that I saw it through to the end.
ABSTRACT
ESSAYS ON GENDER AND HEALTH

Carla Medalia
Jerry A. Jacobs
The relationship between gender and health is complex. Although women live longer
than men in almost every country throughout the world, women also tend to be sicker
than men. While biological sex differences likely contribute to sex gaps in health, crossnational, historical, and life course variation suggest that social factors also play a role.
This dissertation is composed of three chapters which examine social explanations for
gender gaps in mortality and morbidity. The first chapter looks at the relationship
between gender equality in the public sphere, and sex gaps in life expectancy throughout
the world. I find that influence of gender equality on the sex gap in life expectancy
depends on the level of economic development. The second chapter takes an historical
perspective to examine the trend in the sex gap in depression in the United States
between 1971 and 2008. In examining this trend, I find that the sex gap in depression has
decreased over the past forty years, due to a decrease in depression among women that is
primarily attributable to an increase in womens labor force participation and attachment.
In the third chapter, I examine the relationship between gender, aging, and depression
using longitudinal data for the population over age fifty in the United States. In doing so,
I find that age does not increase depression until age 75, after which point depression
vi
increases for both sexes, but particularly for men, leading to a reversal in the sex gap in
depression at the end of the lifespan. Furthermore, while the majority of the age effect on
depression is explained by social and health changes, I conclude that there is a net effect
of age per se on depression after age 75.
vii
TABLE OF CONTENTS
CHAPTER 1: Gender equality, development, and cross-national sex gaps in

life expectancy ................................................................................................................ 1
Abstract ................................................................................................................................................. 1
Theoretical Considerations .............................................................................................................. 5
Mortality and gender research in LDCs ............................................................................................... 5
Mortality and gender research in HDCs .............................................................................................. 7
Data and Methods ............................................................................................................................. 11

Gender equality and sex gaps in mortality ......................................................................................... 12
Results ................................................................................................................................................. 16
Effects of Smoking ............................................................................................................................ 19
Explaining divergence and convergence in the sex gaps in life expectancy ...................................... 22
Discussion .......................................................................................................................................... 24
Tables and Figures ........................................................................................................................... 29
Appendix 1.1 ....................................................................................................................................... 34
References .......................................................................................................................................... 35
CHAPTER 2: Historical changes in the gender gap in depression in the

United States, 1971 to 2008 ....................................................................................... 40
Abstract ............................................................................................................................................... 40
Theoretical Considerations ............................................................................................................ 44
Research on Gender and Depression ................................................................................................. 44
Socio-demographic characteristics associated with depression ......................................................... 46
Objectives ........................................................................................................................................... 52
Data and Methods ............................................................................................................................. 52
Measuring Depression ....................................................................................................................... 52
viii
Survey Waves and Depression Outcomes ......................................................................................... 54

Analysis ............................................................................................................................................. 59
Independent Variables ....................................................................................................................... 60
Methods ............................................................................................................................................. 61
Results ................................................................................................................................................. 62
Objective 1 ......................................................................................................................................... 63
Objective 2 ......................................................................................................................................... 65
Objective 3 ......................................................................................................................................... 69
Discussion .......................................................................................................................................... 74
Tables and figures ............................................................................................................................ 79
Appendix 2 .......................................................................................................................................... 87
References .......................................................................................................................................... 89
CHAPTER 3: Aging, gender and depression ........................................................ 97

Theoretical Considerations .......................................................................................................... 101
Aging and depression....................................................................................................................... 101
Gender, aging and depression .......................................................................................................... 107
Research Questions ....................................................................................................................... 112

Data and Methods ........................................................................................................................... 113
Cohorts............................................................................................................................................. 114
Methods ........................................................................................................................................... 114
Depression ....................................................................................................................................... 115
Independent variables ...................................................................................................................... 116
Findings ............................................................................................................................................. 119

Descriptive results............................................................................................................................ 119
Question 1 ........................................................................................................................................ 122
ix
Question 2 ........................................................................................................................................ 123

Question 3 ........................................................................................................................................ 126
Question 4 ........................................................................................................................................ 128
Question 5 ........................................................................................................................................ 129
Discussion ........................................................................................................................................ 130

Tables and Figures ......................................................................................................................... 136
Appendix 3.1 ..................................................................................................................................... 142
References ........................................................................................................................................ 144
LIST OF TABLES
Table 1.1
Descriptive statistics for LDCs and HDCs
29
Table 1.2
Regression of sex differences in mortality on gender equality

and level of development
30
Table 1.3
Regression of the sub-indexes of gender equality on the sex gap

in life expectancy
31
Table 1.4
Analysis of the sex ratio in the ASDR
31
Table 1.5
Hypotheses for conditional relationship between gender equality

and sex gap in mortality
32
Table 1.6
Regression of male and female life expectancy, separately
32
Appendix 1.1 123 countries in analysis and gender equality score
34
Table 2.1
Objective 2: Beta coefficients and standard errors for logistic

regression of depression on socio-demographic characteristics,
separately by year and sex
84
Table 2.2
Objective 2: Beta coefficients, significance levels and standard

84
errors for regression of depressive severity on socio-demographic
characteristics, separately by year and sex
Table 2.3
Objective 3: Beta coefficients for logistic regression for

predicting depression, pooled years and sexes
85
Table 2.4
Objective 3: Beta coefficients and significance levels for

regression for predicting depressive severity, pooled years
and sexes
86
Table 3.1
Birth years and ages by cohort and survey wave
136
Table 3.2
Fixed effects regression by sex
139
Table 3.3
Age*Sex interactions
141
Appendix 3.1 Summary of literature on age, gender and depression

xi
142-143
LIST OF ILLUSTRATIONS
Figure 1.1
Relationship between gender equality and sex gaps in mortality

by level of development
33
Figure 2.1
Means and 95% confidence intervals for socio-demographic

characteristics by year, males
79
Figure 2.2
Means and 95% confidence intervals for socio-demographic

characteristics by year, females
79
Figure 2.3
Objective 1: Percent with depression (and 95 % confidence

intervals) by sex and year
80
Figure 2.4
Predicted female odds ratios for logistic regression of depression, 80

in models with and without control variables, by year
Figure 2.5
Objective 1: Mean levels of depressive severity (and 95 %

confidence intervals) by sex and year
81
Figure 2.6
Female beta coefficients for regression of depressive severity,

in models with and without control variables, by year
81
Figure 2.7
Objective 2: Odds ratios (and 95% confidence intervals)

for logistic regression of depression on socio-demographic
characteristics by sex and year
82
Figure 2.8
Objective 2: Beta coefficients (and 95% confidence intervals)

for regression of depressive severity on socio-demographic
83
Appendix 2.1 Distribution of depression scores, females 1971
87
Appendix 2.2 Distribution of depression scores, females 2008
87
Appendix 2.3 Distribution of depression scores, males 1971
88
Appendix 2.4 Distribution of depression scores, males 2008
88
Figure 3.1
136
Average CES-D scores by age and sex

(all cohorts and waves)
xii
Figure 3.2
Marital status by sex and age group
137
Figure 3.3
Employment status by sex and age group
137
Figure 3.4
Health by sex and age group
138
Figure 3.5
Predicted CES-D score by sex and age, from fixed

effects regression
138
Figure 3.6
Average CES-D by cohort Males
140
Figure 3.7
Average CES-D by cohort Females
140
xiii
CHAPTER 1: Gender equality, development, and crossnational sex gaps in life expectancy1
Carla Medalia and Virginia W. Chang2
Abstract
Female life expectancy exceeds male life expectancy in almost every country throughout
the world. Nevertheless, cross-national variation in the sex gap suggests that social
factors, such as gender equality, may directly affect or mediate an underlying biological
component. In this paper, we examine the association between gender equality and the
sex gap in mortality. Previous research has not addressed this question from an
international perspective with countries at different levels of development. We examine
131 countries using a broad measure of national gender equality that is applicable in both
Less Developed Countries (LDCs) and Highly Developed Countries (HDCs). We find
that the influence of gender equality is conditional on level of development. While
gender equality is associated with divergence between female and male life expectancies
in LDCs, it is associated with convergence in HDCs. The relationship between gender
equality and the sex gap in mortality in HDCs strongly relates to, but is not explained by,
This paper was previously published: Gender equality, development, and Cross-National Sex Gaps in
Life Expectancy by Carla Medalia and Virginia W. Chang (2011), in the International Journal of
Comparative Sociology, Vol. 52, No. 5, pp. 371-389.
2
This paper was coauthored by Virginia W. Chang, Assistant Professor of Medicine and Assistant
Professor of Sociology at the University of Pennsylvania, Although the published paper was coauthored,
the original idea for the paper, along with the research questions, were conceived solely by me.
Furthermore, I researched and wrote the literature review, collected and cleaned all data, performed all
analyses and wrote all drafts of this paper. Virginias role as a coauthor was to advise, edit and revise the
version of this paper which was published.
sex differences in lung cancer mortality. Finally, we find that divergence in LDCs is
primarily driven by a strong positive association between gender equality and female life
expectancy. In HDCs, convergence is potentially related to a weak negative association
between gender equality and female life expectancy, though findings are not statistically
significant.
Female life expectancy exceeds male life expectancy in almost every country throughout
the world.3 While females higher life expectancy is partly attributable to biological
factors, including hormonal differences (Bird and Rieker, 2008: 389), cross-national
variation in the size of the sex gap suggests that the biological component is likely
conditioned or directly affected by social factors. Gender equality, which refers to parity
between womens and mens positions in society, is one social factor that may help
explain cross-national variation in the sex gap in mortality.
Previous research offers insight into the association between gender equality and
the sex gap in life expectancy, but is limited from a global perspective. First, crossnational research on gender equality and mortality, or the sex gap in mortality, is
typically focused on either less developed countries (LDCs) or highly developed
countries (HDCs). Second, the findings from studies in LDCs are not comparable to
those in HDCs because researchers use different measures or proxies for gender equality
in each setting. For example, studies in LDCs concentrate on educational, political, and
3
According to WHO data from 2008, only Tonga and the Central African Republic had higher male life
expectancies than female.
economic measures of gender equality, while studies in HDCs tend to examine sex
differences in epidemiological risk factors, such as smoking and alcohol consumption
rather than gender equality per se. Third, research in LDCs and HDCs are further
differentiated by the mortality outcomes analyzed. While female or child mortality is the
dependent variable in most LDC studies, sex differences in mortality are more often the
focus in studies of HDCs.
Although research in LDCs and HDCs is not directly comparable, we believe
there is preliminary evidence to suggest that the sex gap in mortality varies by level of
gender equality across nations and, furthermore, that this relationship is conditional on
level of development. For example, gender equality extends female life expectancy in
LDCs (Aden et al., 1997), and the few studies which examine sex differences in life
expectancy find a positive relationship between the sex gap and gender equality (Fuse
and Crenshaw, 2006; Williamson and Boehmer, 1997). Gender equality is also related to
the sex gap in HDCs. For example, as gender equality increases, women have adopted
many of the same risky behaviors, such as smoking and drinking (McKee and
Shkolnikov, 2001; Rogers et al., 2010; Smith, 2004). This trend may have led to a
convergence in sex differences in life expectancy (Bird and Rieker, 2008; Bobak and
Marmot, 1996; Crawley et al., 2008; Degenhardt et al., 2008; McKee and Shkolnikov,
2001; Morley and Hall, 2008). The sex gap also appears to be smaller where there is
more gender equality in the public sphere (Backhans et al., 2007; Crawley et al., 2008;
Helweg-Larsen and Juel, 2000), and may also be exacerbated by women experiencing
stress from work-family conflict (Schulz and Beach, 1999). While studies show that
womens mortality in LDCs is lower where gender equality is higher, less is known about
the nature of this relationship in HDCs. Moreover, it is unclear how male mortality is
associated with gender equality at either level of development. Drawing on the literature,
we hypothesize that the relationship between gender equality and sex differences in life
expectancy is conditional on level of development. More specifically, we hypothesize
that gender equality is associated with a divergence between mens and womens life
expectancies in LDCs and a convergence in HDCs.
In this study, we examine the relationship between gender equality and the sex
gap in mortality in 131 countries, including both HDCs and LDCs. To our knowledge,
this is the first study to use a broad measure of gender equality to examine sex differences
in mortality across a wide range of countries. We use common indicators of public
sphere gender equality, including gender gaps in the economic, political and educational
spheres, which are applicable across different levels of development. In contrast,
epidemiological risk factors are specific to different causes of death, which vary crossnationally and do not reflect the mortality experience in all countries (Link and Phelan,
1995). The analyses are divided into three separate sections. The first section examines
the relationship between gender equality and the sex gap in five mortality outcomes,
including three measures of adult mortality (life expectancy, healthy life expectancy
[HALE], and adult probability of dying between ages 15 and 60) and two measures of
youth mortality (infant and child mortality between ages 1 and 5). In the second section,
we explore the potential contribution of smoking to the sex gap in mortality by

comparing sex differences in lung cancer mortality versus all other causes. In the third
section, we examine the association of gender equality with male and female life
expectancies as separate outcomes to better understand the patterns we observe with
respect to sex gaps in mortality.
We find that greater gender equality is associated with divergence in life
expectancy between the sexes in LDCs but convergence in HDCs. In LDCs, divergence
is primarily driven by a strong positive association between gender equality and female
life expectancy. In HDCs, convergence is possibly attributable to a negative association
between gender equality and female mortality.
Theoretical Considerations
Mortality and gender research in LDCs
Gender equality in LDCs generally increases womens life expectancy, and some studies
suggest that it contributes to divergence between mens and womens mortality. Gender
equality has often been operationalized by factors such as levels of female labor force
participation (LFP), female education, and reproductive health interventions.
Several studies show that the transition of women into the paid labor force
increases female life expectancy relative to males (Baunach, 2003; Fuse and Crenshaw,
2006; Williamson and Boehmer, 1997). This relationship operates through several
pathways. For example, spending time outside the home reduces womens exposure to
the harmful indoor air pollution emitted from burning biomass fuel, which contributes to
5
womens excess morbidity and mortality in LDCs (Murray and Lopez, 2002). In
addition, womens employment in the service sector enhances womens economic value
and discourages female infanticide and neglect (Fuse and Crenshaw, 2006). Decreases in
female mortality at young ages have a profound effect on increasing female life
expectancy at birth (Preston et al., 2001; Shrestha, 2006).
Female education, through improved literacy, enrollment rates, and years
completed, positively influences life expectancy for both sexes (Lee, 2000), and for
women in particular (Williamson and Boehmer, 1997). Like employment, education
increases daughters economic value, thereby reducing parental preferences for sons and
females excess mortality at young ages (McNay, 2005). Education also contributes to
increased female autonomy in the home. Exercised through decision making, more
equitable divisions of household labor and increased age at marriage, female autonomy
improves maternal and child health (J. Caldwell and Caldwell, 1993; Li, 2004).
Caldwells seminal study (1986) shows that the most important route to overall low
mortality in poor countries is the increase in female autonomy.
Gender equality has also increased womens life expectancy through reproductive
health interventions, such as the use of midwives during childbirth (Kabeer, 2005; Shaw,
2006). Midwives decrease both infant and maternal mortality (Frankenberg and Thomas,
2001), thereby, increasing female life expectancy (Williamson and Boehmer, 1997).
Higher contraceptive prevalence (Shen and Williamson, 1997; Williamson and Boehmer,
1997) and lower fertility rates (Fuse and Crenshaw, 2006; Pillai and Gupta, 2006;
Williamson and Boehmer, 1997) also improve female life expectancy. Lastly, gender
equality can improve womens mortality through increased access to other types of health
services and improved nutritional well-being (Potter and Volpp, 1993).
The research summarized above has several limitations with respect our aim of
comparing the relationship between gender equality and the sex gap in mortality crossnationally in LDCs and HDCs. While some studies investigate sex differences in
mortality (Baunach, 2003; Fuse and Crenshaw, 2006; Williamson and Boehmer, 1997),
most of the studies described above examine only womens or child mortality. We
cannot assume that increases in female life expectancy relate to divergence between the
sexes because the same factors that increase female life expectancy may also improve
male life expectancy. For example, maternal health and female education are positively
associated with healthier sons and daughters, and the factors which increase female LFP
and improve womens health and life expectancy may also benefit men. Another caveat
with respect to LDCs is that studies that do examine sex differences typically use only a
small number of LDCs and are therefore limited in generalizability (Baunach, 2003; Fuse
and Crenshaw, 2006; Williamson and Boehmer, 1997).
Mortality and gender research in HDCs

In HDCs, the sex gap in mortality is often attributed to sex differences in epidemiological
risk factors such as tobacco and alcohol use. Improvements in gender equality may lead
women to adopt riskier behaviors and lifestyles previously associated with men (Bird and
Rieker, 2008; Rogers et al., 2010). For example, increases in female smoking rates have
7
reduced the sex difference in mortality by 23 percent (Preston and Wang, 2006: 641).
Increases in female smoking and alcohol consumption have also decreased mens excess
heart disease mortality (Crawley et al., 2008). Another study finds that about threequarters of the sex gap in mortality is due to sex differences in smoking and alcohol
related causes of death, including ischemic heart disease, lung cancer and traumatic
deaths (Wong et al., 2006). In Eastern Europe, where sex differences in life expectancy
are generally considered to be large, the gap is narrowing as womens alcohol and
tobacco consumption approaches mens (Bobak and Marmot, 1996; Degenhardt et al.,
2008; McKee and Shkolnikov, 2001). While most studies explain convergence in the sex
gap in mortality through increases in female smoking rates (Morley and Hall, 2008),
decreases in male smoking-related mortality is also a factor (Zatonski et al., 2007).
Although most studies in HDCs use epidemiological risk factors which may be
related to gender equality, a few studies explicitly examine broader measures of gender
equality in conjunction with sex differences in mortality. For example, Pampel (2001,
2002) examines gender differences in smoking-related mortality using an index of gender
equality. Although he did not find an association, the gender equality index used
included the Total Fertility Rate (TFR) and divorce rate, factors which have not been
identified as standard measures of gender equality and are not highly correlated with
more general measures of gender equality (McNay, 2005). Two Scandinavian studies use
more standard indicators of public sphere gender equality, such as female LFP, the
proportion of parliamentary seats occupied by women, and the ratio of female-to-male
wages. One of these studies finds that gender equality, particularly equality in the
division of labor and economic resources, is related to convergence in morbidity and
mortality between the sexes (Backhans et al., 2007). While gender equality is negatively
associated with mortality for both sexes, they conclude that convergence is attributable to
a stronger negative association between gender equality and female mortality than with
male mortality. Another study posits that as gender equality continues to increase in the
future, and as mens and womens social positions become more similar, further
convergent trends in gender differences in mortality can be expected (Helweg-Larsen
and Juel, 2000: 220).
Gender equality is often associated with an increase in female LFP, which may
have both positive and negative implications for health and mortality. On the one hand,
the increase in womens LFP in the United States over the past 20 years has improved
womens self-rated health (Schnittker, 2007), a finding corroborated by earlier research
(Repetti et al., 1989; Ross and Mirowsky, 1995). Labor force participation may also
reduce female mortality. A study analyzing Wisconsin death certificates from the 1970s
found that death rates among women in the labor force were generally lower than those
of housewives (Passannante and Nathanson, 1985). On the other hand, a decrease in the
gendered specialization of household and labor market production may mean that both
sexes, but especially women, are working a second shift (Hochschild, 1989). Workfamily conflict is related to an increase in stress (Coverman, 1989), which could have a
negative effect on mortality. Stress related to care-giving for an aging parent, for
example, has been found to increase the risk of mortality by as much as 63 percent
(Schulz and Beach, 1999) compared with non-caregivers. This risk may be especially
salient in the many HDCs that have female life expectancies of 80 or older. The negative
impact of care-giving stress is particularly relevant for women, because daughters are
much more likely to care for an aging parent than are sons (Wolf, 1994). Stress from
other causes also contributes to an increase in female mortality. One study found that
women who report high levels of stress are twice as likely to die from a stroke or
coronary heart disease, and 1.5 times more likely to die from cardiovascular disease,
compared with women who reported low stress (Iso et al., 2002). Finally, some research
has shown that as womens presence in the labor market expands, they are more likely to
exhibit the coronary-prone behavior pattern, also known as the type A personality,
which may mitigate sex differences in mortality caused by coronary heart disease
(Crawley et al., 2008; Waldron, 1978). However, other research has questioned the
relationship between the type A behavior pattern and health risks, and instead argued that
underlying hostility levels are more important (Williams et al., 1980). Taken together,
research on female labor force participation and mortality describes a complex
relationship. Some studies find that female labor force participation positively benefits
womens health and mortality, while others find that work-family conflict may
exacerbate the negative association between stress and mortality. However, these studies
do not suggest that this potential negative relationship outweighs the many other positive
benefits of gender equality and womens labor force participation for women and for
10
society in general. Therefore, it remains unclear as to whether the relationship between

gender equality and the sex gap in mortality in HDCs may operate through pathways
other than risky behaviors such as smoking.
Data and Methods

The analysis is conducted in three parts, and the description of the variables, methods
used, and results for each part will be described in turn. Although the mortality
outcomes, provided by the World Health Organization World Health Report (2010), are
available for 193 countries, the analytical sample includes 131 nations, including LDCs
and HDCs (see Appendix 1.1). To distinguish between HDCs and LDCs, we use the
Human Development Index (UN Human Development Report 2010). Of the original 193
countries, we lose 26 countries where HDI is not calculated due to missing data, leaving
167 countries. The average sex difference in life expectancy is not significantly different
for these 26 countries than for the remaining 167. Finally gender equality is missing for
an additional 36 of the remaining countries, 35 of which are LDCs, bringing the sample
down to 131. Comparing the LDCs that are missing gender equality to those that are not,
the average sex difference in life expectancy is one-tenth of a standard deviation lower
among the missing gender equality countries. Thus, for missingness on gender equality
to bias results towards a positive association between gender equality and the sex gap in
life expectancy, gender equality would have to be higher in the non-sample LDCs, which
is unlikely given that they are lower on HDI.
11
Gender equality and sex gaps in mortality

