Shaun Seaman

My current main interest is the treatment of missing data in observational studies, particularly in cohort studies. One major cause of missingness in cohort studies is attrition: individuals dropping out of the study. In addition, data are typically missing for some items even on individuals who remain in the study. The simplest approach to analysing data when some values are missing is to restrict the analysis to complete cases. However, it is known that this can lead to bias unless the data are missing completely at random. Multiple imputation (MI) and inverse probability weighting (IPW) are methods that give consistent estimation under the more general assumption that the data are missing at random. The former requires that the imputation model be correctly specified; the latter, that the missingness model (i.e. the model for the probability that an individual is a complete case) be correctly specified. The more recent, doubly robust methodology offers some protection against misspecified imputation or missingness models. Linear increments offers an alternative approach.

I am interested particularly in the use of MI, IPW and linear increments for handling real cohort data. I work with epidemiological researchers in a variety of MRC units, with the aim of helping improve the way that missing data are handled.

I am also interested in risk prediction and causal inference, and have a long-standing interest in HIV/AIDS epidemiology and injecting drug use epidemiology.