Do episodes of mental health (MH) problems cause future MH problems, and if yes, how strong are these dynamics? We quantify the degree of persistence in MH problems using nationally representative, longitudinal data from Australia and system generalized method of moments (GMM), and correlated random effects approaches are applied to separate true from spurious state dependence. Our results suggest only a moderate degree of persistence in MH problems when assuming that persistence is constant across the MH distribution once individual-specific heterogeneity is accounted for. However, individuals who fell once below a threshold that indicates an episode of depression are up to five times more likely to experience such a low score again a year later, indicating a strong element of state dependence in depression. Low income is a strong risk factor in state dependence for both men and women, which has important policy implications. Copyright © 2013 John Wiley & Sons, Ltd.