On causation in therapeutic research: observational studies, randomised experiments and instrumental variable analysis

Abstract

The counterfactual theory defines a conceptual framework for judging causation of therapeutic interventions. The main assumption underlying the claim that a counterfactual statement holds is that compared groups are exchangeable. In observational studies for therapeutic effects causal claims are hard to justify because the lack of exchangeability between compared groups. Randomised studies have a rather straightforward causal interpretation, based on the randomisation procedure that leads to expected exchangeability between compared groups. The use of instrumental variables can theoretically overcome the problems adherent to observational data, at least to a certain extent. The general idea of instrumental variables is to mimic a randomised trial, by searching for a variable that determines the probability of exposure (treatment) but that is not in other ways associated with the outcome under study. The assumptions for causal claims in IV analyses are strong, but the alternative of avoiding causal statements based on observational data, is less appealing.