Data processing and statistical analysis

Complete data was available for 15,813 persons (88.7%). Missing data was not at random (Little’s MCAR-Test: χ² = 7493.71, p < .001) and highest for job tenure (with 2.8%). For analyses regarding risk factors of detachment difficulties and related health outcomes, we applied multiple data imputation of missing data (k = 5; cf.) [29] and report pooled results among datasets.

Considering information on prevalence, frequencies of responses and corresponding 95% confidence intervals were calculated for all four possible responses in relation to the item on difficulties to detach psychologically from work. Unweighted (raw) and weighted results are reported. Rohrbach and Schmidt [28] describe the weighting algorithm relating to an adjustment of data according to official data from the 2017 micro census. The estimates relating to the two subsamples of teachers are compared with the sample of employees from other occupations.

Odds ratios (OR) and 95% confidence intervals (95% CI) were calculated to examine the relationships of difficulties detaching psychologically from work with demographic, occupational, and work-related variables (overload demands, cognitive demands, emotional demands, contextual resources, and physical demands) using binary logistic regression. First univariate analyses were conducted and psychological detachment difficulties as outcome were regressed on each predicting variable as independent variable. In a second step, a multivariate logistic regression analysis was conducted, integrating all predictors into the model in parallel.

Moreover, ORs and corresponding 95% CIs for difficulties detaching psychologically from work in relation to the five health-related outcomes (sleep disturbances, exhaustion, psychosomatic complaints, musculoskeletal complaints, and sickness absence) were calculated. In a first step, univariate logistic regression analyses were conducted. We compared these results with estimates from additional multivariate logistic regression analyses that were adjusted for demographic, occupational, and work-related variables (see above). We considered parameter estimates with p-values lower .05 as significant. All analyses were conducted using IBM SPSS Statistics 26.0.