How to handle missing data and restore sample representativeness in longitudinal surveys
Speaker(s):
Bio: Richard Silverwood is Associate Professor of Statistics and Chief Statistician at the UCL Centre for Longitudinal Studies. His applied research is mainly within the context of health, in particular the causes and consequences of non-communicable diseases, often taking a life course perspective. He also has methodological interests, including approaches for handling missing data, the analysis of linked survey and administrative data and making causal inferences from observational data.
Bio: Was a social worker Hammersmith SSD 1974-77 Welfare rights adviser National Council for One Parent Families 1977-1989 Welfare rights advocate National Assoc..Citizens Advuice Bureaux 1989-1990 Research Fellow Centre for Longitudinal Studies, 1990-present
Michail Katsoulis
Abstract:
Missing data are common in longitudinal surveys, particularly due to attrition over waves of data collection, and can lead to substantial bias. Principled methods of missing data handling are usually required to obtain unbiased estimates in such settings. In this talk we will briefly cover the relevant missing data theory before discussing missing data methods and their application. There will be a particular emphasis on why and how variables other than those required for the analysis should be included in missing data handling. We will present findings from our recent work across several of the British birth cohorts (1958 National Child Development Study, 1970 British Cohort Study, Next Steps), including the use of linked administrative data. We will demonstrate that with careful analysis it is possible to largely restore sample representativeness despite the presence of selective attrition. There will be plenty of opportunity for questions and discussion.