Analysis with Missing Data in Social Work Research

Missing data occur in virtually all research. Traditional methods for handling missing data, like complete case analyses and mean substitution, can result in the loss of statistical power and biased estimates of parameters. Recent statistical and software developments have provided readily accessible methods of handling missing data that provide more statistical power and less biased estimates than in the past. This paper compares newer techniques and the statistical programs that employ these techniques for imputing data in structural equation modeling analysis. The techniques presented include the expectation-maximization (EM) algorithm, multiple imputations (MI), and raw maximum-likelihood (RML), as employed in the statistical programs EMCOV, NORM, Amos, and Mplus. The assumptions, benefits, and limitations of each method and accompanying statistical programs are discussed.