- Email: [email protected]
Men’s work, Women’s work, and mental health: A longitudinal investigation of the relationship between the gender composition of occupations and mental health
Social Science & Medicine 204 (2018) 16–22
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Men’s work, Women’s work, and mental health: A longitudinal investigation of the relationship between the gender composition of occupations and mental health
T
Allison Milnera,b,∗, Tania Kinga, Anthony D. LaMontagneb, Rebecca Bentleya, Anne Kavanagha a b
Centre for Health Equity, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia Population Health Research Centre, School of Health & Social Development, Deakin University, Geelong, Australia
A R T I C LE I N FO
A B S T R A C T
Keywords: Gender Work Mental health Occupation Job stressors
This longitudinal investigation assesses the extent to which the gender composition of an occupation (e.g., the extent to which an occupation is comprised of males versus females) has an impact on mental health. We used 14 annual waves of the Household Income Labour Dynamics in Australia (HILDA) study to construct a measure representing the gender ratio of an occupation. The outcome measure was the Mental Health Inventory (MHI-5). A Mundlak model was used to compare within and between person effects, after controlling for possible confounders. Results suggest that males and females employed in occupations where their own gender was dominant had better mental health than those in gender-neutral occupations (between person effects). However, within-person results suggested that a movement from a gender-neutral to a male or female dominated occupation was associated with both a decline (females) and improvement (males) in mental health. These results highlight the need for more research on gender specific selection into and out of different occupations in order to progress understandings of gender as a social determinant of health in the work context.
1. Introduction Gender is as an important social determinant of health (Krieger, 2003; Phillips, 2005) constructed through norms, roles and relationships within and between groups of women and men (WHO, 2015). This is separate from sex, defined as a biological construct premised upon biological characteristics enabling sexual reproduction (Phillips, 2005). Gender interacts with other social determinants of health, including education and income, as well as employment and working experiences (Hosseinpoor et al., 2012), and may influence health states directly as well as through interaction with other social determinants. Thus, gender may be considered as a fundamental cause of health and health problems (Link and Phelan, 1995). In the employed population, there is gender patterning across different occupations. For example, in Australia (as in many high income countries), a greater proportion of men are employed in construction related jobs (about 20% of employed males and 4% of employed females), or in higher levels of management (about 16% of employed males and about 9% of employed females) (Australian Bureau of Statistics, 2006). In contrast, females are more likely to be employed in nursing (about 90% of nurses are female) and secretarial work (about
∗
23% of employed females and 6% of employed males) (Australian Bureau of Statistics, 2006). Gender segregation of the workforce first came to the attention of social researchers in the late nineteenth century (Preston, 1999). This phenomenon has persisted across countries and over time. Since the 1960s (which is when women starting entering the workforce in substantial numbers), there has been persistent gender segregation of women into clerical, sales and service occupations in industrialised nations (Preston, 1999). The nuances of gender in the workforce have generally been ignored in epidemiology, which has traditionally focused on working conditions (employment arrangements, working hours, psychosocial stressors) (Bildt and Michélsen, 2002; Plaisier et al., 2007). However, there is some evidence from a limited number of studies that the gender composition of a job (e.g., the extent to which a job is comprised of males versus females) has an impact on health (Elwer et al., 2013; Elwer et al., 2014; Evans and Steptoe, 2002; Hall, 1989; Hensing and Alexanderson, 2004; Mastekaasa, 2005; Sobiraj et al., 2015). In general, this evidence has suggested that working in a job where the other gender is dominant (e.g., males working in a female dominated occupation, and females working in a male dominated occupation) may have damaging effects on psychological health (Elwer et al., 2013; Sobiraj et al., 2015) and be
Corresponding author. Centre for Health Equity; School of Population and Global Health; University of Melbourne; 207 Bouverie Street; Melbourne, 3010, Australia. E-mail address: [email protected] (A. Milner).
https://doi.org/10.1016/j.socscimed.2018.03.020 Received 2 December 2017; Received in revised form 7 March 2018; Accepted 10 March 2018 Available online 12 March 2018 0277-9536/ © 2018 Elsevier Ltd. All rights reserved.
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
annually from over 13,000 individuals within over 7000 households (Wilkins, 2013). The response rate to wave 1 was 66% (Wilkins, 2013). The survey covers a range of dimensions including social, demographic, health and economic conditions using a combination of face-to-face interviews with trained interviewers and a self-completion questionnaire. The initial wave of the survey began with a large national probability sample of Australian households occupying private dwellings (Wilkins, 2013). Interviews were sought in later waves with all persons in sample households who turned 15 years of age. Additional persons have been added to the sample as a result of changes in household composition. Inclusion of these new households is the main way in which the HILDA survey maintains sample representativeness. A top-up sample of 2000 people was added to the cohort in 2011 to allow better representation of the Australian population using the same methodology as the original sample (i.e., a three-stage area-based design) (Watson, 2011). The retention rates for the HILDA survey are above 90% for respondents who have continued in the survey and above 70% for new respondents being invited into the study (Wilkins, 2013). The main variables examined in this study were available in all annual waves of HILDA (2001–2014).
associated with higher sickness absence (Evans and Steptoe, 2002; Hensing and Alexanderson, 2004). Explanations for this have drawn on Kanter's (1993) theory of being in a minority demographic group. Kanter (1993) argues that being in a minority (e.g., female in a male dominated occupation) may be particularly damaging because the affected individual may have heightened visibility and thus be subject to stereotyping. Minority status at work might also affect mental health through mechanisms that include differential working conditions and pay (Blau and Kahn, 2016). Another theory posited by Blalock (1967) focuses on the dominance of the majority group and extent to which the minority group can be considered a threat to power and resources. Thus, as women increase in numbers in male-dominated jobs, they may experience poorer treatment, worse conditions and greater discrimination. Although this theory has also been used to describe the dynamics of gender at work, it is important to acknowledge that this theory was developed to explain race relations. Another perspective is that women and men employed in occupations where the other gender is dominant may experience gender-role conflict because they deviate from normative work-arrangements for male and females (Simon, 1995). A limitation of most past research on the gender composition of the workforce and health is that it has been cross-sectional and/or has not controlled for within person (time invariant) influences (Evans and Steptoe, 2002; Hensing and Alexanderson, 2004; Mastekaasa, 2005; Sobiraj et al., 2015). This is problematic as a comparison between persons (e.g., a female employed in a male dominated versus female dominated occupation) may produce substantially different estimates compared to those that can be found within persons (e.g., a person who changes between a male and female dominated occupation). The analytic approach used in this paper enables us to estimate the differences in mental health between groups defined by the gender dominance of their occupation relative to their own gender. It also allows us to examine how changing the gender dominance of a person's occupation impacts on mental health within-persons, thus capturing the dynamic relationship between gender and work environment. Using 14 waves of longitudinal data from an Australian working population cohort, we create and describe an occupational gender ratio measure across a range of individual and job characteristics (Aim 1). We then assess the association between the occupational gender ratio and mental health, adjusting for known confounders (Aim 2). Following this, we test if changes in mental health occur for people who change from a gender neutral to a male or female dominated occupation across the 14-year study period (Aim 3). Last, we assess whether the relationship between occupational gender ratio and mental health is modified by a person's own gender (Aim 4). This is important considering gender differences in the working conditions and the overall prevalence of common mental health problems, as women are more likely to report mental health problems than men (WHO, 2015). A major contribution of this paper is to build understanding of the role that gendered contexts have in influencing health outcomes, thereby expanding the conceptualisation of gender from being an individual influence on health (e.g., a person's gender), to an environmental influence (e.g., normative expressions of gendered behaviours at work). From a public health perspective, this paper will provide information on whether the gender composition of a person's job may have an independent effect on their mental health. If so, then this would provide a rationale for targeted prevention initiatives in male or female dominated occupations.
2.2. Outcome variable Mental health was assessed using the five-item Mental Health Inventory (MHI-5), a subscale from the Short Form-36 (SF-36) general health measure. The MHI-5 assesses symptoms of depression and anxiety (nervousness, depressed affect) and positive aspects of mental health (feeling calm, happy) in the past 4 weeks. The MHI-5 has reasonable validity and is an effective screening instrument for mood disorders or severe depressive symptomatology in the general population (Gill et al., 2006; Rumpf et al., 2001; Yamazaki et al., 2005) and has been validated as a measure for depression using clinical interviews as the gold standard (Berwick et al., 1991; Cuijpers et al., 2009; Rumpf et al., 2001). The current analyses use the continuous MHI-5 score (scale 1 to 100), with higher scores representing better mental health. Although there is no universally accepted translation of MHI-5 score difference to clinical meaningfulness, a difference of three points on the norm based scale (T-score) has been suggested to reflect a minimally important difference (Ware, 2000), and a difference of four or more on the unstandardised scale has been characterised as indicating a moderate clinically significant effect (Contopoulos-Ioannidis et al., 2009). 2.3. Exposure variable: occupational gender ratio We constructed a measure of whether an occupation was male dominated, female dominated or gender-neutral based on the 2006 census population level statistics from the Australian Bureau of Statistics (ABS) (Australian Bureau of Statistics, 2006). We used the Australian and New Zealand Standard Classification of Occupations (ANZSCO) two-digit occupation (n = 50 occupations) (ANZSCO, 2009), which was the most detailed occupational data available from HILDA. The first step in creating the exposure variable was to calculate the ratio of males to females in each occupational group using the census data. This was a continuous variable running from 0.02 to 81.39. As seen in Supplementary Table 1, there were only 0.02 males to 1 female in the occupation “Personal Assistants and Secretaries”. This was the most heavily female dominant occupation in the exposure. The most heavily male dominated occupation was “Automotive and Engineering Trades Workers” where there were 81.39 males to 1 female. Next, we created a three-level variable representing gender neutral, male dominated or female dominated occupation. If there were 0.50 or fewer males to 1 female ratio in an occupation, then it was classified as female dominated. If there were more than 1.50 males to 1 female in the occupation, then it was classified as male dominated. We created an alternate five level measure based on the quintiles of
2. Methods 2.1. Data source The Household, Income and Labour Dynamics in Australia (HILDA) survey is a longitudinal, nationally representative study of Australian households established in 2001. It collects detailed information 17
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
family conflict, and family to work conflict) by occupational gender ratio and gender.
the continuous measure: heavily female dominated (0–0.43 males to females), moderately female dominated (0.44–0.99 males to females), gender neutral (1.0–1.99 males to females), moderately male dominated (2–2.69 males to females), and heavily male dominated (2.7 or more males to females) and assessed this in sensitivity analysis.
