Date of this Version
Although there are numerous challenges for the investigation of aging-related changes in older adults, statistical analysis with incomplete data and the conceptualization of population processes related to mortality is one of the most difficult. Selective attrition and mortality selection within longitudinal studies on aging are intrinsically related to many aging-related changes and must be carefully considered in the analysis and interpretation of results (e.g., Baltes, 1968; Hofer & Sliwinski, 2006; Schaie, Labouvie, & Barrett, 1973). A key distinction is made between attrition (i.e., selective dropout) and mortality selection (i.e., selective survival) in that attrition affects characteristics of the particular sample under investigation, whereas mortality selection affects both the definition of the population as well as the sample under study (Baltes, 1968). Including time-to-death as a predictor in models for estimating change in outcomes of interest permits conditional inferences to defined populations based on age and survival (and their interaction) and is easily performed when complete data are available for both chronological age and age of death. In most studies, however, complete data for all individuals are not currently available and may not be available for a substantial period of time. The purpose of the current work is to present a two-stage multiple-imputation approach for treating mortality and attrition as distinct processes leading to incomplete data and which permit the use of time-to-death in the predictive models when follow-up is incomplete.