Mortality. We hypothesize that the association between gender equality and crossnational sex gaps in mortality is conditional on level of development. In this first part of
the analysis, we consider five measures of sex gaps in mortality: the difference (in years)
between females and males in (1) life expectancy at birth and (2) healthy life expectancy
at birth (HALE); and the log of the sex ratios for (3) the adult probability of dying
between the ages 15 and 60 (45q15), (4) the Infant Mortality Rate (IMR, 1q0), and (5) the
probability of a child of dying between the ages one and five (4q1). These data derive
from national death registrations from 2007 (HALE) or 2008 that are compiled by the
World Health Organization (WHO) (World Health Statistics 2010, 2010). For all
measures, larger values indicate a female advantage, where sex differences are calculated
by subtracting male from female values, and sex ratios are calculated by dividing male by
female values.
We include the sex gap in the adult probability of dying between ages 15 and 60
in part to control for variation in the age pattern of mortality. Sex gaps in life expectancy
derive from a combination of two factors sex differences in age-specific mortality rates
and the age pattern of mortality. The age pattern of mortality contributes to the sex gap
in life expectancy through the quantity of deaths that occur at specific ages. For example,
when deaths are more concentrated at older ages (a rectangular age pattern of mortality),
the gains in life expectancy produced by declines in age-specific death rates are smaller.
12
Because women have a more rectangular age pattern of mortality than men, the sex
differential in life expectancy could converge even if there were no change in sex
differences in age-specific mortality rates. In a longitudinal study of high income
countries, Glei and Horiuchi (2007) argue that differences in the age pattern of mortality
between men and women may be more important to convergence in the sex gap in life
expectancy than changes in age-specific mortality rates. To address this potential
concern, we examine the sex gap in mortality for a limited age range. Specifically, we
model the log of the sex ratio of the probability of an adult dying between the ages 15 and
60 (45q15). This measure is less sensitive to sex differences in the rectangular age pattern
of mortality than is life expectancy at birth.
Gender equality. We measure gender equality in the public sphere using a
modified version of the Gender Gap Index computed by the World Economic Forum
(WEF) (Hausmann et al., 2010). The original variable measures the gap between women
and men in terms of economic participation and opportunity, political empowerment,
educational attainment, and health and survival. Our measure excludes gender gaps in
health and survival because of its overlap with the dependent variables. Each of the three
domains in this analysis reflects gender inequalities, captured by the ratio of female to
male values, in multiple areas4. Economic participation and opportunity encompasses
The WEF derives each component of the index from the following sources: the International Labour
Organization Key Indicators of the Labour Market and LABORSTA Internet online database; the United
Nations Development Programme Human Development Report; UNESCO Institute for Statistics Education
Indicators, the World Banks World databank: World Development Indicators & Global Development
13
four variables: the ratio of female to male labor force participation; wage equality
between women and men for similar work; the sex ratio of legislators, senior officials and
managers; and the sex ratio of professional and technical workers. Educational
attainment encompasses four variables: sex ratios in a countrys literacy rate, net primary
and secondary enrollment rates, and gross tertiary enrollment rates. Political
empowerment encompasses three variables: the sex ratio for parliamentary seats, the sex
ratio at the ministerial level, and the ratio of the number of years with a female head of
state or government over years with a male head. Each sub-index is calculated by
averaging its variables after they have been normalized by their standard deviations.
To compute the gender equality score for each country, we averaged the scores
for the three dimensions. A value of 100 in the gender equity index would indicate
complete equality between males and females across the three domains, while a score of
0 would indicate complete inequality. The final measure ranges from 28.9 to 81.0, with
an average score of 58.1 across all countries (see Appendix). Of the countries in the
analysis, Yemen, Chad, and Pakistan have the lowest gender equality while the
Scandinavian countries, including Finland, Norway and Iceland, have the highest gender
equality. We also examine each of the three sub-indices separately.
Level of development. Each year, the United Nation (UN) publishes The Human
Development Index (HDI) (Human Development Report, 2010), which measures and
ranks countries by human development. In this study, we use the non-income version of
Finance, The World Economic Forum Executive Opinion Survey, the Inter-Parliamentary Union National
Women in Parliaments and Women in Politics.
14
the HDI, which combines indicators of population health (life expectancy) and education
(mean and expected years of schooling). After carefully examining the ranked countries,
we chose the non-income HDI as opposed to the income version, which also includes
Gross National Income (GNI) per capita. Some countries that have very high GNIs are
significantly lower on other previously mentioned aspects of human development (e.g.,
Qatar, the United Arab Emirates, and Brunei Darussalam). As such, using the income
version of the HDI combines countries that are very different from one another. The
countries with the highest HDI values include New Zealand, Australia, and Norway, and
those with the lowest HDI values include Chad, Mozambique, and Burkina Faso. The
UN ranks countries by their HDI value from highest to lowest human development (1 to
169), and then divides them into four categories of equal size with the following labels:
very high, high, medium, and low human development. From this point forward, we
refer to the countries with very high human development as HDCs, and other countries
as LDCs. To assess how level of development modifies the association between gender
equality and sex differences in life expectancy, we model an interaction between HDC
and gender equality. Exploratory analysis shows that very high is the cut-point that is
most salient for differences in the association of gender equality with sex differences in
life expectancy.
One advantage to using the HDI as an indicator of development is that it includes
life expectancy at birth for both sexes combined. The countries in this analysis vary
widely in life expectancy, from a minimum of 42 years in Zimbabwe to a maximum of 83
15
in Japan. It is possible that the heterogeneous relationship between gender equality and
the sex gap in life expectancy is driven by overall mortality levels in LDCs and HDCs.
In countries where life expectancy is relatively low, there may be more room for life
expectancy to be increased by factors such as gender equality. Because our measure of
development includes life expectancy, it is unlikely that our results are driven by
variation in overall levels of mortality.
Finally, models also control for the log of GNI per capita (Purchasing Power
Parity, from the UN (Human Development Report, 2010)) to account for potential
confounding from wealth.
Results
Table 1.1 shows descriptive statistics for all variables by level of development. Average
sex gaps in mortality are larger in HDCs than LDCs, and life expectancies for both sexes
are higher in HDCs. On average, HDCs are wealthier and have greater gender equality.
Regression results are presented in Table 1.2. Using Ordinary Least Squares
(OLS), we regress each of the four sex gaps in mortality on gender equality, HDC status,
the interaction between gender equality and HDC, and the log of GNI per capita. The
main effect of gender equality in LDCs is represented by the first row of coefficients, and
the estimated effect of gender equality in HDCs, computed as the sum of the main effect
and interaction, is listed at the bottom of the table. To simplify the description of the
results, we refer to a 10 point increase in gender equality when discussing associations
with gender equality. Unless otherwise specified, results are significant at least at the
16
p<.05 level. In Models 1 and 2, the outcome variable is the sex difference in life
expectancy. Model 1 includes gender equality, HDC, the interaction between gender
equality and HDC, and the log of GNI per capita. The interaction term is negative and
highly significant. Results show that a 10 point increase in gender equality is associated
with a 1.5 year increase in the sex difference in life expectancy in LDCs and a weakly
significant 1.1 year decrease in the sex gap in HDCs (p<.10). Therefore, the influence of
gender equality on the sex gap in life expectancy appears to vary by level of
development. In Model 2, we remove the interaction between gender equality and HDC.
Comparing the R-squared parameters between Models 1 and 2 indicates that the
interaction explains 36.5 percent of the overall variation in the sex gap in life expectancy.
The subsequent models in Table 1.2 show the regression of three additional sex
gaps in mortality on the same explanatory variables included in Model 1. In Model 3, a
10 point increase in gender equality is associated with a 1.0 year increase in the sex
difference in HALE in LDCs and a 1.1 year decrease in HDCs. The results shown in
Model 4 indicate that gender equality is associated with divergence in the sex differential
in the adult probability of dying in LDCs and convergence in HDCs. Specifically, a 10
point increase in gender equality contributes to a 10.3 percent increase in the sex ratio in
adult mortality in LDCs, and a 18.0 percent decrease in the sex gap in HDCs. This
suggests that the association we find between gender equality and the sex gap in mortality
is not merely an artifact of the age pattern of mortality.
17
In contrast to the adult outcomes, we do not observe a statistically significant

interaction between gender equality and HDC for either IMR or child mortality. The
relationship between the log of the sex ratio in IMR and gender equality is not significant
at either level of development, shown in Model 4. Finally, in Model 6, gender equality
has a weak positive association with child mortality in LDCs, increasing the sex ratio by
4.1 percent for every 10 point increase in gender equality (p<.10), and a non-significant
association in HDCs. In models excluding the interaction between gender equality and
HDC for the child outcomes, gender equality is not significantly related to the sex ratio in
IMR or youth mortality at either level of development (models not shown).
Figure 1.1 illustrates the relationship between gender equality and each of the four
sex gaps in mortality described above. The regression lines are estimated from models
which control for GNI per capita, and the scatter points are from observed data. In
LDCs, the positive slopes of the regression lines for the three adult outcomes support the
divergence hypothesis that gender equality is associated with an increase in the sex gap in
life expectancy. In HDCs, the weaker but negative slopes of the regression lines are
consistent with convergence between male and female life expectancies and healthy life
expectancies in countries with greater gender equality. The relationship between gender
equality and the sex gaps in the two youth mortality outcomes looks quite different. In
Figure 1.1d, gender equality has almost no visible association with the log of the sex ratio
in IMR at either level of development. We observe a weak slope of gender equality on
the log of the sex ratio in the probability that a child dies between ages 1 and 5 in Figure
18
1.1e for both LDCs and HDCs. In summary, results for adults support the heterogeneous
association between gender equality and sex gaps in mortality. In contrast to the sex gap
in adult mortality, gender equality has a weaker relationship with the sex gap in youth
mortality that is not dependent on level of development.
In Table 1.3, we examine the relationship between the sex gap in life expectancy
with each of the three sub-indices of gender equality: economic participation and
opportunity, educational attainment, and political empowerment. Results indicate that
the overall pattern is consistent with the composite gender equality index. There is a
positive association between each sub-index and the sex gap in life expectancy in LDCs,
though the coefficient for political gender equality is not statistically significant. The
interaction between HDC and gender inequality is negative for all three sub-indices. In
HDCs, both political and economic gender equality are negatively related to the sex gap
in life expectancy, though economic gender equality is not statistically significant. The
only result that deviates from the composite gender equality results is educational gender
equality, which while not statistically significant, has a positive association with the sex
gap in life expectancy in HDCs. This result is not surprising given the very small amount
of variation in educational gender equality among HDCs.
Effects of Smoking
Gender equality may lead to an increase in female smoking behaviors, especially in
HDCs. Many studies have shown that increases in female smoking rates are associated
with declining sex differences in mortality, both longitudinally and cross-nationally.
19
Indeed, gender equality may be related to the sex gap in mortality through its association
with smoking behaviors. To consider this possibility, we examine the contribution of
smoking through a comparison of sex gaps in mortality with and without lung cancer.
While a simple adjustment for lagged smoking rates would be an alternate approach,
lagged measures of smoking are not available for the majority of our sample. Lung
cancer mortality is a commonly used method to estimate deaths attributable to smoking
(Bird and Rieker, 2008; Ezzati and Lopez, 2003; Ezzati et al., 2008; Fenelon and Preston,
Forthcoming; Peto et al., 1992; S. H. Preston et al., 2010). While smoking may lead to
other causes of death, such as deaths from heart disease and other cancers, the vast
majority of lung cancer deaths are attributable to smoking (Office of the Surgeon
General, 2004). To assess whether gender equality is related to the sex gap in life
expectancy above and beyond the association between smoking behaviors and lung
cancer, we compare the Age Standardized Death Rate (ASDR) for lung cancer versus all
other causes of death. By examining the causes of death other than lung cancer, also
referred to as a cause-deleted or associated single decrement (Preston et al., 2001:
80), we ask what mortality would look like in the absence of lung cancer. This allows us
to assess whether the negative effect of gender equality on the sex gap in mortality in
HDCs is driven primarily by the relationship between gender equality and lung cancer, or
if gender equality affects the sex gap in mortality through additional pathways. We use
the ASDR as opposed to the sex gap in the adult probability of dying because the latter is
not available by cause. The data derive from the WHO and refer to the year 2008
20
(Global Burden of Disease, 2010), and are available by sex and country5. The outcome
variables in this part of the analysis are the log of the sex ratio of the ASDR
(males/females) for all cause mortality, lung cancer mortality, and cause-deleted
mortality (mortality in the absence of lung cancer).
The results summarized in Table 1.4 suggest that sex differences in smoking rates
greatly contribute to, but cannot explain, convergence in the sex gap in mortality in
HDCs. A ten point increase in gender equality is associated with a decrease in the sex
ratio in the ASDR for all causes of mortality combined (7 percent) and for causes other
than lung cancer (6 percent). The same ten point increase in gender equality cuts the sex
ratio in the lung cancer ASDR almost in half, by 48.3 percent. Therefore, the interaction
between gender inequality and development is three times larger for lung cancer
mortality relative to that for all-causes and causes other than lung cancer, suggesting that
lung cancer makes major contribution to the convergence of male and female life
expectancies in HDCs. In LDCs, gender equality is positively associated with the sex
ratio in the ASDR for all cause and cause-deleted mortality, but not related to the sex gap
in lung cancer ASDR. These findings also support the notion that an association of
gender equality with sex-related changes to smoking behaviors is more so a phenomenon
in HDCs than LDCs.
5
Because the cause-deleted ASDR is not directly observable, we calculate this outcome by multiplying the
all cause ASDR by the proportion of deaths (ages 60 and over) that are not due to lung cancer. We restrict
the proportion of deaths not due to lung cancer to ages 60 and over as opposed using the proportion for all
ages combined to focus on the effects of smoking. By using ages 60 and older, we are more likely to
remove deaths that are due to smoking than if we removed lung cancer deaths at earlier ages.
21
Explaining divergence and convergence in the sex gaps in life

expectancy
To better understand why gender equality has a conditional relationship with the sex gap
in life expectancy, we estimate the association between gender equality and male and
female life expectancies as separate outcomes. When analyzing male and female life
expectancies, we use the seemingly unrelated regression procedure (sureg in STATA) to
account for any correlation in the error terms due to the fact that estimates come from the
same sample of countries. In order for gender equality to contribute to divergence in
LDCs, females have to benefit more from gender equality than males. On the other hand,
convergence in HDCs is possible only if females benefit less from gender equality than
males. A divergence (or convergence) in the sex gap can be attributed to a variety of
underlying patterns with respect to the relationship between gender equality and female
and male life expectancies (summarized in Table 1.5). For example, we could observe
divergence in the sex gap if gender equality increases both female and male life
expectancies but has a stronger effect for females than males (H1). Alternatively, gender
equality could increase female life expectancy but have no effect on male life expectancy
(H2) or even decrease male life expectancy (H3) in LDCs. We could also observe no
relationship for females and a negative relationship for males (H4). Finally, gender
equality could disadvantage both female and male life expectancy but be worse for males
22
(H5). Explanations for convergence are the opposite of (or mirror) those proposed for a
divergence.
Data for male and female life expectancy come from the WHO World Statistics
Report (2010). In addition to the covariates included in the previous analysis of sex gaps,
we control for HIV prevalence among adults ages 15 to 49. HIV prevalence comes from
the WHO (World Health Organization, 2009), and is missing for 13 countries, reducing
the sample to 118. In prior models for sex gaps in mortality, its inclusion as a covariate
did not modify our results so we excluded it from the final models to preserve the sample
size6.
Table 1.6 shows the results for the regression of female and male life expectancy,
separately. In LDCs, we find that gender equality is positively related to life expectancy
for both sexes. A 10 point increase in gender equality is associated with a 3.7 year
increase in female life expectancy and a 1.4 year increase in male life expectancy
(p<.10). These results support Hypothesis 1 to explain divergence in the sex gap in life
expectancy: gender equality is positively related to life expectancy for both sexes but has
a stronger positive effect on female life expectancy than male life expectancy.
In HDCs, we do not find statistically significant associations between gender
equality and life expectancy for either sex, but the estimate is weakly negative for
females and weakly positive for males. The interaction between gender equality and
HDC is statistically significant for females, implying that there is a different relationship
HIV prevalence has a strong negative association with both female and male life expectancies (r=-0.6 for
both sexes), but a much weaker association with the sex gap in life expectancy (r=-.3)
23
between gender equality and female life expectancy in LDCs than in HDCs. Hence,
though lacking statistical significance, we can speculate that convergence in HDCs may,
to some extent, be driven by a negative association of gender equality with female life
expectancy and a positive association with male life expectancy.
Discussion
In this paper, we find evidence of a heterogeneous relationship between gender equality
and sex differences in life expectancy. In LDCs, gender equality is associated with
divergence in the sex gap in life expectancy: as gender equality increases, women gain
additional years of life over males. In HDCs, on the other hand, gender equality is
associated with convergence in the sex gap in life expectancy. These patterns also
describes the sex difference in healthy life expectancy and the adult probability of dying,
To understand why the association between gender equality and the sex gap in life
expectancy is conditional on level of development, we analyzed male and female life
expectancy separately. In doing so, we found that gender equality contributes to
divergence in LDCs because gender equality has a stronger positive association with
female life expectancy than with male life expectancy. Although researchers typically
assume that the benefits of gender equality are greater for females than males,
convergence in HDCs is only possible if males experience a net benefit. Though lacking
statistical significance, our results suggest that convergence may attributable a negative
association of gender equality with female life expectancy and a positive association with
male life expectancy.
24
Why might gender equality differentially affect female life expectancy in LDCs
vs. HDCs? A vast body of research has established that gender equality has numerous
positive effects on womens quality of life and life expectancy, especially in settings
where they are marginalized and disadvantaged. In this study, gender equality reflects
three areas: economic participation and opportunities, educational attainment, and
political empowerment. We separately examined each of these areas, and found that the
pattern we found for gender equality as a whole is reflected in is parts. In LDCs,
economic and educational gender equality showed a strong positive association with the
sex gap in life expectancy. This finding is consistent with the literature on sex and
mortality in LDCs that show that expansions in female educational and occupational
opportunities improve female mortality outcomes (Baunach, 2003; Fuse and Crenshaw,
2006; Williamson and Boehmer, 1997). As education expands, health services improve,
and labor market opportunities open to women, female autonomy increases and prior
work shows reductions their mortality and increases life expectancy.
For HDCs, we found that composite gender equality, as well as political gender
equality, was associated with a decrease in the sex gap in life expectancy, a finding which
confirms the relationship observed in previous research (Helweg-Larsen and Juel, 2000).
Political gender equality may increase womens access to the right to engage in risky
behaviors already practiced by men, such as smoking and drinking (Bird and Rieker,
2008; Rogers et al., 2010). To further investigate the effects of smoking, we compared
the sex gap in mortality for lung cancer to causes other than lung cancer. We found that
25
gender equality had a strong negative relationship to the sex gap in lung cancer mortality,
a finding consistent with the large body of research on the effects of smoking on
population health (Bobak, 2003; Degenhardt et al., 2008; McKee and Shkolnikov, 2001).
This suggests that smoking behaviors have largely contributed to the convergence in the
sex gap in life expectancy observed in HDCs. However, the analysis also showed that
gender equality is negatively related to the sex gap in mortality from causes other than
lung cancer. Therefore, we cannot exclude the possibility that gender equality is
associated with the sex gap in mortality beyond its association with smoking. One
potential mechanism is the increased stress women may face as a result of working in the
labor force while simultaneously fulfilling care-giving responsibilities, such as was found
(Schulz and Beach, 1999). Smoking could also contribute to convergence in ways that
we were not able to examine in this study. In addition to lung cancer, smoking leads to
other causes of death, such as deaths from other cancers, respiratory diseases, and
cardiovascular diseases (Wong et al., 2006). The remaining association between gender
equality and the sex gap in cause-deleted mortality could be driven by these other causes
of death. Additional research is needed to see if the relationship between gender equality
and the sex gap in mortality persists after all smoking-attributable effects are removed.
The results of this study raise questions regarding the relationship between gender
equality and the sex gap in youth and infant mortality. While we observed a robust
relationship between gender equality and the sex gap in mortality for adults, results are
not as salient for the sex gap in youth and infant mortality. Gender equality has a weak
26
and relatively small positive association with the sex gap in child mortality in LDCs, and
there is no observable association between gender equality and the sex difference in IMR
at either level of development. These findings are consistent with the fact that many of
the mechanisms through which gender equality affects mortality pertain to adults, such as
through educational attainment, occupational opportunities, and health behaviors.
However, previous studies have posited that gender equality may lead to divergence in
the sex gap in youth mortality in LDCs through better and safer delivery, post-natal care,
and reducing a preference for sons (Fuse and Crenshaw, 2006). Further research should
examine why gender equality does not impact the sex gap in IMR and only weakly is
related to the sex gap in youth mortality.
In this study, we find that gender equality is not significantly associated with male
life expectancy in either LDCs or HDCs. Although some improvements in gender
equality affect only female life expectancy (e.g., an increase in midwives reducing
maternal mortality), other aspects of gender equality affect population health more
broadly. Future research should focus attention on the relationship between gender
equality and male mortality, a relationship that is much less understood than that between
gender equality and female mortality.
A limitation of this study is that it uses cross-sectional data. The idea that gender
equality first improves and then later mitigates womens mortality advantage over men
implies that a process occurs over time. Therefore, a longitudinal perspective could
27
clarify how changes in gender equality are related to the changing relationship between
male and female life expectancy.
28
Tables and Figures

Table 1.1. Descriptive statistics for LDCs and HDCs
LDCs (n=89)
Mean
Mortality
Life expectancy at birth (both sexes)
Female life expectancy
Male life expectancy
Sex difference in life expectancy
Sex difference in HALE
Sex ratio in adult probability of dying
Sex ratio in IMR
Sex ratio in child mortality
Sex ratio in all cause ASDR
Sex ratio in cause-deleted ASDR
Sex ratio in lung cancer ASDR
Predictors
Gender equality index
GNI per capita (PPP in international $)
Economic gender equality
Educational gender equality
Political gender equality
HIV prevalence (% of adults)
66.213
68.517
64.056
4.461
2.584
1.580
1.223
1.194
1.392
1.375
3.901
SD
Min
9.107
9.855
8.536
2.861
2.285
0.479
0.177
0.180
0.233
0.222
2.664
55.486
7.236
8710.404 12751.020
0.609
0.127
0.928
0.101
0.128
0.094
2.597
5.099
Max
42
42
42
0
-2
1.005
0.714
0.750
0.905
0.905
0.983
Mean
78
81
78
13
10
3.387
2.000
1.951
2.135
2.118
14.292
78.667
81.571
75.690
5.881
4.238
2.000
1.257
1.233
1.636
1.566
3.652
HDCs (n=42)
SD
Min
3.074
2.401
3.751
2.098
2.093
0.418
0.228
0.210
0.225
0.203
2.037
Max
71
76
66
2
0
1.368
0.750
0.692
1.185
1.185
1.115
83
86
80
12
10
2.964
2.000
2.000
2.214
2.063
9.603
28.943
69.723
63.752
6.810
50.860
80.960
176.000 79426.000 29205.050 11161.430 12844.000 58810.000
0.195
0.879
0.687
0.082
0.497
0.831
0.509
1.000
0.993
0.013
0.938
1.000
0.000
0.410
0.233
0.152
0.031
0.675
0.100
24.800
0.288
0.289
0.100
1.400
Note: Sex differences refer to Female-Male values while sex ratios refer to Male/Female values
Source: Mortality data from WHO; gender equality indices from the WEF; GNI per capita from the UN; and HIV prevalence from WHO
29
Table 1.2. Regression of sex differences in mortality on gender equality and level of development (n=131)
Gender Equality
Gender Equality*HDC
HDC
Log of GNI per capita
Constant
Gender Equality in HDCs
R-sq
Model 1
Model 2
Model 3
LE
0.150***
(0.035)
-0.260***
(0.067)
0.881
(0.672)
0.530*
(0.227)
5.116***
(0.287)
-0.110+
(0.058)
0.244
LE
0.0803*
(0.032)
HALE
0.096**
(0.031)
-0.204***
(0.059)
1.754**
-0.589
0.247
(0.199)
2.948***
(0.252)
-0.108*
(0.051)
0.224
-0.056
-0.659
0.524*
(0.239)
4.929***
(0.298)
0.155
Model 4
Adult
Mortality
0.010**
(0.003)
-0.028***
(0.006)
0.179**
-0.063
0.101***
(0.021)
0.494***
(0.027)
-0.018***
(0.005)
0.405
Model 5
IMR
0.002
(0.002)
0.001
(0.004)
0.028
-0.044
0.000
(0.015)
0.188***
(0.019)
0.003
(0.004)
0.034
Model 6
Youth
Mortality
0.004+
(0.002)
-0.005
(0.004)
0.040
-0.042
0.006
(0.014)
0.173***
(0.018)
-0.001
(0.004)
0.058
Note: Beta coefficients with standard errors in parentheses, + p<.10, * p<.05, ** p<.01, *** p<.001
Source: Mortality data from WHO; gender equality indices from the WEF; GNI per capita and HDC indicator from the UN
30
Table 1.3. Regression of the sub-indexes of gender equality on the sex gap in life
expectancy (n=131)
Model 1
Model 2
Model 3
Economic Participation
and Opportunity
Educational Attainment Political Empowerment
Subindex
6.028**
16.17***
3.029
(2.087)
(2.979)
(2.939)
Subindex*HDC
-10.35+
-10.76
-8.560*
(5.652)
(29.540)
(4.020)
HDC
0.438
1.32
0.657
(0.676)
(1.451)
(0.681)
0.692**
-0.205
0.705**
(0.237)
(0.267)
(0.244)
Constant
4.951***
4.660***
4.918***
(0.296)
(0.270)
(0.320)
Subindex in HDCs
-4.318
5.41
-5.531*
(5.237)
(29.319)
(2.699)
Note: Beta coefficients with standard errors in parentheses, + p<.10, * p<.05, ** p<.01,
*** p<.001, a p = .104
Source: Mortality data from WHO; gender equality indices from the WEF; GNI per
capita and HDC indicator from the UN
Table 1.4. Analysis of the sex ratio in the ASDR (n = 128)
Model 1
Gender Equality
Gender Equality*HDC
HDC
Constant
All causes
0.009***
(0.002)
-0.0159***
(0.004)
0.184***
(0.041)
0.005
(0.014)
0.302*
(0.118)
-0.007*
(0.003)
Model 2
Causes other than
lung cancer
0.008***
(0.002)
-0.014***
(0.004)
0.145***
(0.039)
0.001
(0.013)
0.319**
(0.114)
-0.006+
(0.003)
Model 3
Lung cancer
0.006
(0.009)
-0.0539**
(0.016)
0.265
(0.163)
-0.002
(0.055)
1.202*
(0.475)
-0.048***
(0.014)
*** p<.001
31
Table 1.5. Hypotheses for conditional relationship between gender equality and sex
gap in mortality
Divergence in
LDCs
Hypothesis Females Males
+
+
+
null
1
2
Convergence in
HDCs
Females
Males
null
null
+
+
+
null
Note: Bold signs indicate stronger relationship
Table 1.6. Regression of male and female life expectancy, separately (n = 118)
Gender Equality
Gender Equality*HDC
HDC
HIV prevalence
Constant
Model 1
Female LE
.376***
(0.073)
-.441***
(0.122)
0.672
(1.245)
-.990***
(0.098)
5.757***
(0.470)
73.114***
(0.551)
-0.065
(0.099)
Model 2
Male LE
.145*
(0.073)
-0.087
(0.121)
0.470
(1.240)
-.766***
(0.097)
5.191***
(0.468)
67.542***
(0.548)
0.058
(0.099)
*** p<.001
capita and HDC indicator from the UN; HIV prevalence from the WHO
32
Figure 1.1. Relationship between gender equality and sex gaps in mortality by level
of development
33
Appendix 1.1. 123 countries in analysis and gender equality score