2.5.2. Regression approach Random-effects models are commonly applied to panel data to assess the influence of variables that change over time as well as fixed characteristics associated with an individual (e.g., person stable characteristics, such as gender or early childhood experiences) (Wooldridge, 2016). These models are an advancement on Ordinary Least Square (pooled) models (OLS) as they are able to control for the possibility that residuals in a sample may not be distributed with the same variance (Dieleman and Templin, 2014). However, a central assumption of random-effects models is that the variation between persons is random and uncorrelated with exposure variables included in the model (Greene, 2008). Thus, these models may be biased if the person-level effects are not independent of the included exposure variables. To address this problem and answer Aim 2 and 3 we used randomeffects longitudinal regression models with the Mundlak adjustment, which includes group-means of time-varying variables in the model. While this ostensibly ensures the assumption that observed and unobserved variables are uncorrelated, it also provides the added and significant benefit of separating out “within” and “between” person effects. This means that, in addition to assessing the association between our exposure and mental health (Aim 2), we can examine the consequences for mental health of a person changing exposure status over time (Aim 3) in a single model. This within person analysis also enables us to reduce time invariant confounding and health selection as each person acts as their own control. So, issues such as personality characteristics and stable low levels of mental health are effectively controlled for. Evidence suggests that the results of traditional fixed effects models and Mundlak models are generally very similar (Dieleman and Templin, 2014). These models are different from multi-level models, which are used to assess individuals nested in within higher level units such as schools, or schools nested in a region. The formula associated with our Mundlak model is below (1)
2.4. Other variables For the descriptive analysis, we examined the gender (males and females) and age (18–24, 25 to 29, 30 to 34, 35 to 44, 45 to 54 and 55–64 years) of workers in female/male dominated and gender neutral occupations. We also examined a range of psychosocial job stressors (psychological job demands, job control, job security, and fairness of pay). These variables were previously developed as part of an index of psychosocial job quality (Butterworth et al., 2011) and are expressed as binary variables (where 0 indicates not exposed and 1 indicates exposed). Other characteristics examined included: The skill level of jobs defined according to the ANZSCO one-digit occupation (ANZSCO, 2009), where “3” can be considered as the most skilled and “1” can be considered the least skilled; employment arrangement (permanent, casual or labour hire, fixed-term contract or self-employed), and; the degree to which workers in male dominated, female dominated or gender neutral occupations experienced either work to family conflict (“Because of the requirements of my job, my family time is less enjoyable and more pressured”) or family to work conflict (“Because of my family responsibilities, the time I spend working is less enjoyable and more pressured”). Both of these items were scaled from 1 (strongly disagree) to 7 (strongly agree). For our main research question, we controlled for factors that were likely to represent common causes of mental health and occupational gender ratio. These time-varying variables included: age group, education (less than year 12 (high school), year 12, diploma or certificate, bachelor degree or higher), household structure (couple without children, couple with children, lone parent with children, lone person, and other), and weekly household income (equalised). Other variables such as psychosocial job stressors, employment arrangements, work/family conflict were seen as possible mediators of the relationship between employment in a job of a certain gender orientation and mental health, and were thus excluded. Gender was included as our effect modifier. The Directed Acyclical Graph (DAG) for Aims 2 to 4 can be seen in Fig. 1 (acknowledging there is no consensus on how to represent effect modification in a DAG, gender has been represented as a confounder).
yjn = β (x jn − x j ) + yx j + ωjn
(1)
Where x j is the individual level mean of exposure and confounders over time, and n is time. This shows that β are then mean-differenced values. ωjn is comprised of individual specific and random error. We assessed the effects of the three level occupational gender ratio in our main analysis. As a sensitivity analysis, we conducted a Mundlak model using the five-level occupational gender ratio. As a further sensitivity analysis, we assessed the role of unemployment and being not in the labour force on the relationship between occupational gender ratio and mental health, recognising that the experience of being out of work may influence employment experiences and mental health.
2.5. Analytic approach 2.5.1. Descriptive analysis We conducted cross-tabulations (frequencies and percentages) of the categorical variables and means of continuous variables (work to
2.5.3. Effect modification We tested effect modification by including an interaction term between occupational gender ratio (male-dominated, female-dominated, gender neutral) and gender in separate fixed-effects models. The significance of this relationship was examined by assessing the significance of the interaction term (i.e. beta coefficient) and via investigation of a likelihood ratio test, which examined a model with the interaction term and main effects compared to one that includes main effects only. Results suggested that the relationship between the occupational gender ratio and mental health was different for males and females (Aim 4). Compared to males in gender-neutral occupations, the coefficient for females in female dominated occupations was −1.08 (95% −1.67 to −0.49, P < 0.001) and the coefficient for females in male dominated occupations was −0.86 (95% −1.44 to −0.28, P = 0.004). Results of the likelihood ratio test were also significant (LR chi2(2) = 18.81, P < 0.001). On the basis of this, we stratified
Fig. 1. Directed acyclical graph.
18
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
in either male dominated or gender neutral occupations. This table also shows that there was more casual (i.e., temporary) employment in the female dominated occupations. Those in male or gender neutral occupations were more likely to be self-employed. As can be seen, those in female dominated occupations tended to report lower job control, while male and gender neutral occupations were more likely to report high job insecurity. Those in gender equal and female dominated occupations reported higher levels of work interfering with family, and family interfering with work, than male dominated occupations. It is also worth noting that males in female dominated occupations were slightly over-represented at the highest skill level compared to females. The opposite was true for male dominated occupations (where a higher proportion of those females employed in male-dominated occupations were employed at higher skill levels). About 11,269 of the people in HILDA (56%) changed occupation over the period 2001 to 2014. A description of the occupational gender ratio can be seen in Table 2 for persons (not observations). This describes the shifts between the different categories of the occupational gender ratio by gender. As can be seen, men spent most of their contributed waves in male dominated occupations, while females spent most of their contributed waves in female dominated occupations. 3.2. Analytic results regarding the effect of the occupational gender ratio on mental health Results of adjusted coefficients from the Mundlak model can be seen in Table 2. There are notable differences when results are decomposed within (Aim 3) and between persons (Aim 2) and by gender (Aim 4). For both males and females, concordance between gender and occupation dominance was associated with better health (Aim 2, Table 3). Men in male dominated occupations had better mental health than men in gender-neutral occupations (0.90, 95% CI 0.08 to 1.72, P = 0.031) and females working in a female dominated occupation had better mental health than a female working in a gender-neutral occupation (1.42, 95% CI 0.59 to 2.24, P = 0.001). Different gender effects were observed with changes in occupation (Aim 3). This within persons approach compares average mental health for all time periods when a person was employed in a gender-neutral occupation to average mental for the time periods when that same person was employed in a male or female dominated occupation. Compared to when they were employed in gender neutral occupations, men in female dominated occupations had slightly better mental health (0.51, 95% CI 0.02 to 1.01, P = 0.042). Women who changed occupational gender ratio groups had slightly worse mental health in male or female dominated compared to when they were employed in genderneutral occupations (−0.54, 95% CI -1.04 to −0.03, P = 0.037 and −0.53, 95% CI -0.94 to −0.13, P = 0.009, respectively). The sensitivity analysis using the five level occupational gender ratio (heavily female dominated; moderately female dominated; gender neutral; moderately male dominated; and heavily male dominated) demonstrated similar results as described above (Supplementary Table 2). Sensitivity analysis assessing the effect of unemployment and not being in the labour force can be seen in Supplementary Table 3. Once unemployment and not in the labour force were controlled for, there was a weaker effect of the occupational gender ratio on mental health (however size of coefficients and significance are similar).
Fig. 2. Flow chart describing entry into the analytic sample.
analyses by males and females. 3. Results Fig. 2 describes the sample selection into the study. Of the 20,231 employed people eligible for this analysis, 1176 had missing data leaving a final sample of 19,055 (Aim 2). Cases with missing data were slightly more likely to be younger and lower education, although differences from participants with non-missing data were non-significant. After restricting to those who changed occupation (Aim 3), there was 11,269 people remaining.
