Study design and data collection
Self-reported data were drawn from the 2017 population-based health survey "How are you?" conducted in the Central Denmark Region. In 2017, approximately 23% of the Danish population were living in the Central Denmark Region, and the demographic composition (gender, age, and educational attainment) of the population was similar to that of the total Danish population (37). The survey included a representative population sample of 52,000 citizens aged 16 years and above drawn from the Danish Civil Registration System (38). Participants were invited to complete a web-based or postal questionnaire (39). Those who failed to respond to the survey received up to four reminders. The response rate was 62% (32,417), and 80% of the respondents completed the web questionnaire. The study was approved by the Danish Data Protection Agency (r. no. 2012-58-0006) and the Central Denmark Region (r. no. 1-16-02-593-16). Each participant received written information about the purpose of the survey, and informed consent was obtained from all subjects. All methods were carried out in accordance with relevant guidelines and regulations along with the approval. Data from the survey were linked with national administrative registers using the unique personal identification number assigned to all Danish citizens (38). Register data included age, gender, ethnic background and labour market status.
Perceived stress scale
The level of perceived stress was assessed by the 10-item Perceived Stress Scale (PSS) (40). Based on Lazarus' stress model (41, 42), the PSS is a generic measure of stress in which stress relies upon the person's perception of the stressor as stressful or not. The ten items ask how often in the past month life was appraised as unpredictable, uncontrollable and overloaded (e.g., "In the last month, how often have you been upset because of something that happened unexpectedly?") (40). The items were scored from 0 to 4 ("never", "almost never", "sometimes", "fairly often", or "very often"). An increasing sum score (range 0 to 40) indicates an increasing perceived stress level (40). PSS has satisfying psychometric properties (43, 44) and has been used in a number of population-based studies (e.g. 3, 45-48). In the present study, Cronbach’s α indicated that the scale had a high internal consistency (α=0.87).
Work-related and non-work-related stressors were assessed using nine questions covering seven categories from the Life Event Questionnaire (49). The selection of questions was inspired by The Danish National Birth Cohort (50-52). The questions cover major life events, chronic stressors and daily hassles. Stressors were assessed asking "Within the past 12 months have you felt burdened by some of the following things?" The questionnaire covers work situation, financial circumstances, housing conditions, relationship with partner, relationship with family and friends, own disease, disease among close relatives, deaths among close relatives and other types of burdens. The response categories were "no", "yes, a little", "yes, partly" or "yes, a lot". In the present study, all four response categories of each variable were used.
The survey also includes a more comprehensive questionnaire regarding work that covers quantitative demands, job influence, satisfaction and recognition, work-life conflicts, physical and emotional wear and tear, and physical load (53). However, introducing the more comprehensive questionnaire did not alter the analyses, which is in line with a previous study (54). Thus, in order to keep the model simple, the single question about work situation was preferred.
Perceived social support
Perceived social support was assessed using a single item: "Do you have anyone to talk to if you have problems or need support?" The response format was "yes, always", "yes, mostly", "yes, sometimes" or "no, never or almost never". In the present study, all four response categories of each variable were used. The question was inspired by The MOS Social Support Instrument (55).
Age and gender were assessed using a combination of self-reported and register data. Ethnic background was defined using the Danish Civil Registration System (38). Educational attainment was self-reported and categorised as low (primary school, no further education), medium (upper secondary education, vocational education and/or short higher education) or high (bachelor’s degree or higher level of education) according to the Danish version of the International Standard Classification of Education (56). Students were categorised according to their expected graduation level. Labour market status was assed using a combination of self-reported data and data from the Danish Register for Evaluation of Marginalization (57).
We conducted dominance analysis, a relatively new computer-intensive method, to determine the most important contributors to perceived stress in the population and in stratified analyses (i.e., gender, age, educational attainment and labour market status) (58, 59). The analyses were based on a multiple linear regression model with PSS as the dependent variable and the nine different stressors and perceived social support as predictor variables. Dominance analysis is a method of ranking predictors according to how much each variable contributes to the total variance of the dependent variable in a model. The method is particularly suitable when predictors are intercorrelated, as may be expected in this study. Dominance analysis is an ensemble method based on estimation of all possible regression models (all subset regressions) (60). The average increase in total variance explained by the model (R2) when adding a variable to all possible sub-models quantifies the importance of the variable. Dominance analysis is used for decomposition of the total R2, but it has rarely been used in stress research (61). The dominance analyses in the present study consisted of 1,023 (10^2-1) regression models containing all possible combinations of predictors. In addition to the ten predictors, gender, age, ethnic background and educational attainment were included in all models.
Weights constructed by Statistics Denmark using a model-based calibration approach and including socio-demographic characteristics, income, social benefits and healthcare utilisation were applied in order to reduce sampling and non-response bias (39, 62).
Prior to the analyses, data were screened for missing values. The percentage of missing values was acceptable (0-13%). Missing PSS items ranged from 4.3% (item 1) to 5.1% (item 8). If one, two or three items of the PSS scale were missing, the mean of the available items was used to calculate the scale score (63). If responses to more than three items were missing, the PSS score was regarded as missing. Hence, 1,392 observations (4.3%) were excluded.
For stressors, missing values were treated as "no" if respondents had answered at least one of the nine stressor questions. A similar approach was adopted in the "How are you?" health survey with regards to measures of chronic diseases (64).
PSS was obtained in 95.7% of the respondent (Table 1), and stressors were obtained in 93.3% of the respondents (Table 2). Respondents with missing PSS or stressors were more likely to be 16-24 years old, have an ethnic background other than Danish and an unknown educational attainment or work situation (results not shown).
Stata/SE v16.1 (StataCorp, College Station, TX) was used to prepare the data and perform the descriptive and inferential analyses including dominance analysis using the community-contributed extension DOMIN (65).