28.94 Yemen
53.94 Malaysia
58.79 Viet Nam
62.93 Uganda
39.03 Chad
53.98 Maldives
58.86 Venezuela
63.02 Barbados
41.01 Pakistan
54.27 Mauritius
59.02 United Rep. of Tanzania 63.14 Portugal
43.19 Mali
54.33 Japan
59.05 Ukraine
43.23 Cte d'Ivoire
54.40 Kenya
59.31 Uruguay
63.18 Argentina
43.63 Saudi Arabia
54.50 Belize
59.48 Thailand
63.27 Mongolia
44.21 Benin
54.99 Azerbaijan
59.72 Colombia
63.43 Costa Rica
44.52 Morocco
55.04 Mexico
59.72 Greece
63.93 Luxembourg
45.83 Turkey
55.29 El Salvador
59.74 Peru
64.28 Namibia
46.09 Egypt
55.99 Indonesia
59.82 Honduras
64.48 Australia
46.49 Syrian Arab Rep.
56.08 Brazil
59.85 Botswana
65.39 Trinidad and Tobago
46.72 Iran (Islamic Rep. of) 56.11 Tajikistan
59.87 Croatia
65.68 Mozambique
47.80 Ethiopia
56.13 Zimbabwe
59.93 Singapore
65.68 Canada
48.29 Jordan
56.23 Cyprus
60.32 Kyrgyzstan
66.18 USA
48.49 Algeria
56.51 Georgia
60.44 Israel
66.40 Latvia
48.71 Nigeria
56.80 Malta
60.47 Bulgaria
66.79 Sri Lanka
63.16 Nicaragua
49.05 Qatar
56.84 Angola
60.77 China
66.93 Netherlands
49.08 Nepal
57.01 Hungary
60.86 Chile
67.14 United Kingdom
49.38 Mauritania
57.07 Madagascar
60.99 Estonia
67.50 Belgium
49.43 Cameroon
57.50 Bangladesh
61.02 France
67.78 Germany
49.89 Burkina Faso
57.51 Gambia
61.17 Jamaica
68.21 South Africa
50.52 Guatemala
57.61 Bolivia
61.18 Russian Federation
68.21 Spain
50.77 Fiji
57.68 Armenia
61.20 Poland
68.36 Switzerland
50.86 Bahrain
57.72 Slovakia
61.42 Kazakhstan
69.41 Philippines
51.02 India
57.78 Brunei Darussalam 61.44 TFYR Macedonia
69.72 Lesotho
51.47 Tunisia
57.88 Italy
70.44 Denmark
51.61 Zambia
57.95 Dominican Republic 61.76 Ecuador
71.31 Ireland
52.13 Republic of Korea
58.18 Ghana
61.79 Panama
71.78 New Zealand
52.21 Kuwait
58.19 Paraguay
61.90 Guyana
74.55 Sweden
52.95 Suriname
58.43 Romania
61.93 Austria
77.48 Finland
53.07 Senegal
58.45 Albania
61.45 Slovenia
62.39 Bahamas
79.73 Norway
53.25 United Arab Emirates 58.69 Czech Republic
62.44 Lithuania
80.96 Iceland
53.78 Cambodia
62.82 Republic of Moldova
58.72 Malawi
Note: Sorted by gender equality; bold countries are HDCs
Source: Gender equality data from the WEF; HDC indicator from the UN
34
References
Aden AS, Omar MM, Omar HM, et al. (1997) Excess female mortality in rural Somalia Is inequality in the household a risk factor? Social science & medicine 44(5): 709-715.
Backhans MC, Lundberg M and Mnsdotter A (2007) Does increased gender equality
lead to a convergence of health outcomes for men and women? A study of Swedish
municipalities. Social science & medicine 64(9): 1892-1903.
Baunach DM (2003) Gender, mortality, and corporeal inequality. Sociological Spectrum
23(3): 331-358.
Bird CE and Rieker PP (2008) Gender and Health: The Effects of Constrained Choices
and Social Policies. Cambridge University Press.
Bobak M (2003) Relative and absolute gender gap in all-cause mortality in Europe and
the contribution of smoking. European journal of epidemiology 18(1): 1518.
Bobak M and Marmot M (1996) East-West health divide and potential explanations.
East-West Life Expectancy Gap in Europe. Environmental and Non-Environmental
Determinants. Boston: Kluwer Academic Publishers, 17-44.
Caldwell J (1986) Routes to low mortality in poor countries. Population and
Development Review 12(2): 171-220.
Caldwell J and Caldwell P (1993) Womens Position and Child Mortality and Morbidity
in Less Developed Countries. Womens Position and Demographic Change. Oxford:
Clarendon Press, 122-139.
Coverman S (1989) Role Overload, Role Conflict, and Stress: Addressing Consequences
of Multiple Role Demands. Social Forces 67(4): 965-982.
Crawley SL, Foley LJ and Shehan CL (2008) Gendering Bodies. Lanham: Rowman &
Littlefield Publishers.
Degenhardt L, Chiu W-T, Sampson N, et al. (2008) Toward a global view of alcohol,
tobacco, cannabis, and cocaine use: Findings from the WHO World Mental Health
Surveys. Plos Medicine 5(7): 1053-1067.
Ezzati M and Lopez AD (2003) Measuring the accumulated hazards of smoking: global
and regional estimates for 2000. British Medical Journal 12(1): 79.
35
Ezzati M, Friedman AB, Kulkarni SC and Murray CJL (2008) The Reversal of Fortunes:
Trends in County Mortality and Cross-County Mortality Disparities in the United States.
PLoS Med 5(4): e66.
Fenelon A and Preston SH (Forthcoming) Estimating smoking-attributable mortality in
the United States. Demography.
Frankenberg E and Thomas D (2001) Womens Health and Pregnancy Outcomes: Do
Services Make a Difference? Demography 38(2): 253-265.
Fuse K and Crenshaw EM (2006) Gender imbalance in infant mortality: A cross-national
study of social structure and female infanticide. Social Science & Medicine 62(2): 360374.
Glei DA and Horiuchi S (2007) The narrowing sex differential in life expectancy in highincome populations: Effects of differences in the age pattern of mortality. Population
Studies: A Journal of Demography 61(2): 141.
Global Burden of Disease (2010). World Health Organization. Available
at:http://www.who.int/healthinfo/global_burden_disease/en/.
Global Health Observatory Data Repository (2009). World Health Organization.
Hausmann R, Tyson, Laura D. and Zahidi, Saadia (2010) The Global Gender Gap
Report. Geneva, Swizterland: World Economic Forum.
Helweg-Larsen K and Juel K (2000) Sex differences in mortality in Denmark during half
a century, 1943-92. Scandinavian journal of public health 28(3): 214-221.
Hochschild AR (1989) The Second Shift. New York: Penguin Group.
Human Development Report (2010). United Nations Development Program. Available
at:http://hdr.undp.org/en/reports/global/hdr2010/.
Iso H, Date C, Yamamoto A, et al. (2002) Perceived Mental Stress and Mortality From
Cardiovascular Disease Among Japanese Men and Women: The Japan Collaborative
Cohort Study for Evaluation of Cancer Risk Sponsored by Monbusho (JACC Study).
Circulation 106(10): 1229-1236.
Kabeer N (2005) Gender Equality and Human Development: The Instrumental Rationale.
United Nations Development Program, 1-18.
36
Lee MR (2000) Modernization, gender equality, and mortality rates in less developed
countries. Sociological Spectrum 20(2): 195-220.
Li JH (2004) Gender inequality, family planning, and maternal and child care in a rural
Chinese county. Social science & medicine 59(4): 695-708.
Link BG and Phelan J (1995) Social conditions as fundamental causes of disease. Journal
of Health and Social Behavior 35: 8094.
McKee M and Shkolnikov V (2001) Understanding the toll of premature death among
men in Eastern Europe. British Medical Journal 323(3): 1051-1055.
McNay K (2005) The implications of the demographic transition for women, girls and
gender equality: a review of developing country evidence. Progress in Development
Studies 5(2): 115-134.
Morley KI and Hall WD (2008) Explaining the convergence of male and female smoking
prevalence in Australia. Addiction 103(3): 487-495.
Murray C and Lopez AD (2002) The World Health Report: Reducing Risks, Promoting
Healthy Life. Geneva, Swizterland: World Health Organization.
Office of the Surgeon General (2004) The health consequences of smoking: a report of
the Surgeon General. Washington, D.C.: Dept. of Health and Human Services, Centers
for Disease Control and Prevention, National Center for Chronic Prevention and Health
Promotion, Office on Smoking and Health. Available
at:http://www.cdc.gov/tobacco/data_statistics/sgr/2004/highlights/cancer/index.htm.
Pampel FC (2001) Cigarette Diffusion and Sex Differences in Smoking. Journal of
Health & Social Behavior 42(4): 388.
Pampel FC (2002) Cigarette Use and the Narrowing Sex Differential in Mortality.
Population & Development Review 28(1): 77-104.
Passannante MR and Nathanson CA (1985) Female labor force participation and female
mortality in Wisconsin 1974-1978. Social Science & Medicine 21(6): 655665.
Peto R and Lopez AD (1992) Mortality from tobacco in developed countries: Indirect
estimation from national vital statistics. Lancet 339(8804): 1268.
37
Peto R, Boreham J, Lopez AD, Thun M and Heath C (1992) Mortality from tobacco in
developed countries: indirect estimation from national vital statistics. The Lancet
339(8804): 1268-1278.
Pillai VK and Gupta R (2006) Cross-national analysis of a model of reproductive health
in developing countries. Social science research 35(1): 210-227.
Potter JE and Volpp LP (1993) Sex Differentials in Adult Mortality in Less Developed
Countries: The Evidence and its Explanation. Womens Position and Demographic
Change. Oxford: Clarendon Press, 140-161.
Preston SH, Glei DA and Wilmoth JR (2010) A new method for estimating smokingattributable mortality in high-income countries. International Journal of Epidemiology
39(2): 430 -438.
Preston SH and Wang H (2006) Sex Mortality Differences in the United States: the Role
of Cohort Smoking Patterns. Demography 43(4): 631-646.
Preston SH, Heuveline P and Guillot M (2001) Demography: measuring and modeling
population processes. Wiley-Blackwell.
Repetti RL, Matthews KA and Waldron I (1989) Employment and womens health:
Effects of paid employment on womens mental and physical health. American
Psychologist 44(11): 1394-1401.
Rogers RG, Everett BG, Saint Onge JM and Krueger PM (2010) Social, Behavioral, and
Biological Factors, and Sex Differences in Mortality. Demography 47(3): 555-578.
Ross CE and Mirowsky J (1995) Does Employment Affect Health? Journal of Health
and Social Behavior 36(3): 230-243.
Schnittker J (2007) Working More and Feeling Better: Womens Health, Employment,
and Family Life, 1974-2004. American Sociological Review 72(2): 221 -238.
Schulz R and Beach SR (1999) Caregiving as a Risk Factor for Mortality. JAMA: The
Journal of the American Medical Association 282(23): 2215 -2219.
Shaw D (2006) Womens right to health and the Millennium Development Goals:
Promoting partnerships to improve access. International Journal of Gynecology and
Obstetrics 94: 207-215.
38
Shen C and Williamson JB (1997) Child mortality, womens status, economic

dependency, and state strength: A cross-national study of less developed countries. Social
Forces 76(2): 667-700.
Shrestha LB (2006) Life expectancy in the United States. Library of Congress.
Congressional Research Service.
Smith HL (2004) Response Cohort Analysis Redux. Sociological Methodology 34: 111119.
Waldron I (1978) Sex Differences in Coronary Heart Disease. Journal of Health & Social
Behavior 19(1): 119-121.
Williams RB, Haney TL, Lee KL, et al. (1980) Type A behavior, hostility, and coronary
atherosclerosis. Psychosomatic Medicine 42(6): 539.
Williamson JB and Boehmer U (1997) Female life expectancy, gender stratification,
health status, and level of economic development: A cross-national study of less
developed countries. Social Science & Medicine 45(2): 305-317.
Wolf DA (1994) The elderly and their kin: patterns of availability and access. In: Martin
L and Preston S (eds) Demography of Aging. Washington, DC: National Academy Press,
146-194.
Wong MD, Chung AK, Boscardin WJ, et al. (2006) The contribution of specific causes of
death to sex differences in mortality. Public health reports 121(6): 746-754.
World Health Statistics 2010 (2010). Geneva, Swizterland: World Health Organization.
Available at:http://www.who.int/whosis/whostat/2010/en/index.html.
Zatonski WA, Manczuk M, Powles J and Negri E (2007) Convergence of male and
female lung cancer mortality at younger ages in the European Union and Russia.
European journal of public health 17(5): 450-454.
39
CHAPTER 2: Historical changes in the gender gap in

depression in the United States, 1971 to 2008
Abstract
Depression is a major public health concern that is more prevalent among women than
men. Social factors have been shown to explain some of the sex gap. Furthermore, an
earlier study concluded that the increase in womens employment between the 1950s and
1970s contributed to a decrease in the sex gap in depression. Since then, significant
social changes have occurred in the realms of gender, work, and family. This paper
updates the debate on sex and depression by investigating whether there have been
changes in the sex gap in depression since 1970, and if so, what explains those trends?
Using data from the National Health and Nutrition Examination Survey, I find that
women are less depressed now than forty years ago, and as a result, that the sex gap in
depression has decreased. The narrowing of the sex gap is attributable to the increase in
female employment over this period.
Depression is a major public health concern, affecting approximately 6 to 17 percent of

Americans over their lifetimes, depending on the study (Ronald C. Kessler 2006).
According to the World Health Organization, depression is among the leading causes of
disability worldwide and is the third leading contributor to the global burden of disease
(World Health Organization 2008). However, the risk for depression is not uniformly
distributed between the sexes. Research has consistently shown that women are more
likely to experience depression than men (Clancy and Gove 1974; Gove and Tudor 1973;
40
L. Radloff 1975; Ronald C. Kessler et al. 1993; Ronald C. Kessler 2003; Bird and Rieker
2008). Although estimates of womens excess depression varies, community studies find
that women are between 1.5 and 3 times more likely to be depressed than men (Ronald C.
Kessler 2006). There are many explanations for the sex gap in depression, including
biological and hormonal explanations, comorbidity between depression and other
conditions and diseases, and social explanations, including education, employment status,
marital status, parental status, income, poverty, and race, among others. While all of
these factors are likely to play a role, social factors in particular may help to explain
historical variation in the sex gap in depression. Research has shown that where mens
and womens social roles are similar, their depression levels are also similar (John
Mirowsky and Catherine E. Ross 2003; Ronald C. Kessler and McRae 1981).
One study examined the trend in the relationship between sex and depression
between 1957 and 1976 (Ronald C. Kessler and McRae 1981). Given the sex role
hypothesis, which predicts that the relationship between sex and distress should decline
as sex roles become more comparable, the authors hypothesized that the increase in
womens labor force participation, the increase in womens educational attainment, and
other changes in marital status and parental status would lead to a narrowing of the sex
gap in psychological distress. The findings from this study indicate that although women
had higher levels of distress in both years, the sex gap was larger in the 1950s than in the
1970s. In assessing the relevance of the sex role hypothesis in explaining this trend, the
authors found that the increase in womens labor force participation explained 20 percent
of the changing relationship between gender and distress. Labor force participation was
41
in fact the only compositional change to explain the decrease in sex gap in distress over
this period.
Since 1976, there have been dramatic changes in the United States that have led to
greater equality between men and women. For example, while educational attainment
has increased for both sexes, this trend has been especially true for women. As women
have entered the labor force in greater numbers, fertility rates have declined, and
marriages have been postponed to older ages. Only a few studies have undertaken an
historical perspective to examining the relationship between social changes and changes
in the sex gap in depression (R.C. Kessler, K.A. McGonagle, et al. 1994; Ronald C.
Kessler and McRae 1981; Klumb and Lampert 2004; Meertens, Scheepers, and Tax
2003). However, these studies either refer to an earlier period or have study design
limitations in the methods or unit of analysis. As a result, we do not know what impact
these social changes have had on the sex gap in depression since the 1970s. In short, we
need an update on Kessler and McRaes social examination of the trend in the sex gap in
depression (Ronald C. Kessler and McRae 1981).
In this paper, I document trends in depression over the past 40 years for men and
for women. Furthermore, I investigate whether or not there have been changes in the sex
gap in depression. If so, what explains those trends? In my analysis, I undertake a social
perspective to understanding the changing relationship between sex and depression.
Specifically, I examine the importance of changes in employment, marital status, poverty,
and parental status on trends in depression.
In addition to the literature on sex and depression, this study fits into the body of
research that examines the relationship between social changes and other dimensions of
42
public health. For example, research has shown that female employment has led to
improvements in self-rated health among women, leading to a convergence in the sex gap
in self-rated health (Schnittker 2007). As gender equality increases, women have adopted
many of the same risky behaviors as men, such as smoking and drinking (Bird and Rieker
2008; Rogers et al. 2010; Smith 2004). Such increases in female smoking rates have
reduced the sex difference in mortality by 23 percent (Preston and H. Wang 2006:641),
and increases in female smoking and alcohol consumption have also decreased mens
excess heart disease mortality (Crawley, Foley, and Shehan 2008).
To examine trends in the sex gap in depression, I use data from two waves of the
National Health and Nutrition Examination Survey (NHANES): NHANES I, which was
fielded between 1971 and 1975, and the 2005 to 2008 Continuous NHANES. I find that
the sex gap in depression has decreased over this period, and that convergence in the sex
gap is attributable to the decrease in depression among women. This historical change in
the sex gap in depression could be explained by one or both of the following two factors.
The relationship between social demographic characteristics and the risk of depression
for men and/or women could have changed over this period. In addition, compositional
changes in employment, education, marital status, and other attributes could be at play. I
conclude that womens employment is responsible for the narrowing of the sex gap in
depression because more women work now than did forty years ago and because
employment is more protective against depression for women now than it once was.
43
Research on Gender and Depression
Depression is an important public health concern, affecting millions of Americans a year
(Ronald C. Kessler 2006). Consistently, research has shown that women are more likely
to have depression than men. Estimates of womens excess depression ranges between
1.5 to 3 times the rates of men (Bird and Rieker 2008; Ronald C. Kessler 2003; J.
Mirowsky and C.E. Ross 1989; Van de Velde, Bracke, and Levecque 2010).
Why is there such a robust sex gap in the prevalence of depression? Before
summarizing the explanations offered by the literature, it is important to address a factor
that does not contribute to the sex gap. Research has found that the sex gap in depression
reflects a real difference in health and is not attributable to differences in help-seeking
behaviors or reporting (Bird and Rieker 2008; Veroff, Douvan, and Kulka 1981). Other
research also indicates that there is no evidence for women and men labeling symptoms
and addressing severity differently (Verbrugge 1985). Furthermore, the sex gap in
depression is found in nationally representative surveys in addition to clinical settings.
Explanations used to explain the sex gap generally fall into four categories. One
potential explanation for the gender difference in depression derives from whether the
episode is the first onset of depression or a recurrence of depression. Research has shown
that females larger risks for depression and anxiety in childhood and adolescence puts
them at risk later in life (Piccinelli and Wilkinson 2000). Kessler (2006) argues that
while there is a significant gender difference in the risk of first onset, there are mixed
results for the sex gap in recurring depressive episodes, often depending on the subjects
history of depression. Due to limitations of data, I am unable to address the difference in
44
first onset and recurring episodes of depression. While the sex gap in the rates of
depression may be due to a confluence of first onset and recurring episodes, this studys
focus on historical change and the relationship between social demographic
characteristics and the sex gap in depression should still be relevant.
The second explanation is that there are biological determinants and hormonal
differences between men and women that contribute to the sex gap in depression. For
example, some argue that genetic predisposition (Risch et al. 2009), immune function and
the cardiovascular system (Bird and Rieker 2008; Hemingway and Marmot 1999), may
interact with psychological processes. There is less consistent evidence for the role of
hormones (Ronald C. Kessler 2006; Young and Korszun 2010). Furthermore, in a review
of the literature on gender differences in depression, there was no evidence in for genetic
and biological factors of the sex gap in depression (Piccinelli and Wilkinson 2000).
Third, comorbidity between depression and other diseases may explain part of the
sex gap in depression. Research has established that certain conditions may exacerbate
depressive symptoms. For example, empirical evidence finds that comorbidity between
chronic pain and depression accounts for some of the sex gap in depression (Haley,
Turner, and Romano 1985). Obesity, which is more common among adult women than
men (Hedley et al. 2004) , is also strongly associated with risk of depression (Ma and
Xiao 2009). There is also evidence for comorbidity with other psychiatric disorders; for
example, with depression are more likely to have comorbid anxiety, while depressed men
are more likely to have comorbid substance abuse or dependence (Bird and Rieker 2008;
Breslau, Schultz, and Peterson 1995). In addition to diseases and conditions exacerbating
depression, research also suggests that depression may affect health outcomes. For
45
example, one study found that depression increases the probability of reporting poor self
rated health by 15 percent for both sexes, but that depression has a slightly larger effect
on womens self-rated health than on mens (Case and Paxson 2005). Furthermore,
another study found that sex differences in psychosocial factors likely contribute to the
sex gap in non-fatal chronic disease and milder acute problems, but not to the sex gap in
acute or serious conditions (Verbrugge 1985).
Finally, social explanations include demographic characteristics that make women
more susceptible to experiencing stressful events or interpreting those events in such a
way that leads to a depressive episode. It is likely that all of these explanations,
including ones history of depression, biological factors, comorbid diseases and
conditions, and social factors, work together to determine an individuals tendency
toward depression. Although the other types of explanations may still be relevant to this
topic, it is likely that the dramatic changes in social demographic characteristics over the
past forty years are related to any change in the sex difference in depression.
Socio-demographic characteristics associated with depression

Age. Putting gender aside, the relationship between age and depression is complex and
often contradictory (Jorm 2000). Empirical research has found evidence for all possible
relationships between age and depression: some find that age is associated with an
increased risk of depression (Luppa et al.; Newmann 1989; Sonnenberg et al. 2000);
others find that depression decreases as people age (Bebbington et al. 1998; Cairney and
Wade 2002; Christensen et al. 1999; Jorm 2000; Korten and Henderson 2000; Kroenke
and Spitzer 1998; McGuire et al. 2009); research has found that the relationship between
46
age and depression is nonlinear, and that it follows either a U-shaped (Ronald C. Kessler
et al. 1992; John Mirowsky 1996; John Mirowsky and Catherine E. Ross 1992) or inverse
U-shaped (Newmann 1989) pattern; and finally, studies have concluded that there is no
real trend in the relationship between age and depression, or that the relationship is too
inconsistent to pin down (Djernes 2006; Glaesmer et al. 2011; van Grootheest et al.
1999).
In terms of the relationship between gender, age and depression, the story is just
as unclear. One theoretical explanation for the sex gap in depression is that it is
attributable to both the social and biological implications of the reproductive years. A
few studies have shown that the sex gap in depression emerges during young adulthood
and disappears after women go through menopause (Bebbington et al. 1998; Korten and
Henderson 2000). Another possibility is that women are significantly more likely to
experience depression than men at all ages (Case and Paxson 2005), and that the sex gap
is constant over age (Cairney and Wade 2002; Hopcroft and Bradley 2007; Glaesmer et
al. 2011; Ronald C. Kessler et al. 1992; Ronald C. Kessler and McRae 1981; McGuire et
al. 2009; Sonnenberg et al. 2000; van Grootheest et al. 1999). However, other research
suggests that the sex gap in depression increases as people age (Moen 1996; Luppa et
al.).
Thus, there are many inconsistencies in the literature with respect to age and
depression, and to sex gaps in depression as people age. Some of the reasons for these
inconsistencies include how age is measured, the age range considered, and the
depression outcome examined. One study concluded that the nonlinear age pattern in
depression needs to be taken into account when examining sex gaps in depression,
47
particularly at the oldest ages (Medalia 2012). In this study, the sample population is
between the ages of 25 and 74, and I control for age and age-squared to account for the
non-linear association between age and depression.
Race. Depression is also associated with race, though findings are mixed. One
study found that while whites are more likely to experience major depression than blacks,
blacks are more likely to have depressive symptoms (Jackson and Williams 2006).
Whites having greater rates of major depression than blacks is a consistent finding in the
literature, and whites also appear to be at greater risk for major depression than Mexicans
as well (Riolo et al. 2005). In terms of sex differences in depression, previous research
has shown mixed findings by race. For example, a cross-national study found that the
sex gap in depression was larger among Danes and American whites than among the
Japanese (Oksuzyan et al. 2010). Another study found that while the sex gap in
depression was present for whites, it was not statistically significant for blacks (Mezuk et
al. 2010). In this study, I attempt to mitigate variation by race or ethnicity by controlling
for a dummy variable indicating that the respondent is white. Unfortunately, race was
measured as white/not white in the NHANES I, so this is the only possible race variable
for historical comparison.
Poverty. Poverty has a strong positive association with depression. One study
found that being in poverty doubles the risk of recurrent depressive episodes for both
sexes, but that there were no significant gender differences in the effect of poverty on
depression (Bruce, Takeuchi, and Leaf 1991). Other research focusing on women, find
that women in poverty, and especially mothers, are more likely to be depressed than those
not in poverty (Belle 1990; Belle and Doucet 2003; Kahn et al. 2000). However, yet
48
another study found that the positive association between poverty and depression was
true only for whites, and not for Blacks or Mexicans (Riolo et al. 2005). I control for
whether the respondents family income falls below the poverty line in this study.
Marital status. There is a complex relationship between marital status and
depression. Marital satisfaction reduces distress and improves well-being for both men
and women (Coverman 1989). Typically, being married also protects both sexes from
depression, but the effect may be particularly strong for men (Whisman, Weinstock, and
Tolejko 2006). The sex difference in depression is also larger for married than unmarried
people (Ronald C. Kessler 2006). There may be a more nuanced relationship for other
categories of marital status, like divorce and widowhood. Unfortunately, I am unable to
address this possibility in this study. There are too few widowed, separated, nevermarried and divorced individuals in NHANES I to treat them as separate categories.
Therefore, I control for marital status as whether or not the respondent is married.
In addition to marital status affecting depression outcomes, another possibility is
that depression affects marital status. There could be a selection effect of non-depressed
people into marriage. For example, depressed people could be less likely to marry and
more likely to divorce than non-depressed people. One way to rule out the possibility of
reverse causation is to use longitudinal data as opposed to cross-sectional data (the latter
is used in this study). Another way to reduce the implications of reverse causality is to
focus on historical changes in depression. In this study, I examine the association
between depression and marital status in 1971 and again in 2008. I also observe whether
or not the association between depression and marital status has changed over this period.
If there has been no change in the association between marital status and depression, but
49
marital status differences between 1971 and 2008 explain some of the change in the sex
gap in depression, then it is likely that compositional changes in marital status are
responsible. If this is the case, I can conclude that changes in the sex gap in depression
that are attributable to marital status are due to marital status affecting depression and not
the reversal.
Parental status. Most recent studies show that there is little difference in
depression between parents and non-parents (Evenson and Simon 2005; Bures,
Koropeckyj-Cox, and Loree 2009). However, in the past, there was a stronger
relationship between parental status and depression. Radloff (1975; 1980) found that
parents of both sexes, especially those with children younger than 6, had higher
depression scores than non-parents. Another study found that being a parent is more
emotionally disturbing than not being a parent in 1976 (Veroff et al. 1981). In this study,
the measure of parental status that I include is whether or not the respondent has children
under 18 living in the home. More detailed age measures of children are not available.
Education. Most studies find evidence that education is negatively associated
with depression (Adler et al. 1994; Chen et al. 2005; Dohrenwend et al. 1992; Ronald C.
Kessler, Katherine A. McGonagle, et al. 1994). The effect of education on depression
may also be amplified over age. One study found that the education gradient in
depression increased with age, explained by physical health problems among the low
educated (Miech and Shanahan 2000). In this study, I control for age by using a dummy
variable for having a Bachelors Degree (BA).
Employment. At the individual level, there is a negative relationship between
employment and depression both men and women who work are less depressed than
50
those who do not. In a meta-analysis of 91 studies on work and womens well-being

conducted in the 1980s and 1990s, Klumb and Lampert (2004) calculated that the average
benefit of employment was a 60 percent reduction in womens depression. In an earlier
study, using data from 1976, married women who worked outside the home were
significantly less depressed than housewives. However, for their husbands, the effect
was in the other direction men whose wives worked were more likely to be depressed
than those with housewives (Ronald C. Kessler and McRae 1982). The relationship
between work and depression is most strongly supported by two explanations that
describe unemployment as both cause and consequence of depression (Lennon 2006).
First, depression acts as a barrier to employment; depressed people who do work are also
more likely to take sick days. Second, there is strong evidence supporting the
explanation that unemployment can cause depression.
This study does not specifically address the possibility of reverse causation in
terms of employment. However, like the explanation provided for reverse causation and
marital status, the design of this study, which examines historical trends, mitigates that
possibility. Previous research has demonstrated that changes in the sex gap in labor force
participation are related to changes in the sex gap in depression. For example, one study
examined the trend in the relationship between sex and depression between 1957 and
1976 (Ronald C. Kessler and McRae 1981). They found that while women had higher
levels of symptoms in both years, the sex gap was larger in the 1950s then in the 1970s.
The authors attributed this change in the sex gap in distress to the increase in womens
labor force participation over that period, which explains 20 percent of the changing
relationship between gender and distress.
51
Objectives
This paper will attempt to answer three primary questions.
1. What was the relationship between sex and depression forty years ago, and what
is it today? Has the sex gap in depression changed?
The second and third questions address the change in the sex gap in depression.
2. Has the relationship between socio-demographic characteristics and depression
changed for men and/or women over the past forty years? What factors protected
people from depression, and what factors augment the risk for depression? Are
there the same factors associated with depression for men and women? At both
points in time?
3. How do compositional changes affect trends in the sex difference in depression?
Data and Methods

