Journal of Happiness Studies

, Volume 11, Issue 6, pp 663–682

Gender Differences in the Incidence of Depression and Anxiety: Econometric Evidence from the USA

Authors

    • School of EconomicsUniversity of Ulster
Research Paper

DOI: 10.1007/s10902-009-9155-4

Cite this article as:
Borooah, V.K. J Happiness Stud (2010) 11: 663. doi:10.1007/s10902-009-9155-4
  • 260 Views

Abstract

Using data from the Collaborative Psychiatric Epidemiology Surveys (CPES) for the United States for the period 2001–2003, this paper addresses a vexed question relating to inter-gender differences in depression rates, namely how much of the observed difference in depression rates between men and women may be explained by differences between them in their exposure, and how much may be explained by differences between them in their response, to depression-inducing factors. The contribution of this paper is to propose a method for disentangling these two influences and to apply it to US data. The central conclusion of the paper was differences between men and women in rates of depression and anxiety were largely to be explained by differences in their responses to depression-inducing factors: the percentage contribution of inter-gender response differences to explaining the overall difference in inter-gender probabilities of being depressed was 93 percent for “sad, empty type depression”; 92 percent for “very discouraged” type depression; and 69 percent for “loss of interest” type depression.

Keywords

GenderDepressionAnxietyDecomposition

1 Introduction

Some economists are beginning to question a (arguably, the) fundamental belief that underpins our subject, namely that a better economic performance by a country is in itself, and of itself, a “good thing” (Frank 1997, 1999; Layard 2002, 2003).1 In response to such concerns, studies (both econometric and non-econometric) about the nature of happiness, and about the factors underlying happiness, have mushroomed.2 Since a prominent conclusion of such studies is that mental ill-health is a major reason for being unhappy, this study examines two specific aspects of mental ill-health: depression and anxiety.3

The reasons for the emphasis on depression and anxiety are three-fold. The first is the large number of people who are affected by these two ailments: the Psychiatric Morbidity Survey (2000) estimates that 6 million persons in the United Kingdom suffer from depression or anxiety or both, with 14.4 million persons suffering similar disorders in the USA. The second reason is that depression and anxiety, by preventing many of its afflicted citizens from working, may impose large economic costs on a country: Layard (2006) estimates that the loss in output in the UK due to depression and anxiety is some £12 billion (or 1 per cent of the UK’s national income) per year. Lastly, there is a significantly large gender bias to depression and anxiety with women being much more likely than men to have these conditions: meta-analysis of studies conducted in various countries show that women are roughly twice as likely as men to suffer depression (Noel-Hoeksema 1990; Weissman et al. 1996.

Indeed, it is the unequal distribution of depression and anxiety between men and women that is the focus of this paper. There is a large literature on differences between men and women in their propensities to be depressed and Culbertson (1997), Piccinelli and Wilkinson (2000), Noel-Hoeksema (2001) and Murakumi (2002) inter alia provide a good perspective to this body of work. In explaining why rates of depression are higher for women than for men, Noel-Hoeksema (2001) distinguished between two effects, the quantification of which constitutes the fundamental purpose of this paper.

First, she argued that, compared to men, women might be more likely to be exposed to depression-inducing factors. So, for example: women were more likely than men to be the victims of childhood sexual assault; they were more likely to be trapped in the role of perpetual carers, with their lives sandwiched between caring for their young children and their aged parents; they were more likely to be unequal partners in heterosexual relationships with major, life-changing decisions being made by their male partners; and they were more likely to do atypical and non-standard type work exemplified by temporary or part-time jobs.4

Second, even when men and women were exposed to the same depression-inducing factors, women might be more likely than men to develop depression. This might be due to gender differences in the response to such factors caused inter alia by: biological factors;5 differences in levels of self-esteem between men and women;6 differences between men and women in their respective propensities to introspection and rumination.7

Given these two effects—engendered, respectively, by gender differences in exposure, and in response, to depression-inducing factors—the need is for an integrative model, encompassing both exposure and response effects, to explain differences in depression rates between women and men (Noel-Hoeksema 2001).8 In addition, it would be useful to quantify how much of the observed difference in depression rates between men and women could be explained by differences between them in their exposure, and how much could be explained by differences between them in their response, to depression-inducing factors. The central purpose of this paper is to build such a model and offer such quantification.

2 The Data

The data used in this paper are from the Collaborative Psychiatric Epidemiology Surveys (CPES) for the United States for the period 2001–2003. These data, which are described in some detail in Alegria et al. (2007), present inter alia information on the prevalence of mental disorders and on the personal and social circumstances of the respondents all of whom were 18 years or older. The CPES joins together three nationally representative surveys: the National Comorbidity Survey Replication (NCS-R); the National Survey of American Life (NSAL), and the National Latino and Asian American Study (NLAAS); in consequence, CPES permits analysts to approach analysis of the combined dataset as though it were a single, nationally representative study.

The CPES dataset is organised in different files, each relating to a particular aspect of respondents’ lives, and from these files this paper focused on two: the “Screening” and the “Demographic” files. The “Screening” and “Demographic” instruments were administered to all the respondents in the survey; the instruments pertaining to the other files were only applied to those affected by one or (more) mental disorder. Using information from the Screening file, we defined a person as having experienceddepression if he/she answered “yes” to any of the following questions:
  1. (i)

    Have you ever in your life had a period, lasting several days or longer, when most of the day you felt sad, empty, or depressed?

     
  2. (ii)

    Have you ever in your life had a period, lasting several days or longer, when most of the day you were very discouraged about how things were going in your life?

     
  3. (iii)

    Have you ever in your life had a period, lasting several days or longer, when you lost interest in most things you usually enjoy like work, hobbies, and personal relationships?

