Development of methodology for natural experiments and causal inference
We wrote an accessible overview of natural experiment methods in public health and an overview of the synthetic control method with an applied example. While most focus is on the evaluation of new interventions and policy, our comprehensive review showed the potential for withdrawal of interventions and policy to be just as illuminating. We described the challenges inherent to the design and conduct of economic evaluations of population health interventions alongside NEs, providing a comprehensive framework and outlining a research agenda in this area.
Mendelian randomisation studies make use of genetic variants that are linked to exposures of interest to understand their causal effects on health (and other) outcomes. This approach is an example of applying instrumental variable analysis. Our review provided an up-to-date explanation of how to implement the process, including the latest developments in methodology to investigate, and potentially relax, the underpinning assumptions. Building on this initial work, we have established a collaboration which has attracted research funding from the Health Foundation to understand the causal effects of depression and problematic use on social outcomes.
We used insights from the causal inference literature to give clear guidance on the interpretation of analyses involving multiple indicators of socioeconomic position. It is currently unclear to which population the effect from regression models applies, despite this being a common procedure for natural experiments and causal methods. We used novel methodology to illustrate how to do this. In further work we showed that a causal approach makes converting between different effect measures simple and exact, whereas doing so using the results from standard regression models had been complex and approximate.
Other methodological work has included a book on multilevel modelling for public health and health services research and two books on SAS – one on its use for statistical analysis, the other on its graphics capabilities.