Objectives Based on previous analyses, using ROC curves and correlation, we aim to improve agreement assessments between diverse formulations of the Demand Control (DC) Model for Job Strain in order to test its ability to predict anxiety and depression (AD) in a nationally representative population of workers from Chile, a country transitioning to high economic development and with high prevalence of AD.

Method A weighted national sample of 9503 workers representing the entire Chilean workforce was surveyed during 2010–2011 in Chile. Goldberg Health Questionnaire (12 questions) to assess AD and diverse formulation of the DC model were used as dependent and independent variables respectively. Bland-Altman plots for agreement and Poisson-log models (controlled for demographics) for predictive ability were used to assess each formulation.

Results Good agreement between Log and Quotient formulations. For different formulations, high strain jobs had between 1.7 (quadrant and tertile formulations) and 3.7 (extreme tertile formulation), higher prevalence of AD than low strain jobs. Approximately 12–25% of AD cases might be attributed to increased strain.

Conclusions Predictive ability of the DC model for AD was similar in trend to other studies. Most accurate models (extreme formulations) represent less population and might be impractical. At the cost of excluding many people from the evaluation, the extreme tertile model seems to be the best formulation to predict AD among extreme exposures to job strain within the Chilean working population. Better operationalizations of the DC model should be considered in future surveys to allow international comparisons and guide eventual interventions.