|
Abstract:
|
Imputation is a common pragmatic approach for handling missing data---missing values are replaced by estimates and analyses are conducted on the filled-in data set. The obvious drawback of imputation is that it is "cheating," since it "makes up" data. As a result, inferences from the filled-in data overstate the amount of information, yielding confidence intervals that are too narrow, and tests that do not achieve nominal levels. Multiple imputation addresses this concern by creating multiple data sets with different sets of plausible values imputed. The basic idea and underlying theory of multiple imputation will be reviewed, including multiple imputation combining rules, imputation models, controversies about the method, available software, and pros and cons relative to other missing data approaches.
|
