Using a multiple imputation technique to merge data sets

Abstract

Surprisingly, despite the fact that the information needed to address a research question is often spread across separate surveys, there is not a well established method for merging data sets prior to analysis. This letter presents a relatively straightforward method for merging information from separate surveys of representative samples drawn from the same population. It uses a multiple imputation technique that was originally developed to correct for missing values on items from survey data. The advantage of the process is that it accounts for the variance related to merging separate data sets, as well as the sample variance present in any sample survey. This provides more accurate estimates of the precision of the coefficients of interest than ad hoc alternatives.