Constructing and validating an occupational job strain index based on five Norwegian nationwide surveys of living conditions on work environment

A substantial number of current studies have constructed and evaluated the reliability and validity of a psychosocial JEM. The reliability of the JEM was mainly reported by indicators, such as the internal consistency of the constructed JEM, kappa statistics to test the agreement between individual-based and occupational-based job exposure, and sensitivity and specificity to report the ability of constructed JEM to identify the exposure or non-exposure individuals, respectively. Psychosocial exposures at work are mostly described by the dimensions of Karasek’s models, including job demand and job control, which were commonly reported to have satisfactory internal consistency [7, 8]. A validation of alternative formulations of job strain supported using a continuous index when investigating health outcomes instead of the more common quadrant approach based on dichotomies, which inevitably will lead to loss of information [9]. The performance of the constructed psychosocial JEM varied across countries, which was reported as good for both job demand and job control in Australia [10], low for job control and bad for job demands in France [11], and good for job control and job strain in Finland [12]. The accuracy of detecting job exposure has been reported differently between genders [11, 12], but mostly suggests that the ability to identify psychosocial job exposure is better for women than for men. The reliability of the JEM was found to be different among exposures, which is likely to be higher for job control and job strain than for job demand [13].

The validity of the constructed JEM was tested by evaluating criterion-related validity using large population data. Based on solid evidence about possible links between psychosocial work exposures, especially high job strain (high job demand and low job control), and the risks of ill health, i.e., sickness absence [14], disability pension [15], and cardiovascular diseases [16] and different mental disorders [17, 18], such as risk for depressive symptoms [19, 20] and sleeping problems [21, 22], various health outcomes were taken into account to examine the reproducible likelihood of the constructed JEM compared with individual-based job exposure and the predictive validity of the JEM based on register data. The assessment of psychosocial work factors measured by JEM can also help to answer the question of whether the relationship between exposure and outcome is consistent regardless of the method used [16].

Recent Scandinavian studies have constructed and validated the JEM based on Karasek’s Demand-Control Model (1979), using large population data, such as the Danish JEM based on Work Environment Cohort Study data, including all patients aged 18–65 who received depressive and anxiety disorder treatments [23], the Swedish JEM using a large study population of all individuals aged 30–54 [24], the Finnish JEM utilizing the Health 2000 Study, and the Finnish National Work and Health Surveys [12]. The results showed the ability of a constructed JEM to predict various health outcomes, i.e., anxiety disorders [23], depression [12], sickness absence, and disability pension [24], with different patterns between men and women.

With respect to the context of Norway, a previous study by Hanvold et al. [4] utilized data of the work environment in 2006 and 2009 to construct group-based exposure estimations and to assess psychosocial JEM performance. The constructed JEM showed fair to poor agreement with the different performances between genders, reported to be higher among women than men [4]. The constructed JEM in Hanvold et al.’s study showed a good ability to identify occupations that are exposed to job strain, job control, and job demand. However, this study only investigated the concurrent validity of psychosocial occupational-level job exposure on low back pain.

This study used five waves of the Norwegian nationwide Survey of Living Conditions on work environment. This pooled dataset was used to examine four aspects of reliability (i.e., agreement, consistency, sensitivity, and specificity) associated with the JSI and its dimensions and components. Survey data were further used to assess the construct validity by means of factor analysis and the concurrent validity of the JSI, based on both individual based and occupational based exposures, using individually reported “long term sick leave”, “anxious symptoms”, “depressive symptoms” and “sleeping difficulty symptoms” as health outcomes. Finally, we assessed the predictive validity of the JSI for the entire working-age population in Norway, using register data and “disability benefits”, “mortality”, and “number of long-term sick absence periods” as health indicators. Where appropriate, the analyses were stratified by gender, as current research has shown divergent effects of work stressors on men and women [18].

This study utilized five surveys of the Norwegian nationwide Survey of Living Conditions on work environment from 2006, 2009, 2013, 2016, and 2019, with a total sample of 43,977 individuals. The purpose of using five surveys is to reach a larger number of observations, which may increase the accuracy of JEM performance. Data collection was conducted by Statistics Norway. The personal interviews conducted by telephone with computer assistance are on average 24–30 min long. Less than 0.5% of the interviews were conducted face-to-face. Since 2006, the survey on work environment has been funded by the Ministry of Labor and Social Inclusion to expand the sample and develop the survey as a panel.

