A Comparison of Approximate Bayesian Bootstrap and Weighted Sequential Hot Deck for Multiple Imputation
Creel, D. V. (2011, July). A Comparison of Approximate Bayesian Bootstrap and Weighted Sequential Hot Deck for Multiple Imputation. Presented at JSM 2011, Miami Beach, FL.
To account for missing data, Rubin and Schenker (1986) describe a multiple imputation approach called Approximate Bayesian Bootstrap (ABB) Imputation, which is simpler and more direct computationally than Bayesian Bootstrap Imputation. Several Monte Carlo studies have investigated the properties of ABB and suggested improvements to the ABB procedure. This paper proposes an alternative to ABB for multiple imputation. We will empirically investigate the properties of the ABB alternatives and weighted sequential hot deck (WSHD) for multiple imputation when the missing data mechanism is ignorable and nonignorable. Two different approaches to WSHD will be explored. The first approach uses WSHD to multiply impute using the same donor pool. The second approach uses a two-stage process that selects, with replacement, a new donor pool from the original set of donors and then applies WHSD to the new donor pool. The multiple imputation procedures will be assessed using bias, variance, mean squared error, and coverage.