     

Similarly, a person was defined as having experiencedmild anxiety if he/she answered the following question in the affirmative: have you ever in your life had an attack of fear or panic when all of a sudden you felt very frightened, anxious, or uneasy? By extension, severe anxiety was defined as answering yes to the following question: Have you ever had an attack when all of a sudden you became very uncomfortable, you either became short of breath, dizzy, nauseous, or your heart pounded, or you thought that you might lose control, die, or go crazy?9

For the NCS-R, a total of 9,282 adult interviews were completed: 7,963 with the main respondent and 1,589 interviews with the second adult in the household; in addition, 554 interviews were conducted with a sample of non-respondents using a shortened version of the instrument. The final response rate was 70.9 percent for primary respondents and 80.4 percent for secondary respondents. For the NSAL, the overall response rate was 71.5 percent while, for the NLAAS, the response rate was 75.7 percent.10

Table 1 shows that 44 percent (of the 5,862 women analysed), compared to 35 percent (of the 4,227 men analysed), had felt “sad, empty, or depressed”; 44 percent of women, compared to 38 percent of men, had felt “very discouraged”; 33 percent of women, compared to 29 percent of men, had “lost interest in most things”; 40 percent of women, compared to 32 percent of men, had experienced mild anxiety; and 10 percent of women, compared to 8 percent of men, had experienced severe anxiety.11 So, for every facet of depression and anxiety, women were more likely than men to have experienced that condition with the gender gap being largest for feeling “sad, empty, depressed” and for mild anxiety and smallest for “losing interest” and for severe anxiety. Since the data refer to self-reported depression or anxiety, the possibility is that gender differences in depression rates may be the result of men responding to stress through alternative modes such as antisocial behaviour and alcohol abuse (Kessler et al. 1994; Meltzer et al. 1995).
Table 1

Proportion of men and women who have been depressed/anxious, by type of condition

Percentage who for a period lasting several days or longer

Men (4,227)

Women (5,862)

All persons (10,089)

Felt sad, empty, or depressed

35

44

40.5

Felt very discouraged about how things were going in life

38

44

41.3

Lost interest in most things they usually enjoyed like work, hobbies, and personal relationships

29

33

31.6

Mild anxiety

32

40

36.7

Severe anxiety

8

10

7.9

Notes:

Proportions were computed for persons who reported non-missing values for all the variables used in the logistic regressions (Table 4, 5, 6, 7, and 8): 10,089 persons of whom 5,862 were women and 4,227 were men

Depression is defined as a positive response to any of the following questions

(iv) Have you ever in your life had a period, lasting several days or longer, when most of the day you felt sad, empty, or depressed?

(v) Have you ever in your life had a period, lasting several days or longer, when most of the day you were very discouraged about how things were going in your life?

(vi) Have you ever in your life had a period, lasting several days or longer, when you lost interest in most things you usually enjoy like work, hobbies, and personal relationships?

Mild anxiety is defined as answering yes to the following question: Have you ever in your life had an attack of fear or panic when all of a sudden you felt very frightened, anxious, or uneasy?

Severe anxiety is defined as answering yes to the following question: Have you ever had an attack when all of a sudden you became very uncomfortable, you either became short of breath, dizzy, nauseous, or your heart pounded, or you thought that you might lose control, die, or go crazy?

Table 2 shows the distribution of depression and anxiety by non-gender attributes. The highest rates of depression and anxiety were for White persons (“sad”: 50 percent), followed by Hispanics (“sad”: 46 percent) and the lowest rates were for Asians (“sad”: 31 percent). People below the age of 30 had markedly higher rates of depression and anxiety than the over 60s (“sad”: respectively, 43 and 34 percent) and those who were married or cohabiting (“sad”: 35 percent) had markedly lower rates of depression and anxiety compared to the never married or the separated/divorced/widowed (“sad”: 44 and 48 percent, respectively). Better-off persons (those whose income-to-poverty line ratio was higher than the mean ratio) had slightly lower rates of depression and anxiety (“sad”: 39 percent) compared to poorer persons12 (“sad”: 41 percent).
Table 2

Distribution of depression and anxiety by attribute

Percentage with the condition ↓

Sad

Discouraged

Loss of interest

Mild anxiety

Severe anxiety

Overall sample

41

41

32

37

8

Race, Ancestry

    Asian

31

32

24

30

5

    Black

40

41

33

35

9

    Hispanic

46

44

33

41

8

    White

50

55

41

45

12

Age

    18–30

43

46

36

37

8

    31–45

41

42

31

37

7

    46–60

42

42

33

39

8

    >60

34

31

22

31

11

Marital status

    Married or cohabiting

35

36

27

34

7

    Separated, divorced, widowed

48

46

37

40

10

    Never married

44

46

37

39

7

Resources (average score = 3.57)

