This is an exploratory study, aiming at identifying possible mechanisms in a population, using the advanced APC analyses. We used annual information from Statistics Sweden’s Longitudinal Integration Database for Health Insurance and Labour Market Studies (LISA) concerning sex, age, type of occupation, emigration, income (from work, unemployment benefit, SA, DP, social security benefits, or student benefit), and number of days with SA/DP benefits. All people in Sweden are included in LISA for all years living here from 1990, including immigrants (however, not people seeking asylum in Sweden (e.g., refugees) before assessed as fulfilling the criteria for asylum). For 1985, we used information from the People and Housing Census of 1985 (FoB85).
Our aim was to study differences between birth cohorts (1929–1983), ages (20–56), and periods (1985–1993, 1990–1998, 2003–2011). Each individual was intended to be followed up 8 years after their occupation was registered, and we have followed the tradition in an age cohort analysis (Keyes et al. 2010). For each of these three periods, we included all people living in Sweden registered as working in an occupation from three population-based cohorts (in 1985 N = 3,183,549; in 1990 N = 3,372,152; and in 2003 N = 3,565,579). Those who had emigrated or died during the 8-year follow-up were excluded from period 1985–1993 (n = 124,819; 3.92%), period 1990–1998 (n = 149,675; 4.44%), and period 2003–2011 (n = 152,509; 4.27%).
The categorizations of level of occupational gender segregation of occupation in 1985, 1990, and 2003, respectively, were based on the gender distribution among all the occupations, at a three-digit level (112 occupations in 2003; 111 in 1985 and 1990). The Standard for Swedish Occupational Classification (SSYK-96) at the three-digit level was used (SSYK-96 closely follows the International Standard Classification of Occupations, ISCO-88 and the ISCO (COM), used in the statistical publications from the European Union (Bettio and Verashchagina 2009). Information about peoples’ occupations at inclusion in respective population cohort was for 1985 obtained from FoB85 and for 1990 and 2003 from LISA. The reason for this selection of years was that LISA had no occupational information for the years 1991–2002. Occupations were grouped into five categories according to whether they were numerically dominated by women or men. The classifications used were similar to other studies (Kumlin 2010): extremely female dominated (≥ 90% women), e.g., office secretaries, nursing, and midwifery professionals; female dominated (≥ 60– < 90% women), e.g., personal care, helpers, and cleaners; gender integrated (≥ 40– < 60% women), e.g., public service administrative professionals, secondary education teaching professionals; male dominated (≥ 10– < 40% women), e.g., finance and sales professionals, directors, and chief executives; extremely male dominated (< 10% women), e.g., transport workers, building workers (Gonäs et al. 2018).
The individuals were followed up regarding their labour market position or main source of income at 8 years after inclusion, that is in 1993, 1998, and 2011, respectively. An 8-year follow-up was used because this was the only alternative given by the data to get a comparable follow-up period for all three periods.
The employment situation at follow-up was classified according to a way that Statistics Sweden had classified people as being employed or not employed, based on the size of their annual income from work. This categorization was refined by using complementary data concerning different social security benefits (parental leave, student benefit, unemployment, SA, DP, old-age pension) (Wikman et al. 2012).
According to this classification, each individual was assigned to one of the following categories: (a) employment (including self-employed), (b) parental leave benefit, (c) student benefit, (d) SA, (e) DP, (f) unemployment benefit, (g) social assistance benefit, (h) unknown (no type of registered income or benefit), (i) old-age pension. These nine categories were then combined into two dichotomized outcome variables. For description of the subgroups and dichotomization, see Gonäs et al. 2018.
The first outcome concerned employment status and was dichotomized as:
Employed, including self-employed (a) and those on parental leave (b)
Not employed, i.e., (c) those with student benefit, (d) SA, (e) DP, (f) unemployment benefit, (g) social assistance benefit, (h) unknown, no type of registered income, and (i) old-age pension.
The second outcome concerned sickness absence benefits, dichotomized as:
Sickness absence or disability pension, i.e., categories (d) and (e)
All other categories, i.e., (a) employed (including self-employed), (b) parental leave, (c) student benefit, (f) unemployment benefit, (g) social assistance benefit, (h) unknown, no type of registered income, and (i) old-age pension.
First, rates of employment or levels of SA/DP at 8-year follow-up were calculated for each period and each occupational group, stratified by gender.
Second, rates of employment and rates of SA/DP were calculated for the five occupational categories of gender segregation, together with age, and period (presented in two types of diagrams).
Finally, an APC analysis was conducted regarding age (A), period (P), and cohort (C) effects. For this, we used mean polish approach (Tukey 1977; Selvin 2004; Keyes et al. 2010). This approach evaluates the remaining covariation (residuals) between birth cohorts and outcomes when covariations in age and period have been considered and excluded. Age (A) was retained as a continuous variable (28–64 years). The period (P) refers to the three 8-year periods: 1985–1993, 1990–1998, and 2003–2011, respectively. Cohort (C) was based on birth cohorts, 1929–1983.
The residuals from the APC analysis were then regressed on birth cohort membership using univariate logistic regression analysis with the dependent variables “employment” respectively “SA/DP”.
For different birth cohorts, the remaining differences in employment, SA, and DP were calculated when covariations in age and period were eliminated.
The analysis quantifies the influence of observed birth cohort effects on employment respective SA/DP, using the average of residuals in each period as reference points. These average values were assumed to be a constant component associated with each birth cohort. A value greater than zero indicates higher than average influence and less than zero indicates lower than average influence from age/period effects on employment rate and SA or DP rate. For estimated birth cohort effects, the results of the univariate logistic regression are presented as odds ratios (OR) with 95% confidence intervals (CI).
All data used in the study were de-identified by Statistics Sweden before being made available to the research team. The project was approved by the Regional Ethical Review Board in Stockholm, Sweden (no. 2007/762-31 2012/863-32).