4. Discussion This study indicates that: 1) across the working population of the HILDA cohort, being employed in an occupation where your own gender is dominant is associated with slightly higher mental health; 2) among the people who have changed jobs and occupational gender ratio categories, the shift into an occupation where the other gender is dominant is associated with varying (small) effects that are modified by
3.1. Descriptive results of the characteristics of male, female or gender neutral occupations Table 1 describes characteristics of those employed in male, female or gender neutral occupations (Aim 1). People in female dominated occupations tended to be younger (20.24% are 18–24 years) than those 19
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
Table 1 Characteristics of those employed in male dominated, female dominated or gender neutral occupations. Gender
Gender neutral Male obs = 11,546 (49%)
Female dominated Female obs = 12,010 (51%)
Age 18 to 24 years 12% 13.7% 25 to 29 years 11.1% 12.7% 30 to 34 years 11.8% 11.1% 35 to 44 years 24.6% 25.8% 45 to 54 years 24.9% 24.2% 55 to 64 years 15.5% 12.5% Occupational skill level Low 9.9% 16.4% Medium 31.3% 28.5% High 58.8% 55.1% Employment arrangement Casual/labour hire 10.5% 18.1% Fixed term 7% 8.5% Permanent 59% 55% Self employed 23.5% 18.5% High job demands and complexity No 46.7% 53.9% Yes 53.3% 46.1% Low job control No 56.8% 51% Yes 43.2% 49% High job insecurity No 64.2% 66.8% Yes 35.7% 33.2% Low fairness of pay No 76.5% 75.2% Yes 23.5% 24.8% Work interfering with family (1 = strongly disagree, 7 = strongly agree) Mean 3.30 3.17 (95% CI) 3.25, 3.35 3.12, 3.22 Family interfering with work (1 = strongly disagree, 7 = strongly agree) Mean 3.13 3.30 (95% CI) 3.09, 3.18 3.25, 3.35
Male dominated
Male obs = 8252 (21%)
Female obs = 30,481 (78%)
Male obs = 34,357 (80%)
Female obs = 8765 (20%)
22% 10.6% 10.6% 21% 21.5% 14.2%
18.6% 9.9% 9.6% 24.7% 24.8% 12.5%
13.2% 11.1% 11.3% 26.4% 24.3% 13.6%
9.8% 11.6% 12.4% 27.2% 25.9% 13%
24.1% 42.2% 33.7%
15.6% 56.8% 27.6%
32.5% 35.5% 32%
31.1% 18.4% 50.5%
22.6% 9.6% 60.8% 7.1%
24.5% 8.9% 59.9% 6.8%
11.7% 6.4% 61.0% 20.1%
16.7% 8.1% 59.4% 15.9%
48.1% 51.9%
54.1% 45.9%
46.5% 53.5%
49% 51%
39% 61%
35% 65%
51.6% 48.4%
50.3% 49.7%
71.3% 28.7%
72.4% 27.6%
62.5% 37.5%
65.7% 34.3%
73.4% 26.6%
73.1% 26.9%
75.9% 24.1%
76.1% 23.9%
3.37 3.31, 3.42
3.06 3.03, 3.09
3.31 3.28, 3.33
3.21 3.15, 3.27
3.16 3.11, 3.21
3.11 3.09, 3.14
3.12 3.10, 3.15
3.33 3.28, 3.39
Notes: obs = observations. Table 2 Movement between male dominated, female dominated or gender neutral occupations, by gender.
Gender neutral Female dominated Male dominated Total movements
Table 3 Mundlak regression model, gender composition of occupations and mental health, stratified by gender, HILDA, 2001 to 2014.
Males (total persons = 9596)
Females (total persons = 9459)
Coefficient
Persons
% of total
Persons
% of total
Males
3548 2702 7331 13,581
26% 20% 54%
3914 7058 2896 13,868
28% 51% 21%
Within persons (5517 persons, 34,637 observations) Gender neutral 0 Female dominated 0.51 0.02 Male dominated 0.33 −0.06 Between persons (9596 persons, 54,155 observations) Gender neutral 0 Female dominated 0.04 −1.05 Male dominated 0.90 0.08
gender. Below we summarise the main results of the paper by separately discussing the between and within persons effects, as well as the descriptive results of the psychosocial characteristics of male dominated, female dominated and gender neutral occupations.
Lower CI
Upper CI
P value
1.01 0.72
0.042 0.099
1.13 1.72
0.943 0.031
−0.13 −0.03
0.009 0.037
2.24 1.12
0.001 0.971
Females Within persons (5752 persons, 34,950 observations) Gender neutral 0 Female dominated −0.53 −0.94 Male dominated −0.54 −1.04 Between persons (9459 persons, 51,256 observations) Gender neutral 0 Female dominated 1.42 0.59 Male dominated 0.02 −1.08
4.1. Gender dominance in working conditions (Aim 1) Over 20 years ago, the possibility of the ‘glass elevator’ was raised as an outcome of gender diversity in the workplace. This describes the rapid promotion of males in female dominated occupations comparative to females in the same job (Williams, 2013). While our study focuses on occupational skill level rather than promotion, our results suggest that females may also experience a ‘glass elevator’ in maledominated occupations. At the same time, females are much more likely to be working in casual/temporary/precarious jobs, and have lower levels of job control, particularly in female dominated jobs. This is
Notes: models adjusted for age, household structure, household income, household structure, education; Lower CI = lower confidence at 95% significance; Upper CI = upper CI at 95% significance; P value = significance test at 95%. Persons in the within persons analysis represented those persons who changed their occupation over the period 2001 to 2014.
20
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
(n = 50 occupational codes). Hence, there may be a number of different effects within specific 3- or 4-digit occupations (e.g., male nurses, female fire-fighters) that we are not detecting at the aggregate level. Another issue related to our definition of exposure is that it ignores the sequence and direction of change (e.g., whether a person entered into a male dominated occupation and then moved to gender neutral or vice versa) (Clougherty et al., 2011; Elwer et al., 2014), but rather compares average mental health in different exposure categories to the reference. We also did not have data on non-binary aspects of gender, recognising that gender is a complex construct not confined to male/female classifications only. It is also possible that there are generational effects in the influence of occupational gender ratios on mental health, although our exploratory analysis of age as an effect modifier was null. Differential misclassification is also a possibility. For example, it may be that workplace norms around home life may lead to systematic over- or underreporting of work-to-family conflict in jobs with a certain gender composition. The strengths of this study were its large sample size, and rigorous methodological approach to considering within and between person effects. We had limited missing data in the sample, thus we suggest there was minimal selection bias. In saying this, those with missing data were more likely to have lower education and be younger, which impacts our ability to generalise findings to these groups. It is also important to acknowledge that we found relatively small effects on the MHI. These effects will not likely be related to clinical differences in symptomology of depression or anxiety. However, it is important to consider that occupational gender ratio likely operates as a wider contextual factor, and is therefore a more distal (but still important) influence on mental health.
consistent with what has been found in past research (Campos-Serna et al., 2013) and may reflect the typology of female jobs as being focused around hospitality, administrative support or caring, which are more likely to be casualised and have lower occupational status than male dominated jobs (Campos-Serna et al., 2013). Males were much more likely to report work interfering with family than females. This was particularly the case in male dominated jobs, possibly because these are less amenable to balancing the demands of work and family (Maume and Houston, 2001). Females were more likely to report family interfering with work in gender neutral and male-dominated occupations. Again, attitudes towards the balancing of work and family demands may play a role here. 4.2. The association between occupational gender ratio and mental health (Aim 2) Looking at the between persons results, it appears that working in an occupation where your own gender is dominant is associated with better mental health than working in a job that is gender neutral. This finding is similar to what has been found in past research (Elwer et al., 2013; Sobiraj et al., 2015). These results may be explained by Kanter's (1993) theory, which suggests that being part of the majority group of an occupation means that you are subject too less stereotyping and polarisation. We would also argue that individuals who are working in occupations where their own gender is in the majority experience greater endorsement of their work-related gendered-identities (Abrahamsson, 2006; Acker, 1990; Lester, 2008). It is also important to acknowledge the likelihood of self-selection into gender-dominated or neutral occupations, in that workers who identified more strongly with cultural norms around masculinity/femininity may be drawn to more male/female dominated occupations, respectively (Sobiraj et al., 2015).
5. Conclusion In conclusion, the present study suggests that the gender composition of a person's job may have an independent effect on their mental health. This relationship appears to be nuanced and further research is required. In particular, we would suggest the need for further investigation into gender specific selection into occupations upon entry into the labour market, as well as the movement between and selection out of occupations by gender. At the same time, there is a pressing need for a greater understanding of the relationship between gender, employment and working conditions among researchers, policy makers, and employers. These findings add to the growing understanding of gender as a social determinant of health in the work context, which will only grow in importance as more and more women participate in paid employment.
4.3. Within-person differences between gender neutral and male or female dominated occupations and mental health (Aim 3) The fixed effects estimates of our analyses suggested that moving between a gender neutral and female dominated occupation was associated with better mental health for males making this transition. For females, being in either a female or male dominated job was associated with lower mental health compared to when those same individuals were employed in a gender-neutral occupation. While our withinperson results are more methodologically robust than a between persons approach, we must also acknowledge that these results do not generalise to the working population who remain stably employed in one occupational gender ratio category. It is also possible that there are time-varying confounders that we were not able to include in our analysis, such as reasons for changing occupations (e.g., voluntary versus involuntary), industrial sector, life-stage, or area-of-residence. We would encourage further research into the specific reasons for shifting between occupations with different gender compositions. While there is limited longitudinal research for us to compare to our findings to, past research from cross-sectional studies suggest that those who work in non-gender dominant occupations may be substantially different from those who work in occupations where their own gender is dominant. For example, Evans and Steptoe (2002) suggest that male and female nurses reported higher “feminine traits” (emotional expressivity) than male or female accountants, who instead reported greater instrumentality (i.e., traditionally masculine psychological traits) scores, which is more commonly associated with masculinity. This supports our contention that sub-samples in the between versus within-persons analyses are non-comparable.
