Missing Data

An important indicator of data quality is the fraction of missing data. Missing data (also called "item non-response") means that for some reason data on particular items or questions are not available for analysis. In practice, many researchers tend to solve this problem by restricting the analysis to complete cases through "listwise" deletion of all cases with missing data on the variables of interest. However, this results in loss of information, and therefore estimates will be less efficient. Furthermore, there is the possibility of systematic differences between units that respond to a particular question and those that do not respond—that is, item nonresponse error. If this is the case, the basic assumptions necessary for analyzing only complete cases are not met, and the analysis results may be severely biased.

Modern strategies to cope with missing data are imputation and direct estimation. Imputation replaces the missing values with plausible estimates to make the data set complete. Direct estimation means that all available (incomplete) data are analyzed using a maximum likelihood approach. The increasing availability of user-friendly software will undoubtedly stimulate the use of both imputation and direct estimation techniques.

However, a prerequisite for the statistical treatment of missing data is to understand why the data are missing. For instance, a missing value originating from accidentally skipping a question differs from a missing value originating from reluctance of a respondent to reveal sensitive information. Finally, the information that is missing can never be replaced. Thus, the first goal in dealing with missing data is to have none. Prevention is an important step in dealing with missing data. Reduction of item nonresponse will lead to more information in a data set, to more data to investigate patterns of the remaining item nonresponse and select the best corrective treatment, and finally to more data on which to base imputation and a correct analysis.