A Review of Current Software for Handling Missing Data

When we deal with a large data set with missing data, we have to undertake two important tasks. First, it is important to inspect the pattern of missingness. This can provide very practical information. For instance, we may find that most of the missing values concern only one specific variable. If this variable is not central to our analysis problem, we may delete it from our analysis, rather than keeping it, and therefore having to delete many cases. We would also like to know if the missingness forms a pattern, or if it is related to some of our observed variables. For if we discover a system in the missingness pattern, we may try to include that in our statistical analyses. The second task is to produce sound estimates of the parameters of interest, despite the incompleteness of the data. There are two major approaches to this problem. One is to make the data complete by imputing the missing values, and then do the analysis on the completed data. The other is to use a method, typically a likelihood-based procedure, that allows us to model incomplete data directly. Modern software assists us in both tasks. For inspecting the pattern of missingness, either we can use the standard procedures in a statistical package like SPSS, or the specialized procedures made available in the SPSS procedure Missing Value Analysis (MVA). For imputation and direct modeling of incomplete data, we need specialized software. This contribution reviews some of the options in generally available software like SPSS MVA, SOLAS, and NORM.