The sample of the Norwegian nationwide Survey of Living Conditions on work environment was randomly drawn from the population aged 18–69 years, which represented active working-age people in the country. In the 2006 survey, the number of observations was 12,550 (with 67.2% response rate); in the 2009 survey, the number was 12,555 (with 61% response rate); in the 2013 survey, the number was 10,857 (with 53.1% response rate); in the 2016 survey, the number was 10,665 (with 52.6% response rate); and in the 2019 survey, the number of observations was 11,212 (with 57% response rate).

In 2007, the register data population consisted of people aged 18–55 who had a valid occupational code. In total, this included 1,589,535 individuals. Tables 1 and 2 below show the background characteristics of the study population. In both the survey data and register data, the number of men was slightly higher than women (23,062 men and 20,915 women in the survey data, 819,232 men and 770,303 women in the register data). The survey data had a lower proportion of respondents aged 25–44 but a higher proportion of respondents aged 45–69 (43.6% of the total sample aged 25–44 and 46.2% of the total sample aged 45–69) than that of the register data (56.2% and 29.2% of the total sample, respectively). The respondents in the survey data have a higher educational level than the population in the register data, as 42.5% of total respondents in the survey data have college or university education, compared with 34.1% in the register data. However, the distribution of the major occupational groups in both samples was not likely to be different.

As shown in Table 1, 16% of respondents in our survey data had experienced long-term sick leave during the previous 12 months. The percentage of respondents who experienced different mental health symptoms was 2.7% for anxiety, 2.3% for depression, and 8.0% for sleeping difficulty. More women than men in our survey sample reported different mental health problems.

As presented in Table 2, there is a low percentage receiving disability benefits, and mortality is low, with 0.3% and 1.2% of the study population, respectively. Approximately 27% of our register study sample took ten long-term sick leave periods or more during 2008 and 2015, 35.9% of women compared to 18.6% of men.

In this paper, we investigated the reliability and validity of our constructed psychosocial JEM, i.e., the JSI. These assessments involved comparisons of individual job strain with occupational job strain, and of their respective psychosocial dimensions and components, as well as an appraisal of the reliability and criterion validity of the occupational JSI itself. Measured by kappa, agreement between individual-based and occupation-based psychosocial exposures was poor to fair. However, the internal consistency of the two dimensions that make up occupation-based job strain, job demand, and job control was clearly acceptable. According to the factor analysis, the construct validity of the JEM was also fully acceptable. As for concurrent validity, assessed by the survey data, individual- and occupation-based job strain were significantly associated with anxiety and depression for men. For women, the significant associations between job strain and anxiety and between job strain and depression were observed only for individual-based job strain but not for occupation-based job strain. With respect to predictive validity, occupation-based job strain was significantly related to all three health outcomes (disability, sick leave, and mortality) in the register data for both genders.

The results pertaining to the reliability of the JSI were somewhat mixed. The measures that compared individual exposures and occupation-based exposures (kappa, sensitivity, and specificity) tended to be poor, although they varied. On the other hand, the measure of consistency of the two dimensions of job strain performed well. The interpretation of the results related to agreement, sensitivity, and specificity is not straightforward since no gold standard exists. In other words, since individual psychosocial estimates cannot be perceived as the gold standard, poor agreement is subject to several interpretations. This may imply that occupation-based results are far from the mark, but it may equally be that they are close to the mark due to systematic bias in the individual estimates. Hence, due to these interpretive challenges, we would argue that poor agreement and occasionally low sensitivity and specificity do not provide evidence implying that our measures of occupation-based job strain, or job demand or job control were unreliable [36, 37].

Our positive results regarding the predictive validity of the JSI corresponded well with previous studies examining the validity of the JEM in other countries, such as the French psychosocial JEM [11] and the Finnish psychosocial JEM [12]. Since the ultimate purpose of this paper was to construct a validated measure of occupation-based psychosocial work environments for use in register data, we find this specific result rather assuring. We are inclined to put more trust in this finding than in the findings emanating from the analysis of the survey data, which were more mixed. Evidence pertaining to future outcomes (the predictive aspect) is generally considered more robust than evidence related to associations established in cross-sectional data (the concurrent aspect) because of the common variance problem [36]. See also the discussion of limitations below.

