Missing data is a prevalent issue in many fields, yet only about half of published studies mention dropouts, and less than 20% of those studies incorporated dropouts into their analyses (Ladouceur, Gosselin, Laberge, & Blaszcynski, 2001). Nearly all common statistical analyses assume complete data yet many statistics books do not deal with missingness. Proper missing data techniques have therefore gone mostly ignored by researchers until recently (Rubin, 1996).
Three major problems with incomplete data are: (1) loss of information or power due to loss of data; (2) complication during data management and analysis, partially because of limitations with standard statistical software; and (3) potential marked bias because of systematic differences between observed and missing values (Barnard & Meng, 1999).
Recent advancements in software have made missing data analyses easier and more prolific. This is critically important because gatekeepers in research, such as grant reviews, regulatory agencies, and journal reviewers, are evermore critical of the treatment of missing data.
But when are data truly missing? Many surveys incorporate “skip patterns”, for example, where a respondent may be instructed to skip a subsequent question based on a certain response to a key beginning question in the section. Overall, missing data fall into a similar category of latent variables –true values exist, but we cannot always accurately determine the exact value (D. F. Heitjan & Rubin, 1991; Schafer & Graham, 2002).