    Income-to-needs-ratio: hhinc/poverty line > mean

39

39

29

36

7

    Income-to-needs-ratio: hhinc/poverty line < mean

41

42

33

37

9

Years of education

    Education high

40

41

30

37

6

    Education medium

41

43

35

38

8

    Education moderate

40

41

31

37

8

    Education low

41

40

30

35

9

Employment status

    Employed

39

40

30

36

7

    Unemployed

47

50

40

40

10

    Inactive

43

41

34

39

10

Immigration status

    US born

43

46

36

39

10

    Non-US born

37

35

26

34

6

US region

    North East

44

45

36

42

9

    Mid West

45

48

38

41

10

    West

36

37

28

35

6

    South

41

41

31

34

8

Health, childhood, society

    Good physical health

37

38

28

34

7

    Bad physical health

54

54

44

47

14

    Childhood traumas

52

55

44

47

12

    No childhood trauma

29

28

19

27

5

    Church member

39

40

31

36

8

    No church membership

41

42

32

37

8

Disability scores

    WHO-DAS: cognitive score > mean = 1.19

76

80

72

69

28

    WHO-DAS: cognitive score < mean = 1.19

37

38

28

34

7

    WHO-DAS: social interaction score  > mean = 0.85

78

81

76

68

24

    WHO-DAS: social interaction score < mean = 0.85

38

39

29

35

7

Persons born in the USA were considerably more likely to have experienced depression and anxiety, compared to non-US born persons, (“sad”: 43 and 37 percent, respectively) and persons living in the west of the USA (“sad”: 36 percent) were markedly less likely to have experienced depression and anxiety than persons living elsewhere in the USA (“sad”: above 40 percent). Rates of depression and anxiety were impervious to education level but there was a strong link between such rates and employment status: people who were unemployed had markedly higher rates of depression and anxiety than those in employment (“sad”: 47 and 39 percent, respectively). This is consistent with much of the literature on the connection between unemployment and depression (Clark and Oswald 1994; Clark et al. 2008). However, the direction of causation is open to question: does unemployment cause depression or are depressed persons more likely to be made unemployed? Böckerman and Ilmakunnas (2009), using panel data for Finland, suggest that persons with lower levels of self-assessed health were more likely to become unemployed.

Compared to those in bad physical health (“sad”: 54 percent), persons in good physical health (“sad”: 37 percent)—and, compared to those who had known childhood trauma (“sad”: 52 percent),13 persons who had not experienced childhood trauma (“sad”: 29 percent)—had much lower rates of depression and anxiety. Lastly, there appeared to be a strong association between cognitive and social disability14 and rates of depression and anxiety.

Table 3 shows the distribution of the attributes, noted above, between men and women. Compared to the female part of the sample, a larger proportion of males were: Asian (23 versus 18 percent); married (59 versus 45 percent); employed (73 versus 61 percent); lived in the West (30 versus 24 percent); and had an income-to-poverty line score higher than the mean score. Also, compared to the female part of the sample, a smaller proportion of males were: separated/divorced/widowed (16 versus 28 percent); unemployed (7 versus 9 percent); US born (54 versus 60 percent). So, on all these counts, the gender distribution of attributes was biased towards higher rates of depression and anxiety for women.
Table 3

Distribution of attributes between women and men

Percentage of women and men with attribute ↓

Women (58%)

Men (42%)

Race, Ancestry

    Asian

18

23

    Black

49

40

    Hispanic

25

28

    White

8

9

Age

    18–30

27

28

    31–45

36

35

    46–60

23

23

    >60

14

14

Marital status

    Married or cohabiting

45

59

    Separated, divorced, widowed

28

16

    Never married

27

25

Resources (average score)

    Income-to-needs-ratio: hhinc/poverty line

3.16

4.15

Years of education

    Education high

21

24

    Education Medium

25

23

    Education moderate

29

29

    Education low

25

24

Employment status

    Employed

61

73

    Unemployed

9

7

    Inactive

30

20

Immigration status

    US born

60

54

    Non-US born

40

45

US region

    North East

24

21

    Mid West

9

8

    West

24

30

    South

43

41

Health, childhood, society

    Good physical health

77

82

    Childhood traumas

46

55

    Church membership

36

23

Disability scores (average)

    WHO-DAS: cognitive score

1.37

0.96

    WHO-DAS: social interaction score

0.94

0.61

Notes:

1. Asian: Vietnamese, Filipino, Chinese, all other Asian; Hispanic: Puerto Rican, Mexican, all other Hispanic; Black: African American, Afro-Caribbean

2. Education high: 16 or more years of education; Education medium: 13–15 years; Education moderate: 12 years; Education low: 0–11 years

3. Childhood traumas: yes to any of the following: fidgety childhood; frequently in trouble with adults for 6 months or more during childhood or adolescence; lying, stealing as child or teenager; run away frequently, played truant, or stayed out late as child or teenager; separation anxiety for 1 month or more as child

4. WHO-DAS: World Health Organisation Disability Assessment Score: higher the score, greater the disability

Conversely, compared to men: a larger proportion of women were Black (49 versus 40 percent) and a smaller proportion were Hispanic (25 versus 28 percent); and a smaller proportion of women had experienced childhood trauma (46 versus 55 percent). So, on all these counts, the gender distribution of attributes was biased towards lower rates of depression and anxiety for women.

The preceding discussion raises two issues. First, what was the contribution of each of the factors, listed in Table 2, to the likelihood of a person experiencing depression and anxiety, after controlling for the other factors? Table 2, and the discussion based on it, refer to the contributions in the absence of any imposed controls. This question was answered in the context of an estimated logit model the results from which are discussed in the next section. The second issue relates to the aggregate contribution that differences in the distribution of the different attributes between men and women made to gender differences in rates of depression and anxiety. Section 4 addresses this question in the context of a decomposition model originally developed by Blinder (1973) and Oaxaca (1973) for measuring discrimination in the labour market.

3 Estimates from a Logistic Model of Depression and Anxiety

We estimated a logistic model for a dependent variable Yi such that Yi = 1, if the person (i = 1…N) has had a condition (depression, anxiety), Yi = 0, otherwise. The model was estimated on a vector of variables, Xij being the value of the jth variable for the ith person (j = 1…J).15 A natural question to ask from the logistic model is how the probability of having a particular condition would change in response to a change in the value of one of the condition-affecting factors. These probabilities are termed marginal probabilities.

