Methodological developments that allow researchers to more realistically model the data are crucial to the production of more robust research findings. MiSoC researchers Paul Clarke and Yanchun Bao have made a number of contributions to the analysis of panel data. They have produced a detailed comparison of the strengths and weaknesses of four different techniques from the social science, economics and biostatistics literature (presented at the 2014 ESRC Research Methods Festival). They have adapted a robust semiparametric approach from biostatistics called structural mean models to panel studies so that, for example, the causal effects of state transitions can be efficiently estimated, and bias from the ‘initial conditions’ problem adjusted for. This is an extension of another contribution in which they showed how nonlinear structural mean models can be straightforwardly estimated using the generalized method of moments from econometrics (Stats Science paper with Frank 2015).
Another contribution is to the use of instrumental variables for the decomposition of causal effects into direct and indirect effects through mediated pathways: results regarding the identification, interpretation and estimation of these models has been set out (BMJ Open paper; Reframed study report accepted for EME Journal). They have also been involved in the development of a novel approach to allow for people changing household when analysing household panel data.