Contributions The article and design was conceived by AM, who also conducted analysis. TK checked results and all authors contributed to interpretation of results. AM drafted the manuscript with feedback from all authors. All authors contributed to the final draft of the manuscript. Financial support AM is funded by a Victorian Health and Medical Research Fellowship. Conflicts of interest We declare no conflicts of interest. Acknowledgments
4.4. Limitations This paper uses unit record data from the Household, Income and Labour Dynamics in Australia HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of
One limitation of our paper relates to the classification of occupational gender ratio, which was based on broad occupational categories 21
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
Social Services DSS) and is managed by the Melbourne Institute of Applied Economic and Social Research Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either DSS or the Melbourne Institute. The data used in this paper was extracted using the Add-On Package PanelWhiz for Stata. PanelWhiz (http://www.PanelWhiz.eu) was written by Dr. John P. Haisken-DeNew ([email protected]).
PLoS One 9 e110257. Elwer, S., Harryson, L., Bolin, M., Hammarstrom, A., 2013. Patterns of gender equality at workplaces and psychological distress. PLoS One 8, e53246. Elwer, S., Johansson, K., Hammarstrom, A., 2014. Workplace gender composition and psychological distress: the importance of the psychosocial work environment. BMC Publ. Health 14, 241. Evans, O., Steptoe, A., 2002. The contribution of gender-role orientation, work factors and home stressors to psychological well-being and sickness absence in male- and female-dominated occupational groups. Soc. Sci. Med. 54, 481–492. Gill, S.C., Butterworth, P., Rodgers, B., Anstey, K.J., Villamil, E., Melzer, D., 2006. Mental health and the timing of men's retirement. Soc. Psychiatr. Psychiatr. Epidemiol. 41, 515–522. Greene, W.H., 2008. Econometric Analysis. Pearson/Prentice Hall, Upper Saddle River, N.J. Hall, E.M., 1989. Gender, work control, and stress: a theoretical discussion and an empirical test. Int. J. Health Serv. 19, 725–745. Hensing, G., Alexanderson, K., 2004. The association between sex segregation, working conditions, and sickness absence among employed women. Occup. Environ. Med. 61, e7. Hosseinpoor, A.R., Stewart Williams, J., Amin, A., Araujo de Carvalho, I., Beard, J., Boerma, T., et al., 2012. Social determinants of self-reported health in women and men: understanding the role of gender in population health. PLoS One 7, e34799. http://dx.doi.org/10.1371/journal.pone.0034799. Kanter, R.M., 1993. Men and Women of the Corporation, second ed. Harper Collins, Basic Books, New York. Krieger, N., 2003. Genders, sexes, and health: what are the connections–and why does it matter? Int. J. Epidemiol. 32, 652–657. Lester, J., 2008. Performing gender in the workplace. Community Coll. Rev. 35, 277–305. Link, B.G., Phelan, J., 1995. Social conditions as fundamental causes of disease. J. Health Soc. Behav. 80 Spec No:80–94. Mastekaasa, A., 2005. Sickness absence in female- and male-dominated occupations and workplaces. Soc. Sci. Med. 60, 2261–2272. Maume, D.J., Houston, P., 2001. Job segregation and gender differences in work-family spillover among white-collar workers. J. Fam. Econ. Issues 22, 171–189. Phillips, S.P., 2005. Defining and measuring gender: a social determinant of health whose time has come. Int. J. Equity Health 4, 11. Plaisier, I., de Bruijn, J.G., de Graaf, R., ten Have, M., Beekman, A.T., Penninx, B.W., 2007. The contribution of working conditions and social support to the onset of depressive and anxiety disorders among male and female employees. Soc. Sci. Med. 64 (2), 401–410. Preston, J.A., 1999. Occupational gender segregation: trends and explanations. Q. Rev. Econ. Finance 39, 611–624. Rumpf, H.J., Meyer, C., Hapke, U., John, U., 2001. Screening for mental health: validity of the MHI-5 using DSM-IV Axis I psychiatric disorders as gold standard. Psychiatr. Res. 105, 243–253. Simon, R.W., 1995. Gender, Multiple Roles, Role Meaning, and Mental Health. pp. 182. Sobiraj, S., Rigotti, T., Weseler, D., Mohr, G., 2015. Masculinity ideology and psychological strain: considering men's social stressors in female-dominated occupations. Psychol. Men Masc. 16, 54–66. Ware Jr., J.E., 2000. SF-36 health survey update. Spine (Phila Pa 1976) 25 (24), 3130–3139. Watson, N., 2011. Methodology for the HILDA Top-up Sample. Hilda Project Technical Paper Series No 1/11. Melbourne Institute, Melbourne. WHO, 2015. Media Centre: Gender. Fact Sheet N°403. http://www.who.int/ mediacentre/factsheets/fs403/en/. Wilkins, R., 2013. Families, Incomes and Jobs, Volume 8: a Statistical Report on Waves 1 to 10 of the Household, Income and Labour Dynamics in Australia Survey. Melbourne Institute of Applied Economic and Social Research Faculty of Business and Economics, Melbourne. Williams, C.L., 2013. The glass escalator, revisited. Gend. Soc. 27, 609–629. Wooldridge, J.M., 2016. Advanced panel data methods. In: Wooldridge, J.M. (Ed.), Introductory Econometrics: a Modern Approach, 6th Edition. (pp. 434–498). Cengage Learning, Boston, MA. Yamazaki, S., Fukuhara, S., Green, J., 2005. Usefulness of five-item and three-item Mental Health Inventories to screen for depressive symptoms in the general population of Japan. Health Qual. Life Outcome 3, 48.
Appendix A. Supplementary data Supplementary data related to this article can be found at http://dx. doi.org/10.1016/j.socscimed.2018.03.020. Footnote page ABS Australian Bureau of Statistics ANZSCO Australian and New Zealand Standard Classification of Occupations HILDA Household Income Labour Dynamics in Australia MHI-5 Mental Health Inventory- 5 References Abrahamsson, L., 2006. Exploring construction of gendered identities at work. In: Billett, S., Fenwick, T., Somerville, M. (Eds.), Work, Subjectivity and Learning: Understanding Learning through Working Life (pp. 105–121). Springer Netherlands, Dordrecht. Acker, J., 1990. Hierarchies, Jobs, Bodies: a Theory of Gendered Organizations. pp. 139. ANZSCO, 2009. Australian and New Zealand Standard Classification of Occupations, First Edition, Revision 1. Australian Bureau of Statistics, Canberra. Australian Bureau of Statistics, 2006. 2006 Census - Labour Force. Occupation 06 (ANZSCO) (OCC06P) - 2 Digit by Sex Male/Female (SEXP) Counting: Persons, Place of Usual Residence. Viewed 1 April 2017. Berwick, D.M., Murphy, J.M., Goldman, P.A., Ware Jr., J.E., Barsky, A.J., Weinstein, M.C., 1991. Performance of a five-item mental health screening test. Med. Care 29, 169–176. Bildt, C., Michélsen, H., 2002. Gender differences in the effects from working conditions on mental health: a 4-year follow-up. Int. Arch. Occup. Environ. Health 75, 252–258. Blalock, H.M., 1967. Toward a Theory of Minority-group Relations. Wiley, New York. Blau, F.D., Kahn, L.M., 2016. The Gender Wage Gap: Extent, Trends, and Explanations. The Institute for the Study of Labor IZA DP No. 9656. Butterworth, P., Leach, L.S., Strazdins, L., Olesen, S.C., Rodgers, B., Broom, D.H., 2011. The psychosocial quality of work determines whether employment has benefits for mental health: results from a longitudinal national household panel survey. Occup. Environ. Med. 68, 806–812. Campos-Serna, J., Ronda-Pérez, E., Artazcoz, L., Moen, B.E., Benavides, F.G., 2013. Gender inequalities in occupational health related to the unequal distribution of working and employment conditions: a systematic review. Int. J. Equity Health 12, 57. http://dx.doi.org/10.1186/1475-9276-12-57. Clougherty, J.E., Eisen, E.A., Slade, M.D., Kawachi, I., Cullen, M.R., 2011. Gender and sex differences in job status and hypertension. Occup. Environ. Med. 68, 16–23. Contopoulos-Ioannidis, D.G., Karvouni, A., Kouri, I., Ioannidis, J.P.A., 2009. Reporting and interpretation of SF-36 outcomes in randomised trials: systematic review. BMJ 338, a3006. Cuijpers, P., Smits, N., Donker, T., ten Have, M., de Graaf, R., 2009. Screening for mood and anxiety disorders with the five-item, the three-item, and the two-item Mental Health Inventory. Psychiatr. Res. 168, 250–255. Dieleman, J.L., Templin, T., 2014. Random-effects, fixed-effects and the within-between specification for clustered data in observational health studies: a simulation study.