For discrete variables, the marginal probabilities refer to changes in the probabilities consequent upon a move from the residual category for that variable to the category in question, the values of the other variables remaining unchanged. For continuous variables, the marginal probabilities refer to changes in the probabilities (of having the conditions) consequent upon a unit change in the value of the variable, the values of the other variables remaining unchanged. Tables 4, 5, and 6 show the estimated marginal probabilities from the logistic model for, respectively: “felt sad, empty depressed”; “felt very discouraged about how things were going in life”, and “lost interest in most things one usually enjoyed” and Tables 7 and 8 show the estimated marginal probabilities from the logistic model for, respectively, “mild anxiety” and “severe anxiety”.
Table 4

Marginal probabilities from “sad, empty, depressed” logistic equation

 

Entire sample (10,087)

Female sample (5,861)

Male sample (4,226)

Mg. prb

Z val

Mg. prb

Z val

Mg. prb

Z val

Sex

Male

−0.097

−8.98

    

Race, Ancestry (residual: White)

    Asian

−0.142

−5.8

−0.127

−3.6

−0.160

−4.93

    Black

0.122

6.27

0.088

3.29

0.161

5.98

    Hispanic

0.025

1.08

0.014

0.43

0.069

2.2

Age (residual > 60 years)

    18–30

0.083

3.72

0.079

2.73

0.092

2.63

    31–45

0.089

4.38

0.073

2.75

0.113

3.57

    46–60

0.095

4.6

0.083

3.13

0.111

3.42

Marital Status (residual: never married)

    Married or cohabiting

−0.054

−3.81

−0.045

−2.37

−0.058

−2.72

    Separated, divorced, widowed

0.058

3.32

0.043

1.97

0.096

3.24

Resources

    Income-to-needs-ratio: hhinc/poverty line

0.002

1.35

0.001

0.6

0.003

1.25

Years of education (residual, low education)

    Education high

0.087

4.81

0.083

3.48

0.087

3.23

    Education Medium

0.035

2.16

0.031

1.43

0.037

1.49

    Education moderate

0.022

1.45

0.022

1.12

0.020

0.88

Employment status (residual: inactive)

    Employed

0.022

1.5

0.005

0.26

0.066

2.64

    Unemployed

0.031

1.39

0.050

1.78

−0.003

−0.09

Immigration status (residual: born abroad)

    US born

0.001

0.1

0.013

0.74

−0.018

−0.91

US region (residual: South)

    North East

0.017

1.19

0.017

0.89

0.017

0.79

    Mid West

0.011

0.56

−0.006

−0.22

0.038

1.21

    West

−0.036

−2.13

−0.024

−1.04

−0.041

−1.76

Health, childhood, society

    Good physical health

−0.133

−9.31

−0.140

−7.72

−0.118

−5.18

    Childhood traumas

0.231

21.9

0.225

15.94

0.230

15.13

    Church membership

−0.026

−1.81

−0.039

−2.07

−0.002

−0.1

Disability scores

    WHO-DAS: cognitive score

0.013

7.7

0.015

6.51

0.009

4.15

    WHO-DAS: social interaction score

0.011

5.16

0.012

4.12

0.010

3.04

Notes:

1. Depression is defined as answering yes to the following question: Have you ever in your life had a period, lasting several days or longer, when most of the day you felt sad, empty, or depressed?

2. Asian: Vietnamese, Filipino, Chinese, all other Asian; Hispanic: Puerto Rican, Mexican, all other Hispanic; Black: African American, Afro-Caribbean

3. Education high: 16 or more years of education; Education medium: 13–15 years; Education moderate: 12 years; Education low: 0–11 years

4. Childhood traumas: yes to any of the following: fidgety childhood; frequently in trouble with adults for 6 months or more during childhood or adolescence; lying, stealing as child or teenager; run away frequently, platyed truant, or stayed out late as child or teenager; separation anxiety for 1 month or more as child

5. WHO-DAS: World Health Organisation Disability Assessment Score: higher the score, greater the disability

6. The rows of the table with typeface in italics are those variables for which the difference between the female and male marginal probabilities was significantly different from zero at the 20% (or less) level of significance

Table 5

Marginal probabilities from “very discouraged” logistic equation

 

Entire sample (10,087)

Female sample (5,861)

Male sample (4,226)

Mg. prb

Z val

Mg. prb

Z val

Mg. prb

Z val

Sex

    Male

−0.072

−6.50

    

Race, Ancestry (residual: White)

    Asian

−0.180

−7.55

−0.186

−5.54

−0.179

−5.37

    Black

0.156

7.92

0.117

4.33

0.197

7.15

    Hispanic

0.096

4.30

0.068

2.17

0.129

4.14

Age (residual > 60 years)

    18–30

0.137

6.01

0.121

4.05

0.163

4.52

    31–45

0.129

6.20

0.118

4.36

0.146

4.46

    46–60

0.122

5.77

0.127

4.65

0.117

3.52

Marital status (residual: never married)

    Married or cohabiting

−0.044

−3.09

−0.052

−2.69

−0.030

−1.4

    Separated, divorced, widowed

0.044

2.49

0.028

1.27

0.078

2.61

Resources

    Income-to-needs-ratio: hhinc/poverty line

0.002

1.00

0.004

1.53

0.000

−0.06

Years of education (residual, low education)

    Education high

0.096

5.25

0.089

3.66

0.102

3.7

    Education medium

0.060

3.61

0.066

3.02

0.047

1.86

    Education moderate

0.031

2.04

0.037

1.81

0.023

0.99

Employment status (residual: inactive)

    Employed

0.015

1.00

0.000

0.02

0.045

1.76

    Unemployed

0.050

2.19

0.063

2.22

0.025

0.68

Immigration status (residual: born abroad)

    US born

0.034

2.56

0.048

2.67

0.014

0.71

US region (residual: South)

    North East

0.035

2.45

0.033

1.73

0.040

1.81

    Mid West

0.035

1.74

0.041

1.56

0.027

0.85

    West

−0.012

−0.68

0.008

0.36

−0.029

−1.19

Health, childhood, society

    Good physical health

−0.135

−9.32

−0.141

−7.66

−0.122

−5.21

    Childhood traumas

0.249

23.72

0.246

17.48

0.250

16.24

    Church membership

0.024

1.62

0.060

3.15

0.030

1.32

Disability scores

    WHO-DAS: cognitive score

0.017

8.84

0.019

7.22

0.015

5.16

    WHO-DAS: social interaction score

0.010

4.51

0.011

3.7

0.009

2.52

Notes:

Depression is defined as answering yes to the following question: Have you ever in your life had a period, lasting several days or longer, when most of the day you were very discouraged about how things were going in your life?

