Solutions for Missing Data in Structural Equation Modeling
Rufus Lynn Carter · Research & Practice in Assessment · 2006
Many times in both educational and social science research it is impossible to collect data that is complete. When administering a survey, for example, people may answer some questions and not others. This missing data causes a problem for researchers using structural equation modeling (SEM) techniques for data analyses. Because SEM and multivariate methods require complete data, several methods have been proposed for dealing with these missing data.What follows is a review of several methods currently used, a description of strengths and weaknesses of each method, and a proposal for future research. Methods for Dealing with Missing Data Listwise Deletion Listwise deletion is an ad hoc method of dealing with missing data in that it deals with the missing data before any substantive analyses are done. It is considered the easiest and simplest method of dealing with missing data (Brown, 1983). It involves removing incomplete cases (record with missing data on any variable) from the dataset. This means the researcher removes all the records that have missing data on any variable. Depending on the sample size and number of variables this can result in a great reduction in the sample size available for data analysis. Listwise deletion assumes that the data are missing completely at random (MCAR). Data are missing completely at random when the probability of obtaining a particular pattern of missing data is not dependant on the values that are missing and when the probability of obtaining the missing data pattern in the sample is not dependant on the observed data (Rubin, 1976). An advantage in using listwise deletion is that all analyses are calculated with the same set of cases. Pairwise Deletion