Sample covariance matrices constructed with pairwise deletion for randomly missing data were used in a simulation with three sample sizes and five levels of missing data (up to 50%). Parameter estimates were unbiased, parameter variability was largely explicable, and no sample covariance matrices were nonpositive definite except for 50% missing…

After reviewing the multiple fit indices in structural equation models, evidence on their behavior is presented through simulation studies in which sample size, estimation method, and model misspecification varied. Two sampling studies, with and without known populations, are presented, and implications for the use of fit indices are discussed.…

Relative fit indices using the null model as the reference point in computation may differ across estimation methods, as this article illustrates by comparing maximum likelihood, ordinary least squares, and generalized least squares estimation in structural equation modeling. The illustration uses a covariance matrix for six observed variables…

A Monte Carlo simulation study investigated the effects on 10 structural equation modeling fit indexes of sample size, estimation method, and model specification. Some fit indexes did not appear to be comparable, and it was apparent that estimation method strongly influenced almost all fit indexes examined, especially for misspecified models. (SLD)

Conducted a meta-analysis focusing on the explanation of empirical Type I error rates for six principal classes of estimators. Generally, chi-square tests for overall model fit were found to be sensitive to nonnormality and the size of the model for all estimators, with the possible exception of elliptical estimators with respect to model size and…

A Monte Carlo simulation study was conducted to investigate the effects of sample size, estimation method, and model specification on structural equation modeling (SEM) fit indices. Based on a balanced 3x2x5 design, a total of 6,000 samples were generated from a prespecified population covariance matrix, and eight popular SEM fit indices were…

This study empirically investigated bootstrap bias estimation in the area of structural equation modeling (SEM). Three correctly specified SEM models were used under four different sample size conditions. Monte Carlo experiments were carried out to generate the criteria against which bootstrap bias estimation should be judged. For SEM fit indices,…

In this Monte Carlo study, the effects of four factors on structural equation modeling (SEM) fit indices and parameter estimates were investigated. The 14,400 samples generated were fitted to 3 SEM models with different degrees of model misspecification. Effects of data nonnormality, estimation method, and sample size are noted. (SLD)

Alternative models for a large dataset were analyzed by asymptotically distribution-free, generalized least squares, maximum likelihood, and ordinary least squares estimation methods, comparing incremental and nonincremental fit indexes. Incremental methods were quite unstable across estimation methods. This phenomenon is explained. (SLD)

Used simulation to compare the ability of maximum likelihood (ML) and generalized least-squares (GLS) estimation to provide theoretic fit in models that are parsimonious representations of a true model. The better empirical fit obtained for GLS, compared with ML, was obtained at the cost of lower theoretic fit. (Author/SLD)

A Monte Carlo study was conducted to assess the effects of some potential confounding factors on structural equation modeling (SEM) fit indices and parameter estimates for both true and misspecified models. The factors investigated were data nonnormality, SEM estimation method, and sample size. Based on the fully crossed and balanced 3x3x4x2…

Basic philosophical and statistical issues in structural equation modeling (SEM) are reviewed, including model conceptualization, identification, and parameter estimation and data-model-fit assessment and model modification. These issues should be addressed before the researcher uses any of the new generation of SEM software. (SLD)

Goodness of fit indexes developed by R. P. McDonald (1989) and Satorra-Bentler scale correction methods (A. Satorra and P. M. Bentler, 1988) were studied. The Satorra-Bentler index is shown to have the least error under each distributional misspecification level when the model has correct structural specification. (SLD)

LISREL 8 invokes a ridge option when maximum likelihood or generalized least squares are used to estimate a structural equation model with a nonpositive definite covariance or correlation matrix. Implications of the ridge option for model fit, parameter estimates, and standard errors are explored through two examples. (SLD)

Implications of ignoring correlated error on parameter estimates in some simple structural equation models are examined. It is shown analytically and empirically through simulation that ignoring positive between-construct correlated error overestimates the structural parameter linking the two constructs. Effects become more pronounced with…