Dynamic predictors of attrition in Understanding Society
|Day:||Thu 4 Jul|
Non-random attrition can introduce substantial bias into the results from longitudinal studies. Identifying predictors of attrition can alleviate the problem by informing recruitment and retention strategies and by identifying potential auxiliary variables that should be measured and included in statistical corrections for missingness. Previous research has identified a range of predictors of attrition; however, more research is needed to understand the effects of predictors of attrition that vary over the course of panel studies - such as changes in employment or health status, or the break-up of families - on different stages of the response process. In this study we use the 8 waves of the UKHLS sample to examine a range of candidate time-varying predictors of non-contact and refusal given contact. Results are discussed in terms of their implications for understanding why attrition occurs, how panel studies can be designed to minimise or mitigate the effects of selective attrition, and appropriate statistical corrections to correct for selective attrition.