The data used for this study come from multiple waves of the National Health and
Examination Survey (NHANES). The NHANES is a large nationally representative
survey that contains interview and examination components at the individual level, as
well as household-level information. I use data from NHANES I (1971 to 1975) and the
Continuous NHANES (2005 to 2008). The sample is restricted to those ages 25 to 74 for
whom depression measures could be calculated.
Measuring Depression
Survey research on depression generally uses one of two types of depression
instruments. On the one hand are surveys which ask respondents a relatively small
number of questions (between 5 and 30) to determine depression severity and probable
52
Major Depression. Depression severity instruments include questions about the

respondents mood (including sadness and depression), somatic symptoms or malaise
(such as energy level), social isolation (nobody likes me), and positive affect (enjoying
life). These instruments are often used to understand the relationship between social
factors and differences in depression between subgroups (Newmann 1989). On the other
hand are diagnostics tools, which involve a much longer battery of specific questions
pertaining to specific occurrences of depression throughout ones life. These instruments
are often used to diagnose people with a specific disorder, such as Major Depressive
Episode (MDE), according to the criteria established by the Diagnostic and Statistical
Manual of Mental Disorders (DSM). Depending on which type of depression instrument
is used, studies come to very different conclusions regarding the prevalence of depression
in the population (Newmann 1989).
In performing this historical analysis, it is crucial that the indicators of depression
I use are comparable to one another and belong to the same type of depression
instruments. In this study, I use the former type of depression instrument described,
which are based on questions to determine depressive severity and probable depression.
NHANES I uses the Center for Epidemiologic Studies Depression Scale (CES-D), and
the Continuous NHANES uses the Patient Health Questionnaire (PHQ-9). Both of these
instruments have been demonstrated to perform as valid and reliable indicators of
depression severity and are both highly predictive of medically diagnosed Major
Depression. Studies that have compared the CES-D to the PHQ-9 found that the
measures are highly correlated and when used in regressions on a variety of sociodemographic characteristics and other clinical depression measures, and that both
53
instruments produce similar and not statistically significantly different scores (Milette et
al.). Another study also concluded that the PHQ-9 and CES-D measure the same
concept, and that they are highly correlated (r=.83) (Dbouk, Arguedas, and Sheikh 2007).
While each wave uses a different survey instrument, they share many of the same
questions as well as overall indicators of major depression. In this paper, I analyze two
sets of outcome variables. First, I create a dichotomous variable that indicates probable
major depression by using a cut-point on each scale. From this point forward, I will refer
to this measure as depression. Clinicians and other researchers typically use a cut-point
on the scale to determine who is depressed and who is not. The cut-points used in this
study are described below. Second, I examine each survey instrument as a complete,
continuous variable (CES-D for NHANES I and the PHQ-9 for the Continuous
NHANES). I will refer to this measures as depressive severity, since it indicates the
preponderance of depressive symptoms.
This is not the first study to use a historical perspective to examine sex differences
in depression. A study by Kessler and McRae (1981) analyzed the trend in the sex gap in
depression between 1957 and 1976, using five different surveys and four different
depression instruments. Although there was cross-wave variation in questions and
response categories, the authors argued that by analyzing scales, they could still study the
relative changes in depression for men and women (Ronald C. Kessler and McRae 1981).
Survey Waves and Depression Outcomes

NHANES I was collected between 1971 and 1975. A subset of 3,059 adults between the
ages of 25 and 74 received the Center for Epidemiologic Studies Depression Scale (CES54
D), created by the National Institute for Mental Health (NIMH). The CES-D consists of
20 items that ask the respondent how frequently during the past week he experienced a
particular feeling or symptom associated with depression. Responses range from 0 to 3,
where 0 indicates rarely or none of the time (less than 1 day), 1 is some or a little of the
time (1 to 2 days), 2 is occasionally or a moderate amount of the time (3 to 4 days), and 3
is most or all of the time (5 to 7 days). The total instrument is therefore scored from 0 to
60, with 60 indicating the most severe depression. The scale is reliable, and has high
internal consistency between the items (Chronbachs alpha usually around .85)
(McDowell 2006). It is also considered a valid instrument for determining major
depression, and it covers seven out of nine DSM IV symptoms of a Major Depressive
Episode (McDowell 2006).
Four of the items on the CES-D are worded positively (e.g., I enjoyed life),
while the remaining sixteen questions are worded in a negative direction (e.g., I felt
depressed). Documentation for NHANES I urged caution when using the four
positively worded items in the CES-D, because they were not always consistent with the
negatively worded items. However, this is a common feature of the CES-D, and is part
of its design. Empirical evidence suggests that the positive items on the CES-D form a
separate dimension, rather than being the inverse of the negative items (McDowell
2006:350). While non-depressed individuals may experience some of the negative
symptoms, it is thought that the depressed people would not experience the
corresponding positive feelings. Therefore, I use the entire CES-D score for both the
continuous and dichotomous outcome measures.
55
Various cut-points on the CES-D have been recommended to determine

depression, ranging from 16 to 25 (Dbouk et al. 2007; Haringsma et al. 2004; Myers and
Weissman 1980; Turk and Okifuji 1994; Wada et al. 2007; Watnick et al. 2005;
Zonderman, Costa, and McCrae 1989). The selection of a cut-point is important for
estimating the prevalence of depression in a population. Sixteen is the most commonly
used cut-point, especially in clinical settings, although a higher point may be beneficial
in primary care settings to reduce the number of false positives (McDowell 2006:352).
This is true for a couple of reasons. In clinical settings, patients have already selfselected into seeing a mental health professional, so lower scores may still indicate
depression. Secondly, the primary goal of using the CES-D in a clinical setting is to
diagnose and potentially treat a patient. False positives would be preferable to false
negatives, because misdiagnosing people who actually have major depression could be
detrimental to their health. For the purposes of this study, however, a low cut-point
would lead to an overestimation of depression, which could affect the conclusions I make
with regard to historical changes in the sex gap in depression. I considered using several
different cut-points, but after examining the cumulative distribution of each possible
score, I selected 20 as the cut-point, which corresponded to 9.9 percent having
depression. If I had used a cut-point of 16, 15.9 percent would have depression, which is
much too high for a nationally representative population.
The Continuous NHANES was fielded between 2005 and 2008. The mental
health module, the Patient Health Questionnaire (PHQ-9), was administered to 10,214
individuals between 18 and 85 years old. To facilitate comparison between survey
waves, the sample was limited to the 7,622 people between ages 25 and 74. The PHQ-9
56
consists of 9 questions that ask the respondent how frequently they have been bothered
by problems such as little interest in doing things, feeling down, depressed, or
hopeless, or feeling bad about yourself. Responses to these 9 questions are scored
from 0 to 3, where 0 is not at all, 1 is several days, 2 is more than half the days, and 3 is
nearly every day. The score for each question is summed to create a total scale of 0 to
27, where 27 is the highest level of depression. The PHQ-9 includes a tenth question that
asks about the difficulty these problems have caused, which, like in most non-clinical
studies that use the measure, is not included in the scale. The questions on the PHQ-9
derive from the DSM-IV criteria for major depressive disorder (MDD); the instrument is
therefore a valid indicator of major depression (Dbouk et al. 2007).
In this study, the cut-point selected for depression on the PHQ-9 was 10 on the
scale from 0 to 27. This cut-point was selected because it is commonly used and a
standard cut-point (Milette et al.). Previous research has found that when comparing the
PHQ-9 and CES-D, the cut-point used to determine depression may matter for the
prevalence of depression in the study (Dbouk et al. 2007; Milette et al.). Using higher
cut-points on the PHQ-9 may lead to more agreement between the two measures (Dbouk
et al. 2007).
The selection of cut-points in this study was determined after substantial
consideration. Lower cut-points would result in higher levels of people having
depression, while higher cut-points would result in lower levels of people having
depression. Therefore, when comparing two different surveys, it is important that the
cut-points accurately reflect as closely as possible depression levels at both points in
time. That being said, while the selection of cut-points is particularly important for
57
determining absolute changes in depression over the past 40 years, it is less relevant to
comparing relative changes in depression between men and women. Since the same cutpoint is used within each survey for men and women, the sex gap in depression at either
point in time, as well as the change in the sex gap over time, should not be driven by the
particular cut-point used. One way to support this argument is to examine the
distribution of depression scores for men and women using both surveys. Appendices 2.1
through 2.4 show that the distribution of scores is similar between females at both points
in time and for males at both points in time. The primary difference in the distributions is
that more men and women in 2008 report having zero symptoms of depression than in
1971, which are well below the cut-points. Furthermore, in addition to examining cutpoints, I use a continuous depression severity score to analyze trends in the sex gap in
depression over time. Using both of these outcome measures provides a sensitivity
analysis to help overcome the use of different survey instruments in determining
depression.
The two outcome measures of depression used in this study, depression and
depressive severity, represent the two best options to analyze depression using different
survey instruments. Another possible way to analyze depression was to examine specific
questions across survey waves. This alternative was not feasible for two reasons. First,
questions are worded differently and include different key words in each wave. For
example, to register depressive affect, NHANES I asks respondents how often I felt
depressed, whereas the Continuous NHANES asks I felt down, depressed, or
hopeless. In terms of physiological symptoms of depression, NHANES I asks how
often My sleep was restless, whereas the Continuous NHANES asks if the respondent
58
had Trouble sleeping or sleeping too much. These are just a few examples to illustrate
that while both survey instruments measure the same concept, the questions are not
worded identically across surveys. Therefore, it would be impossible to accurately
compare these specific questions to make conclusions regarding trends in the sex gap
over time. The second reason I did not analyze specific questions is that taken separately,
they do not measure depression. While each survey instrument was developed to
measure depression when used as a whole, individual responses do not indicate
depression. For example, someone could have trouble sleeping but experience no other
symptoms of depression. If I compared these incompatible questions across waves I
would only be able to generalize my results to difficulty sleeping, and not to depression,
which is the goal of this paper.
Analysis
To recap, in this paper, I analyze two sets of outcome variables, both of which derive
from the mental health instrument in the NHANES surveys: the CES-D and the PHQ-9.
First, I analyze a dichotomous variable that indicates depression, based on a cut-point of
the original score. The cut-points chosen were selected based on their applicability for
the general population. For the CES-D, the cut-point was 20 on the original scale of 0 to
60, and for the PHQ-9, the cut-point chosen was 10 on the scale from 0 to 27. To analyze
these dichotomous variables, I use logistic regression for survey analysis. Second, I
examine both the CES-D and the PHQ-9 as a continuous variable. Although the response
categories for both instruments are comparable and range from 0 to 3, the CES-D
contains 20 questions while the PHQ-9 contains 9 questions. I rescaled the CES-D so
59
that both scales range between 0 and 27 so that the results can be interpreted. To do this, I
used a linear transformation (see Equation 2.1).
27
Equation 2.1.
60
Because the NHANES has a complex survey design, I use survey specific procedures to
account for weights, and nesting within cluster and strata (National Health and Nutrition
Examination Survey 1982, 2006, 2011b, 2011a).
Independent Variables
To examine sex differences in depression between 1971 and 2008, it is important to
account for socio-demographic changes which may underlie any changes in the sex gap.
In addition to the models that control for sex and survey year, the other independent
variables include age and age-squared (to account for non-linear relationship between age
and depression), race (measured as white or not white due to the limited race questions in
the NHANES I), poverty (below 100 percent of the poverty threshold for each year),
marital status (married or not), parental status (having children under the age of 18 in the
home), educational attainment (BA or not), and employment status (currently has a job or
not).
Most of these variables are measured the same way in each survey. However, an
explicit indicator for parental status was not available in NHANES I. To generate
parental status, I used the household roster (unfortunately, the ages of household
members was not provided), and the relationship of the respondent to the head of
household. A respondent was coded as being a parent if they were married and either the
head of household or the spouse of the head and if the household contained more than 2
60
people, or if the respondent was the head, unmarried, and the household contained more
than 1 person. They are coded as not being a parent when they live alone or with nonrelatives, if they are the child or other relative. Additionally, a married adult who is the
head or spouse of head in a household with two members was considered not a parent.
Methods
The results section is broken down into three parts to answer the three questions posed at
the beginning of the paper. To recap, the first objective asks, What was the relationship
between sex and depression forty years ago, and what is it today? Has the sex gap in
depression changed? These questions are answered looking descriptively at the mean
levels of depression and depressive symptoms by sex and year. The second objective is
to examine the relationship between various social demographic characteristics and
depression for men and women at both points in time. Has the relationship between
socio-demographic characteristics and depression changed for men and/or women over
the past forty years? What factors protected people from depression, and what factors
augment the risk for depression? Are the same factors associated with depression for
men and women? At both points in time? To address these questions, I separately
examine depression and depressive severity for men and women by year (four analytical
groups), adjusting for socio-demographic characteristics. The third objective is: How do
compositional changes affect the sex difference in depression? To study compositional
changes, I pool together both sexes and both years of data. I include female and survey
year in the model, as well as the interaction between female and year. By adding in
61
control variables individually and then all together, I can examine what demographic
shifts are responsible for the convergence in the sex gap I observe.
Results
In this paper, I examine the association between gender and depression in light of social
changes that have occurred between 1971 and 2008. Figures 1 and 2 illustrate these
demographic shifts for men and women. The average age of the population has increased
by about 1 year for both sexes, but this difference is not statistically significant as the
confidence intervals overlap. The percentage of the population that is white has
decreased from about 89-90 percent to 72 percent for both sexes. In 1971, men ages 25
to 74 were more likely to have a BA than women, 17.8 percent versus 12.6 percent. In
2008, the percentage of the population with a BA has increased to 28.3 percent for males
and 29.1 percent for females, resulting in a convergence and reversal in the sex gap in
education. The percentage of the population that is currently married has decreased for
both sexes over this period. In 1971, 85.0 percent of males were married, as were 72.3
percent of females. By 2008, 74.1 percent of males and only 67.0 percent of females
were married. Thus, there was a decrease for both sexes, but a larger decrease in the
proportion married among men than among women. Like being married, being a parent
of children in the home is also less common in 2008 than it was in 1971. In 1971, 57
percent of both sexes had children living at home. By 2008, 42.9 percent of men and
45.3 percent of women were still parents. Having a job decreased slightly but
significantly for men over this time period. In 1971, 82.9 percent of men had a job, and
by 2008, 77.9 did. For women, there has been a massive increase in the percentage of
62
women with a job. In 1971, among women ages 25 to 74, 47.4 percent had a job. In
2008, 64.4 percent of women had a job. This shift decreased the sex gap in employment,
but did not eliminate it, as men are still more likely to have a job. Finally, being in
poverty has increased significantly for men, from 6.9 to 10.8 percent, but has not
significantly increased for women. In summary, the largest socio-demographic shifts
observed over this period include an increase in female employment, a decrease in the
percentage of men and women who were parents and married, and a decrease in the
percentage of the population who was white, though whites still represent the majority.
Objective 1
The first question this paper asks, what were the levels of depression for men and women
in 1971, and what are those levels in 2008? What about the sex gap in depression? To
answer these questions, I compare the mean levels of major depression for men and
women at both points in time. Figure 2.3 shows the means and 95 percent confidence
intervals for the percentage of the population with major depression. A striking finding
is that depression has become less prevalent among women, but has remained unchanged
for men. In 1971, 13.7 percent of women had depression. In 2008, 8.9 percent of women
were currently depressed. This is an absolute decrease of 4.8 percent, which translates
into a 35.0 percent reduction in depression over these years. During the same period,
mens average levels of depression decreased from 5.6 percent in 1971 to 5.1 percent in
2008. However, the confidence intervals overlap, so the decrease of 0.5 percent in the
percent of depression among men is not statistically significant.
63
Change in the sex gap in depression is measured by combining the trends

observed for men and women. Between 1971 and 2008, the sex gap in depression
decreased from 8.1 percent to 3.8 percent. This is an absolute decrease of 4.3 percent,
which means that over this 35 year period, the sex gap in depression was cut by more
than half, specifically by 53.1 percent.
Figure 2.4 illustrates another way to examine change in the sex gap in depression.
Here, logistic regressions were performed separately by year. In the null model, the only
predictor variable is female. In the full model, all socio-demographic characteristics are
included as covariates. In 1971, females were 2.7 times more likely to be currently
depressed without controls, and 1.8 times more likely with controls. In 2008, women are
1.8 times more likely than men to have depression in the null model and 1.5 times more
likely in the full model. In both the null and full models, women are more likely to have
depression than men, and in both models, the sex gap decreases over time. However, the
cumulative effect of socio-demographic controls had a larger impact in explaining the
relationship between sex and depression in 1971 than in 2008. In 1971, the control
variables explained 32.6 percent of the variation in the sex gap in depression. By 2008,
the same socio-demographic controls explained only 14.1 percent of the variation in the
sex gap in depression.
In addition to examining cut-points to represent having depression, I look at the
sex gap in depressive severity, measured by the gap in the total depression scores seen in
Figure 2.5. Depression severity scores range from 0 to 27. Men in 1971 had an average
depression severity of 3.2, while in 2008, mens average severity score decreased to 2.3.
This decrease of 0.8 points is statistically significant. For women, the average depression
64
severity score in 1971 was 4.4, and by 2008, their depression severity score significantly
decreased to 3.4 points. The sex gap in depressive symptoms in 1971 was 1.3 points, and
was 1.0 in 2008. Therefore, there was a slight decrease in the sex gap in depressive
severity over this period, due to a decrease in depressive severity for both sexes but a
larger decrease for women than men. As Figure 2.6 illustrates, however, the time trend in
the sex gap in depressive severity appears to be completely explained by the inclusion of
socio-demographic control variables in the model.
The answer to the question posed in the first objective is that while both sexes
report lower levels of depressive severity now, only women are less likely to have
depression in 2008 than in 1971. Furthermore, there is a significant decrease in the sex
gap in depression, but there does not appear to be a decrease in the sex gap in depressive
severity.
Objective 2
The second part of this paper is about the sex difference in the relationship between
social characteristics and depression over time. What was the relationship between sex,
social factors and depression forty years ago, and what do those relationships look like
today? Have these relationships changed over the past forty years? In other words, this
question tries to address the change in the sex gap in depression by asking if there have
been changes in the causes of depression over this period. Have certain social factors
become either less or more important in predicting depression? To answer these
questions, I look at the effects of social characteristics separately by sex and year, as
65
illustrated in Figure 2.7 and in Table 2.1 (depression), and in Figure 2.9 and in Table 2.2
(depressive severity).
Figure 2.7 show the odds ratios and 95 percent confidence intervals for the
primary socio-demographic characteristics calculated from the full model where age, agesquared, the dummy variable for white, as well as poverty, marital status, parental
status, education, and employment status were also included in the model. Table 2.1
shows the beta coefficients, standard errors, and statistical significance for the same
models. In models not shown, the social characteristics were added separately (along
with controls for age and race). These results support the findings for the full models
shown in Figure 2.7, with one exception: poverty had a larger effect on depression for
both sexes and at both points in time when no other variables were included in the model
than it does in the full model.
Figure 2.7 illustrates the finding that being in poverty is depressing for both men
and women. The sex gap in the effect of poverty on depression is not significant at either
point in time. Being in poverty approximately doubles the likelihood of being depressed
for both sexes in 1971 and in 2008. The change over time in the effect of poverty on
depression within each sex is not statistically significant.
As predicted, being married decreases the likelihood of having depression.
Married women are between 47 and 45 percent less likely to be depressed than unmarried
women in 1971 and 2008, respectively. Married men are 60 percent less likely to be
depressed than unmarried men in 1971 and 41 percent less likely in 2008. However,
there is no significant sex difference in the effect of marital status on depression at either
66
point in time. Furthermore, there is no significant change over time in the effect of
marital status on depression for either sex, including for men.
Parental status has no significant relationship with depression for women in either
1971 or 2008. While not significant, the direction of the effect of motherhood on
depression is positive in 1971 and negative in 2008. For men, being a parent has a
positive effect on depression in 1971, but a negative effect on depression in 2008.
Specifically, fathers were 83 percent more likely to be depressed than non-fathers in
1971, but were 31 percent less likely to be depressed than non-fathers in 2008. While the
effect of being a father is significant in both waves, the change in the effect of fatherhood
on depression is not as strong. Additional analyses (not shown) test the significance of
the interaction between parent and year for men, which is significant at the p<.10 level.
The effect of being a parent on depression is not significantly different for men and
women within each wave. With respect to the time trend, there is no significant time
trend in the effect of parental status for women, but the effect of being a father has
flipped from exacerbating depression in 1971 to protecting against depression in 2008 for
men.
Education is negatively associated with depression risk, as seen in the fact that
having a BA protects against depression for both sexes and at both points in time. In
1971, having a bachelors degree reduced depression by 74 percent for women and by 8
percent for males, though the effect of education on depression for males is not
significant. Although this is big difference, the confidence intervals for men are
particularly large for BA so the sex difference in 1971 is not statistically significant. In
2008, education reduced depression by 66 percent for women and by 62 percent for men;
67
the sex difference in the effect of education on depression in 2008 is also not significant.
Furthermore, there is no significant time trend for either sex.
Finally, having a job is protective against depression for both sexes and at both
points in time. In 1971, males with a job were 70 percent less likely to be depressed than
unemployed men; in 2008, the protective effect of employment increased to 77 percent.
The time trend for males is not significant, however. For females, on the other hand,
there was a significant increase in the effect of having a job on depression between 1971
and 2008. In 1971, women with a job were 36 percent less likely to be depressed, but by
2008, they were 72 percent less likely to be depressed. To verify my results, I also
analyzed these models when both waves of data were combined. In this analysis (not
shown), I additionally included a variable for survey year and an interaction between job
and year. The increase in the protective effect of having a job on depression is significant
at the p<.001 level for females.
In this section, I asked whether there were any sex differences in the effects of
social characteristics on depression at either point in time. Only one characteristic
exhibited a sex gap: parental status has a stronger protective effect against depression for
males than for females in 2008. In addition to sex differences, have been any changes in
the effects of socio-demographic characteristics on having major depression for either
men or women over time? The results show that there are two instances where there has
been a change in the effects of social characteristics on depression. First, there was a
significant change in the effect of being a parent for men, which went from exacerbating
to protecting against depression over the past 40 years. Second, the effect of having a job
is more protective against depression for women in 2008 than it was in 1971.
68
The relationship between social characteristics and depressive severity for males
and females at both points in time is shown in Figure 2.8 and Table 2.2. There are no
significant sex differences in the effect of social characteristics on depressive severity in
either 1971 or 2008. Like the results for depression show, there are two cases where
there is a time trend in the association between a social factor and depression. Being a
parent protects males in 2008, whereas there was an insignificant positive effect of
fatherhood on depressive severity in 1971. For females, the effect of having a job has
become significantly more protective over time.
The relationship between social factors and depressive symptoms has remained
relatively constant over time for both sexes. The two exceptions for both depression and
depressive severity are parental status for males and employment status for females.
However, given that the trend in the sex gap in depression was driven by trends in
womens depression, it is unlikely that the change in the effect of fatherhood has
contributed to the decrease in the sex gap in depression. On the other hand, it is possible
that the change in employment status contributed to the trend in the sex gap in
depression. In order to better understand the decrease in the sex gaps in depression and
depressive severity, I examine compositional changes in the population in the next
section.
Objective 3
The third question asks, How do compositional changes affect the sex difference in
depression? For example, have womens greater educational attainment and increased
labor force participation reduced the sex difference in depression? To answer this
69
question, data for both sexes and survey years was pooled and analyzed using logistic
regression for survey analysis where the outcome is depression and regression where the
outcome is depressive severity; the predictors include a dummy variable for female,
survey year, and the interaction between female and year. By adding in the same sociodemographic characteristics described in the previous section, I examine whether or not
there are changes in the level and significance of the interaction. In doing so, the
compositional changes that are responsible for the decrease in the sex gap in depression
between 1971 and 2008 are identified.
In Table 2.3, the beta coefficients, standard errors, and statistical significance
when predicting depression are shown. Model 1 includes the dummy variables for
female and year, the interaction female*year, age, age-squared, and the dummy for
white. Models 2 through 6 add each of the socio-demographic characteristics, being in
poverty, married, having a BA, being a parent, and having a job, one at a time. Finally,
Model 7 includes all variables together.
The first thing to notice in this table is that the coefficient for being female is
positive and statistically significant at the p<.001 level for all of the models. Because this
model also adjusts for year and the interaction female*year, the female coefficient can be
interpreted as the sex gap in depression in 1971. Therefore, the fact that the variable
female is statistically significant in all models implies that sex differences in social
characteristics cannot completely explain the sex gap in depression in 1971. The second
variable in the table is year, which can be interpreted as the change in the likelihood of a
male being depressed between 1971 and 2008. The coefficient for year is significant and
negative across all eight models which implies that men are less likely to be depressed in
70
2008 than they were in 1971, even controlling for changes in socio-demographic
characteristics.
The third variable entered into the models is the interaction between female and
year, which can be used to identify if there are changes in the sex gap in depression over
time. The interaction is statistically significant in Models 1 through 5, but becomes
insignificant in Models 6, which adjusts for employment status, and Model 7, the full
model. This means that compositional changes in having a job can explain the significant
decrease in the sex gap between 1971 and 2008. In order to better understand the change
in the sex gap in depression, I compare the female*year coefficient from Model 1 to each
of the following models and calculate the percentage of the change in the sex gap in
depression that is explained by controlling for each socio-demographic characteristic.
Model 2 controls for being in poverty. Adjusting for changes in the percentage of
men and women who are in poverty explains 11.1 percent of the trend in the sex gap in
depression between 1971 and 2008. This finding makes sense given that poverty
exacerbates depression, and that there was an increase in the percentage of males who
were in poverty since 1971. The next model, Model 3, adjusts for changes in marital
status over the period. Controlling for the percentage of the population who is currently
married explains 10.1 percent of the decrease in the sex gap in depression since 1971.
Recall that being married is protective against depression and that there has been a
decline in marriage for both sexes, but particularly for males, since 1971. If there had not
been a decrease in the proportion married, the decrease in the sex gap in depression over
the past 40 years would have been slightly smaller.
71
In Model 4, changes in the percentage of the population with a BA explain 12.1

