Causal Modelling and Evaluation

Much of applied health and social research is aimed at estimating causal effects of risk factors or interventions (treatments, programmes, policies etc.). This includes analyses and assessments carried out by research organisations to inform their strategies and help with their management and funding decisions (policy and programme evaluation). However, it is not always clear whether the quantity targeted in a particular analysis/assessment has a causal interpretation and even if that is the case, whether it can be estimated without bias from the study data. For example, findings from observational studies are frequently criticised for lacking confounding control.

This course will review statistical designs and analyses that enable valid causal effect estimation, including Propensity Scoring and Mendelian Randomisation in observational studies, methods for dealing with non-compliance in trials, Mediation Analysis and some Quasi-experimental designs. We will focus on methods that are easily accessible to the applied researcher. Throughout methods will be motivated and demonstrated with real data examples from health research. The course will use STATA although analyses are easily translatable to other general-purpose statistical software packages.