On the problem of treatment-confounder feedback in longitudinal psychology

Causation occurs in time. Therefore, investigating the relationship between cause and effect requires time series data. Causality is also dynamic. Where there is treatment-confounder feedback, the relationship between cause and effect cannot be identified using standard regression methods, including multi-level regression and structural equation models. Instead, special methods – "G-methods" – are needed. Here, I use three causal graphs to describe a problem of treatment- confounder feedback, and direct readers to G-methods for its solution.