Jan Digutsch, Richard N Landers, Hannah M Schade, Hans-Joachim Lincke, Matthias Nuebling, Clemens Stachl, Gerhard Rinkenauer, Yannick A Metzler
The job demands-resources model suggests that job demands (e.g., workload) and resources (e.g., autonomy) relate to employee well-being through both main and interactive effects. Although the predictive validity of main effects is well established, the validity of the proposed interaction effects, buffering and boosting, remains unclear. Prior research has examined only selected combinations of job characteristics, leaving open the possibility that unexplored interactions may still be informative. In this study, we conducted a comprehensive analysis of 24 job characteristics and all possible nonlinear and higher order interactions among them using machine learning methods in a large, quota-representative sample of the German working population (N = 100,000) stratified by industry, gender, age, and employment status. We compared the out-of-sample predictive performance of models capturing main effects (linear mixed-effects models) with models capable of modeling complex interactions (boosting algorithms). The inclusion of interaction effects did not improve predictive accuracy beyond models based solely on main effects. These findings challenge the validity of buffer and boost effects as global mechanisms across jobs and organizations, as interactions among job characteristics did not meaningfully contribute to predicting either burnout or work engagement. Instead, job characteristics appear to exert only additive effects. (PsycInfo Database Record (c) 2026 APA, all rights reserved).