Causal modeling of epidemiological data on psychiatric disorders

This paper reviews the logic of causal inference from epidemiological data. I maintain that the clearest causal statements can be made when the philosophical causal principles of association, direction and isolation are upheld in epidemiological research. After reviewing the argument by Holland that only experimental manipulation affords clear causal claims, I examine the utility of structural equation models and longitudinal methods for making causal claims from non-experimental data. This examination leads to the conclusion that mental health epidemiologists should begin to incorporate intervention trials into the last phases of their research programmes when they want to make strong causal claims.