percent of the decrease in the sex gap in depression between 1971 and 2008. Over this
period, there was an increase in educational attainment, particularly for women. Had
educational attainment not increased, the decrease in the sex gap in depression would
have been 12.1 percent smaller. Model 5 adjusts for parental status. Since 1971, there
was a decrease in the percentage of men and women who had children living in the home
from about 57 percent to about 45 percent in 2008. As shown in previous Figures and
Tables, the relationship between depression and parental status for men and women at
both points in time is confusing. It appeared as though being a parent may have
increased depression in 1971 (though the effects were not significant for either sex), but
that being a parent reduced the likelihood of depression in 2008 (significant only for
men). That being said, controlling for parental status increases the time trend in the sex
gap by 13.1 percent. Therefore, the change in the sex gap over the past 40 years would
have been larger if there had not been a decrease in the proportion of men and women
who are parents.
Employment status is adjusted for in Model 6. Controlling for changes in the
proportion of the population with a job explains 66.7 percent of the decrease in the sex
gap in depression between 1971 and 2008, and completely eliminates the statistical
significance of the time trend. As shown in Figures 1 and 2, the percentage of men with a
job decreased slightly since 1971, while the percentage of women with a job increased by
over one-third. Furthermore, as shown in Figure 2.7, having a job protects people against
depression, and there has been an increase in the protective effect for women over the
period. Therefore, the increase in womens labor force participation, together with the
72
enhanced protection that women gain from employment today, has greatly diminished the
sex gap in depression since 1971.
Finally, Model 7 includes all of the previously described social characteristics.
Adjusting for all of these factors in the full model explains 87.9 percent of the trend in the
sex gap in depression. Like in Model 6, the trend in the sex gap is no longer statistically
significant once I control for changes in the composition of social characteristics for men
and women. Therefore, the answer the third objective is that the decrease in the sex gap
in depression between 1971 and 2008 is primarily attributable to the increase in female
labor force participation over the past 40 years.
In addition to examining changes in the likelihood of having depression, I also
analyze depressive severity in Table 2.4. Once again, the coefficient for being female is
significant and positive across all models, indicating that women have higher levels of
depressive severity than men in 1971. Second, the negative and significant year
coefficient implies that there has been a decrease in depressive severity for men between
1971 and 2008. However, the year*female interaction is not statistically significant in
any of the models. This result indicates that depressive severity has decreased for both
men and women and that the sex gap in depressive severity has not changed significantly
over the past forty years. Model 1 shows that in 1971, females average depressive
severity scores were 1.3 points higher than males. In 2008, females scores were 1.0
higher than males on average. Between 1971 and 2008, males average depressive
symptoms scores decreased by 0.8 points. Therefore, although there was a 0.3 decrease
in the sex gap between 1971 and 2008, the change in the sex gap in depression is not
statistically significant. It is worth noting, however, that controlling for employment
73
status in Model 6 reduced females excess depressive severity in 2008 to 0.8 points.
None of the other socio-demographic characteristics have this large of an effect on
changes in the sex gap in depressive severity. Therefore, while there has not been a
statistically significant decline in the sex gap in depressive severity between 1971 and
2008, the observable decline is primarily attributable to compositional changes in
womens employment.
Discussion
Depression is a major health problem that affects both sexes, but particularly women.
Most research finds that women are between 1.5 and 3 times as likely as men to have
depression. While biological sex differences likely contribute to the sex gap in
depression, evidence that womens excess depression varies across subgroups, ages,
countries, and over time suggests that social factors also play a role. In order to better
understand the social factors that contribute to the sex gap in depression, this study
examined trends in the sex gap in depression between 1971 and 2008. I asked whether
there were trends in the sex gap in depression, and if so, why. In examining these trends,
I found that there was a decrease in the sex gap in depression over the past 40 years. As
illustrated in Figure 2.4, without controlling for any social characteristics, women in 1971
were 2.8 times more likely than men to have depression. By 2008, the sex gap was
reduced and women were only 1.8 times more likely to be depressed than men.
Why did the sex gap in depression decrease? Because the sex gap in depression is
a product of male and female depression levels, changes in the sex gap in depression can
be understood by examining trends in depression for both sexes. In this study, I found
74
that 5.5 percent of men had depression in 1971, compared to 5.2 percent in 2008. This
decrease in depression is small and not statistically significant. For women, on the other
hand, 13.7 percent were depressed in 1971, but only 8.9 percent had depression in 2008.
This means that over the past forty years, the percentage of women with depression was
reduced by over one-third. Therefore, the decline in depression among women led to a
narrowing of the sex gap in depression.
In addition to examining male and female trends, I examined the contribution of
social factors to the trend in the sex gap in depression between 1971 and 2008. What role
do changes in the sex differences in social characteristics play? As shown in Figure 2.5, I
found that socio-demographic characteristics explained 35.0 percent of the sex gap in
depression in 1971, but only 11.7 percent of the variation in the sex gap in depression in
2008. This suggests that the majority of the time trend in the sex gap in depression
appears to be driven by sex differences in socio-demographic characteristics.
I argued that the change in the sex gap in depression could be due to one or both
of the following factors. First, the relationship between socio-demographic
characteristics and depression could have changed. I investigated this possible
explanation for the trend in the sex gap in depression by separately analyzing depression
for women and men in 1971 and in 2008. This explanation was largely unsupported by
the data. Surprisingly, there has been little to no change in the majority of these
relationships over time. As shown in Figure 2.7, there are only two cases where the
relationship between a social factor and depression changed for either sex over the past
forty years. The first example is that the effect of parental status on depression reversed
for men between 1971 and 2008: fatherhood went from exacerbating to protecting against
75
depression. This change in the effect of fatherhood on depression was significant, but
only marginally. Furthermore, recall that there has been no meaningful change in
depression among men since 1971. Therefore, I conclude that the change in the effect of
parental status on mens depression is not driving the decrease in the sex gap in
depression. I also observed a change in the effect of employment status for women,
where the protective effect of having a job grew stronger over the past forty years.
Because the trend in the sex gap in depression was driven by a decrease in womens
depression, it is possible that the increase in the protective effect of employment on
depression could have contributed to a decrease in the sex gap between 1971 and 2008.
The second explanation is that compositional factors led to a decrease in the sex
gap in depression. There were many dramatic changes in education, marital status,
parental status, and employment status for both men and women over the past forty years.
I found that the increase in female employment since 1971 explained the vast majority of
the narrowing sex gap in depression. While other factors may contribute, changes in
employment alone explain over two-thirds of the decrease in the sex gap in depression,
rendering the time trend insignificant.
Ultimately, I conclude that female employment is responsible for the decrease in
the sex gap in depression between 1971 and 2008 for two reasons. First, womens labor
force participation increased dramatically, from 47 percent to 64 percent. This absolute
change of 17 percent corresponds to a relative increase of over 36 percent in womens
labor force participation. The second way that womens employment contributes to the
narrowing of sex differences in depression is that employment is more protective against
76
depression today than it was forty years ago. Together, these two forces contributed to
over a 50 percent reduction in the sex gap in depression between 1971 and 2008.
In addition to examining depression, I analyzed depression severity as a way to
provide support for my empirical findings. When I look at depressive severity, a slightly
different story emerges. Between 1971 and 2008, both men and women experienced an
average decrease in depressive severity of about 1 point on a 27 point scale. Although
the sex gap decreased slightly, the change over time was not significant. However, like
for depression, I found that womens employment explains the change in the sex gap in
depressive severity between 1971 and 2008.
The results from this study support previous research on time trends in the sex gap
in depression. Earlier research found that between the 1950s and 1970s, sex differences
in depression declined due to the increase in female labor force participation over that
period (Ronald C. Kessler and McRae 1981). It is not surprising that increases in
womens employment have reduced the disparity in depression between the sexes. As
women have entered the workforce in increasing numbers, gender equality in the public
sphere has improved. Other research has shown that gender equality leads to a
diminished depressive symptoms in women (Chen et al. 2005). In addition to mental
health outcomes, increases in womens employment has improved womens self-rated
health (Schnittker 2007).
This paper used two survey instruments, the CES-D and PHQ-9 to estimate the
prevalence of major depressive and severity of depressive symptoms. While these
surveys are similar, they are not the same. To improve our understanding of the
historical trend in the sex gap in depression, it would be beneficial to have nationally
77
representative data for the same mental health instrument. At the time being, such data
are not available. If new data becomes available that uses the CES-D, it would facilitate
comparison to the data collected in the NHANES I.
This paper shows that the sex gap in depression has decreased between 1971 and
2008. Over this period, female labor force participation dramatically increased, which
led to a subsequent decrease in the prevalence of depression for women. This is an
important finding, and suggests that if women continue to increase their levels of
employment, that the sex gap may be further diminished. However, there remains a sex
gap in depression as well as a sex gap in employment. If the sex gap in employment was
eliminated and an equal percentage of men and women were employed, what would
happen to the sex gap in depression? Future research should address this question.
The sex gap in depression is smaller now than it once was, but it remains. Could
any other socio-demographic factors not examined in this paper help to further reduce the
gap? In addition to social explanations for sex differences in depression, biological
factors and comorbidities with other diseases explain part of the gap. Ideally, research
would be able to look at all of these explanations together, and to examine interactions
between social factors and other types of explanations.
78
Tables and figures

Figure 2.1. Means and 95% confidence intervals for socio-demographic
characteristics by year, males.
100
90
Mean or percent
80
70
60
50
40
30
20
10
0
1971 2008 1971 2008 1971 2008 1971 2008 1971 2008 1971 2008 1971 2008
Age
White
BA
Married
Parent
Job
Poverty
Source: National Health and Nutrition Examination Survey (NHANES 1, 1971-1975;

Continuous NHANES 2005-2008)
Figure 2.2. Means and 95% confidence intervals for socio-demographic
characteristics by year, females.
100
90
80
Mean or percent
70
60
50
40
30
20
10
0
1971 2008 1971 2008 1971 2008 1971 2008 1971 2008 1971 2008 1971 2008
Age
White
BA
Married
Parent
Job
Poverty

79
Figure 2.3. Objective 1: Percent with depression (and 95 % confidence intervals) by

sex and year
16
Percent with depression
14
12
10
8
6
13.7
8.9
4
2
5.6
5.1
1971
2008
0
1971
2008
Females
Males

Figure 2.4. Predicted female odds ratios for logistic regression of depression, in
models with and without control variables, by year
3.0
2.7
2.5
1.8
Odds Ratio
2.0
1.5
1.8
null
1.5
full
1.0
0.5
0.0
1971
2008

80
Mean level of depressive severity
Figure 2.5. Objective 1: Mean levels of depressive severity (and 95 % confidence

intervals) by sex and year
5
4.5
4
3.5
3
2.5
2
1.5
1
0.5
0
4.4
3.4
3.2
2.3
1971
2008
1971
Females
2008
Males

Figure 2.6. Female beta coefficients for regression of depressive severity, in models
with and without control variables, by year
1.4
1.3
1.2
Beta coefficients
1.0
1.0
0.8
0.6
null
0.7
0.7
1971
2008
full
0.4
0.2
0.0

81
Figure 2.7. Objective 2: Odds ratios (and 95% confidence intervals) for logistic regression of depression on socio-demographic
5.5
5.0
4.5
4.0
3.5
3.0
2.5
1.99
2.16
1.90
2.0
2.04
1.83
1.32
1.5
1.0
1971
2008
Poverty
1971
2008
Married
1971
2008
Parent
1971
2008
BA
0.30
0.28
0.23
Females
Males
0.34
Males
Females
Males
Females
Males
Females
0.29
Males
0.64
0.38
Females
0.59
0.92
0.69
Males
0.76
Females
0.55
Males
Females
Males
Females
Males
Females
0.0
0.40
Females
0.53
Males
0.5
1971
2008
Job
Source: National Health and Nutrition Examination Survey (NHANES 1, 1971-1975; Continuous NHANES 2005-2008)
82
Figure 2.8. Objective 2: Beta coefficients (and 95% confidence intervals) for regression of depressive severity on sociodemographic characteristics by sex and year
3.0
2.5
2.0
1.60
1.49
1.54
1.5
0.99
1.0
0.45
0.5
0.29
0.0
-0.5
-0.41 -0.42
-0.58
-1.0
-1.16
-1.5
-0.53
-0.82
-0.97 -1.03
-0.74
-1.01
-1.07
-1.50
-2.0
-1.89
-2.04
-2.5
1971
2008
Poverty
1971
2008
Married
1971
2008
Parent
1971
2008
BA
1971
2008
Job
Source: National Health and Nutrition Examination Survey (NHANES 1, 1971-1975; Continuous NHANES 2005-2008)
83
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
Males
Females
-3.0
Table 2.1. Objective 2: Beta coefficients and standard errors for logistic regression
of depression on socio-demographic characteristics, separately by year and sex
Male
Female
1971
Intercept
Age
Age*Age
White
Poverty
Married
Parent
BA
Job
2008
SE
Sig
5.99
0.21
0.00
0.49
0.77
0.92
0.60
0.09
1.19
(2.09)
(0.10)
(0.00)
(0.34)
(0.41)
(0.29)
(0.30)
(0.46)
(0.33)
**
*
*
6.02
0.22
0.00
0.17
+
0.71
** 0.53
*
0.37
0.97
*** 1.48
SE
(1.26)
(0.05)
(0.00)
(0.18)
(0.16)
(0.16)
(0.17)
(0.36)
(0.23)
1971
Sig
SE
*** 1.73 (0.94)

*** 0.03 (0.05)
*** 0.00 (0.00)
0.11 (0.21)
*** 0.69 (0.38)
** 0.63 (0.26)
*
0.28 (0.20)
** 1.23 (0.47)
*** 0.45 (0.24)
2008
Sig
SE
Sig
5.12
0.21
0.00
0.04
0.64
0.60
0.28
1.08
1.29
(0.96)
(0.05)
(0.00)
(0.14)
(0.13)
(0.14)
(0.16)
(0.22)
(0.14)
***
***
***
+
*
**
+
***
***
+
***
***
+ p<.10, * p<.05, ** p<.01, *** p<.001

Table 2.2. Objective 2: Beta coefficients, significance levels and standard errors for
regression of depressive severity on socio-demographic characteristics, separately
by year and sex
Male
Female
1971
Intercept
Age
Age*Age
White
Poverty
Married
Parent
BA
Job
2008
SE
Sig
3.98
0.06
0.00
0.33
1.49
0.97
0.29
0.82
1.07
(0.69)
(0.04)
0.00
(0.21)
(0.45)
(0.25)
(0.19)
(0.26)
(0.39)
***
0.28
0.23
0.00
0.13
**
0.99
*** 0.58
0.42
** 0.74
** 2.04
SE
(0.54)
(0.03)
0.00
(0.13)
(0.19)
(0.16)
(0.13)
(0.11)
(0.28)
1971
Sig
***
***
***
***
**
***
***
2008
SE
Sig
4.91
0.05
0.00
0.61
1.60
1.16
0.45
1.50
0.53
(1.28)
(0.05)
0.00
(0.41)
(0.51)
(0.32)
(0.27)
(0.27)
(0.28)
***
0.76
0.25
0.00
0.07
**
1.54
*** 1.03
0.41
*** 1.01
+
1.89
SE
(0.91)
(0.04)
0.00
(0.17)
(0.22)
(0.19)
(0.18)
(0.21)
(0.21)
Sig
***
***
***
***
*
***
***
+ p<.10, * p<.05, ** p<.01, *** p<.001

84
Table 2.3. Objective 3: Beta coefficients for logistic regression for predicting
depression, pooled years and sexes (in two parts)
Intercept
Female
Year
Year*Female
Age
Age*Age
White
Poverty
Married
BA
Parent
Job
SE
Sig
SE
Sig
SE
Sig
4.11
0.40
0.20
0.10
0.09
0.00
0.41
(0.76)
(0.04)
(0.08)
(0.04)
(0.03)
0.00
(0.10)
***
***
**
*
**
**
***
4.71
0.38
0.20
0.09
0.10
0.00
0.22
1.18
(0.69)
(0.04)
(0.07)
(0.04)
(0.03)
0.00
(0.11)
(0.15)
***
***
**
*
**
**
*
***
4.11
0.36
0.23
0.09
0.11
0.00
0.31
(0.75)
(0.05)
(0.07)
(0.04)
(0.03)
0.00
(0.10)
***
***
**
*
**
**
**
0.70
(0.09)
Table 2.3. Continued
5
B
SE
Intercept
4.32 (0.60)
Female
0.41 (0.04)
Year
0.20 (0.08)
Year*Female 0.11 (0.04)
Age
0.10 (0.03)
Age*Age
0.00 0.00
White
0.41 (0.11)
Poverty
Married
BA
Parent
0.08 (0.08)
Job
Sig
***
***
**
**
***
***
***
B
4.06
0.25
0.17
0.03
0.13
0.00
0.40
6
SE
(0.76)
(0.05)
(0.08)
(0.05)
(0.04)
0.00
(0.10)
Sig
B
*** 4.21
*** 0.23
*
0.17
0.01
*** 0.15
*** 0.00
*** 0.12
0.71
0.67
0.90
0.03
1.15 (0.11) *** 1.00
3.96
0.39
0.13
0.09
0.09
0.00
0.32
***
1.14
7
SE
(0.68)
(0.05)
(0.07)
(0.05)
(0.03)
0.00
(0.10)
(0.18)
(0.10)
(0.21)
(0.09)
(0.12)
SE
Sig
(0.76)
(0.04)
(0.08)
(0.04)
(0.03)
0.00
(0.10)
(0.21)
***
***
+
*
**
**
**
Sig
***
***
*
***
***
***
***
***
***
+ p<.10, * p<.05, ** p<.01, *** p<.001

85
***
Table 2.4. Objective 3: Beta coefficients and significance levels for regression for
predicting depressive severity, pooled years and sexes (in two parts)
Intercept
Female
Year
Year*Female
Age
Age*Age
White
Poverty
Married
BA
Parent
Job
SE
Sig
SE
Sig
SE
2.23
1.26
0.93
0.23
0.06
0.00
0.50
(0.52)
(0.14)
(0.14)
(0.17)
(0.02)
0.00
(0.12)
***
***
***
**
**
***
1.53
1.18
0.97
0.17
0.08
0.00
0.24
2.11
(0.54)
(0.13)
(0.14)
(0.16)
(0.02)
0.00
(0.12)
(0.22)
**
***
***
2.51
1.12
1.04
0.17
0.09
0.00
0.37
(0.50)
(0.14)
(0.13)
(0.16)
(0.02)
0.00
(0.12)
1.12
(0.11)
***
***
*
***
Sig
*** 2.39
*** 1.20
*** 0.78
0.16
*** 0.06
*** 0.00
** 0.36
***
1.25
SE
Sig
(0.51)
(0.14)
(0.14)
(0.16)
(0.02)
0.00
(0.11)
(0.13)
***
***
***
Table 2.4. Continued
B
Intercept
2.05
Female
1.26
Year
0.95
Year*Female 0.23
Age
0.08
Age*Age
0.00
White
0.47
Poverty
Married
BA
Parent
0.17
Job
5
SE
(0.51)
(0.14)
(0.14)
(0.17)
(0.02)
0.00
(0.12)
Sig
B
*** 2.53
*** 0.72
*** 1.01
0.11
*** 0.13
*** 0.00
*** 0.45
(0.10) +
6
SE
(0.52)
(0.15)
(0.14)
(0.16)
(0.02)
0.00
(0.11)
Sig
B
*** 2.38
*** 0.59
*** 1.00
0.20
*** 0.15
*** 0.00
*** 0.07
1.46
1.00
0.95
0.05
1.58 (0.16) *** 1.33
7
SE
(0.51)
(0.15)
(0.13)
(0.15)
(0.02)
0.00
(0.11)
(0.21)
(0.10)
(0.12)
(0.10)
(0.15)
Sig
***
***
***
***
***
***
***
***
***
+ p<.10, * p<.05, ** p<.01, *** p<.001

86
**
**
**
***
Appendix 2
Appendix 2.1. Distribution of depression scores, females 1971

Appendix 2.2. Distribution of depression scores, females 2008

87
Appendix 2.3. Distribution of depression scores, males 1971

Appendix 2.4. Distribution of depression scores, males 2008

88
References
Adler, Nancy E. et al. 1994. Socioeconomic status and health: The challenge of the
gradient. American Psychologist 49:1524.
Bebbington, P.E. et al. 1998. The influence of age and sex on the prevalence of
depressive conditions: report from the National Survey of Psychiatric Morbidity.
Psychological Medicine 28(1):919.
Belle, Deborah. 1990. Poverty and womens mental health. American Psychologist
45(3):385389.
Belle, Deborah, and Joanne Doucet. 2003. Poverty, Inequality, and Discrimination as
Sources of Depression Among U.S. Women. Psychology of Women Quarterly 27(2):101
113.
Bird, Chloe E., and Patricia P. Rieker. 2008. Gender and health: the effects of
constrained choices and social policies. New York: Cambridge University Press.
Breslau, Naomi, Lonni Schultz, and Edward Peterson. 1995. Sex differences in
depression: a role for preexisting anxiety. Psychiatry Research 58(1):112.
Bruce, Martha Livingston, David T. Takeuchi, and Philip J. Leaf. 1991. Poverty and
Psychiatric Status: Longitudinal Evidence From the New Haven Epidemiologic
Catchment Area Study. Arch Gen Psychiatry 48(5):470474.
Bures, Regina M., Tanya Koropeckyj-Cox, and Michael Loree. 2009. Childlessness,
Parenthood, and Depressive Symptoms Among Middle-Aged and Older Adults. Journal
of Family Issues 30(5):670 687.
Cairney, John, and Terrance J. Wade. 2002. The influence of age on gender differences
in depression. Social Psychiatry and Psychiatric Epidemiology 37(9):401408.
Case, Anne, and Christina Paxson. 2005. Sex Differences in Morbidity and Mortality.
Demography 42(2):189214.
Chen, Ying-Yeh, S. V Subramanian, Doloros Acevedo-Garcia, and Ichiro Kawachi.
2005. Womens Status and Depressive Symptoms: A Multilevel Analysis. Social
Science & Medicine 60(1):4960.
Christensen, H. et al. 1999. Age differences in depression and anxiety symptoms: a
structural equation modelling analysis of data from a general population sample.
89
Clancy, Kevin, and Walter R. Gove. 1974. Sex Differences in Mental Illness: An
Analysis of Response Bias in Self-Reports. The American Journal of Sociology
80(1):205216.
Coverman, Shelley. 1989. Role Overload, Role Conflict, and Stress: Addressing
Consequences of Multiple Role Demands. Social Forces 67(4):965982.
Crawley, Sara L, Lara J Foley, and Constance L Shehan. 2008. Gendering Bodies.
Lanham: Rowman & Littlefield Publishers.
Dbouk, Nader, Miguel R. Arguedas, and Aasim Sheikh. 2007. Assessment of the PHQ-9
as a Screening Tool for Depression in Patients with Chronic Hepatitis C. Digestive
Diseases and Sciences 53(4):11001106.
Djernes, JK. 2006. Prevalence and predictors of depression in populations of elderly: a
review. Acta Psychiatrica Scandinavica 113(5):372387.
Dohrenwend, BP et al. 1992. Socioeconomic status and psychiatric disorders: the
causation-selection issue. Science 255(5047):946 952.
Evenson, Ranae J., and Robin W. Simon. 2005. Clarifying the Relationship Between
Parenthood and Depression. Journal of Health and Social Behavior 46(4):341 358.
Glaesmer, H., S. Riedel-Heller, E. Braehler, L. Spangenberg, and M. Luppa. 2011. Ageand Gender-Specific Prevalence and Risk Factors for Depressive Symptoms in the
Elderly: A Population-Based Study. International Psychogeriatrics 23(08):12941300.
Gove, Walter R., and Jeannette F. Tudor. 1973. Adult Sex Roles and Mental Illness.
The American Journal of Sociology 78(4):812835.
van Grootheest, D. S., A.T.F. Beekman, M. I. Broese van Groenou, and D. J. H. Deeg.
1999. Sex differences in depression after widowhood. Do men suffer more? Social
Psychiatry and Psychiatric Epidemiology 34(7):391398.
Haley, William E., Judith A. Turner, and Joan M. Romano. 1985. Depression in chronic
pain patients: Relation to pain, activity, and sex differences. Pain 23(4):337343.
Haringsma, R., G. I. Engels, A. T. F. Beekman, and Ph. Spinhoven. 2004. The criterion
validity of the Center for Epidemiological Studies Depression Scale(CES-D) in a sample
of self-referred elders with depressive symptomatology. International Journal of
Geriatric Psychiatry 19(6):558563.
90
Hedley, Allison A. et al. 2004. Prevalence of Overweight and Obesity Among US

