Imputation of missing values in the case of a multiple item instrument measuring alcohol consumption

Abstract

Missing values in survey instruments are a common problem for survey researchers. It is aggravated in the case of instruments used to measure alcohol consumption: they usually consist of item batteries from which summary measures, such as grams of pure alcohol per day, are constructed, and a missing value (for example, quantity or frequency) in regard to a single item for only one of several beverages results in a missing summary measure across all of the beverages, though the values for the remaining items are known. The present paper examines different approaches to imputation of missing values, feasible with standard statistical software packages. Hot-deck imputation is shown to have certain advantages, but even single-value imputation (for example, median imputation) results in values that are comparable to those of the other three imputation methods.