22
Contents lists available at ScienceDirect
Social Science & Medicine journal homepage: www.elsevier.com/locate/socscimed
Men’s work, Women’s work, and mental health: A longitudinal investigation of the relationship between the gender composition of occupations and mental health
T
Allison Milnera,b,∗, Tania Kinga, Anthony D. LaMontagneb, Rebecca Bentleya, Anne Kavanagha a b
Centre for Health Equity, Melbourne School of Population and Global Health, The University of Melbourne, Melbourne, Australia Population Health Research Centre, School of Health & Social Development, Deakin University, Geelong, Australia
A R T I C LE I N FO
A B S T R A C T
Keywords: Gender Work Mental health Occupation Job stressors
This longitudinal investigation assesses the extent to which the gender composition of an occupation (e.g., the extent to which an occupation is comprised of males versus females) has an impact on mental health. We used 14 annual waves of the Household Income Labour Dynamics in Australia (HILDA) study to construct a measure representing the gender ratio of an occupation. The outcome measure was the Mental Health Inventory (MHI-5). A Mundlak model was used to compare within and between person effects, after controlling for possible confounders. Results suggest that males and females employed in occupations where their own gender was dominant had better mental health than those in gender-neutral occupations (between person effects). However, within-person results suggested that a movement from a gender-neutral to a male or female dominated occupation was associated with both a decline (females) and improvement (males) in mental health. These results highlight the need for more research on gender specific selection into and out of different occupations in order to progress understandings of gender as a social determinant of health in the work context.
1. Introduction Gender is as an important social determinant of health (Krieger, 2003; Phillips, 2005) constructed through norms, roles and relationships within and between groups of women and men (WHO, 2015). This is separate from sex, defined as a biological construct premised upon biological characteristics enabling sexual reproduction (Phillips, 2005). Gender interacts with other social determinants of health, including education and income, as well as employment and working experiences (Hosseinpoor et al., 2012), and may influence health states directly as well as through interaction with other social determinants. Thus, gender may be considered as a fundamental cause of health and health problems (Link and Phelan, 1995). In the employed population, there is gender patterning across different occupations. For example, in Australia (as in many high income countries), a greater proportion of men are employed in construction related jobs (about 20% of employed males and 4% of employed females), or in higher levels of management (about 16% of employed males and about 9% of employed females) (Australian Bureau of Statistics, 2006). In contrast, females are more likely to be employed in nursing (about 90% of nurses are female) and secretarial work (about
∗
23% of employed females and 6% of employed males) (Australian Bureau of Statistics, 2006). Gender segregation of the workforce first came to the attention of social researchers in the late nineteenth century (Preston, 1999). This phenomenon has persisted across countries and over time. Since the 1960s (which is when women starting entering the workforce in substantial numbers), there has been persistent gender segregation of women into clerical, sales and service occupations in industrialised nations (Preston, 1999). The nuances of gender in the workforce have generally been ignored in epidemiology, which has traditionally focused on working conditions (employment arrangements, working hours, psychosocial stressors) (Bildt and Michélsen, 2002; Plaisier et al., 2007). However, there is some evidence from a limited number of studies that the gender composition of a job (e.g., the extent to which a job is comprised of males versus females) has an impact on health (Elwer et al., 2013; Elwer et al., 2014; Evans and Steptoe, 2002; Hall, 1989; Hensing and Alexanderson, 2004; Mastekaasa, 2005; Sobiraj et al., 2015). In general, this evidence has suggested that working in a job where the other gender is dominant (e.g., males working in a female dominated occupation, and females working in a male dominated occupation) may have damaging effects on psychological health (Elwer et al., 2013; Sobiraj et al., 2015) and be
Corresponding author. Centre for Health Equity; School of Population and Global Health; University of Melbourne; 207 Bouverie Street; Melbourne, 3010, Australia. E-mail address: [email protected] (A. Milner).
https://doi.org/10.1016/j.socscimed.2018.03.020 Received 2 December 2017; Received in revised form 7 March 2018; Accepted 10 March 2018 Available online 12 March 2018 0277-9536/ © 2018 Elsevier Ltd. All rights reserved.
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
annually from over 13,000 individuals within over 7000 households (Wilkins, 2013). The response rate to wave 1 was 66% (Wilkins, 2013). The survey covers a range of dimensions including social, demographic, health and economic conditions using a combination of face-to-face interviews with trained interviewers and a self-completion questionnaire. The initial wave of the survey began with a large national probability sample of Australian households occupying private dwellings (Wilkins, 2013). Interviews were sought in later waves with all persons in sample households who turned 15 years of age. Additional persons have been added to the sample as a result of changes in household composition. Inclusion of these new households is the main way in which the HILDA survey maintains sample representativeness. A top-up sample of 2000 people was added to the cohort in 2011 to allow better representation of the Australian population using the same methodology as the original sample (i.e., a three-stage area-based design) (Watson, 2011). The retention rates for the HILDA survey are above 90% for respondents who have continued in the survey and above 70% for new respondents being invited into the study (Wilkins, 2013). The main variables examined in this study were available in all annual waves of HILDA (2001–2014).
associated with higher sickness absence (Evans and Steptoe, 2002; Hensing and Alexanderson, 2004). Explanations for this have drawn on Kanter's (1993) theory of being in a minority demographic group. Kanter (1993) argues that being in a minority (e.g., female in a male dominated occupation) may be particularly damaging because the affected individual may have heightened visibility and thus be subject to stereotyping. Minority status at work might also affect mental health through mechanisms that include differential working conditions and pay (Blau and Kahn, 2016). Another theory posited by Blalock (1967) focuses on the dominance of the majority group and extent to which the minority group can be considered a threat to power and resources. Thus, as women increase in numbers in male-dominated jobs, they may experience poorer treatment, worse conditions and greater discrimination. Although this theory has also been used to describe the dynamics of gender at work, it is important to acknowledge that this theory was developed to explain race relations. Another perspective is that women and men employed in occupations where the other gender is dominant may experience gender-role conflict because they deviate from normative work-arrangements for male and females (Simon, 1995). A limitation of most past research on the gender composition of the workforce and health is that it has been cross-sectional and/or has not controlled for within person (time invariant) influences (Evans and Steptoe, 2002; Hensing and Alexanderson, 2004; Mastekaasa, 2005; Sobiraj et al., 2015). This is problematic as a comparison between persons (e.g., a female employed in a male dominated versus female dominated occupation) may produce substantially different estimates compared to those that can be found within persons (e.g., a person who changes between a male and female dominated occupation). The analytic approach used in this paper enables us to estimate the differences in mental health between groups defined by the gender dominance of their occupation relative to their own gender. It also allows us to examine how changing the gender dominance of a person's occupation impacts on mental health within-persons, thus capturing the dynamic relationship between gender and work environment. Using 14 waves of longitudinal data from an Australian working population cohort, we create and describe an occupational gender ratio measure across a range of individual and job characteristics (Aim 1). We then assess the association between the occupational gender ratio and mental health, adjusting for known confounders (Aim 2). Following this, we test if changes in mental health occur for people who change from a gender neutral to a male or female dominated occupation across the 14-year study period (Aim 3). Last, we assess whether the relationship between occupational gender ratio and mental health is modified by a person's own gender (Aim 4). This is important considering gender differences in the working conditions and the overall prevalence of common mental health problems, as women are more likely to report mental health problems than men (WHO, 2015). A major contribution of this paper is to build understanding of the role that gendered contexts have in influencing health outcomes, thereby expanding the conceptualisation of gender from being an individual influence on health (e.g., a person's gender), to an environmental influence (e.g., normative expressions of gendered behaviours at work). From a public health perspective, this paper will provide information on whether the gender composition of a person's job may have an independent effect on their mental health. If so, then this would provide a rationale for targeted prevention initiatives in male or female dominated occupations.
2.2. Outcome variable Mental health was assessed using the five-item Mental Health Inventory (MHI-5), a subscale from the Short Form-36 (SF-36) general health measure. The MHI-5 assesses symptoms of depression and anxiety (nervousness, depressed affect) and positive aspects of mental health (feeling calm, happy) in the past 4 weeks. The MHI-5 has reasonable validity and is an effective screening instrument for mood disorders or severe depressive symptomatology in the general population (Gill et al., 2006; Rumpf et al., 2001; Yamazaki et al., 2005) and has been validated as a measure for depression using clinical interviews as the gold standard (Berwick et al., 1991; Cuijpers et al., 2009; Rumpf et al., 2001). The current analyses use the continuous MHI-5 score (scale 1 to 100), with higher scores representing better mental health. Although there is no universally accepted translation of MHI-5 score difference to clinical meaningfulness, a difference of three points on the norm based scale (T-score) has been suggested to reflect a minimally important difference (Ware, 2000), and a difference of four or more on the unstandardised scale has been characterised as indicating a moderate clinically significant effect (Contopoulos-Ioannidis et al., 2009). 2.3. Exposure variable: occupational gender ratio We constructed a measure of whether an occupation was male dominated, female dominated or gender-neutral based on the 2006 census population level statistics from the Australian Bureau of Statistics (ABS) (Australian Bureau of Statistics, 2006). We used the Australian and New Zealand Standard Classification of Occupations (ANZSCO) two-digit occupation (n = 50 occupations) (ANZSCO, 2009), which was the most detailed occupational data available from HILDA. The first step in creating the exposure variable was to calculate the ratio of males to females in each occupational group using the census data. This was a continuous variable running from 0.02 to 81.39. As seen in Supplementary Table 1, there were only 0.02 males to 1 female in the occupation “Personal Assistants and Secretaries”. This was the most heavily female dominant occupation in the exposure. The most heavily male dominated occupation was “Automotive and Engineering Trades Workers” where there were 81.39 males to 1 female. Next, we created a three-level variable representing gender neutral, male dominated or female dominated occupation. If there were 0.50 or fewer males to 1 female ratio in an occupation, then it was classified as female dominated. If there were more than 1.50 males to 1 female in the occupation, then it was classified as male dominated. We created an alternate five level measure based on the quintiles of
2. Methods 2.1. Data source The Household, Income and Labour Dynamics in Australia (HILDA) survey is a longitudinal, nationally representative study of Australian households established in 2001. It collects detailed information 17
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
family conflict, and family to work conflict) by occupational gender ratio and gender.
the continuous measure: heavily female dominated (0–0.43 males to females), moderately female dominated (0.44–0.99 males to females), gender neutral (1.0–1.99 males to females), moderately male dominated (2–2.69 males to females), and heavily male dominated (2.7 or more males to females) and assessed this in sensitivity analysis.
