Variables

Physical function, the dependent variable, was assessed as the self-reported ability to stand without support (yes/no), walk up and down stairs without difficulty (yes/no), walk 100 m fairly briskly without difficulty (yes/no), and get up from a kitchen chair without using the support of your arms (yes/no), as well as self-reported problems with balance indoors (yes/no). The answers were combined into an index ranging 0 to 5, where 5 indicated the greatest physical function. Because few people scored 0, 1, or 2, these categories were combined.

Psychosocial working conditions were assessed on the basis of self-reported main occupation in 1991 and self-reported occupations during working life. The occupations were coded using a psychosocial job exposure matrix (JEM) [39, 40]. Occupational history was assessed with retrospective questions about the respondent’s first occupation that lasted at least six months and all occupations held by the respondent one by one in temporal order thereafter [41]. First occupation and occupation at age 25, 30, 35, 40, 45, and 50 (based on occupational history), and occupation in 1991 were then matched with the JEM. Both the matrix and the occupation variables in LNU were coded using the 1980 Nordic version of the three-digit International Standard Occupational Classification Codes. The JEM was created on the basis of a random sample of 12,084 Swedish workers from the 1977 and 1979 Swedish Survey of Living Conditions (ULF). The ULF surveys contained items specific to the job demand-control model. To create the matrix, average scores that ranged from 0 to 10 for job demands and control were generated separately for women and men for 262 occupations. The occupation-based job control scale consisted of a linear composite of 12 items: influence over the planning of work, setting of work pace, how time is used in work, selection of supervisor, selection of co-workers, planning of work breaks, planning of vacations, flexible work hours, varied task content, varied work procedures, opportunity to learn new things, and experience of personal fulfillment on the job. The job demand scale consisted of a linear composite of 2 items constructed on the basis of responses to two items, i.e., psychologically demanding (taxing) work and hectic work.

In our analytic sample, occupation-based psychological job demand scores in 1991 ranged from 1.3 to 8.2 (mean = 4.9, SD = 1.4), and occupation-based job control scores from 1.8 to 8.2 (mean = 5.3, SD = 1.2). Job variables were created for each person’s first occupation and their occupation every five years thereafter until baseline in 1991. Demands and control were measured on continuous scales to best capture the variability in how active the active jobs were and how stressful the high strain jobs were. Jobs with a “demand” score above 5 (range 0–10) were considered high demands and jobs with a “control” score above 5 (range 0–10), high control.

Active jobs. We first re-scaled the demands and control variables by subtracting 5 and coding values below 0 as 0. That is, scores of 5 or lower on the job demands and job control variable were coded as 0. Second, we summed the transformed demands and control variables to create the active job variable, in which jobs with demands and/or control values of 5 or below were classified as non-active and all other jobs as active. Non-active jobs included high strain jobs (those with high demands and low control), low strain jobs (those with low demands and high control), and passive jobs (those with low demands and low control). High strain jobs, low strain jobs, and passive jobs all had a score of 0 in this variable, and all scores above 0 indicated some level of active jobs (5.7 indicated the most active jobs in our analytic sample). See Fig. 1.

The construct of the active job variable. Source: Adapted from Karasek & Theorell (1990) and modified by authors

High strain jobs. The original control variable was reversed so that a score of 10 indicated the lowest level of control and 0 the highest level of control. In a next step, the demand and control variables were re-scaled by subtracting 5 and coding values below 0 as 0. That is, scores of 5 or lower on the job demands and/or job control variable (i.e., low job demands or high job control) were coded as 0. The re-scaled demands and control variables were then summed to create the high strain job variable. Non-high strain jobs (i.e., low strain jobs, active jobs, and passive jobs) all had a score of 0 in this variable, and all scores above 0 indicated some level of job strain (3.3 indicated the highest degree of job strain in our analytic sample). See Fig. 2.

The construct of the high strain job variable. Source: Adapted from Karasek & Theorell (1990) and modified by authors

Data on occupational history and occupation at baseline were used to construct trajectories of active jobs and trajectories of high strain jobs. Random effects growth curve models were used to calculate within-person change. Random effects allow for variation between participants in the individual intercept and slope [42]. The intercept of the trajectories was divided in low and high via a median split. The slope was divided into downward (−), stable (0), and upward (+) slope groups. A slope was considered upward if there was an increase of more than a half standard deviation above zero and downward if there was a decrease below zero. The intercept and slope of the trajectories were then combined to create a variable with five categories, all within the same quadrant: low starting point and downward slope (low/−), low starting point and stable slope (low/0), low starting point and upward slope (low/+), high starting point and downward slope (high/−), high starting point and stable (high/0), and high starting point and upward slope (high/+). Theoretically there should have been six categories, but no participants fell in the category low/−. The other four categories were then compared to the category low/0.