In October 2015, Professor Bianca De Stavola and Dr Rhian Daniel from the Centre for Statistical Methodology at the London School of Hygiene & Tropical Medicine came to Glasgow to present their work on concepts and methods in causal mediation analysis. The event was run jointly with the MRC/CSO Social & Public Health Sciences Unit at the University of Glasgow. The intent was to present and compare two approaches: the classical approach used in the social and behavioural sciences and a new approach based on counterfactuals and novel estimation methods.
There were four parts within the three-hour afternoon session. Part one introduced the language of causality and part two introduced causal diagrams using Directed Acyclic Graphs (DAGs). These two sessions effectively set the scene and established the basic concepts that were required for the later sessions. Part three discussed traditional mediation analysis and part four took us through a more formal approach to causality that Dr Daniel and Professor De Stavola have been working on.
They acknowledged that the study of mediation is well established in the social sciences and is sometimes attempted in epidemiology. There was discussion about the developments in modern causal inference that have highlighted the limitations of classical approaches. Dr Daniel then explained new definitions based on counterfactuals and novel estimation methods that deal with specific challenges such as intermediate confounding. They concluded by arguing that the greater rigour and generality permitted by the potential outcomes framework is worth investigating further.
There was a discussion about how to implement the different methods. The afternoon demonstrated that causal mediation analysis is a hot topic which is evolving and improving.