2.5.2. Regression approach Random-effects models are commonly applied to panel data to assess the influence of variables that change over time as well as fixed characteristics associated with an individual (e.g., person stable characteristics, such as gender or early childhood experiences) (Wooldridge, 2016). These models are an advancement on Ordinary Least Square (pooled) models (OLS) as they are able to control for the possibility that residuals in a sample may not be distributed with the same variance (Dieleman and Templin, 2014). However, a central assumption of random-effects models is that the variation between persons is random and uncorrelated with exposure variables included in the model (Greene, 2008). Thus, these models may be biased if the person-level effects are not independent of the included exposure variables. To address this problem and answer Aim 2 and 3 we used randomeffects longitudinal regression models with the Mundlak adjustment, which includes group-means of time-varying variables in the model. While this ostensibly ensures the assumption that observed and unobserved variables are uncorrelated, it also provides the added and significant benefit of separating out “within” and “between” person effects. This means that, in addition to assessing the association between our exposure and mental health (Aim 2), we can examine the consequences for mental health of a person changing exposure status over time (Aim 3) in a single model. This within person analysis also enables us to reduce time invariant confounding and health selection as each person acts as their own control. So, issues such as personality characteristics and stable low levels of mental health are effectively controlled for. Evidence suggests that the results of traditional fixed effects models and Mundlak models are generally very similar (Dieleman and Templin, 2014). These models are different from multi-level models, which are used to assess individuals nested in within higher level units such as schools, or schools nested in a region. The formula associated with our Mundlak model is below (1)
2.4. Other variables For the descriptive analysis, we examined the gender (males and females) and age (18–24, 25 to 29, 30 to 34, 35 to 44, 45 to 54 and 55–64 years) of workers in female/male dominated and gender neutral occupations. We also examined a range of psychosocial job stressors (psychological job demands, job control, job security, and fairness of pay). These variables were previously developed as part of an index of psychosocial job quality (Butterworth et al., 2011) and are expressed as binary variables (where 0 indicates not exposed and 1 indicates exposed). Other characteristics examined included: The skill level of jobs defined according to the ANZSCO one-digit occupation (ANZSCO, 2009), where “3” can be considered as the most skilled and “1” can be considered the least skilled; employment arrangement (permanent, casual or labour hire, fixed-term contract or self-employed), and; the degree to which workers in male dominated, female dominated or gender neutral occupations experienced either work to family conflict (“Because of the requirements of my job, my family time is less enjoyable and more pressured”) or family to work conflict (“Because of my family responsibilities, the time I spend working is less enjoyable and more pressured”). Both of these items were scaled from 1 (strongly disagree) to 7 (strongly agree). For our main research question, we controlled for factors that were likely to represent common causes of mental health and occupational gender ratio. These time-varying variables included: age group, education (less than year 12 (high school), year 12, diploma or certificate, bachelor degree or higher), household structure (couple without children, couple with children, lone parent with children, lone person, and other), and weekly household income (equalised). Other variables such as psychosocial job stressors, employment arrangements, work/family conflict were seen as possible mediators of the relationship between employment in a job of a certain gender orientation and mental health, and were thus excluded. Gender was included as our effect modifier. The Directed Acyclical Graph (DAG) for Aims 2 to 4 can be seen in Fig. 1 (acknowledging there is no consensus on how to represent effect modification in a DAG, gender has been represented as a confounder).
yjn = β (x jn − x j ) + yx j + ωjn
(1)
Where x j is the individual level mean of exposure and confounders over time, and n is time. This shows that β are then mean-differenced values. ωjn is comprised of individual specific and random error. We assessed the effects of the three level occupational gender ratio in our main analysis. As a sensitivity analysis, we conducted a Mundlak model using the five-level occupational gender ratio. As a further sensitivity analysis, we assessed the role of unemployment and being not in the labour force on the relationship between occupational gender ratio and mental health, recognising that the experience of being out of work may influence employment experiences and mental health.
2.5. Analytic approach 2.5.1. Descriptive analysis We conducted cross-tabulations (frequencies and percentages) of the categorical variables and means of continuous variables (work to
2.5.3. Effect modification We tested effect modification by including an interaction term between occupational gender ratio (male-dominated, female-dominated, gender neutral) and gender in separate fixed-effects models. The significance of this relationship was examined by assessing the significance of the interaction term (i.e. beta coefficient) and via investigation of a likelihood ratio test, which examined a model with the interaction term and main effects compared to one that includes main effects only. Results suggested that the relationship between the occupational gender ratio and mental health was different for males and females (Aim 4). Compared to males in gender-neutral occupations, the coefficient for females in female dominated occupations was −1.08 (95% −1.67 to −0.49, P < 0.001) and the coefficient for females in male dominated occupations was −0.86 (95% −1.44 to −0.28, P = 0.004). Results of the likelihood ratio test were also significant (LR chi2(2) = 18.81, P < 0.001). On the basis of this, we stratified
Fig. 1. Directed acyclical graph.
18
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
in either male dominated or gender neutral occupations. This table also shows that there was more casual (i.e., temporary) employment in the female dominated occupations. Those in male or gender neutral occupations were more likely to be self-employed. As can be seen, those in female dominated occupations tended to report lower job control, while male and gender neutral occupations were more likely to report high job insecurity. Those in gender equal and female dominated occupations reported higher levels of work interfering with family, and family interfering with work, than male dominated occupations. It is also worth noting that males in female dominated occupations were slightly over-represented at the highest skill level compared to females. The opposite was true for male dominated occupations (where a higher proportion of those females employed in male-dominated occupations were employed at higher skill levels). About 11,269 of the people in HILDA (56%) changed occupation over the period 2001 to 2014. A description of the occupational gender ratio can be seen in Table 2 for persons (not observations). This describes the shifts between the different categories of the occupational gender ratio by gender. As can be seen, men spent most of their contributed waves in male dominated occupations, while females spent most of their contributed waves in female dominated occupations. 3.2. Analytic results regarding the effect of the occupational gender ratio on mental health Results of adjusted coefficients from the Mundlak model can be seen in Table 2. There are notable differences when results are decomposed within (Aim 3) and between persons (Aim 2) and by gender (Aim 4). For both males and females, concordance between gender and occupation dominance was associated with better health (Aim 2, Table 3). Men in male dominated occupations had better mental health than men in gender-neutral occupations (0.90, 95% CI 0.08 to 1.72, P = 0.031) and females working in a female dominated occupation had better mental health than a female working in a gender-neutral occupation (1.42, 95% CI 0.59 to 2.24, P = 0.001). Different gender effects were observed with changes in occupation (Aim 3). This within persons approach compares average mental health for all time periods when a person was employed in a gender-neutral occupation to average mental for the time periods when that same person was employed in a male or female dominated occupation. Compared to when they were employed in gender neutral occupations, men in female dominated occupations had slightly better mental health (0.51, 95% CI 0.02 to 1.01, P = 0.042). Women who changed occupational gender ratio groups had slightly worse mental health in male or female dominated compared to when they were employed in genderneutral occupations (−0.54, 95% CI -1.04 to −0.03, P = 0.037 and −0.53, 95% CI -0.94 to −0.13, P = 0.009, respectively). The sensitivity analysis using the five level occupational gender ratio (heavily female dominated; moderately female dominated; gender neutral; moderately male dominated; and heavily male dominated) demonstrated similar results as described above (Supplementary Table 2). Sensitivity analysis assessing the effect of unemployment and not being in the labour force can be seen in Supplementary Table 3. Once unemployment and not in the labour force were controlled for, there was a weaker effect of the occupational gender ratio on mental health (however size of coefficients and significance are similar).
Fig. 2. Flow chart describing entry into the analytic sample.
analyses by males and females. 3. Results Fig. 2 describes the sample selection into the study. Of the 20,231 employed people eligible for this analysis, 1176 had missing data leaving a final sample of 19,055 (Aim 2). Cases with missing data were slightly more likely to be younger and lower education, although differences from participants with non-missing data were non-significant. After restricting to those who changed occupation (Aim 3), there was 11,269 people remaining.