Table 6

Marginal probabilities from “loss of interest” logistic equation

 

Entire sample (10,087)

Female sample (5,861)

Male sample (4,226)

Mg. prb

Z val

Mg. prb

Z val

Mg. prb

Z val

Sex

    Male

−0.033

−3.24

    

Race, Ancestry (residual: White)

    Asian

−0.111

−5.18

−0.140

−4.86

−0.088

−2.80

    Black

−0.092

−5.19

−0.081

−3.3

−0.105

−4.20

    Hispanic

−0.065

−3.33

−0.054

−1.97

−0.082

−2.99

Age (residual > 60 years)

    18–30

0.176

7.55

0.159

5.16

0.207

5.64

    31–45

0.146

7.03

0.141

5.12

0.159

4.91

    46–60

0.154

7.08

0.152

5.34

0.163

4.75

Marital status (residual: never married)

    Married or cohabiting

−0.040

−3.07

−0.041

−2.33

−0.031

−1.56

    Separated, divorced, widowed

0.054

3.26

0.034

1.65

0.097

3.37

Resources

    Income-to-needs-ratio: hhinc/poverty line

0.003

1.69

0.005

2.22

0.000

0.12

Years of Education (residual, low education)

    Education high

0.079

4.46

0.063

2.67

0.093

3.51

    Education Medium

0.072

4.54

0.063

2.99

0.080

3.27

    Education moderate

0.023

1.59

0.008

0.43

0.040

1.83

Employment status (residual: inactive)

    Employed

0.059

4.26

0.041

2.32

0.091

3.79

    Unemployed

0.007

0.34

0.000

−0.01

0.018

0.54

Immigration status (residual: born abroad)

    US born

0.031

2.55

0.053

3.18

0.001

0.06

US region (residual: South)

    North East

0.043

3.16

0.050

2.76

0.032

1.59

    Mid West

0.039

2.06

0.021

0.85

0.069

2.26

    West

−0.010

−0.64

0.001

0.04

−0.015

−0.66

Health, childhood, society

    Good physical health

−0.124

−8.81

−0.133

−7.33

−0.110

−4.92

    Childhood traumas

0.224

22.56

0.212

15.56

0.237

16.74

    Church membership

0.014

1.04

0.038

2.19

0.024

1.16

Disability scores

    WHO-DAS: cognitive score

0.016

10.03

0.016

7.86

0.016

6.20

    WHO-DAS: social interaction score

0.010

5.27

0.013

5.05

0.004

1.80

Notes:

Depression is defined as answering yes to the following question: Have you ever in your life had a period, lasting several days or longer, when you lost interest in most things you usually enjoy like work, hobbies, and personal relationships?

Table 7

Marginal probabilities from the “mild anxiety” logistic equation

 

Entire sample (10,087)

Female sample (5,861)

Male sample (4,226)

Mg. prb

Z val

Mg. prb

Z val

Mg. prb

Z val

Sex

    Male

−0.089

−8.58

    

Race, ancestry (residual: White)

    Asian

−0.110

−4.70

−0.124

−3.78

−0.107

−3.34

    Black

−0.101

−5.44

−0.092

−3.6

−0.110

−4.2

    Hispanic

0.016

0.74

0.012

0.38

0.053

1.79

Age (residual > 60 years)

    18–30

0.011

0.53

0.017

0.63

0.059

1.79

    31–45

0.033

1.70

0.035

1.36

0.036

1.23

    46–60

0.056

2.86

0.048

1.86

0.071

2.34

Marital status (residual: never married)

    Married or cohabiting

−0.017

−1.27

−0.019

−1.02

−0.007

−0.36

    Separated, divorced, widowed

0.004

0.22

0.009

0.45

−0.011

−0.41

Resources

    Income-to-needs-ratio: hhinc/poverty line

0.001

0.80

0.001

0.54

0.001

0.58

Years of education (residual, low education)

    Education high

0.094

5.32

0.079

3.37

0.113

4.25

    Education Medium

0.059

3.68

0.038

1.79

0.088

3.57

    Education moderate

0.050

3.39

0.044

2.27

0.055

2.45

Employment status (residual: inactive)

    Employed

−0.015

−1.10

−0.005

−0.28

−0.032

−1.37

    Unemployed

0.006

0.27

0.004

0.14

0.007

0.22

Immigration status (residual: born abroad)

    US born

0.023

1.78

0.031

1.8

0.004

0.21

US region (residual: South)

    North East

0.071

5.12

0.051

2.78

0.100

4.62

    Mid West

0.044

2.25

0.023

0.9

0.078

2.51

    West

0.025

1.50

0.031

1.36

0.025

1.06

Health, childhood, society

    Good physical health

0.086

6.24

0.050

2.82

0.141

6.29

    Childhood traumas

0.198

19.31

0.188

13.42

0.210

14.36

    Church membership

−0.005

−0.37

−0.009

−0.47

−0.005

−0.21

Disability scores

    WHO-DAS: cognitive score

0.011

8.40

0.017

8.04

0.005

3.03

    WHO-DAS: social interaction score

0.006

3.74

0.006

2.82

0.005

2.48

Notes:

Mild anxiety is defined as answering yes to the following question: Have you ever in your life had an attack of fear or panic when all of a sudden you felt very frightened, anxious, or uneasy?

