According to a growing body of cross-sectional studies, a low income is associated with poor mental health. Although income is a widely studied determinant of mental health, the results derived from longitudinal research have been conflicting. Research attention has focused recently on bias resulting from time-invariant unobserved variables such as personality, genetics and life experiences. Such variables could drive the relationship via indirect selection, but they are challenging to measure and include in a model. Fixed-effects models based on within-person variation provide a solution in terms of controlling for this type of heterogeneity. They have been used in a small number of survey studies on income and mental health, with varying results.

We assess the association between psychotropic drug use and individual income longitudinally, controlling for both unobserved individual differences, and observed time-variant time trend, age, marital status and employment status. The register-based data comprises an 11% nationally representative random sample of Finnish residents aged 30 to 63 years, followed annually from 1996 to 2013: the final sample includes 389,428 individuals and 4,722,359 person-years. We estimate ordinary-least-squares and fixed-effects models, the latter controlling for all unobserved time-invariant individual characteristics. Adjusted for observed sociodemographic confounders, the doubling of income decreased the likelihood of psychotropic drug use by 0.63 percentage points in the ordinary-least-squares model (ß= -0.0091, 95% confidence interval: -0.0102, -0.0080; P < 0.001), and by 0.10 percentage points in the fixed-effects model (ß= -0.0014, 95% confidence interval: -0.0022, -0.0006; P < 0.001). These associations were slightly weaker for women and individuals with a high baseline income. The contemporaneous association observed between income and psychotropic drug use was enduring - albeit very small - even when time-invariant individual differences and sociodemographic changes were accounted for.