A Hot-Deck Multiple Imputation Procedure for Gaps in Longitudinal Event Histories Based on Multivariate Predictive Mean Matching

• Activity Number: 380
• Type: Contributed
• Date/Time: Tuesday, August 4, 2009 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Chia-Ning Wang*+ and Roderick J.A. Little and Bin Nan and Sioban Harlow
• Companies: University of Michigan and University of Michigan and University of Michigan and University of Michigan
• Keywords: missing data ; hot-deck imputation ; predictive mean matching ; longitudinal analysis ; event history

Abstract

In longitudinal studies, interest often concerns the relationship between an intermediate state and to a final event, such as an occurrence of a disease-related nonfatal event and death, respectively. However, a difficulty arises in deciding the time of the intermediate state when subjects have gaps in their event history. This issue can be solved by imputing the missing information for gaps. Predictive mean matching is used to impute by matching gaps to completely recorded histories, conditional on pertinent longitudinal characteristics and the time of the final event (which may be censored.) We apply this approach to menstrual history data, where the intermediate state is a marker of menopausal transition and the final event is menopause. The influence of different choices of PMM models for selecting matched gaps is assessed, and extensions to other missing data settings are outline.