Table 8

Marginal probabilities from the “severe anxiety” logistic equation

 

Entire sample (10,087)

Female sample (5,861)

Male sample (4,226)

Mg. prb

Z val

Mg. prb

Z val

Mg. prb

Z val

Sex

    Male

−0.010

−1.66

    

Race, Ancestry (residual: White)

      

    Asian

−0.020

−1.54

−0.030

−1.65

−0.007

−0.41

    Black

−0.012

−1.19

−0.007

−0.45

−0.015

−1.12

    Hispanic

−0.010

−0.90

−0.014

−0.84

−0.004

−0.28

Age (residual > 60 years)

    18–30

0.020

1.94

0.003

0.15

0.034

2.92

    31–45

−0.022

−2.28

−0.014

−0.97

−0.027

−2.23

    46–60

0.014

1.60

0.002

0.13

0.027

2.6

Marital status (residual: never married)

    Married or cohabiting

0.013

1.55

0.020

1.67

0.002

0.22

    Separated, divorced, widowed

0.018

1.60

0.025

1.62

0.006

0.43

Resources

    Income-to-needs-ratio: hhinc/poverty line

0.000

0.44

0.000

−0.16

0.001

0.92

Years of education (residual, low education)

    Education high

0.010

1.00

0.002

0.14

0.023

2.13

    Education Medium

0.005

0.61

0.010

0.73

−0.002

−0.18

    Education moderate

0.006

0.73

0.004

0.32

0.017

1.85

Employment status (residual: inactive)

    Employed

−0.011

−1.27

−0.004

−0.33

−0.020

−1.55

    Unemployed

0.010

0.74

0.016

0.87

0.001

0.05

Immigration status (residual: born abroad)

    US born

0.026

3.42

0.019

1.74

0.031

3.08

US region (residual: South)

    North East

0.013

1.47

0.011

0.87

0.014

1.12

    Mid West

0.009

0.78

0.007

0.46

0.009

0.55

    West

−0.005

−0.53

−0.005

−0.35

0.000

0.01

Health, childhood, society

    Good physical health

−0.048

−4.76

−0.055

−3.98

−0.037

−2.68

    Childhood traumas

0.059

8.13

0.052

4.97

0.063

6.59

    Church membership

−0.009

−1.21

−0.015

−1.44

0.000

0.04

Disability scores

    WHO-DAS: cognitive score

0.002

2.74

0.002

1.55

0.002

3.37

    WHO-DAS: social interaction score

0.001

1.78

0.004

3.96

0.003

2.00

Notes:

Severe anxiety is defined as answering yes to the following question: Have you ever had an attack when all of a sudden you became very uncomfortable: you either became short of breath, dizzy, nauseous, or your heart pounded, or you thought that you might lose control, die, or go crazy?

The marginal probabilities are first shown for the model estimated over the entire sample (10,087 persons: 5,861 women and 4,226 women) and then for the model estimated over, respectively the female and male subsamples. The associated z-values are shown alongside the marginal probabilities: a z value exceeding 1.96 indicates that the coefficient was significantly different from zero at a 5% significance level. The rows of the table with typeface in italics are those variables for which the difference between the female and male marginal probabilities was significantly different from zero at the 20% (or less) level of significance.

So, for example, in Table 4, looking at the estimates obtained from the entire sample, ceteris paribus the average probability of men having felt “sad, empty, depressed” was 9.7 percentage points lower than that for women. The results estimated across the female and male subsamples show that there were six statistically significant (significance level: 20% or less) differences between women and men: between Black women and men; between Hispanic women and men; between men and women who were separated, divorced, or widowed; between employed women and men; between women and men who had had childhood traumas; and between women and men with respect to their WHO-DAS cognitive scores.

Compared to the probability of a white person having felt “sad, empty, depressed”, the corresponding probability for a black person was ceteris paribus 12.2 points lower: row “Black”, column 2 of Table 4. However, compared to the probability of a white woman having felt “sad, empty, depressed”, the corresponding probability for a black woman was ceteris paribus 8.8 points lower and, compared to the probability of a white man having felt “sad, empty, depressed”, the corresponding probability for a black man was ceteris paribus 16.1 points lower: row “Black”, columns 3 and 5, respectively, of Table 4. The results further suggest that the marginal probability for black women (8.8 points) was significantly different from that for black men (16.1 points).

A persistent feature of the results reported in Tables 4, 5, 6, 7, and 8 was the importance of race: ceteris paribus Asians, followed by Blacks, were much less likely to have experienced depression and anxiety than White persons (see above discussion): see Dunlop et al. (2003). Another important feature of the results was marital status: ceteris paribus compared to never married persons, separated/divorced/widowed persons were more likely (in the case of feeling “sad, empty, depressed” by 5.8 points) to have experienced depression (though not anxiety). The effect of marital status on depression, however, varied by gender: compared to never married men, men who were separated/divorced/widowed were significantly more likely to have experienced all three forms of depression (“sad, empty, depressed”: 9.6 points; “very discouraged”: 7.8 points; “loss on interest”: 9.7 points); compared to never married women, women who were separated/divorced/widowed were significantly more likely to have only experienced feeling “sad, empty, depressed” (by 4.3 points) without any significant difference in the likelihood of feeling “very discouraged” or a “loss of interest”.

A third feature of the results was the importance of childhood trauma in determining whether a person, male or female, would experience depression and anxiety as an adult: compared to their counterparts who had not known childhood trauma, men and women who had were more likely, by approximately 22 percentage points, to have experienced all three forms of depression and, by approximately 20 points, to have experienced mild anxiety.

4 Gender Decomposition of the Probabilities of Depression and Anxiety

The Oaxaca (1973) and Blinder (1973) method (hereafter, O–B) of decomposing differences between groups, in their respective mean values, into a “discrimination” and a “characteristics” component is, arguably, the most widely used decomposition technique in economics. This method has been extended from its original setting within regression analysis, to explaining group differences in probabilities derived from models of discrete choice with a binary dependent variable and estimated using logit/probit methods (Nielsen 1998).

The O–B decomposition (and its extension) is formulated for situations in which the sample is subdivided into two mutually exclusive and (collectively exhaustive) groups, such as, for example, men and women. Then, one may decompose the difference in, for example, average wages between men and women—or the difference between men and women in their average probabilities of being depressed—into two parts, one due to gender differences in the coefficient vectors and one due to gender differences in the attribute (or variable) vectors.