Children, Adolescents, and Adults, 1999-2002. JAMA: The Journal of the American
Medical Association 291(23):2847 2850.
Hemingway, Harry, and Michael Marmot. 1999. Psychosocial factors in the aetiology
and prognosis of coronary heart disease: systematic review of prospective cohort
studies. BMJ 318(7196):1460 1467.
Hopcroft, Rosemary L., and Dana Burr Bradley. 2007. The Sex Difference in
Depression Across 29 Countries. Social Forces 85(4):1483 1507.
Jackson, Pamela Braboy, and David R. Williams. 2006. Culture, Race/Ethnicity, and
Depression. Pp. 328359 in Women and depression: a handbook for the social,
behavioral, and biomedical sciences, edited by Corey L. M. Keyes and Sherryl H.
Goodman. New York: Cambridge University Press.
Jorm, A. F. 2000. Does Old Age Reduce the Risk of Anxiety and Depression? A Review
of Epidemiological Studies Across the Adult Life Span. Psychological Medicine
30(01):1122.
Kahn, Robert S, Paul H Wise, Bruce P Kennedy, and Ichiro Kawachi. 2000. State
income inequality, household income, and maternal mental and physical health: cross
sectional national survey. BMJ: British Medical Journal 321(7272):13111315.
Kessler, R.C., K.A. McGonagle, et al. 1994. Sex and depression in the national
comorbidity survey. II: Cohort effects* 1. Journal of Affective Disorders 30(1):1526.
Kessler, Ronald C. 2003. Epidemiology of women and depression. Journal of Affective
Disorders 74(1):513.
Kessler, Ronald C., Katherine A. McGonagle, et al. 1994. Sex and depression in the
national comorbidity survey. II: Cohort effects. Journal of Affective Disorders 30(1):15
26.
Kessler, Ronald C. 2006. The Epidemiology of Depression among Women. Pp. 2237
in Women and depression: a handbook for the social, behavioral, and biomedical
sciences, edited by Corey L. M. Keyes and Sherryl H. Goodman. New York: Cambridge
University Press.
91
Kessler, Ronald C., Cindy Foster, Pamela S. Webster, and James S. House. 1992. The
Relationship Between Age and Depressive Symptoms in Two National Surveys.
Psychology and Aging 7(1):119126.
Kessler, Ronald C., K A McGonagle, M Swartz, D G Blazer, and C B Nelson. 1993. Sex
and depression in the National Comorbidity Survey. I: Lifetime prevalence, chronicity
and recurrence. Journal of Affective Disorders 29(2-3):8596.
Kessler, Ronald C., and James A. McRae. 1982. The Effect of Wives Employment on
the Mental Health of Married Men and Women. American Sociological Review
47(2):216227.
Kessler, Ronald C., and James A. McRae. 1981. Trends in the Relationship Between
Sex and Psychological Distress: 1957- 1976. American Sociological Review 46(4):443
452.
Klumb, Petra L., and Thomas Lampert. 2004. Women, work, and well-being 19502000:: a review and methodological critique. Social Science & Medicine 58(6):1007
1024.
Korten, Ailsa, and Scott Henderson. 2000. The Australian National Survey of Mental
Health and Well-Being. The British Journal of Psychiatry 177(4):325 330.
Kroenke, K, and RL Spitzer. 1998. Gender differences in the reporting of physical and
somatoform symptoms. Psychosomatic Medicine 60(2):150 155.
Lennon, Mary Clare. 2006. Women, Work, and Depression: Conceptual and Policy
Issues. Pp. 309327 in Women and depression: a handbook for the social, behavioral,
and biomedical sciences, edited by Corey L. M. Keyes and Sherryl H. Goodman. New
York: Cambridge University Press.
Luppa, M. et al. Age- and gender-specific prevalence of depression in latest-life
Systematic review and meta-analysis. Journal of Affective Disorders (DOI).
Ma, Jun, and Lan Xiao. 2009. Obesity and Depression in US Women: Results From the
2005-2006 National Health and Nutritional Examination Survey. Obesity 18(2):347
353.
McDowell, Ian. 2006. Measuring health: a guide to rating scales and questionnaires.
New York: Oxford University Press US.
92
McGuire, L.C., T.W. Strine, R.S. Allen, L.A. Anderson, and A.H. Mokdad. 2009. The
patient health questionnaire 8: current depressive symptoms among US older adults, 2006
Behavioral Risk Factor Surveillance System. American Journal of Geriatric Psych
17(4):324.
Medalia, Carla. 2012. Age, Gender, and Depression. Dissertation, Philadelphia, PA:
University of Pennsylvania.
Meertens, Vivian, Peer Scheepers, and Bert Tax. 2003. Depressive symptoms in the
Netherlands 1975-1996: a theoretical framework and an empirical analysis of sociodemographic characteristics, gender differences and changes over time. Sociology of
Health and Illness 25(2):208231.
Mezuk, Briana et al. 2010. Reconsidering the Role of Social Disadvantage in Physical
and Mental Health: Stressful Life Events, Health Behaviors, Race, and Depression.
American Journal of Epidemiology 172(11):1238 1249.
Miech, Richard Allen, and Michael J. Shanahan. 2000. Socioeconomic Status and
Depression over the Life Course. Journal of Health and Social Behavior 41(2):162176.
Milette, Katherine, Marie Hudson, Murray Baron, Brett D. Thombs, and Canadian
Scleroderma Research Group. Comparison of the PHQ-9 and CES-D depression scales
in systemic sclerosis: internal consistency reliability, convergent validity and clinical
correlates. Rheumatology.
Mirowsky, J., and C.E. Ross. 1989. Social causes of psychological distress. Aldine de
Gruyter.
Mirowsky, John. 1996. Age and the Gender Gap in Depression. Journal of Health and
Social Behavior 37(4):362380.
Mirowsky, John, and Catherine E. Ross. 1992. Age and Depression. Journal of Health
and Social Behavior 33(3):187205.
Mirowsky, John, and Catherine E. Ross. 2003. Social causes of psychological distress.
New York: Aldine de Gruyter.
Moen, Phyllis. 1996. A life course perspective on retirement, gender, and well-being.
Journal of Occupational Health Psychology;Journal of Occupational Health Psychology
1(2):131144.
93
Myers, JK, and MM Weissman. 1980. Use of a self-report symptom scale to detect
depression in a community sample. Am J Psychiatry 137(9):10811084.
National Health and Nutrition Examination Survey. 1982. A Statistical Methodology for
Analyzing Data from a Complex Survey: The first National Health and Nutrition
Examination Survey. Hyattsville, MD: U.S. Department of Health and Human Services
Public Health Service.
National Health and Nutrition Examination Survey. 2006. Analytic and Reporting
Guidelines. Hyat: National Center for Health Statistics.
National Health and Nutrition Examination Survey. 2011a. NHANES - Continuous
NHANES Web Tutorial - Home.
National Health and Nutrition Examination Survey. 2011b. NHANES I Web Tutorial Home.
Newmann, Joy P. 1989. Aging and depression. Psychology and Aging 4:150165.
Oksuzyan, Anna et al. 2010. Cross-national comparison of sex differences in health and
mortality in Denmark, Japan and the US. European Journal of Epidemiology 25:471
480.
Piccinelli, Marco, and Greg Wilkinson. 2000. Gender differences in depression. The
British Journal of Psychiatry 177(6):486 492.
Preston, Samuel H., and Haidong. Wang. 2006. Sex Mortality Differences in the United
States: The Role of Cohort Smoking Patterns. Demography 43(4):631646.
Radloff, Lenore. 1975. Sex differences in depression. Sex Roles 1(3):249265.
Radloff, Lenore Sawyer. 1980. Depression and the empty nest. Sex Roles 6(6):775
781.
Riolo, Stephanie A., Tuan Anh Nguyen, John F. Greden, and Cheryl A. King. 2005.
Prevalence of Depression by Race/Ethnicity: Findings From the National Health and
Nutrition Examination Survey III. Am J Public Health 95(6):9981000.
Risch, Neil et al. 2009. Interaction Between the Serotonin Transporter Gene (5HTTLPR), Stressful Life Events, and Risk of Depression. JAMA: The Journal of the
American Medical Association 301(23):2462 2471.
94
Rogers, Richard G., Bethany G. Everett, Jarron M. Saint Onge, and Patrick M. Krueger.
2010. Social, Behavioral, and Biological Factors, and Sex Differences in Mortality.
Demography 47(3):555578.
Schnittker, Jason. 2007. Working More and Feeling Better: Womens Health,
Employment, and Family Life, 1974-2004. American Sociological Review 72(2):221
238.
Smith, H.L. 2004. Response: cohort analysis redux. Sociological Methodology
34(1):111119.
Sonnenberg, C. M, A. T. F Beekman, D. J. H Deeg, and W. Tilburg. 2000. Sex
differences in late-life depression. Acta Psychiatrica Scandinavica 101(4):286292.
Turk, Dennis C., and Akiko Okifuji. 1994. Detecting depression in chronic pain
patients: Adequacy of self-reports. Behaviour Research and Therapy 32(1):916.
Van de Velde, Sarah, Piet Bracke, and Katia Levecque. 2010. Gender differences in
depression in 23 European countries. Cross-national variation in the gender gap in
depression. Social Science & Medicine 71(2):305313.
Verbrugge, Lois M. 1985. Gender and Health: An Update on Hypotheses and
Evidence. Journal of Health and Social Behavior 26(3):156182.
Veroff, Joseph, Elizabeth Douvan, and Richard A. Kulka. 1981. The Inner American: A
Self Portrait from 1957 to 1976. New: Basic Books, Inc., Publishers.
Wada, K. et al. 2007. Validity of the Center for Epidemiologic Studies Depression Scale
as a screening instrument of major depressive disorder among Japanese workers.
American journal of industrial medicine 50(1):812.
Watnick, Suzanne, Pei-Li Wang, Theresa Demadura, and Linda Ganzini. 2005.
Validation of 2 Depression Screening Tools in Dialysis Patients. American Journal of
Kidney Diseases 46(5):919924.
Whisman, Mark A., Lauren M. Weinstock, and Tolejko. 2006. Marriage and
Depression. Pp. 219240 in Women and depression: a handbook for the social,
behavioral, and biomedical sciences, edited by Corey L. M. Keyes and Sherryl H.
Goodman. New York: Cambridge University Press.
World Health Organization. 2008. The Global Burden of Disease: 2004 Update.
Switzerland: World Health Organization.
95
Young, E, and A Korszun. 2010. Sex, trauma, stress hormones and depression. Mol
Psychiatry 15(1):2328.
Zonderman, Alan B., Paul T. Costa, and Robert R. McCrae. 1989. Depression as a Risk
for Cancer Morbidity and Mortality in a Nationally Representative Sample. JAMA: The
Journal of the American Medical Association 262(9):1191 1195.
96
CHAPTER 3: Aging, gender and depression

Is aging depressing? Popular belief, as well as a general consensus in the fields of
gerontology and mental health, holds that aging is associated with an increased risk of
depression (Newmann 1989). Most arguments for the positive association between age
and depression pertain to the aging process: aging is usually associated with diminished
physical and cognitive health, often comes with transitions from living with a partner to
being widowed, living alone, and possibly living in a nursing home, retirement and
reductions in economic stability. All of these factors are positively associated with
depression, so it would make sense if aging led to an increase in depression.
In addition to this gross effect of the aging process, aging itself could be
depressing: there could be a net effect of age that remains after adjusting for these other
life transitions. For example, older age could be more depressing because peoples social
networks narrow as their peers die, and limited social support has been shown to be
associated with increased depression among the elderly (La Gory and Fitpatrick 1992).
Another part of the net age effect could include diminished cognitive function and
increased dementia, which are also depressing (Wu, Schimmele, and Chappell 2012).
Additional explanations for the net effect of age on depression may include changes in
living situations, such as being institutionalized (Snowdon 2001; Valvanne et al. 1996),
or lacking transportation (La Gory and Fitpatrick 1992). Depression among the elderly is
also associated with comorbid anxiety (Gottfries 1998). While there are a variety of
reasons for a net age effect, the goal of this paper is not to select among possible
97
mechanisms. Instead, I seek to establish whether there is a direct effect of aging net of
other factors.
However, empirical research on aging and depression is inconclusive to say the
least, and contradictory at worst. A few studies do find that depression increases with
age (Luppa et al.; Sonnenberg et al. 2000), while others show the opposite that
depression actually decreases as people age (Bebbington et al. 1998; Cairney and Wade
2002; Christensen et al. 1999; Jorm 2000; Korten and Henderson 2000; Kroenke and
Spitzer 1998; McGuire et al. 2009). Still other research shows that the relationship
between age and depression is nonlinear, and that it follows either a U-shaped (Kessler et
al. 1992; Mirowsky 1996; Mirowsky and Ross 1992) or inverse U-shaped (Newmann
1989) pattern. Finally, other studies find that there is no real relationship between age
and depression, or that the relationship is too inconsistent to pin down (Djernes 2006;
Glaesmer et al. 2011; van Grootheest et al. 1999).
Research is unclear on the relationship between age and depression for many
reasons, including using small and unrepresentative data sets, operationalizing age in
various ways, considering different age ranges, and measuring depression with diverse
instruments. Even if all of these dissimilarities could be addressed, our understanding of
the relationship between age and depression would remain uncertain for several reasons.
First, the vast majority of research uses cross-sectional data as opposed to longitudinal
data. Cross-sectional data allows for comparisons of different people at different ages,
while longitudinal data can capture the effect of age itself on the same individual over
98
time. This is potentially problematic because some research suggests that there may be
cohort differences in the relationship between age and depression (Bebbington 1996;
Kessler et al. 1992; Luppa et al.; Newmann 1989). Second, research has not been able to
disentangle the effect of the aging process from the potential effect of age per se on
depression. This distinction between the gross effect of the aging process, which
includes the association between age and diminished health, widowhood, and retirement,
among other life changes, and the net effect of age itself on depression, is important for
understanding how, and how much, caretakers can address depression among the elderly.
Much of the research suggests that the increase in depression associated with age is
actually attributable to decreases in health. Some research goes as far as saying that age,
or maturity, actually protects people from depression once controls for health changes
are factored in (Djernes 2006; Mirowsky and Ross 1992; van Grootheest et al. 1999).
However, because these assertions derive from cross-sectional data, they are
inconclusive. We need to examine longitudinal data to fully understand the relationship
between age and depression.
What about the role that gender plays in the relationship between age and
depression? Research on gender and health has consistently found that women are more
likely to be depressed than men. While part of the explanation may be biological,
historical and cultural variation suggests that there is a social component to the sex gap in
depression. Many social characteristics are related to an individuals risk of depression,
including age, race, education, poverty, marital status, and employment status. Part of
99
the sex gap in depression is explained by sex differences in these risk and protective
factors. Whether or not the sex gap in depression remains constant over the lifespan is a
topic of much debate. Like the literature on aging and depression, there are many
theoretical and empirical inconsistencies regarding the relationship between gender, age
and depression.
One hypothesis is that the sex gap in depression is attributable to both the social
and biological implications of the reproductive years. Some research has shown that the
sex gap in depression emerges during young adulthood and disappears after women go
through menopause (Bebbington et al. 1998; Korten and Henderson 2000). Another
hypothesis is that the sex gap in depression increases as people age, because older women
are at a greater risk of becoming widowed and falling into poverty than are older men
(Moen 1996; Luppa et al.). Yet another hypothesis is that the sex gap in depression
should remain constant over the life course, given the evidence that throughout their
lives, women are consistently at a greater risk for depression (Cairney and Wade 2002;
Hopcroft and Bradley 2007; Glaesmer et al. 2011; Kessler et al. 1992; McGuire et al.
2009; Sonnenberg et al. 2000; van Grootheest et al. 1999).
To summarize, research on aging, gender, and depression is contradictory and
leaves us unable to fully understand these complex relationships. Do people become
more or less depressed as they age? What explains the change in depression that occurs
with age? Do changes in marital status, employment, and health explain the brunt of the
change in depression? Or, does age per se have a direct effect on depression? What
100
happens to the sex gap in depression as people pass through their middle and elderly
years? And finally, are there cohort differences in the age trend in depression? We do
not know the answers to these questions because of data and methodological limitations
of previous research. In this paper, I examine the relationship between gender, aging, and
depression using longitudinal and nationally representative data from the Health and
Retirement Study (HRS). In doing so, I find that age per se does not increase depression
until age 75, after which point depression increases for both men and women. While
social and health changes explain the majority of the increase in depression with age, I
find that age itself has a net effect on depression levels. The sex gap in depression
decreases after age 75 because the aging process has a stronger effect on mens
depression than on womens. And finally, I find no support for the claim that the age
trend in depression is due to cohort differences, which has been a major criticism of
cross-sectional research. Instead, the measurement of age itself seems to be an important
underlying explanation for the contradictory research on aging, gender and depression. A
nonlinear measure of age, such as the age spline used in this study, more accurately
represents the trend in depression over the middle to elderly years.
Aging and depression
As previously noted, empirical research on aging and depression is not only inconclusive,
but also contradictory. The following discussion of the literature is summarized in
Appendix 3.1. A few studies find that depression increases with age. For example, a
101
meta-analysis of studies that examine the relationship between age and gender on
depression during latest life (ages 75 and older) found that late-life depression is common
and often increases up to the very oldest ages (Luppa et al.). However, they note that this
increase could be explained by an increase in risk factors and that age itself may not be
the cause of the increase in depression (i.e., that there is no net age effect). In this metaanalysis, age is treated as a categorical variable, using five-year age increments. Another
study, which examined Dutch adults between the ages of 55 and 85, also found an
increase in depression for both sexes over age, using five-year age categories
(Sonnenberg et al. 2000).
However, many more studies find that, contrary to popular belief, depression
actually decreases as people age. For example, in another meta-analysis of the literature,
Jorm (2009) found that aging per se reduces depression. This pattern was not initially
evident, and only emerged after controlling for other risk factors. In other words, while
the gross age effect was unclear, Jorm found a consistent negative relationship between
depression and the net effect of age. In a study of Canadians ages 20 and older, the risk
for major depressive episodes was found to decrease for both sexes over age (Cairney and
Wade 2002). The study dichotomized age at 55; for both sexes, adults over age 55 were
about 50 percent less likely to be depressed than adults under age 55. Another study, of
adults aged 65 and older in the United States, dichotomized age at 75 (McGuire et al.
2009). They found that adults between ages 65 and 74 were more likely to be depressed
than adults aged 75 or older, but did not specifically examine net versus gross age effects.
102
Other studies operationalize age differently. A study using an Australian sample and tenyear age categories found that age is negatively correlated with depression (Christensen
et al. 1999). Another Australian study, which also measured age in ten-year age
categories, also found a decrease in depressive symptoms at older ages (Korten and
Henderson 2000). One criticism of this study is that they are missing elderly people who
are institutionalized, who are most likely more depressed than those who are not
institutionalized (Snowdon 2001).
Another common pattern is that age has a nonlinear association with depression.
For example, in a study of US adults covering the age range 19 to 96, Kessler et al.
(1992) found that depression followed a U-shaped pattern: depression was higher at age
25, lowest at age 50, and then higher around age 75. Mirowsky and Ross (1992) found
similar results in their study of US ages 18 to 90. They argued that the age pattern is not
a perfect U-shaped relationship: young adults are more depressed than middle aged
adults, but the elderly are the most depressed. This study expands upon the analysis of
age and depression by examining the trend in conjunction with other factors that
contribute to depression, and conclude that being older is not in itself depressing.
Still other research finds that there is no real trend in the relationship between age
and depression. For example, in a study of German elderly aged 60 to 85 and five-year
age categories, there was no clear age pattern to depression for either men or women
(Glaesmer et al. 2011). Two other studies, one of Italian elderly and another of Dutch
elderly, found that there was no effect of age on the depression when controlling for other
103
variables (Minicuci et al. 2002; van Grootheest et al. 1999). In a meta-analysis of

research on depression among Caucasian elderly populations, several trends were
observed: some studies showed that age was positively associated with depression, others
that the association was negative, and others found no trend with age (Djernes 2006).
So, what are we to make of the relationship between age and depression?
Inconsistencies in the literature derive from several factors. First, the majority of studies
on aging and depression focus on small community surveys or even clinical samples, so
the findings are not necessarily generalizable to the general population. Second, the age
range considered varies from study to study: some look at a wide range of ages, from 18
to 91 (thus having a small sample of elderly respondents), while others focus on the older
adult population, for example, from ages 55 or 65 to 85. The conclusions we draw
regarding the relationship between age and depression are likely to differ depending on if
aging means comparing young adults to middle aged adults to the elderly, or if aging
refers to passing through the last several decades of life. Third, age is operationalized
differently, and often poorly, in many studies. In my review of the literature, I was
surprised by how many studies treated age as a categorical variable (eight of the studies
cited above) or a binary variable (three studies). Only five of the studies measured age
using age and age-squared.
Putting age aside, a fourth reason for all of the disparities is the measurement of
depression. Depression is often measured in two general ways: either as a summary
score of depression severity, as in surveys like the Center for Epidemiological Studies
104
Depression Scale (CES-D) and the Patient Health Questionnaire-Nine (PHQ-9), or it is

more formally diagnosed in a clinical way, with respect to the criteria specified by the
Diagnostic and Statistical Manual of Mental Disorders (DSM), in surveys such as the
Composite International Diagnostic Interview (CIDI). As it turns out, studies that use the
former depression severity instrument find a very different relationship between
depression and age than do studies that use the diagnostic measure of clinical depression
(Newmann 1989).
In her review of research on depression and aging, Newmann (1989), Newmann
describes two general trends in the relationship between age and depression: a U-shaped
pattern and an inverse U-shaped pattern. She finds that studies that use measures of
depression severity, such as the CES-D, often find a U-shaped relationship between age
and depression. Studies that use diagnostic indicators of clinical depression, on the other
hand, find an inverse U-shaped pattern. The operationalization of depression is important
to studying the relationship between aging and depression. Proponents of depression
severity scores are often interested in social factors associated with differences in
depression between subgroups (Newmann 1989). She argues that the rise in depression
toward the end of life may reflect the types of questions used in the CES-D. The CES-D
incorporates four categories of questions, including depressive affect (feeling sad,
depressed, or blue), somatic symptoms or malaise (lacking energy, could not get going,
sleep problems), social isolation (nobody likes me), and positive affect (enjoying life).
These dimensions may result in a potential age bias: a number of investigators have
105
argued that composite scale scores may be disproportionately inflated among elderly
persons as a consequence of various types of physical malaise that older persons
commonly experience more than their younger counterparts (Newmann 1989:161).
However, several studies have investigated this claim, separating out the depressive
affect and somatic symptoms questions in the CES-D and have concluded that there is no
separate age pattern between these dimensions, and that somatic symptoms do not
explain the trend in the relationship between age and depression that they observe
(Kessler et al. 1992; Mirowsky and Ross 1992). If the disparity in the age pattern of
depression is not attributable to somatic symptoms, what is it attributable to?
As previously noted, studies that use clinical diagnostic indicators of depressive
disorders generally find an inverse U-shaped relationship between age and depression
over the lifespan. In other words, they find that depression decreases during the elderly
years. Newmann (1989) argues that this may be explained by why and how the CIDI is
used to diagnose depression. Proponents of the diagnostic approach are trying to figure
out who needs help from mental health professionals, and what kind of help they need.
As a result, they aim to diagnose people with a specific disorder, such as Major
Depressive Episode (MDE). According to the DSM, before someone is diagnosed with a
MDE, a number of exclusionary criteria have to be ruled out so that the depression is not
due to, for example, a physical illness or condition, medication, drugs or alcohol, or
bereavement. These exclusionary criteria have been called into question because they
may tend to produce underestimates of the prevalence of depression (Goldney, Fisher,
106
and Hawthorne 2004; Slade and Andrews 2002), especially in the elderly population
since they are more likely to attribute their depression to physical illness (Knuper and
Wittchen 1994; Newmann 1989) or bereavement (Corruble et al. 2009). Note that not all
studies that use the CIDI as a measure of depression use these exclusionary criteria.
Aging per se, or poor health associated with aging? Another question that
remains regarding the relationship between aging and depression is does aging per se
have an effect on depression, or does the effect of aging operate through its association
with other life changes? Much of the research suggests that the increase in depression
associated with age is actually attributable to diminished health. Some studies even argue
that there is a protective net effect of age, or maturity, on depression that becomes clear
when you control for health changes (Djernes 2006; Mirowsky and Ross 1992; van
Grootheest et al. 1999). For example, one study concludes that With all the functional
and social statuses adjusted, predicted depression drops throughout the lifetime. The
residual decline in depression suggests an underlying benefit of maturity. (Mirowsky
and Ross 1992:201). However, given that these studies use cross-sectional data, they
may not be able to accurately disentangle the effects of aging and declining health on
depression. Longitudinal data is better suited to answering this question, because it
enables an examination of health changes and aging on depression outcomes.
Gender, aging and depression

