Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.

Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.

The relative performance of full -information maximum likelihood estimation for missing data in structural equation models

Craig Kyle Enders, University of Nebraska - Lincoln

Abstract

A Monte Carlo study was designed to compare the performance of four missing data methods in structural equation models: full information maximum likelihood (FIML), listwise deletion, pairwise deletion, and similar response pattern imputation. The effects of three independent variables were examined (factor loading magnitude, sample size, and missing data rate) on four outcome measures: convergence problems, parameter estimate bias, parameter estimate efficiency, and model goodness-of-fit. The performance of the four estimation methods was assessed in two different simulations. In the first, missing values were missing completely at random (MCAR), while the second simulated missing at random (MAR) data. Results indicated that FIML estimation was superior across all conditions of the design. FIML estimates were unbiased and more efficient than the other methods across both the MCAR and MAR simulations. In addition, FIML yielded the lowest rate of convergence problems and provided near-optimal Type 1 error rates across both simulations. When data were MCAR, pairwise deletion generally yielded satisfactory performance. Parameter estimates were unbiased and convergence rates and efficiency were only slightly worse than FIML. However, inflated Type 1 error rates were observed, particularly when the magnitude of factor loadings was high. However, under MAR pairwise deletion parameter estimates were substantially biased. Both listwise deletion and similar response pattern imputation performed substantially worse than FIML across most conditions of the study. The methods yielded high rates of nonconvergence, and inefficient parameter estimates relative to FIML. These efficiency differences became more pronounced as the missing data rate increased. Under MAR, both methods yielded substantially biased parameter estimates. Although listwise deletion rejection rates were acceptable, those of similar response pattern imputation were substantially inflated. Based on the results of this study, it is recommended that applied researchers discontinue the use of popular ad hoc methods such as listwise and pairwise deletion in favor of FIML. ^

Subject Area

Education, Tests and Measurements|Psychology, Psychometrics

Recommended Citation

Enders, Craig Kyle, "The relative performance of full -information maximum likelihood estimation for missing data in structural equation models" (1999). ETD collection for University of Nebraska - Lincoln. AAI9929197.
http://digitalcommons.unl.edu/dissertations/AAI9929197

Share

COinS