4. Discussion This study indicates that: 1) across the working population of the HILDA cohort, being employed in an occupation where your own gender is dominant is associated with slightly higher mental health; 2) among the people who have changed jobs and occupational gender ratio categories, the shift into an occupation where the other gender is dominant is associated with varying (small) effects that are modified by
3.1. Descriptive results of the characteristics of male, female or gender neutral occupations Table 1 describes characteristics of those employed in male, female or gender neutral occupations (Aim 1). People in female dominated occupations tended to be younger (20.24% are 18–24 years) than those 19
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
Table 1 Characteristics of those employed in male dominated, female dominated or gender neutral occupations. Gender
Gender neutral Male obs = 11,546 (49%)
Female dominated Female obs = 12,010 (51%)
Age 18 to 24 years 12% 13.7% 25 to 29 years 11.1% 12.7% 30 to 34 years 11.8% 11.1% 35 to 44 years 24.6% 25.8% 45 to 54 years 24.9% 24.2% 55 to 64 years 15.5% 12.5% Occupational skill level Low 9.9% 16.4% Medium 31.3% 28.5% High 58.8% 55.1% Employment arrangement Casual/labour hire 10.5% 18.1% Fixed term 7% 8.5% Permanent 59% 55% Self employed 23.5% 18.5% High job demands and complexity No 46.7% 53.9% Yes 53.3% 46.1% Low job control No 56.8% 51% Yes 43.2% 49% High job insecurity No 64.2% 66.8% Yes 35.7% 33.2% Low fairness of pay No 76.5% 75.2% Yes 23.5% 24.8% Work interfering with family (1 = strongly disagree, 7 = strongly agree) Mean 3.30 3.17 (95% CI) 3.25, 3.35 3.12, 3.22 Family interfering with work (1 = strongly disagree, 7 = strongly agree) Mean 3.13 3.30 (95% CI) 3.09, 3.18 3.25, 3.35
Male dominated
Male obs = 8252 (21%)
Female obs = 30,481 (78%)
Male obs = 34,357 (80%)
Female obs = 8765 (20%)
22% 10.6% 10.6% 21% 21.5% 14.2%
18.6% 9.9% 9.6% 24.7% 24.8% 12.5%
13.2% 11.1% 11.3% 26.4% 24.3% 13.6%
9.8% 11.6% 12.4% 27.2% 25.9% 13%
24.1% 42.2% 33.7%
15.6% 56.8% 27.6%
32.5% 35.5% 32%
31.1% 18.4% 50.5%
22.6% 9.6% 60.8% 7.1%
24.5% 8.9% 59.9% 6.8%
11.7% 6.4% 61.0% 20.1%
16.7% 8.1% 59.4% 15.9%
48.1% 51.9%
54.1% 45.9%
46.5% 53.5%
49% 51%
39% 61%
35% 65%
51.6% 48.4%
50.3% 49.7%
71.3% 28.7%
72.4% 27.6%
62.5% 37.5%
65.7% 34.3%
73.4% 26.6%
73.1% 26.9%
75.9% 24.1%
76.1% 23.9%
3.37 3.31, 3.42
3.06 3.03, 3.09
3.31 3.28, 3.33
3.21 3.15, 3.27
3.16 3.11, 3.21
3.11 3.09, 3.14
3.12 3.10, 3.15
3.33 3.28, 3.39
Notes: obs = observations. Table 2 Movement between male dominated, female dominated or gender neutral occupations, by gender.
Gender neutral Female dominated Male dominated Total movements
Table 3 Mundlak regression model, gender composition of occupations and mental health, stratified by gender, HILDA, 2001 to 2014.
Males (total persons = 9596)
Females (total persons = 9459)
Coefficient
Persons
% of total
Persons
% of total
Males
3548 2702 7331 13,581
26% 20% 54%
3914 7058 2896 13,868
28% 51% 21%
Within persons (5517 persons, 34,637 observations) Gender neutral 0 Female dominated 0.51 0.02 Male dominated 0.33 −0.06 Between persons (9596 persons, 54,155 observations) Gender neutral 0 Female dominated 0.04 −1.05 Male dominated 0.90 0.08
gender. Below we summarise the main results of the paper by separately discussing the between and within persons effects, as well as the descriptive results of the psychosocial characteristics of male dominated, female dominated and gender neutral occupations.
Lower CI
Upper CI
P value
1.01 0.72
0.042 0.099
1.13 1.72
0.943 0.031
−0.13 −0.03
0.009 0.037
2.24 1.12
0.001 0.971
Females Within persons (5752 persons, 34,950 observations) Gender neutral 0 Female dominated −0.53 −0.94 Male dominated −0.54 −1.04 Between persons (9459 persons, 51,256 observations) Gender neutral 0 Female dominated 1.42 0.59 Male dominated 0.02 −1.08
4.1. Gender dominance in working conditions (Aim 1) Over 20 years ago, the possibility of the ‘glass elevator’ was raised as an outcome of gender diversity in the workplace. This describes the rapid promotion of males in female dominated occupations comparative to females in the same job (Williams, 2013). While our study focuses on occupational skill level rather than promotion, our results suggest that females may also experience a ‘glass elevator’ in maledominated occupations. At the same time, females are much more likely to be working in casual/temporary/precarious jobs, and have lower levels of job control, particularly in female dominated jobs. This is
Notes: models adjusted for age, household structure, household income, household structure, education; Lower CI = lower confidence at 95% significance; Upper CI = upper CI at 95% significance; P value = significance test at 95%. Persons in the within persons analysis represented those persons who changed their occupation over the period 2001 to 2014.
20
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
(n = 50 occupational codes). Hence, there may be a number of different effects within specific 3- or 4-digit occupations (e.g., male nurses, female fire-fighters) that we are not detecting at the aggregate level. Another issue related to our definition of exposure is that it ignores the sequence and direction of change (e.g., whether a person entered into a male dominated occupation and then moved to gender neutral or vice versa) (Clougherty et al., 2011; Elwer et al., 2014), but rather compares average mental health in different exposure categories to the reference. We also did not have data on non-binary aspects of gender, recognising that gender is a complex construct not confined to male/female classifications only. It is also possible that there are generational effects in the influence of occupational gender ratios on mental health, although our exploratory analysis of age as an effect modifier was null. Differential misclassification is also a possibility. For example, it may be that workplace norms around home life may lead to systematic over- or underreporting of work-to-family conflict in jobs with a certain gender composition. The strengths of this study were its large sample size, and rigorous methodological approach to considering within and between person effects. We had limited missing data in the sample, thus we suggest there was minimal selection bias. In saying this, those with missing data were more likely to have lower education and be younger, which impacts our ability to generalise findings to these groups. It is also important to acknowledge that we found relatively small effects on the MHI. These effects will not likely be related to clinical differences in symptomology of depression or anxiety. However, it is important to consider that occupational gender ratio likely operates as a wider contextual factor, and is therefore a more distal (but still important) influence on mental health.
consistent with what has been found in past research (Campos-Serna et al., 2013) and may reflect the typology of female jobs as being focused around hospitality, administrative support or caring, which are more likely to be casualised and have lower occupational status than male dominated jobs (Campos-Serna et al., 2013). Males were much more likely to report work interfering with family than females. This was particularly the case in male dominated jobs, possibly because these are less amenable to balancing the demands of work and family (Maume and Houston, 2001). Females were more likely to report family interfering with work in gender neutral and male-dominated occupations. Again, attitudes towards the balancing of work and family demands may play a role here. 4.2. The association between occupational gender ratio and mental health (Aim 2) Looking at the between persons results, it appears that working in an occupation where your own gender is dominant is associated with better mental health than working in a job that is gender neutral. This finding is similar to what has been found in past research (Elwer et al., 2013; Sobiraj et al., 2015). These results may be explained by Kanter's (1993) theory, which suggests that being part of the majority group of an occupation means that you are subject too less stereotyping and polarisation. We would also argue that individuals who are working in occupations where their own gender is in the majority experience greater endorsement of their work-related gendered-identities (Abrahamsson, 2006; Acker, 1990; Lester, 2008). It is also important to acknowledge the likelihood of self-selection into gender-dominated or neutral occupations, in that workers who identified more strongly with cultural norms around masculinity/femininity may be drawn to more male/female dominated occupations, respectively (Sobiraj et al., 2015).
5. Conclusion In conclusion, the present study suggests that the gender composition of a person's job may have an independent effect on their mental health. This relationship appears to be nuanced and further research is required. In particular, we would suggest the need for further investigation into gender specific selection into occupations upon entry into the labour market, as well as the movement between and selection out of occupations by gender. At the same time, there is a pressing need for a greater understanding of the relationship between gender, employment and working conditions among researchers, policy makers, and employers. These findings add to the growing understanding of gender as a social determinant of health in the work context, which will only grow in importance as more and more women participate in paid employment.
4.3. Within-person differences between gender neutral and male or female dominated occupations and mental health (Aim 3) The fixed effects estimates of our analyses suggested that moving between a gender neutral and female dominated occupation was associated with better mental health for males making this transition. For females, being in either a female or male dominated job was associated with lower mental health compared to when those same individuals were employed in a gender-neutral occupation. While our withinperson results are more methodologically robust than a between persons approach, we must also acknowledge that these results do not generalise to the working population who remain stably employed in one occupational gender ratio category. It is also possible that there are time-varying confounders that we were not able to include in our analysis, such as reasons for changing occupations (e.g., voluntary versus involuntary), industrial sector, life-stage, or area-of-residence. We would encourage further research into the specific reasons for shifting between occupations with different gender compositions. While there is limited longitudinal research for us to compare to our findings to, past research from cross-sectional studies suggest that those who work in non-gender dominant occupations may be substantially different from those who work in occupations where their own gender is dominant. For example, Evans and Steptoe (2002) suggest that male and female nurses reported higher “feminine traits” (emotional expressivity) than male or female accountants, who instead reported greater instrumentality (i.e., traditionally masculine psychological traits) scores, which is more commonly associated with masculinity. This supports our contention that sub-samples in the between versus within-persons analyses are non-comparable.
