Direction of dependence analysis for pre-post assessments using non-Gaussian methods: a tutorial

ObjectiveWe introduced methods for solving causal direction of dependence between variables observed in pre- and post-psychotherapy assessments, showing how to apply them and investigate their properties via simulations. In addition, we investigated whether changes in depressive symptoms drive changes in social and occupational functioning as suggested by the phase model of psychotherapy or vice versa, or neither.MethodAs a Gaussian (normal-distribution) model is unidentifiable here, we used an identifiable linear non-Gaussian structural vector autoregression model, conceptualizing instantaneous effects as during-psychotherapy causation and lagged effects as pre-treatment predictors of change. We tested six alternative estimators in six simulation settings that captured different real-world scenarios, and used real psychotherapy data from 1428 adult patients (Finnish Psychotherapy Quality Registry; assessments on Patient Health Questionnaire-9 and Social and Occupational Functioning Assessment Schedule).ResultsThe methodology was successful in identifying causal directions in simulated data. The real-data results provided no evidence for single direction of dependence, suggesting shared or reciprocal causation.ConclusionsA powerful new tool was presented to investigate the process of psychotherapy using observational data. Application to patient data suggested that depression symptoms and functioning may reciprocate or reflect third variables instead of one predominantly driving the other during psychotherapy.

Automated summary

This study introduced a method for determining the causal direction of dependence between variables observed before and after psychotherapy, and investigated their properties through simulations. The findings suggest that changes in depressive symptoms may not solely drive changes in social and occupational functioning during psychotherapy, but instead there may be reciprocal causation or the influence of other variables.