The attribute contribution is computed by asking what the average male-female difference in probabilities would have been if the difference in attributes between men and women had been evaluated using a common coefficient vector. The critical question though is: what should be this common coefficient vector? Typically, two separate computations of the attribute contribution are provided using, respectively, the male and the female coefficient vectors as the common vector.

Column 1 of Table 9 shows the difference between men and women in the average proportions with a particular condition. On average, compared to men, women were: more likely by 8.9 percentage points to have felt “sad, empty, depressed”; more likely by 6.4 points to have felt “very discouraged”; more likely, by 3.9 points to have felt a “loss of interest”; more likely by 7.7 points to have had mild anxiety; and more likely by 2.0 points to have had severe anxiety.
Table 9

The decomposition of the proportions of men and women who have experienced depression and anxiety: all coefficients

 

1

2

3

4

5

6

7

Sad, empty, depressed

0.442 − 0.353 = 0.089

0.442 − 0.443 = −0.001

0.443 − 0.353 = 0.09

0.364 − 0.353 = 0.011

0.442 − 0.364 = 0.078

0.006

0.443 − 0.360 = 0.083

Discouraged

0.439 − 0.375 = 0.064

0.439 − 0.442 = −0.003

0.442 − 0.375 = 0.067

0.385 − 0.375 = 0.010

0.439 − 0.385 = 0.054

0.005

0.440 − 0.381 = 0.059

Loss of interest

0.332 − 0.293 = 0.039

0.332 − 0.326 = 0.006

0.326 − 0.293 = 0.033

0.310 − 0.293 = 0.017

0.332 − 0.310 = 0.022

0.012

0.330 − 0.303 = 0.027

Mild anxiety

0.399 − 0.322 = 0.077

0.399 − 0.404 = −0.005

0.404 − 0.322 = 0.082

0.315 − 0.322 = −0.007

0.399 − 0.315 = 0.084

−0.013

0.401 − 0.318 = 0.09

Severe anxiety

0.103 − 0.083 = 0.020

0.103 − 0.097 = 0.006

0.097 − 0.083 = 0.014

0.085 − 0.083 = 0.002

0.103 − 0.085 = 0.018

0.004

0.100 − 0.084 = 0.016

Notes:

Column 1: Observed difference. Difference between the proportions of women and men, who have experienced the condition, \( P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{M} ) \)

Column 2: Attribute Difference. Difference between the proportion of women who have experienced the condition and the proportion of men who would have experienced the condition if male attributes had been evaluated using female coefficients,\( P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{W} ) \)

Column 3: Coefficient difference. Difference between columns 1 and 2 (Residual), \( \left\{ {P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{M} )} \right\} - \left\{ {P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{W} )} \right\} = P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{M} ) \)

Column 4: Attribute Difference. Difference between the proportion of women who would have experienced the condition if female attributes had been evaluated using male coefficients and the proportion of men who have experienced the condition, \( P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{M} ) - P({\mathbf{X}}^{M}, \,{\varvec{\hat{\upbeta }}}^{M} ) \)

Column 5: Coefficient difference. Difference between columns 1 and 4 (Residual), \( \left\{ {P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{M} )} \right\} - \left\{ {P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{M} ) - P({\mathbf{X}}^{M},\, {\varvec{\hat{\upbeta }}}^{M} )} \right\} = P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{W} ) - P({\mathbf{X}}^{W},\, {\varvec{\hat{\upbeta }}}^{M} ). \)XW and XM are the attribute vectors, and βW and βM are the coefficient vectors, for women and men, respectively

Column 6: Attribute difference obtained as the weighted average of columns 2 and 4, the weights being the proportions of men (column 2) and women (column 4) in the sample

Column 7: Coefficient difference obtained as the difference between men and women in their “synthetic” probabilities of experiencing that condition

Column 2 of Table 9 shows the amount of the overall gap that is due to the attributes effect when female and male attributes are both evaluated using female coefficients; similarly, column 4 of Table 9 shows the amount of the overall gap that is due to the attributes effect when female and male attributes are both evaluated using male coefficients. Three points need to be made about the attributes effect:
  1. 1.

    The size of the attributes effect differs according to whether female or male coefficients are used in the evaluation. For example, for “loss of interest” depression, the attributes effect based on female coefficients explains 15 percent—while the attributes effect based on male coefficients explains 44 percent—of the overall difference in inter-gender rates.

     
  2. 2.

    For some conditions, the attributes effect predicts that, if male attributes were evaluated at female coefficients, then, compared to women, a higher proportion of men would have had that condition. Feeling “sad, empty, depressed”, feeling “very discouraged”, and mild anxiety are three such conditions.

     
  3. 3.

    In general, the contribution of the attributes effect towards explaining the overall difference in inter-gender rates was small. The largest contributions were recorded when male coefficients were used in the evaluation of female and male attributes and these were 16 percent for “very discouraged” depression and 44 percent for “loss of interest” depression.

     

The problem with the O–B method of decomposition is that the decomposition is anchored either by treating men as women (column 2, Table 9) or women as men (column 4, Table 9). More recently, Borooah and Iyer (2005) have proposed a method of decomposition which combines both “anchors” into a single decomposition formula.

Denote, by \( \bar{P}^{W} {\text{ and }}\bar{P}^{M} \), the average probabilities of having experienced a condition, computed over all the persons in the sample, when their individual attribute vectors (the Xik) are all evaluated using the coefficient vectors of, respectively women (βW) and men (βM); in other words, \( \bar{P}^{W} {\text{ and }}\bar{P}^{M} \) are the average probabilities of having had a condition, computed over the entire sample, when all the persons in the sample are treated as, respectively, women and men. The difference between the probabilities, \( \bar{P}^{W} - \bar{P}^{M} \), represents the “response effect” because it is entirely the consequence of differences between women and men in their (coefficient) responses to a given vector of attributes.