Contributions The article and design was conceived by AM, who also conducted analysis. TK checked results and all authors contributed to interpretation of results. AM drafted the manuscript with feedback from all authors. All authors contributed to the final draft of the manuscript. Financial support AM is funded by a Victorian Health and Medical Research Fellowship. Conflicts of interest We declare no conflicts of interest. Acknowledgments
4.4. Limitations This paper uses unit record data from the Household, Income and Labour Dynamics in Australia HILDA) Survey. The HILDA Project was initiated and is funded by the Australian Government Department of
One limitation of our paper relates to the classification of occupational gender ratio, which was based on broad occupational categories 21
Social Science & Medicine 204 (2018) 16–22
A. Milner et al.
Social Services DSS) and is managed by the Melbourne Institute of Applied Economic and Social Research Melbourne Institute). The findings and views reported in this paper, however, are those of the author and should not be attributed to either DSS or the Melbourne Institute. The data used in this paper was extracted using the Add-On Package PanelWhiz for Stata. PanelWhiz (http://www.PanelWhiz.eu) was written by Dr. John P. Haisken-DeNew ([email protected]).
PLoS One 9 e110257. Elwer, S., Harryson, L., Bolin, M., Hammarstrom, A., 2013. Patterns of gender equality at workplaces and psychological distress. PLoS One 8, e53246. Elwer, S., Johansson, K., Hammarstrom, A., 2014. Workplace gender composition and psychological distress: the importance of the psychosocial work environment. BMC Publ. Health 14, 241. Evans, O., Steptoe, A., 2002. The contribution of gender-role orientation, work factors and home stressors to psychological well-being and sickness absence in male- and female-dominated occupational groups. Soc. Sci. Med. 54, 481–492. Gill, S.C., Butterworth, P., Rodgers, B., Anstey, K.J., Villamil, E., Melzer, D., 2006. Mental health and the timing of men's retirement. Soc. Psychiatr. Psychiatr. Epidemiol. 41, 515–522. Greene, W.H., 2008. Econometric Analysis. Pearson/Prentice Hall, Upper Saddle River, N.J. Hall, E.M., 1989. Gender, work control, and stress: a theoretical discussion and an empirical test. Int. J. Health Serv. 19, 725–745. Hensing, G., Alexanderson, K., 2004. The association between sex segregation, working conditions, and sickness absence among employed women. Occup. Environ. Med. 61, e7. Hosseinpoor, A.R., Stewart Williams, J., Amin, A., Araujo de Carvalho, I., Beard, J., Boerma, T., et al., 2012. Social determinants of self-reported health in women and men: understanding the role of gender in population health. PLoS One 7, e34799. http://dx.doi.org/10.1371/journal.pone.0034799. Kanter, R.M., 1993. Men and Women of the Corporation, second ed. Harper Collins, Basic Books, New York. Krieger, N., 2003. Genders, sexes, and health: what are the connections–and why does it matter? Int. J. Epidemiol. 32, 652–657. Lester, J., 2008. Performing gender in the workplace. Community Coll. Rev. 35, 277–305. Link, B.G., Phelan, J., 1995. Social conditions as fundamental causes of disease. J. Health Soc. Behav. 80 Spec No:80–94. Mastekaasa, A., 2005. Sickness absence in female- and male-dominated occupations and workplaces. Soc. Sci. Med. 60, 2261–2272. Maume, D.J., Houston, P., 2001. Job segregation and gender differences in work-family spillover among white-collar workers. J. Fam. Econ. Issues 22, 171–189. Phillips, S.P., 2005. Defining and measuring gender: a social determinant of health whose time has come. Int. J. Equity Health 4, 11. Plaisier, I., de Bruijn, J.G., de Graaf, R., ten Have, M., Beekman, A.T., Penninx, B.W., 2007. The contribution of working conditions and social support to the onset of depressive and anxiety disorders among male and female employees. Soc. Sci. Med. 64 (2), 401–410. Preston, J.A., 1999. Occupational gender segregation: trends and explanations. Q. Rev. Econ. Finance 39, 611–624. Rumpf, H.J., Meyer, C., Hapke, U., John, U., 2001. Screening for mental health: validity of the MHI-5 using DSM-IV Axis I psychiatric disorders as gold standard. Psychiatr. Res. 105, 243–253. Simon, R.W., 1995. Gender, Multiple Roles, Role Meaning, and Mental Health. pp. 182. Sobiraj, S., Rigotti, T., Weseler, D., Mohr, G., 2015. Masculinity ideology and psychological strain: considering men's social stressors in female-dominated occupations. Psychol. Men Masc. 16, 54–66. Ware Jr., J.E., 2000. SF-36 health survey update. Spine (Phila Pa 1976) 25 (24), 3130–3139. Watson, N., 2011. Methodology for the HILDA Top-up Sample. Hilda Project Technical Paper Series No 1/11. Melbourne Institute, Melbourne. WHO, 2015. Media Centre: Gender. Fact Sheet N°403. http://www.who.int/ mediacentre/factsheets/fs403/en/. Wilkins, R., 2013. Families, Incomes and Jobs, Volume 8: a Statistical Report on Waves 1 to 10 of the Household, Income and Labour Dynamics in Australia Survey. Melbourne Institute of Applied Economic and Social Research Faculty of Business and Economics, Melbourne. Williams, C.L., 2013. The glass escalator, revisited. Gend. Soc. 27, 609–629. Wooldridge, J.M., 2016. Advanced panel data methods. In: Wooldridge, J.M. (Ed.), Introductory Econometrics: a Modern Approach, 6th Edition. (pp. 434–498). Cengage Learning, Boston, MA. Yamazaki, S., Fukuhara, S., Green, J., 2005. Usefulness of five-item and three-item Mental Health Inventories to screen for depressive symptoms in the general population of Japan. Health Qual. Life Outcome 3, 48.
Appendix A. Supplementary data Supplementary data related to this article can be found at http://dx. doi.org/10.1016/j.socscimed.2018.03.020. Footnote page ABS Australian Bureau of Statistics ANZSCO Australian and New Zealand Standard Classification of Occupations HILDA Household Income Labour Dynamics in Australia MHI-5 Mental Health Inventory- 5 References Abrahamsson, L., 2006. Exploring construction of gendered identities at work. In: Billett, S., Fenwick, T., Somerville, M. (Eds.), Work, Subjectivity and Learning: Understanding Learning through Working Life (pp. 105–121). Springer Netherlands, Dordrecht. Acker, J., 1990. Hierarchies, Jobs, Bodies: a Theory of Gendered Organizations. pp. 139. ANZSCO, 2009. Australian and New Zealand Standard Classification of Occupations, First Edition, Revision 1. Australian Bureau of Statistics, Canberra. Australian Bureau of Statistics, 2006. 2006 Census - Labour Force. Occupation 06 (ANZSCO) (OCC06P) - 2 Digit by Sex Male/Female (SEXP) Counting: Persons, Place of Usual Residence. Viewed 1 April 2017. Berwick, D.M., Murphy, J.M., Goldman, P.A., Ware Jr., J.E., Barsky, A.J., Weinstein, M.C., 1991. Performance of a five-item mental health screening test. Med. Care 29, 169–176. Bildt, C., Michélsen, H., 2002. Gender differences in the effects from working conditions on mental health: a 4-year follow-up. Int. Arch. Occup. Environ. Health 75, 252–258. Blalock, H.M., 1967. Toward a Theory of Minority-group Relations. Wiley, New York. Blau, F.D., Kahn, L.M., 2016. The Gender Wage Gap: Extent, Trends, and Explanations. The Institute for the Study of Labor IZA DP No. 9656. Butterworth, P., Leach, L.S., Strazdins, L., Olesen, S.C., Rodgers, B., Broom, D.H., 2011. The psychosocial quality of work determines whether employment has benefits for mental health: results from a longitudinal national household panel survey. Occup. Environ. Med. 68, 806–812. Campos-Serna, J., Ronda-Pérez, E., Artazcoz, L., Moen, B.E., Benavides, F.G., 2013. Gender inequalities in occupational health related to the unequal distribution of working and employment conditions: a systematic review. Int. J. Equity Health 12, 57. http://dx.doi.org/10.1186/1475-9276-12-57. Clougherty, J.E., Eisen, E.A., Slade, M.D., Kawachi, I., Cullen, M.R., 2011. Gender and sex differences in job status and hypertension. Occup. Environ. Med. 68, 16–23. Contopoulos-Ioannidis, D.G., Karvouni, A., Kouri, I., Ioannidis, J.P.A., 2009. Reporting and interpretation of SF-36 outcomes in randomised trials: systematic review. BMJ 338, a3006. Cuijpers, P., Smits, N., Donker, T., ten Have, M., de Graaf, R., 2009. Screening for mood and anxiety disorders with the five-item, the three-item, and the two-item Mental Health Inventory. Psychiatr. Res. 168, 250–255. Dieleman, J.L., Templin, T., 2014. Random-effects, fixed-effects and the within-between specification for clustered data in observational health studies: a simulation study.
22