The Effect of Using Different Imputation Methods for Economic Variables in Aging Surveys

We study the sensitivity of analyses on income and wealth to the methods of imputation for missing data. We do so by implementing a conditional hot-deck imputation, based on the method used by the English Longitudinal Study of Ageing, in the Health and Retirement Study. We compare marginal and joint distributions and regression models estimated from data using these imputations to the same statistics estimated from the RAND HRS data, which use a much more elaborate imputation method, and to estimates from data using an intermediate method. Although many results are qualitatively or quantitatively similar, there are also notable differences.