Borooah and Iyer (2005) have shown that these synthetic probabilities can be used to resolve the ambiguity of the O–B formulation since: \( \bar{Y}^{W} - \bar{Y}^{M} = (\bar{P}^{W} - \bar{P}^{M} ) + {\text{ the weighted average of the two attribute effects }} \)where the two attribute effects are shown in columns 2 and 4 of Table 9, the weights being the proportions of women and men in the sample.

Column 7 of Table 9 shows the difference between women and men in their synthetic probabilities of experiencing a particular condition. Remembering that these differences represent inter-gender response differences, the percentage contribution of such differences to explaining the overall difference in probabilities was: 93 percent for “sad, empty depression”; 92 percent for “very discouraged” depression; 69 percent for “loss of interest” depression; 116 percent for “mild anxiety”; and 80 percent for “severe anxiety”.

5 Conclusions

The contribution of this paper to the extensive literature on gender differences in depression rates was to apply the Oaxaca–Blinder decomposition methodology to quantifying the contribution of differences in exposure, and differences in response, to inter-gender differences in depression rates. Needless to say, the division of the sample could have been by factors other than gender and, in the context of depression and anxiety, it might be useful to examine inter-racial differences in the propensity towards these conditions. As the data in Table 2 shows, reported rates of depression and anxiety are considerably lower for Asians than for Whites. Is this due to the fact that Asians are, perhaps, relatively sheltered from depression/anxiety inducing factors? That the sample of Asians contained relatively more males? Or is it because Asian responses to such factors (these responses perhaps engendered by a Confucian stoicism which makes it culturally less acceptable to admit to emotional frailty than might be the case for those with liberal, Western roots) are relatively more muted than the responses of people of other races? Similar observations might be made about regional divisions: does living in sunny California offer greater protection from factors associated with depression and anxiety compared to living in the rust belts of the Mid West? Or are people living in California better able to cope with such factors?

This paper offered a methodology—with a long and distinguished pedigree in economics—which is capable of providing answers to questions in which responsibility needed to be apportioned between exposure and response. However, it might be pertinent to conclude this paper by pointing to a limitation of this methodology. It should be emphasised that the response effect was defined as a residual: it was what could not be explained by differences between men and women in their exposure to the various “depression-influencing” factors. Consequently, the empirical results are only as good as the variables used in the logit regression; with a different set of variables the exposure/response split might have been different.

The relevance of this observation to the present analysis is that several of the variables used could not be better nuanced. For example, marital breakdown was a “depression-inducing” factor but the data were silent on the circumstances surrounding the breakdown; similarly, differences between men and women in the nature of their employment, or in the nature of their work-home balance, could not be elaborated upon. These examples provide arguments for marrying mental health information to a richer set of data on individual circumstances.

Footnotes
1

For a public policy approach to the pursuit of happiness.

 
2

Inter alia Blanchflower and Oswald (2002), Clark (1996, 1999, 2001), Clark and Oswald (1994), Easterlin (1974, 1987, 2001), Frank (1985, 1997, 1999), Frey and Stutzer (2002), Hirsch (1976), Layard (2002, 2003), Oswald (1997), Scitovsky (1976).

 
3

For example, Borooah (2006) in a study of Northern Ireland reported that only four percent of those with severe mental health problems described themselves as happy and 60 percent described themselves as unhappy; equally tellingly, only 32 percent of those whose mental health problems were not severe described themselves as happy—the same proportion as those with severe heart problems who regarded themselves as happy.

 
4

On this last point, see Mangan (2000).

 
5

The hypothalamic-pituitary-adrenal (HPA) axis plays a major role in regulating stress responses and, compared to men, women are more likely to have a dysfunctional HPA responses to stress (Weiss et al. 1999).

 
6

With the consequence that conflicts in, or the ending of, relationships were more likely to produce depression in women than in men.

 
7

A greater propensity to rumination in response to stress increases the risk of developing depression (Noel-Hoeksema et al. 1999).

 
8

Another possibility is that gender differences in depression rates may be the result of men responding to stress through alternative modes such as antisocial behaviour and alcohol abuse (Kessler et al. 1994; Meltzer et al. 1995).

 
9

A problem with self-reported information is that of recall. If it is a natural instinct to suppress memories of unpleasant events, and if the young are more likely to be susceptible to depression and anxiety, then older persons in a sample are likely to "forget" that they were depressed or anxious when they were young while, for the younger persons in the sample such memories are likely to be vivid. Consequently, in a cross-section of people, older, compared to younger, respondents would report lower rates of depression or anxiety purely for reasons of differences in recall.

 
10

Gender specific response rates were not provided.

 
11

The proportions were computed for persons who reported non-missing values for all the variables used in the logistic regressions (Tables 4, 5, 6, 7, and 8): 10,089 persons of whom 5,862 were women and 4,227 were men.

 
12

Those whose income-to-poverty line ratio was lower than the mean ratio.

 
13

Childhood traumas were any of the following: fidgety childhood; frequently in trouble with adults for 6 months or more during childhood or adolescence; lying, stealing as child or teenager; ran away frequently, played truant, or stayed out late as child or teenager; had separation anxiety, for 1 month or more, as a child.

 
14

Both of these were measured by a person’s World Health Organisation Disability Assessment Score (WHO-DAS): the higher the score, the greater the disability.

 
15

The logit equation is \( {\frac{{\Pr (Y_{i} = 1)}}{{1 - \Pr (Y_{i} = 1)}}} = \exp \{ \sum\nolimits_{j = 1}^{J} {X_{ij} } \beta_{j} \} = \exp \{ z_{i} \} \) for J coefficients, βj and for observations on J variables.

 
16

That is, from applying different coefficient vectors to a given vector of attributes.

 

Copyright information

© Springer Science+Business Media B.V. 2009