Somewhat surprisingly, the occupation-based job strain indicated an elevated risk of anxiety and depression among men but not among women. This does not agree with earlier results showing that higher levels of anxiety and depression were typically reported for women rather than for men [38, 39]. There are two plausible explanations for this gender difference. First, women may be more familiar with working conditions in high-stress and female-dominated occupations than men, such as teachers, social workers, and nurses [40]. Hence, women may tend to underreport their exposure at work compared with men, while mental health outcomes are reported to be higher for women than for men (see descriptive statistic results, page 9–10). There is also evidence that male nurses report more work-related disturbances than female nurses [41], and men working in traditional female jobs may perceive a higher level of social stress than women due to their internalization of the masculinity role [42]. Second, there is evidence of gender differences in job satisfaction, i.e., that men have more difficulties in achieving job satisfaction and are also more willing to express frustration with working conditions than women [43]. Thus, our results suggest that an occupation-based JSI may enhance the ability to identify gender differences in the effect of job strain on health outcomes better than an individual JSI.

Although our results support the idea that a JEM is a reproducible and efficient method for examining work-related health risks in epidemiological studies, some limitations should be considered. The JEMs were converted from individual exposure measurements, which may lead to errors in the JEM assignments due to the imprecise information of exposure for each job and other errors in job coding and duration for individuals [44]. Furthermore, one may argue that JEM is only helpful when job demands within an occupation are comparable, and because JEM assigns the same exposure estimates to all workers with identical job titles, which may affect inter-individual variability, especially in cases where workers have specific tasks [3], or in the case of digitalization of jobs. Another caveat using the JEM developed by the survey data is the risk of differential misclassification. The risk of misclassification is likely to increase when exposure and health outcomes are assessed simultaneously. The individual characteristics of the workers may additionally contribute to the error in self-reported questionnaires in the sense that workers who constantly “complain about everything” may overreport their working exposures and the situation of their health, while another group who “complain about nothing” may underreport their occupational environment and health [36]. This approach may also increase subjective bias and the threats of false positive results, as it reflects the individual perception of the work exposure and health outcomes [45] in cases where workers with health problems tend to report a higher degree of psychosocial exposures than healthy workers. Hence, despite the fact that JEM may provide more objective measures for occupational exposure than self-reported information, this method cannot be seen as a gold standard measure for examining job exposure at work [36, 37]. As discussed above, neither method can. Our study only constructed a JEM based on Norwegian data; thus, it is only appropriate for generalization within Norway and countries that share the same conditions as Norway. To achieve a better applicable JEM, the idea of constructing an international-level JEM (Job Exposure Matrix International-JEMINI) should be further developed [46].

We used the same approach as Hanvold et al. [4] and Solovieva [12] when defining the exposed versus non-exposed as basis for constructing the JEM. Using the median as a cut-off point may, however, be somewhat arbitrary. Thus, in further developments of the JEM one should experiment with different cut-off points to identify possible thresholds for increased health risks.

The JSI could have been constructed by dividing demands by decision latitude which, in contrast to our chosen approach, would have given distributions approaching second degree functions (hyperboles). The advantage of such an approach is the avoidance of defining subjects in extreme "active" and "passive" groups. With our chosen approach there is a risk of labeling subjects as exposed to job strain, who have extremely high demands and rather high control as well and in the other end those who have low demands and extremely low control. However, we have no reason to believe this being an issue of any significance for the results presented in this paper. Dividing demands by decision latitude would exclude more of such problem cases. In further development of the JSI, dividing demands by decision latitude should also be tested.

The Norwegian labor force remains gender- and class-segregated [47]. Our study also indicated that men and women have distinct patterns of psychosocial job exposure that may stem from certain occupations, such as nursing. Although current scholarship has documented evidence of the relationship between job strain, occupational class, and gender [48], few studies have used JEMs. The question of how the risks for different health outcomes are explained by job exposures differentiated by gender and occupational class remains unanswered in our study. Hence, one recommendation is that future research on occupational epidemiology should consider both gender and occupational class when investigating the risk of occupational exposure to health.