Strategies for data analysis with two types of missing values.

摘要

Conventional multiple imputation (MI) replaces the missing values in a dataset by m > 1 sets of simulated values. I explore a two-stage extension of MI in which the missing data are partitioned into two parts and imputed N = mn times in a nested fashion. Two-stage MI divides the missing information into two components of variability, lending insight when the missing values are of two qualitatively different types. Point estimates and standard errors from the N complete-data analyses are consolidated by simple rules derived by analogy to nested analysis of variance.;When two types of missing values are present in the same dataset, new questions about the probabilistic mechanisms generating the two types, and the conditions under which these mechanisms can be partially or completely ignored. I extend Rubin's concepts of missing at random and ignorability to accommodate two types of missing values. A new concept of partial ignorability is explored. Two real-data examples are used to demonstrate the implementation of two-stage MI: one in which the missing data mechanisms are completely ignored, another in which they are partially ignored and partially modeled.