Sample and design

We used data from the Longitudinal Aging Study Amsterdam (LASA). LASA is an ongoing, prospective cohort study, based on a representative sample of the older population in the Netherlands. LASA focuses on the determinants, trajectories and consequences of changes in physical, cognitive, emotional, and social functioning in older adults aged 55years or older. Measurements are conducted approximately every three years and include a main face-to-face computer assisted interview, a face-to-face computer assisted medical interview in which clinical measurements are performed and additional questions are asked, and a self-administered questionnaire. The study received approval by the medical ethics committee of the VU University medical center. Signed informed consent was obtained from all study participants. Sampling, response and procedures are described in detail elsewhere [28].

For the current study, we adopted a lagged-effect design, because we expected that with ageing, older workers would increasingly be affected by their gender role (our main determinant) and working conditions (our moderator/mediator), and that this would result in higher depressive symptoms scores in the course of time. Thus, we assumed a temporal precedence of gender roles and working conditions, as opposed to an immediate effect on depressive symptoms. Accordingly, data from 20122013 (T1) and 20152016 (T2) were used. At T1, 1023 respondents participated in the LASA study. We excluded those who did not have a paid job at T1 (n=395), those who did not participate at T2 (n=93), and those who did not have a paid job at T2 (n=222). We ended up with a sample of 313 older workers.

Our outcome measure was depressive symptoms, measured using the Center for Epidemiologic Studies Depression Scale (CES-D) [29]. The CES-D is a 20-item self-report scale ranging from 0 to 60, with higher scores reflecting more depressive symptoms. The outcome was measured at T2 (2015/2016).

All independent variables were measured at T1 (2012/2013).

We included biological sex, derived from the population registers, as an independent variable.

We constructed a gender index, based on the work of Smith and Koehoorn [9] on gender roles in the labour market. Smith and Koehoorn included four gender items in their index: responsibility for caring for children, occupation segregation, number of working hours, and level of education. Because in our sample of older workers responsibility for caring for children was not applicable, we chose to include a measure of informal caregiving. Providing informal care is much more common among women compared to men and is seen as a more feminine role [30, 31]. As suggested by Smith and Koehoorn, we also included a measure of household responsibilities. Furthermore, Smith and Koehoorn suggested to include a measure for primary earner status. Unfortunately this information was not available in our data. We therefore chose to include income in our index. While Smith and Koehoorn use relative measures (relative to the partner) for educational level and number of working hours, we use absolute measures for these items, because we consider absolute measures to reflect broader societal gender roles rather than gender roles within the household.

The gender index consisted of the sum score of six items: number of working hours, income, occupation segregation, level of education, informal caregiving, and time spent on household chores. For each gender item, a higher score represents more femininity and a lower score represents more masculinity.

Respondents were asked about their number of working hours per week. Responses were categorised into quartiles and recoded so that a higher score (i.e. a lower number of working hours) represents more femininity.

To assess the income of the household, respondents were asked what their monthly household income was, choosing from 24 categories, with the lowest category being 454-567 and the highest category 5446 or more. To ensure comparability of income between persons with and without a partner in the household, income was multiplied by 0.7 for respondents with a partner in the household. The factor 0.7 is the inverse of the squareroot of 2, i.e., the number of household members. This correction makes the incomes of all respondents equivalent to one-person household incomes [32]. Income was categorised into quartiles and recoded so that a higher score (i.e. a lower income) represents more femininity.

Occupation segregation was measured by the percentage of female workers in the sector. Using data from Statistics Netherlands, we assigned each sector to one of four categories in accordance with Smith and Koehoorn [9]: (0)25% female workers, (1) 2650% female workers, (2) 5175% female workers, and (3)76% female workers.

Respondents were asked about their highest completed level of education. We used the International Standard Classification of Education 2011 [33] to categorise educational level into three groups: (0) low (up to lower secondary education, ISCED 02), (1) intermediate (upper secondary education or post-secondary non-tertiary education, ISCED 34), and (2) high (short cycle tertiary and higher, ISCED 56). Again, scores were recoded so that a higher score (i.e. lower educational level) reflects more femininity.

Respondents were asked if they recently provided help with household chores to somebody outside the own household, and whether the respondent provided help with personal care to somebody inside or outside the own household. If so, questions were asked about the intensity (hours) of care. Informal caregiving was categorised into (0) not giving informal care, (1) giving<8h of informal care per week, and (2) giving8h of informal care per week.

Respondents were also asked about the time spent on light and heavy household chores. Time in minutes per day, averaged across the past 14days, was categorised into quartiles.

The gender index ranged from 022 and was dichotomised at the median into masculine (scores 07) and feminine (scores 822) to enable comparison of its association with depressive symptoms with the association of biological sex with depressive symptoms.

We used a written questionnaire to obtain data on working conditions [34]. Respondents could answer (1) never, (2) sometimes, (3) often, or (4) always to all questions on working conditions.

To measure physical demands five items were used: use of force, using tools that cause vibration or shaking, working in an uncomfortable position , standing for a long time, and kneeling down or squatting. Psychological demands consisted of two items: working very fast, and having to do a lot of work. For cognitive demands, six items were used: think of solutions, learn new things, requires creativity, requires thinking intensively, requires focus, and requires attention. Autonomy was measured with three items: control over how to do the job, control over sequence of tasks, and control over when to take time off. For variation in tasks one item was used: having variation in tasks. And for social support four items were included: help and support of colleagues, colleagues willing to listen to work related problems, help and support of supervisor, and supervisor willing to listen to work related problems.

Sum scores were made for each type of working conditions and scores were dichotomised using the median due to non-linearity.

Age was derived from the population registers.

Multiple imputation (MICE) was used to deal with missing values, which were assumed to be missing at random. All independent, control and the outcome variables were included in the imputation process and the number of imputations was set to 30, based on the percentage of missing values (28%) [35]. To assess to what extent the separate gender items as well as the gender index are associated with sex, we conducted logistic regression analyses [36]. We used Structural Equation Modeling (SEM) to estimate the associations visualised in Fig. 1. All analyses were adjusted for age. Separate models were examined for sex and gender. We used tobit regression analyses to estimate the associations of sex/gender and the working conditions with depressive symptoms, because the depressive symptoms scale is skewed to the right due to the detection limit at the lower end of the scale. Tobit models account for this left-censoring by assuming a normal distribution that is cut off (censored) at zero.

Visual representation of the moderation and mediation models

To test whether gender/sex is a moderator in the association between working conditions and depressive symptoms, we built models with an interaction between sex/gender and the working conditions (Fig.1A). In case of a statistically significant interaction, the association between the working conditions and depressive symptoms varies across sexes/genders.

To investigate whether working conditions explain the association between sex/gender and depressive symptoms, we built single mediator analyses (Fig.1B). To estimate the c paths (total effect of sex/gender on depressive symptoms) and the b paths (effect of the mediators on depressive symptoms, while controlling for sex/gender), we used tobit regression analyses, and for the a paths (the effect of sex/gender on the mediators), we conducted logistic regression analyses. We used causal mediation analyses to estimate the indirect effects [37]. We used bootstrapping techniques (500 repetitions) to calculate the 95% confidence intervals around the indirect effects. All analyses were carried out in Stata version 14.

See more here:
Sex and gender differences in depressive symptoms in older workers: the role of working conditions - BMC Public Health - BMC Public Health

Comments are closed.