What about the role that gender plays in the relationship between age and depression?
Before answering that question, let me first summarize where research stands on gender
107
and depression in general. A predominant and consistent finding is that women are more
likely to be depressed than men. While part of the sex gap explanation may be
biological, historical and cultural variation suggests that there is a social component to
the sex gap in depression. Many social characteristics are related to an individuals risk
of depression, including employment status, marital status, education, and poverty,
among others. Due to the fact that these characteristics are distributed differently
between men and women, part of the sex gap in depression is explained by sex
differences in these factors. Employment is probably the most notable social explanation
for the sex gap in depression. Between the 1950s and 1970s, the increase in womens
labor force participation explained 20 percent of the decreased sex gap in depression that
was observed over that period (Kessler and McRae 1981). Between 1970 and the
present, the additional increase in womens labor force participation explained 100
percent of the diminished sex gap in depression observed over the past 40 years (Medalia
2012). In addition to employment, women are less likely to be married, more likely to
fall into poverty, and experience worse health than men, all risk factors for depression.
Not only are women more likely to experience personal hardships, but they are also more
susceptible to adverse events that occur within their social network, since they often have
more extensive and stronger social ties to their networks (Kessler and McLeod 1984).
Although most would agree that more women are exposed to these risks than are men,
there is disagreement as to whether or not men and women experience these hardships
differently. Some argue that women are doubly disadvantaged when it comes to these
108
undesirable life events: women are simultaneously more exposed to these hardships and
also are more emotionally responsive to them (Kessler and McLeod 1984). However,
other research finds that not being currently married (Sonnenberg et al. 2000; van
Grootheest et al. 1999), or having a lower income (Sonnenberg et al. 2000), have a more
profound effect on mens depression than on womens. Clearly, more research is needed
to sort out the role that social factors play in determining the sex gap in depression.
Research is divided as to whether or not the sex gap in the risk of depression
remains constant over the lifespan. One hypothesis is that the sex gap in depression is
attributable to both the social and biological implications of the reproductive years. In a
study of the British population ages 16 to 64, Bebbington et al. (1998) finds that women
are more likely to be depressed than men before age 55, though there is some variation in
the sex gap. After age 55, however, they find that men are more likely to be depressed
than women. In other words, they find evidence not only for convergence in the sex gap,
but also for a reversal. Additional research has found evidence for a convergence in the
sex gap in depression over the lifespan (Korten and Henderson 2000). Retirement may
also contribute to a decrease in the sex gap in depression as people age. Although it is
changing, historically, men have been more attached to the labor force than women, so
they are at greater risk of retiring than are women. Furthermore, retiring may trigger
more depression in men than in women, since research shows that decreases in income
have a stronger negative effect on mens depression than on womens (Sonnenberg et al.
109
2000). As a result, depression may increase more for men than women after retirement,
leading to a narrowing of the sex gap in depression.
Another hypothesis is that the sex gap in depression increases as people age
(Moen 1996). As people enter their elderly years, they often experience significant life
changes that may put them at greater risk for depression. The elderly are more
susceptible to becoming widowed and to falling into poverty. Gender is integrally
related to these processes. Women are more likely to become widowed as their husbands
die, since male life expectancy is lower than female life expectancy. Elderly women are
also more likely to be in poverty as men, in part because of losing a spouse that may have
supported them, but also because women tend to earn less and work less than men, and
therefore have less savings to support them during old age. Thus, if the risk of depression
does increase toward the end of life, these factors may contribute to a larger increase for
women than for men, leading to further divergence in the sex gap in depression.
Empirical support for this pattern is inconclusive, but in a meta-analysis of research on
aging and depression, the majority of the community studies analyzed found evidence for
an increase in the sex gap in depression as people passed through their elderly years
(Luppa et al.).
Yet another hypothesis is that the sex gap in depression should remain constant
over the life course. While it may be true that women are more at risk for becoming
widowed and falling into poverty at old ages, they are also at a greater risk during their
middle years, since, for example, they are less likely to be employed than men
110
(Mirowsky and Ross 1992). Indeed, several studies find no age trend in the sex gap in
depression (Cairney and Wade 2002; Hopcroft and Bradley 2007; Glaesmer et al. 2011;
Kessler and McRae 1981; Kessler et al. 1992; McGuire et al. 2009; Sonnenberg et al.
2000; van Grootheest et al. 1999). Ultimately, like the research on depression and aging,
it is unclear what the relationship is between sex and depression as people age through
their elderly years.
Potential contamination of age by cohort effects. Some research suggests there is
a need to explore the role that birth cohort may play in the relationship between age and
depression (Bebbington 1996; Kessler et al. 1992; Luppa et al.; Newmann 1989). For
example, in a summary of the literature, Bebbington (1996) finds a consistent decline in
depression with age in a number of studies, suggesting a cohort effect of increasing levels
of, and possibly earlier onset of, depression. Other studies point out that the age effect
they observe could be alternatively attributable to a cohort effect (Jorm 2000; Kessler et
al. 1992), and that longitudinal data could help disentangle the perennial age-periodcohort problem.
Social status, health and depression among older populations. People of all ages
are less likely to be depressed if they are married, employed, and in good health puts.
There is less research on the effects of these social factors on depression specifically
among older adults. However, the findings of this research are summarized here. One
study found that the effect of no longer/not being married (compared to currently being
married), was significantly greater for men than women it increased mens risk of
111
depression by about 300 percent and womens risk by about 200 percent (Sonnenberg et
al. 2000). Similarly, another study found that widowhood had a stronger association with
mens depression than womens (van Grootheest et al. 1999). Having a lower income
increased depression among both sexes (McGuire et al. 2009), but particularly for men
(Sonnenberg et al. 2000). Being employed full-time was less depressing than other
employment statuses, including part-time employment (Mirowsky and Ross 1992).
Unsurprisingly, poor health, which encompasses self-rated health, chronic or physical
illness, functional limitations, cognitive impairments, and disability, increases depression
for both sexes (Djernes 2006; McGuire et al. 2009; Sonnenberg et al. 2000).
Research Questions
In this study, I ask five primary research questions.
1. How does the gross aging process affect depression severity? Is this process the
same for men and women? Based on the literature, there are several possible
relationships. Aging could be associated with both linear and nonlinear increases
and decreases in depression, or there could be no observable trend over age.
2. How do other transitions, like changes in marital status, employment status, and
levels of health, affect depression? Do these factors affect depression similarly
for men and women?
3. Do these factors explain the effect of age on depression? Alternatively, does age
have an effect on depression that is net of these other life changes? The literature
112
suggests that entire relationship between aging and depression may be explained
by the association between age and other life transitions.
4. Are there any significant cohort differences in the age pattern of depression? The
literature suggests that cohort differences may explain cross-sectional age patterns
in depression.
5. Finally, what happens to the sex gap in depression as people age? Does the sex
gap widen, shrink, or remain constant throughout the later years of adult life?
Again, the literature is not clear on what to expect with respect to the relationship
between age, sex, and depression.
Data and Methods

The data for this study derive from the Health and Retirement Study (HRS). The HRS is
a longitudinal and nationally representative panel survey of Americans over the age of
50. The HRS data are often used to examine changes in labor force participation, marital
status, and health changes among individuals ages 50 and older. The longitudinal design
of the survey allows for analysis of the effect of aging on depression; as opposed to
comparing different birth cohorts of individuals at different ages at one point in time, I
follow individuals as they age through time. Among many other questions, the HRS
surveys respondents about their income and work status, a variety of physical, mental and
cognitive health measures. The data used in this study come from the eight survey waves
collected every two years between 1994 (wave 2) and 2008 (wave 9). Wave 1 was
113
excluded because it does not include the variable of primary interest, the CES-D short
form.
Cohorts
The HRS was collected for five birth cohorts, also shown in Table 3.1. The oldest cohort
in the sample is the AHEAD cohort, who was born before 1924, and was originally part
of another study before being subsumed under the HRS. The first wave of data available
for the AHEAD cohort is from wave 3 (1996). The next cohort, CODA (Children of the
Depression), were born between 1924 and 1930, and data is available on this cohort
starting with wave 4 (1998). The HRS (born between 1931 and 1941), has data from the
beginning, in wave 2 (1994). The WB (War Babies) cohort were born between 1942 and
1947, and they were included in the sample beginning with wave 4 (1998). Finally, the
EBB (Early Baby Boomers) cohort was born between 1948 and 1953, and they were
added to the sample more recently, beginning with wave 7 (2004).
Methods
The data used in this study are longitudinal, as there are up to eight observations per
individual. In order to account for the fact that I observe repeated observations on
individuals, and that the error terms are likely to be correlated, I use fixed effects
regression methods. Fixed effects regression includes a dummy variable for each
individual in the sample. As a result, the models estimate only within-individual change
in depression over time. Fixed effects regression requires that all predictor variables
included in the model must also be time-varying. These include factors such as age,
114
change in marital status, employment status, and health. With fixed effects, I am able to
ask, Does aging increase ones level of depression? Do life changes, such as retiring,
becoming widowed, or falling into poor health augment depressive symptoms? Does age
have a net effect on depression after adjusting for changes in marital status, employment
status, and health? In other words, fixed effects methods measure the effect of aging
and other transitions on depression. Because this method only analyzes within-person
variation, fixed effects methods also control for all time invariant characteristics of an
individual, such as their personality, whether or not they experienced depression at a time
before the observed study period, their education, race, and all other stable
characteristics.
Depression
The depression measures on the HRS are a modified version of the Center for
Epidemiological Studies Depression Scale (CES-D) (Radloff 1977). Following up the
earlier discussion of the types of depression instruments used in surveys, the CES-D
measures depression severity and was not designed as a clinical diagnostic tool. The
original CES-D consists of 20 items that ask the respondent how frequently during the
past week he or she experienced a particular feeling or symptom associated with
depression. Respondents indicate how often they experienced a particular feeling in the
previous week, with responses ranging from 0 to 3, where 0 indicates rarely or none of
the time (less than 1 day), 1 is some or a little of the time (1 to 2 days), 2 is occasionally
or a moderate amount of the time (3 to 4 days), and 3 is most or all of the time (5 to 7
115
days). The scale is reliable, and has high internal consistency between the items
(Chronbachs alpha usually around .85) (McDowell 2006). The CES-D is also
considered a valid instrument for determining major depression, and it covers seven out
of nine DSM IV symptoms of a Major Depressive Episode (McDowell 2006).
The modified version of the CES-D used in the HRS includes eight of the original
twenty CES-D questions. Another modification is that the HRS version of the CES-D
worded the questions slightly differently from the original module, asking respondents to
affirm or negate the presence of the feeling during much of the week. This approach to
scoring, called the presence approach, has been determined to perform better than the
alternative persistence approach method of dichotomizing the original CES-D
categories (Perzynski and Townsend 2002). Although the original CES-D is not
available, the modified module has the advantage that it is asked in the same exact way in
every survey wave.
Independent variables
Age. When all survey waves and cohorts are combined, respondents are between age 51
and 115 years old7. However, due to the small sample size of respondents over the age of
95, age is top-coded at 95. Figure 3.1 illustrates the general trend between age and
average depression levels for both sexes. The relationship between age and depression is
nonlinear, but not in the way that previous research has suggested. Instead of a U-shaped
In the regression models, age is divided by ten so that a one-unit increase in the dependent variable is
associated with an additional decade of life.
116
or inverse U-shaped pattern, age has very little variation until age 75, after which
depression increases. To account for this nonlinearity, I use a spline to divide the effects
of age into two groups the middle-old are between 51 and 75 years old, and the
oldest-old, who are 76 years old and older.
Sex. Sex is measured as a dummy variable for males and females. It is included in
the analysis differently in different models. Most of the analysis is run separately by sex;
however, at the end of the paper, sex is interacted with age in order to examine gender
differences in the age pattern of depression.
Other social characteristics. In order to account for the association between
aging and transitions into and out of other social statuses, I control for several individual
characteristics in this study. Because of the nature of the analysis, I can only capture the
effect of variables that change over the period of observation. Therefore, stable
characteristics like race, ethnicity, and education are not included in the analysis.
However, the fixed effects regression controls for these variables by comparing each
individual to him or herself. Since I am interested in sex differences in the age trend in
depression, I analyze most models separately by sex. In addition to age, the social
demographic factors include the following variables: marital status, employment status,
and health. Marital status is broken down into the following categories: currently married
or living with a partner is the reference group, and effects are estimated for never being
married, becoming separated or divorced, and becoming widowed. Employment status
117
includes being employed full time (the reference group), being employed part time, being
partially retired, fully retired, not in the labor force, disabled, and unemployed.
Note that in addition to changes in marital status and employment status affecting
depression, it is the reverse is also possible. Becoming depressed could erode the quality
of ones relationship, leading to divorce; furthermore, evidence suggests that depression
could cause people to withdraw from the labor force or take more sick days even if
theyre working (Lennon 2006). Since I observe respondents every two years, I am
unable to determine whether reverse causation influences my findings. However, most
research concludes that generally, changes in these social statuses are more likely to
affect depression outcomes than the reverse.
Health. Because of the high correlation between depression and physical health, I
will also control for health status. There are several ways to measure health status
available in the HRS, including self-rated health, the number of Activities of Daily
Living (ADLs) and Instrumental Activities of Daily Living (IADLs) the person has some
difficulty with, and the number of chronic health conditions. While self-rated health is
generally a good indicator of overall health status, since it is self-reported, it is strongly
associated with depression. In this study, I measure health in the following three ways.
First, the number of ADLs, including bathing, dressing, eating, getting into bed, and
walking, with which the respondent experiences at least some difficulty (ranges from 0 to
5). Second, the number of IADLs, including using the phone, money, and handling
medications, with which the respondent experiences at least some difficulty (ranges from
118
0 to 3). Third, number of chronic health conditions acquired since the last interview
(ranges from 0 to 4)8.
Findings
Descriptive results
Depression. In order to understand the relationship between aging and depression, I
begin by graphically presenting mean CES-D scores by age and sex in Figure 3.1. This
figure combines all cohorts and waves of data, so the effect of aging is potentially
contaminated by period and cohort effects. In Figure 3.1, there is a clear nonlinear
relationship between age and the mean level of depression for both males and females.
Average depression severity remains level until about age 75 for both sexes. Between
ages 51 and 75, the average depression level for males is 1.2, and 1.8 for females. Since
there is no age trend in depression for either sex, the sex gap in depression for the middleold is constant at around 0.42 points, meaning that females have an average depression
score that is 50 percent higher than males in this age group. After age 75, an age trend
emerges, and there is an increase in depression severity for both sexes for the oldest-old.
Between ages 76 and 95, the average depression score is 1.7 for males and 2.0 for
females. This implies that the rate of increase in the average depression score is faster for
males than females: the rate of increase for males between these two age periods was
In addition to the health outcomes examined in this study, additional health measures may also be
useful to consider in future research. For example, specific types of diseases, such as hypertension, heart
disease, and diabetes, may be particularly likely to affect depression outcomes.
119
40.8 percent and was 20.9 percent for females. At the very end of the lifespan, the
observed data indicates a convergence in the sex gap in depression.
Social and health statuses. Before delving into the analysis, I describe the
breakdown of marital status, employment status, and health, by sex and age group.
Figure 3.2, which presents the proportion of the sample by marital status, shows there is a
much larger decrease in the proportion of married females than males between the
middle-old and oldest-old age groups. For the middle-old, 82 percent of males and 66
percent of females are currently married. For the oldest-old, 68 percent of males and
only 27 percent of females are married. Corresponding to the decrease in the proportion
of married individuals, there is a dramatic increase in the proportion of widows and
widowers between these age groups. For the middle-old, 5 percent of males and 17
percent of females were widowed; for the oldest-old, 24 percent of males and 63 percent
of females are widowed on average. This trend is probably a product of two factors.
First, women are often married to men who are older then themselves. Second, life
expectancy is longer for women than men. The other marital statuses included in Figure
3.6, divorced or separated and never married, do not change as dramatically between the
age groups.
Figure 3.3, which depicts employment status differences by sex and age group,
shows that there are also many changes in employment status between the 51 to 75 year
olds and 76 to 95 year olds. The percentage of full time employed people decreases from
38 percent for males and 28 percent for females to 3 percent and 1 percent respectively.
120
The percentage of part-time workers also decreases between these age groups, from 4
percent (males) and 9 percent (females) to 1 percent for both sexes. In terms of
retirement, the percentage of both sexes who are partially retired decreases between the
age groups while the percentage of fully retired males and females increases. For the
middle-old, 42 percent of males and 36 percent of females were retired. For the oldestold, 88 percent of males and 71 percent of females were fully retired. The percentage of
unemployed individuals was only one percent among the middle-old, while no one in the
oldest-old group reported that they are unemployed. This makes sense given the national
retirement age of 65. The two remaining categories reflect not being in the labor force
due to disability and other reasons. The percentage of disabled, not working individuals
is low for both sexes and age groups. An interesting gender pattern is observed for not in
the labor force due to other reasons. While only 1 percent of males are in this category at
both age groups, 15 percent of middle-old females and 23 percent of oldest-old females
respond that they are not in the labor force for other reasons. This is a subsuming
category, but the gender disparity probably reflects the fact that women are more likely to
consider themselves to be homemakers than are men.
Finally, Figure 3.4 illustrates age trends in health by sex. Females are more likely
than males to have at least some difficulty with more ADLs and IADLs than males at
both points in time. Not surprisingly, the percentage of both sexes who report difficulty
with these tasks increases as people age. However, according to the data, there is little to
121
no gender gap in the number of new conditions reported, although this number also
increases with age.
Question 1
How does the aging process affect depression? Is the gross relationship between aging
and depression the same for males and females? To answer these questions, fixed
effects regression of depression scores on age are modeled separately by sex. This model
includes no additional control variables; therefore, it describes the gross association
between age and depression. As a reminder, the effect of age is modeled separately for
the two age groups, the middle-old (ages 51 to 75), and the oldest-old (ages 76 to 95).
Results are included in Table 3.2 (Baseline Model), as well as depicted in Figure 3.5. As
is shown in the figure, age has a slight positive effect on depression for middle-old males
and females9. Between ages 51 and 75, depression increases by 0.09 points per decade
for males, which amounts to 0.23 points over the 25 year age range. For females,
depression increases by 0.11 points per decade, which amounts to 0.28 points over the 25
year age range. In terms of relative difference, this change is associated with a 10.2
percent increase in depression for males and 11.3 percent increase for females between
9
In order to plot the predicted depression scores by age, I need an intercept. However, SAS, the
program used to analyze the data in this paper, does not compute an intercept for fixed effects regression
models. Other statistical packages, including STATA, do compute this statistic; the intercept is adopted
from the random effects regression model. I follow this method and borrow the intercept for this figure
from a random effects regression model where there are no predictor variables (null model).
122
age 51 and 75. Since the gross effect of age on depression is almost the same for both
sexes, the sex gap remains constant for the middle-old, where females have higher
depression scores than males.
Comparing the two age groups, Figure 3.5 shows that the gross association
between age and depression is stronger for the oldest-old than for the middle-old.
According to the baseline model in Table 3.2, per decade, age is associated with a 0.68
point increase in depression among oldest-old males and a 0.43 point increase in
depression among oldest-old females. Between the ages of 76 and 95, depression
increases by 1.36 points for males and 0.86 points for females. This absolute gross
change corresponds to a 56.9 percent increase among oldest-old males and a 39.2 percent
increase for females. Therefore, the gross effect of age on depression among the oldestold is greater for males than for females. While the sample size at these oldest ages is
relatively smaller, especially for men, these findings indicate that toward the very end of
the life span, males experience higher depression levels than females.
Question 2
How do other factors affect depression? Is this the same for both sexes? Table 3.2
includes two models which show the results of the fixed effects regressions of CES-D on
age (baseline) and all time-varying covariates together. As a reminder, the full model
includes age (for both age groups), marital status, employment status, and health. In
models not shown, the effects of changes in marital status, employment status, and health
are very similar between models where the variables are entered separately or
123
simultaneously. I will now describe the effects of changes in social and health statuses
on depression, reserving the discussion net age effects for the next results section.
Marital status transitions have a slightly larger effect on depression severity for males
than for females. Becoming divorced increases depression scores by 0.57 points for
males and by 0.46 points for females. Widowhood increases depression by 0.72 points
for males and 0.53 points for females. Therefore, being married is the most protective
status for both sexes, and widowhood is slightly more depressing than divorce. A sex
difference exists when considering the never married category, which has no effect on
males depression but increases females depression by 0.60 points. While significant,
the effect of being never married should be viewed critically, since a very small
percentage of both sexes say they are never married just 3.0 percent for each sex.
In terms of employment status, there is no significant difference in depression for
either sex between being employed full time (the reference category) and two other
statuses: being employed part time and being partially retired. However, when
comparing full time employment to the remaining employment statuses, the protective
effect of working becomes clear. Fully retiring increases depression by 0.13 points for
both males and females; being unemployed increases depression by 0.40 points for males
and 0.34 points for females; not working because of disability increases depression by
0.33 points for males and 0.47 points for females; and finally, not being in the labor force
for other reasons increases depression by 0.30 points for males and 0.18 points for
females. As previously mentioned, the association between employment and depression
124
could potentially be due to reverse causation, which cannot be differentiated in this study.
However, taken together, fully retiring and unemployment affect male and female
depression levels about the same, while not working due to disability affects females
more than males and not working due to other reasons affects males more than females.
In other words, there is heterogeneity in the impact of employment status transitions on
depression for males and females.
In terms of health, it is not surprising that any health problems, whether it be
experiencing some difficulty with ADLs, IADLs, or having an increasing number of new
conditions, is more depressing than being in better health. As a reminder, in this dataset,
ADLs range from 0 to 5; IADLs range from 0 to 3; and the number of new health
conditions since the last interview ranges from 0 to 4. For males, having at least some
difficulty with each additional ADL increases depression by 0.34 points, meaning that
having some difficulty with 3 ADLs increases depression by 1.02 points. For females,
each additional ADL increases depression by 0.28 points, and having difficulty with 3
ADLs increases depression by 0.84 points. Comparing the magnitude of the effects of
having difficulties with ADLs to IADLs on depression, IADLs increase depression less
than ADLs do for both sexes. Each additional IADL that the respondent says they
experience at least some difficulty with increases depression by 0.18 points for males and
0.15 points for females. The effect of having new health conditions on depression is even
smaller than the effect of IADLs: each additional new health condition increases
depression by 0.09 points for males and 0.11 points for females. In sum, ADLs lead to
125
larger increases in depression than do IADLs or health conditions, and there is a slightly
stronger relationship between health changes overall and male depression than with
female depression.
Question 3
Do these factors explain the relationship between age and depression? Is there an effect
of age on depression net of these other life transitions? As previously noted, many
studies have hypothesized that depression should increase as people age, since aging is
associated with diminished physical and cognitive health, becoming widowed, and no
longer working, all factors associated with an increase in depression. Does the
relationship between age and depression operate through these other pathways, or is there
also a direct (net) effect of age on depression? In other words, how much of aging is due
to age per se and not attributable to the correlation between age and other factors.
The results for the middle-old are shown in Table 3.2. In the baseline model,
there is a very small positive gross effect of age on depression for males and females
among the middle-old. When all variables are adjusted for in the full model, the net
effect of age on depression becomes statistically insignificant for middle-old males.
Therefore, there appears to be no net effect of age on depression among middle-old
males. For middle-old females, on the other hand, adjusting for changes in social statuses
and health contributes to a reversal in the effect of age. While the gross effect of age was
positive, the net effect of age becomes slightly negative. This indicates that there may be
126
a slight protective net effect of age on depression among middle-old females after
controlling for life transitions.
Although the gross age effect was very small for the middle-old, the gross effect
of age on depression is larger for the oldest-old. As indicated in Figure 3.5 and the
baseline model from Table 3.2, age has a positive gross association with depression for
both sexes among the oldest-old. The full model in Table 3.2 shows that the net effect of
age on depression remains significant after controlling for all covariates. How much of
the gross age effect is explained by these life transitions? How much of the net age effect
remains? To answer these questions, I compare age coefficients from the full model to
the baseline model, and calculate the percentage of the gross age effect that is explained
by adjusting for marital status changes, employment status changes, and health changes
(see Equation 3.1).
Eq.3. 1
( D 95B D 76B ) ( D 95F D 76F )

