Handling non-response in UK longitudinal cohort studies

In this seminar, Richard Silverwood will discuss multiple imputation as a method of handling non-response in UK longitudinal cohort studies.

Non-response is common in longitudinal studies, reducing efficiency and introducing the potential for bias. Principled methods, such as multiple imputation, are generally required to obtain unbiased estimates in studies subject to missingness - which is not completely at random. Such methods often rely on a ‘missing at random’ assumption, the plausibility of which can be improved by the inclusion of predictors of non-response in the analysis. 

Richard will present a systematic, data-driven approach used to identify predictors of non-response at each sweep of the 1958 National Child Development Study. The identified predictors of non-response are across a number of broad categories and include some not previously identified in the literature. This approach utilises the identified predictors of non-response within multiple imputation analyses. 

Richard will demonstrate that using this approach makes it possible to restore sample representativeness – relative to both internal (baseline) levels and to external reference data – in key variables. He will propose that judiciously chosen predictors of non-response should be included in future analyses using principled methods to reduce bias due to non-response. 

Quantitative Social Science seminar series 

Quantitative Social Science (QSS) hosts a weekly seminar series, where invited speakers present work of relevance to the research interests of the group.

Links

• Centre for Longitudinal Studies
• Centre for Quantitative Social Science
• Department of Social Science