Investigating Parallel Analysis in the Context of Missing Data: ASimulation Study Comparing Six Missing Data Methods

摘要

Exploratory factor analysis is a statistical method commonly used inpsychological research to investigate latent variables and to developquestionnaires. Although such self-report questionnaires are prone to missingvalues, there is not much literature on this topic with regard to exploratoryfactor analysis—and especially the process of factor retention. Determining thecorrect number of factors is crucial for the analysis, yet little is known abouthow to deal with missingness in this process. Therefore, in a simulation study,six missing data methods (an expectation–maximization algorithm, predictive meanmatching, Bayesian regression, random forest imputation, complete case analysis,and pairwise complete observations) were compared with respect to the accuracyof the parallel analysis chosen as retention criterion. Data were simulated forcorrelated and uncorrelated factor structures with two, four, or six factors;12, 24, or 48 variables; 250, 500, or 1,000 observations and three differentmissing data mechanisms. Two different procedures combining multiply imputeddata sets were tested. The results showed that no missing data method was alwayssuperior, yet random forest imputation performed best for the majority ofconditions—in particular when parallel analysis was applied to the averagedcorrelation matrix rather than to each imputed data set separately. Completecase analysis and pairwise complete observations were often inferior to multipleimputation.