D B D B
95
76
When all variables are included in the model simultaneously, the association
between age and depression is reduced by 62.5 percent for males and 80 percent for
females. This means that 37.5 percent of the net age effect for males and 20 percent of
the net age effect on depression for females remains unexplained. Therefore, age has a
net effect on depression that is independent of changes in marital status, employment
status, and health. For males, age increases depression by 0.03 points for every
additional year after age 75, implying that aging 10 years leads to a predicted increase in
127
depression score of 0.30 points. For females older than age 75, depression increases by
0.02 points for every additional year, equivalent to 0.20 points over 10 years. While
these effects are not very large, they do over the long run lead to increases in depression
that are commensurate with or greater than the effects of changes in marital status,
employment status, and health.
In this study, I examine both the gross effect of age, which includes changes in
marital status, employment status, and health, as well as the net effect of age, which is the
effect of age that remains after controlling for these other changes. What factors could be
included in this net effect of age on depression? One possible explanation is that people
could become more depressed as they age because theyre friends are dying around them,
and they feel increasingly alone. Decreases in cognitive function may also be associated
with both age and depression. Moving into a nursing home or into assisted living may
also be related to both age and depression. Another possibility is that proximity to death
may be associated with both age and depression. In additional models, not shown, I also
adjust for mortality by controlling for the wave before the respondent dies. In these
analyses, the net age effect actually becomes slightly stronger for both males and females
(and for both age groups), while part of the effect of health on depression is absorbed by
proximate mortality.
Question 4
Do cohort differences explain the age pattern in depression? Some research indicates
that there may be cohort differences in depression levels, and argues that the relationship
128
between age and depression cannot be distinguished from potential cohort effects using
cross-sectional data. Since this study uses longitudinal data, it is better suited to
addressing this problem. The data presented in Figures 6 and 7 show the observed
average levels of depression by age and cohort for males and females, respectively. The
pattern observed does not vary significantly by cohort. At a given age, cohort variation
in the level of depression is small and not statistically significant. Therefore, I find no
support for the claim that cohort differences in depression are driving the association
between age and depression. Furthermore, there is no evidence of a cohort*age
interaction on depression, unlike previous research has suggested.
Question 5
What explains the sex difference in the age trend in depression at the oldest ages?
Because the sex gap in depression remained constant over age for the middle-old, here, I
focus on the convergence and reversal in the sex gap in depression among the oldest-old.
In order to answer this question, males and females are examined together. In addition to
age, an interaction between age and sex is included in the fixed effects regression to
capture the change in the sex gap in depression as people age. Note that fixed effects
regression prevents the inclusion of sex itself in the model since it is a stable
characteristic; however, the interaction between a stable characteristic (sex) and time
variant characteristic (age) is possible to model. Table 3.3 shows the effect of age on
depression for males, the effect of age on depression for females, and the sex gap in the
effect of age on depression, where female scores are subtracted from male scores. The
129
results are presented for five models, one that includes only age and the age*sex
interaction, while subsequent models additionally control for marital status, employment
status, health, and all variables combined. In this way, I am able to determine if the sex
gap in the age trend in depression is attributable to other factors. In the first model, only
age and the age*sex interaction are included. Each additional decade after age 75
contributes to a 0.82 point increase in depression scores for males and a 0.51 point
increase for women. The sex gap in the gross effect of age on depression is -0.31 points
per decade. Although the sex gap in the effect of age on depression is very small, it is
statistically significant. In the second model, marital status is also included. While
marital status explains some of the effect of age on depression for both sexes, it does not
mediate the coefficient for the sex gap in the age effect, which remains almost constant at
-0.29 points. The same is true for employment status, health, and the final model where
all variables are combined. The factors which explain some of the gross age effect on
depression for males and females do not explain the narrowing or reversal of the age gap
in depression among the oldest-old.
Discussion
This study uses longitudinal data to examine the relationship between aging, gender, and
depression among a nationally representative population of US adults over the age of 50.
Given data and methodological limitations of previous research, it was unclear what I
would find. The first conclusion this study demonstrates is that the relationship between
the gross effect of the aging process and depression depends on how old someone is
130
specifically, it is conditional on being younger or older than age 75. For adults between
the ages of 51 and 75, age has very small positive association with depression. After age
75, the association between age and depression becomes much steeper. Therefore, there
is a nonlinear relationship between age and depression. This finding is most similar to
the conclusions reached by a small number of previous studies (Kessler et al. 1992;
Mirowsky 1996; Mirowsky and Ross 1992; Newmann 1989). While the positive
relationship between age and depression has been previously documented (Luppa et al.;
Sonnenberg et al. 2000), the story is more complex than a straightforward linear
relationship. The results of this study contradict the conclusions reached by studies
which found that depression decreased with age, even among the elderly (Cairney and
Wade 2002; Christensen et al. 1999; Kroenke and Spitzer 1998; McGuire et al. 2009),
and those that found no relationship between age and depression (Glaesmer et al. 2011;
van Grootheest et al. 1999).
In general, women have higher levels of depression than men. However, the sex
gap in depression narrows, and eventually reverses, among adults over the age of 75. In
my review of the literature, only one other study comes to the conclusion that the sex gap
in depression should decrease over age (Korten and Henderson 2000). I conclude that the
gross association of age and depression is stronger among elderly men than it is among
elderly women. These results do not support the findings reached by studies which found
no age trend in the sex gap in depression (Cairney and Wade 2002; Hopcroft and Bradley
2007; Glaesmer et al. 2011; Kessler and McRae 1981; Kessler et al. 1992; McGuire et al.
131
2009; Sonnenberg et al. 2000; van Grootheest et al. 1999). This study also does not
support the hypothesis that the sex gap in depression is a product of the biological and
social implications of the reproductive years (Bebbington et al. 1998). I find that the sex
gap in depression decreases until well after menopause, and not around age 55. Finally,
these results contradict the findings from studies that suggest that the sex gap in
depression increases as people age (Luppa et al.; Moen 1996).
As people age, they may stop working, become divorced or widowed, and may
experience diminished physical health. These transitions, which are simultaneously
associated with aging and an increase in depression, completely explain the slight
increase in depression that occurs as men age from 51 to 75. In fact, after controlling for
these life changes, there appears to be a slight negative association between age per se
and depression for women. Like previous research has suggested, there may be a
beneficial effect of maturity in preventing depression (Djernes 2006; Mirowsky and
Ross 1992; van Grootheest et al. 1999). However, the maturity benefit only goes so far.
After age 75, there is a negative net effect of age on depression that remains for both
sexes after adjusting for transitions in marital status, employment status, and health. The
most important factor explaining the age effect among the elderly is health, which alone
explains about half of the gross relationship between age and depression.
However, it should be taken into consideration that my focus on marital status,
employment status, and health changes does not exhaust the possible changes that occur
as people age. Other factors are not examined, such as how many friends are dying,
132
hospitalizations, and relocation to nursing homes. These factors are likely to exacerbate
ones risk of depression, and may help to explain more of the effect of age on depression
among. It is probable that the more factors you can control for, the more of the age effect
that is explained. This is interesting, but somewhat beside the point. The fact is that
changes in marital status, employment status, health, social ties, living situation, almost
inevitably change as people age. They are in fact part of the aging process. The effect of
aging on depression may be mitigated or postponed as people live longer and healthier,
but it seems unlikely that the risk for depression will ever disappear completely for the
very old. As a result, caretakers and healthcare providers should emphasize the
recognition and treatment of depression in elderly loved ones and patients.
This study suggests that aging is more depressing for elderly men than for elderly
women. Although women are on average more depressed than men for most of the
lifespan, there may be a reversal in the sex gap among those over age 90. Why would
men take aging harder than women? It is possible that men have a latent tendency toward
depression that is suppressed by the protective influence of marriage, employment, and
good health. When these benefits go away, they become more depressed. However, the
negative effect of age on depression is not completely explained by these other life
transitions. Almost 40 percent of the age effect on depression remains after adjusting for
these factors. Due to small sample sizes of the very old, we cannot be completely
confident in this estimate. Future research should examine depression among
133
nonagenarians and centenarians. However, it appears as though for both sexes, but
particularly for men, it is important to monitor depression levels as people age.
The vast majority of research on aging and depression uses cross-sectional data.
As a result, these studies compare depression among different cohorts, as opposed to
observing the relationship between aging and depression for the same individuals.
Previous research has suggested cross-sectional methods are not able to disentangle
potential cohort differences in depression from the effects of age on depression. In this
study, I use longitudinal data, following several cohorts of people as they age. This
method allows me to measure the relationship between aging and depression for
individuals. To further investigate the role that birth cohort may play, I separately
examined the relationship between age and depression by cohort. I found that there were
no differences in the age pattern of depression by cohort. As more data is collected by
the Health and Retirement Study, it will be possible to follow more adults as they age
into their elderly years. Future research should use this data to provide more information
about potential cohort effects. However, the results from this study do not reaffirm the
concern with cross-sectional studies that cohort effects contaminate the effect of age on
depression. Instead, it appears as though the measurement of age is more important for
the different conclusions reached by previous research with respect to the effect of age on
depression.
One limitation of this study is that it only uses one measure of depression the
short form CES-D. Other measures of depression could be used, including more
134
diagnostic indicators such as the CIDI. It may be the case that the patterns observed here
for levels of depression do not correspond to experiences of major depressive episodes.
Additionally, this paper focuses on the population of adults ages 51 and older, which
leaves questions regarding the relationship between age and depression at younger ages.
Longitudinal data for younger adults would be helpful.
Ultimately, this study illustrates the importance of the measurement of age when
studying depression among older adults. The nonlinear association between age and
depression needs to be taken into account in future research.
135
Tables and Figures

Table 3.1. Birth years and ages by cohort and survey wave
AHEAD
Wave
2
3
4
5
6
7
8
9
Birth years
Survey
Year
1994
1996
1998
2000
2002
2004
2006
2008
<1924
72-95*
75-95*
77-95*
79-95*
81-95*
83-95*
85-95*
CODA
19241930
HRS
19311941
WB
19421947
EBB
19471953
68-74
70-76
72-78
74-80
76-82
78-84
53-63
55-65
57-67
59-69
61-71
63-73
65-75
67-77
51-56
53-58
55-60
57-62
59-64
61-66
51-56
53-58
55-60
*Note: due to the small sample size, respondents over the age of 95 are treated as being
95 years old.
Source: Health and Retirement Survey
Figure 3.1. Average CES-D scores by age and sex (all cohorts and waves)
2.5
2.3
Mean CES-D score
2.1
1.9
1.7
1.5
Male
1.3
Female
1.1
0.9
0.7
0.5
51
56
61
66
71
76
Age

136
81
86
91
Figure 3.2. Marital status by sex and age group

0.90
0.80
Proportion
0.70
0.60
51-75 Males
0.50
51-75 Females
0.40
76-95 Males
0.30
76-95 Females
0.20
0.10
0.00
Married
Separated or Widowed
divorced
Never
married
Proportion
Figure 3.3. Employment status by sex and age group

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00
51-75 Males
51-75 Females
76-95 Males
76-95 Females
137
Figure 3.4. Health by sex and age group

0.80
0.70
Mean
0.60
0.50
51-75 Males
0.40
51-75 Females
76-95 Males
0.30
76-95 Females
0.20
0.10
0.00
ADLs (0-5)
IADLs (0-3)
Conditions (0-4)

Figure 3.5. Predicted CES-D score by sex and age, from fixed effects regression
4
Predicted depression
3.5
3
2.5
Males
Females
1.5
1
0.5
51
56
61
66
71
76
Age
81
138
86
91
Table 3.2. Fixed effects regression by sex

Baseline: age only, no
Full model: all variables
controls
Male
Female
Male
Female
B SE Sig B SE Sig B
SE Sig B
SE Sig
Age, in decades
Age (51-75)
0.09 (0.02) *** 0.11 (0.01) *** -0.03 (0.02)
-0.06 (0.02) ***
Age (76-95)
0.68 (0.04) *** 0.43 (0.03) *** 0.33 (0.04) *** 0.20 (0.04) ***
Marital status (ref=married)
Divorced
0.57 (0.04) *** 0.46 (0.04) ***
Widowed
0.72 (0.04) *** 0.53 (0.03) ***
Never married
0.08 (0.15)
0.60 (0.11) ***
Employment status (ref=employed full time)
Employed part time
0.01 (0.04)
0.01 (0.03)
Partly retired
-0.02 (0.03)
-0.05 (0.03)
Retired
0.13 (0.03) *** 0.13 (0.03) ***
Unemployed
0.4 (0.07) *** 0.34 (0.07) ***
Disabled
0.33 (0.06) *** 0.47 (0.05) ***
Not in LF
0.3 (0.09) ** 0.18 (0.03) ***
Health
ADL
0.34 (0.01) *** 0.28 (0.01) ***
IADL
0.18 (0.02) *** 0.15 (0.02) ***
Number of conditions
0.09 (0.01) *** 0.11 (0.01) ***
Note: *=p<.05, **=p<.01, ***=p<.001
139
Figure 3.6. Average CES-D by cohort Males

2.5
2.3
2.1
1.9
AHEAD: <1924
1.7
CODA: 1924-1930
1.5
HRS: 1931-1941
1.3
WB: 1942-1947
1.1
EBB: 1948-1953
0.9
0.7
0.5
51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95

Figure 3.7. Average CES-D by cohort Females
2.5
2.3
2.1
1.9
AHEAD: <1924
1.7
CODA: 1924-1930
1.5
HRS: 1931-1941
1.3
WB: 1942-1947
1.1
EBB: 1948-1953
0.9
0.7
0.5
51 53 55 57 59 61 63 65 67 69 71 73 75 77 79 81 83 85 87 89 91 93 95
140
Table 3.3. Age*Sex interactions

Age
Age, in decades (males)
Age*female
Divorced
Widowed
Never married
Employed part time
Partly retired
Retired
Unemployed
Disabled
Not in LF
ADL
IADL
# of conditions
Age, in decades (females)
Marital Status
Employment Status
Health
All variables
SE
Sig
SE
Sig
SE
Sig
SE
Sig
SE
Sig
0.82
-0.31
(0.05)
(0.06)
***
***
0.67
-0.29
0.98
0.58
1.10
(0.05)
(0.06)
(0.08)
(0.05)
(0.24)
***
***
***
***
***
0.78
-0.28
(0.05)
(0.06)
***
***
0.58
-0.3
(0.06)
(0.07)
***
***
(0.15)
(0.13)
(0.12)
(0.60)
(0.18)
(0.13)
(0.06)
(0.07)
(0.08)
(0.05)
(0.25)
(0.17)
(0.14)
(0.13)
(0.59)
(0.18)
(0.14)
(0.02)
(0.03)
(0.02)
***
***
***
***
***
0.04
0.18
0.31
0.77
0.36
0.37
0.41
-0.27
0.91
0.56
1.14
-0.07
0.16
0.31
0.70
0.35
0.41
0.26
0.06
0.08
*
*
**
0.26
0.06
0.08
0.51
0.38
0.50
Note: *=p<.05, **=p<.01, ***=p<.001

141
0.28
(0.02)
(0.03)
(0.02)
***
*
***
0.14
*
*
**
***
*
***
Appendix 3.1 Summary of literature on age, gender and depression
Paper
Type
Cou Natl
ntry rep? Depvar
Bebbingtonetal.(1998)
XS1
UK
CairneyandWade(2002)
Christensenetal.(1999)
XS
XS
Djernes(2006)
MA2 +4
HopcroftandBradley(2007)
Glaesmeretal.(2011)
Jorm(2000)
Kessleretal.(1992)
CA
AU
XS
XS
DE
MA
XS
+
US
KortenandHenderson(2000) XS
AU
CISR6
No
Yes
No
CIDISF
+
No
Yes
CESD
Yes
11
10
CIDI
US
+
No
No
Auth.
+
McGuireetal.(2009)
XS
US
Yes
PHQ813
Mirowsky(1996)
MirowskyandRoss(1992)
XS
XS
US
US
Y/N
Y/N
Gross
Gross
N/A
N/A25
N/A
+
15
A+Asq
None
20
26
N/A
21
27
N/A
1996 A+Asq
N/A
Gross
Converge
1891 10yrcat
75
5yrcat
N/A
N/A
N/A
75,<75
N/A
N/A
N/A
N/A
N/A
None
10yrcat
16
14
19
65
CESD
CESDSF
18
3065 +
18+
12
XS
MA
Reversal
6085 5yrcat
KroenkeandSpitzer(1998)
Luppaetal(Forthcoming)
Gross24
18
PHQ9
No
1664 55,<55 (F)17
60
1Q
Yes
Netage Agetrend
trend
insexgap?
15
55,<55
1879 10yrcat
Yes
Gross
age
Age Age
range measure trend
1998 Log
1890 Log
U
22
Newmann(1989)
Sonnenbergetal.(2000)
MA
XS
US
NL
No
Yes
+
CESD
+
+
5585 5yrcat
vanGrootheestetal.(1999)
XS
NL
No
CESD
5585 10yrcat ns23
See notes (next page)

142
U,
Notes: 1 Cross-sectional; 2 Meta analysis; 3 Cross-sectional plus one longitudinal followup; 4 Indicates many or various outcomes, measures, etc.; 5 Indicates both because several
surveys were used; 6 CIS-R: Revised Clinical Interview Schedule; 7 CIDI-SF: Composite
International Diagnostic Interview Short Form; 8 Did you feel depressed or very
unhappy?; 9 PHQ-9: Patient Health Questionnaire-9; 10 CES-D Center for
Epidemiological Studies Depression Scale; 11 CIDI: Composite International Diagnostic
Interview; 12 Author's own patient questionnaire; 13 Patient Health Questionnaire-8; 14
CES-D Short form; 15 Age and age-squared; 16 Falling component (ln(age-17) and rising
component (age-18); 17 Decrease, only for females; 18 Decrease; 19 U-shaped curve; 20 No
trend; 21 Inconsistent trends across several studies; 22 Two trends: a U-shaped trend (CESD), and inverse U-shaped trend (diagnostic interviews); 23 Decrease, but not significant; 24
Net effect of age is same/similar to gross effect of age; 25 Not available/tested; 26 No net
effect of age; 27 Net age effect remains but is smaller than gross effect.
143
References
Bebbington, P.E. et al. 1998. The influence of age and sex on the prevalence of
depressive conditions: report from the National Survey of Psychiatric Morbidity.
Bebbington, P.E. 1996. The origins of sex differences in depressive disorder: Bridging
the gap. International Review of Psychiatry 8(4):295.
Cairney, John, and Terrance J. Wade. 2002. The influence of age on gender differences
in depression. Social Psychiatry and Psychiatric Epidemiology 37(9):401408.
Christensen, H. et al. 1999. Age differences in depression and anxiety symptoms: a
structural equation modelling analysis of data from a general population sample.
Corruble, Emmanuelle, Virginie-Anne Chouinard, Alexia Letierce, Philip A P M
Gorwood, and Guy Chouinard. 2009. Is DSM-IV bereavement exclusion for major
depressive episode relevant to severity and pattern of symptoms? A case-control, crosssectional study. The Journal of Clinical Psychiatry 70(8):10911097.
Djernes, JK. 2006. Prevalence and predictors of depression in populations of elderly: a
review. Acta Psychiatrica Scandinavica 113(5):372387.
Glaesmer, H., S. Riedel-Heller, E. Braehler, L. Spangenberg, and M. Luppa. 2011. Ageand Gender-Specific Prevalence and Risk Factors for Depressive Symptoms in the
Elderly: A Population-Based Study. International Psychogeriatrics 23(08):12941300.
Goldney, R. D., L. J. Fisher, and G. Hawthorne. 2004. Depression and the CIDI. The
British Journal of Psychiatry 185(6):518 519.
La Gory, Mark, and Kevin Fitpatrick. 1992. The Effects of Environmental Context on
Elderly Depression. Journal of Aging and Health 4(4):459479.
Gottfries, C. G. 1998. Is there a difference between elderly and younger patients with
regard to the symptomatology and aetiology of depression? International Clinical
Psychopharmacology 13(5):S13.
van Grootheest, D. S., A.T.F. Beekman, M. I. Broese van Groenou, and D. J. H. Deeg.
1999. Sex differences in depression after widowhood. Do men suffer more? Social
Psychiatry and Psychiatric Epidemiology 34(7):391398.
144
Hopcroft, Rosemary L., and Dana Burr Bradley. 2007. The Sex Difference in
Depression Across 29 Countries. Social Forces 85(4):1483 1507.
Jorm, A. F. 2000. Does Old Age Reduce the Risk of Anxiety and Depression? A Review
of Epidemiological Studies Across the Adult Life Span. Psychological Medicine
30(01):1122.
Kessler, Ronald C., Cindy Foster, Pamela S. Webster, and James S. House. 1992. The
Relationship Between Age and Depressive Symptoms in Two National Surveys.
Psychology and Aging 7(1):119126.
Kessler, Ronald C., and Jane D. McLeod. 1984. Sex Differences in Vulnerability to
Undesirable Life Events. American Sociological Review 49(5):620631.
Kessler, Ronald C., and James A. McRae. 1981. Trends in the Relationship Between
Sex and Psychological Distress: 1957- 1976. American Sociological Review 46(4):443
452.
Knuper, Brbel, and Hans-Ulrich Wittchen. 1994. Diagnosing major depression in the
elderly: Evidence for response bias in standardized diagnostic interviews? Journal of
Psychiatric Research 28(2):147164.
Korten, Ailsa, and Scott Henderson. 2000. The Australian National Survey of Mental
Health and Well-Being. The British Journal of Psychiatry 177(4):325 330.
Kroenke, K, and RL Spitzer. 1998. Gender differences in the reporting of physical and
somatoform symptoms. Psychosomatic Medicine 60(2):150 155.
Lennon, Mary Clare. 2006. Women, Work, and Depression: Conceptual and Policy
Issues. Pp. 309327 in Women and depression: a handbook for the social, behavioral,
and biomedical sciences, edited by Corey L. M. Keyes and Sherryl H. Goodman. New
York: Cambridge University Press.
Luppa, M. et al. Forthcoming. Age- and gender-specific prevalence of depression in
latest-life Systematic review and meta-analysis. Journal of Affective Disorders.
McDowell, Ian. 2006. Measuring health: a guide to rating scales and questionnaires.
New York: Oxford University Press US.
McGuire, L.C., T.W. Strine, R.S. Allen, L.A. Anderson, and A.H. Mokdad. 2009. The
patient health questionnaire 8: current depressive symptoms among US older adults, 2006
145
Behavioral Risk Factor Surveillance System. American Journal of Geriatric Psych

17(4):324.
Medalia, Carla. 2012. Historical Changes in the Gender Gap in Depression in the United
States, 1974 to 2008. Dissertation, Philadelphia, PA: University of Pennsylvania.
Minicuci, Nadia, Stefania Maggi, Mara Pavan, Giuliano Enzi, and Gaetano Crepaldi.
2002. Prevalence Rate and Correlates of Depressive Symptoms in Older Individuals.
The Journals of Gerontology Series A: Biological Sciences and Medical Sciences
57(3):155 161.
Mirowsky, John. 1996. Age and the Gender Gap in Depression. Journal of Health and
Social Behavior 37(4):362380.
Mirowsky, John, and Catherine E. Ross. 1992. Age and Depression. Journal of Health
and Social Behavior 33(3):187205.
Moen, Phyllis. 1996. A life course perspective on retirement, gender, and well-being.
Journal of Occupational Health Psychology;Journal of Occupational Health Psychology
1(2):131144.
Newmann, Joy P. 1989. Aging and depression. Psychology and Aging 4:150165.
Radloff, Lenore Sawyer. 1977. The CES-D Scale. Applied Psychological Measurement
1(3):385 401.
Slade, T., and G. Andrews. 2002. Exclusion criteria in the diagnostic classifications of
DSM-IV and ICD-10: revisiting the co-occurrence of psychiatric syndromes.
Psychological medicine 32(7):12031211.
Snowdon, J. 2001. Prevalence of depression in old age. The British Journal of
Psychiatry 178(5):476 477.
Sonnenberg, C. M, A. T. F Beekman, D. J. H Deeg, and W. Tilburg. 2000. Sex
differences in late-life depression. Acta Psychiatrica Scandinavica 101(4):286292.
Valvanne, Jaakko, Kati Juva, Timo Erkinjuntti, and Reijo Tilvis. 1996. Major
Depression in the Elderly: A Population Study in Helsinki. International
Psychogeriatrics 8(03):437443.
Wu, Zheng, Christoph M Schimmele, and Neena L Chappell. 2012. Aging and Late-Life
Depression. Journal of Aging and Health 24(1):328.
146

Essays On Gender and Health

Uploaded by

Copyright:

Available Formats

Essays On Gender and Health

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Essays On Gender and Health

Uploaded by

Copyright:

Available Formats

University of Pennsylvania

Essays on Gender and Health

Follow this and additional works at: http://repository.upenn.edu/edissertations

This paper is posted at ScholarlyCommons. http://repository.upenn.edu/edissertations/546

Essays on Gender and Health

Doctor of Philosophy (PhD)

depression, gender equality, life expectancy

Demography, Population, and Ecology | Sociology

This dissertation is available at ScholarlyCommons: http://repository.upenn.edu/edissertations/546

ESSAYS ON GENDER AND HEALTH

This dissertation is dedicated to my Nana,

ESSAYS ON GENDER AND HEALTH

CHAPTER 1: Gender equality, development, and cross-national sex gaps in

Data and Methods ............................................................................................................................. 11

CHAPTER 2: Historical changes in the gender gap in depression in the

Survey Waves and Depression Outcomes ......................................................................................... 54

CHAPTER 3: Aging, gender and depression ........................................................ 97

Research Questions ....................................................................................................................... 112

Findings ............................................................................................................................................. 119

Question 2 ........................................................................................................................................ 123

Discussion ........................................................................................................................................ 130

Descriptive statistics for LDCs and HDCs

Regression of sex differences in mortality on gender equality

Regression of the sub-indexes of gender equality on the sex gap

Analysis of the sex ratio in the ASDR

Hypotheses for conditional relationship between gender equality

Regression of male and female life expectancy, separately

Appendix 1.1 123 countries in analysis and gender equality score

Objective 2: Beta coefficients and standard errors for logistic

Objective 2: Beta coefficients, significance levels and standard

Objective 3: Beta coefficients for logistic regression for

Objective 3: Beta coefficients and significance levels for

Birth years and ages by cohort and survey wave

Fixed effects regression by sex

Appendix 3.1 Summary of literature on age, gender and depression

Relationship between gender equality and sex gaps in mortality

Means and 95% confidence intervals for socio-demographic

Means and 95% confidence intervals for socio-demographic

Objective 1: Percent with depression (and 95 % confidence

Predicted female odds ratios for logistic regression of depression, 80

Objective 1: Mean levels of depressive severity (and 95 %

Female beta coefficients for regression of depressive severity,

Objective 2: Odds ratios (and 95% confidence intervals)

Objective 2: Beta coefficients (and 95% confidence intervals)

Appendix 2.1 Distribution of depression scores, females 1971

Appendix 2.2 Distribution of depression scores, females 2008

Appendix 2.3 Distribution of depression scores, males 1971

Appendix 2.4 Distribution of depression scores, males 2008

Average CES-D scores by age and sex

Marital status by sex and age group

Employment status by sex and age group

Health by sex and age group

Predicted CES-D score by sex and age, from fixed

Average CES-D by cohort Males

Average CES-D by cohort Females

we explore the potential contribution of smoking to the sex gap in mortality by

Mortality and gender research in HDCs

society in general. Therefore, it remains unclear as to whether the relationship between

+ p<.10, * p<.05, p<.01, * p<.001

+ p<.10, * p<.05, p<.01, * p<.001

+ p<.10, * p<.05, p<.01, * p<.001

+ p<.10, * p<.05, p<.01, * p<.001