Reducing Attrition Bias using Targeted Refreshment Sampling and Matching

This paper examines the possibility of reducing attrition bias in panel data using targeted refreshment sampling and matched imputation. The targeted refreshment sampling approach consists of collecting new data from the original sampling population from individuals who would never usually respond to surveys. Using propensity score matching and imputation in conjunction with refreshment sampling it is suggested that the dropouts from a panel can effectively be 'replaced'. The procedure allows us to identify underlying joint distributions in the data. The method is illustrated using data from the Youth Cohort Surveys in the UK which suffer 45% attrition in the second wave. A comparison of the results of this method with other techniques for attrition modeling suggest that the technique could be an effective way to overcome a substantial part of the bias associated with attrition.