Advancements in causal modelling theory and the elucidation of counterfactual principles make it much easier to establish:
- whether it is even possible to ascribe a causal relationship from a set of data
- what variables need to be controlled/not controlled in order to achieve it
We understand the latest techniques – see our 2011 publication (Virchow, Kay, et al) where we implemented a genetic algorithm propensity score combined to standard regression techniques (doubly robust method).
We can apply a host of leading edge techniques dependent on the data characteristics (dealing with issues such as time varying endogenous covariates) as well as separating direct from indirect/mediation effects. Causal modelling methods are not restricted just to observational data – they can be used effectively on RCTs (recognising and correcting for deficiencies that arise with any human study even if randomised).
Developments in this field are happening all the time – following them is our favourite pursuit.