ERIC Number: ED445031
Record Type: Non-Journal
Publication Date: 2000-Apr
Reference Count: N/A
A Comparison of Methods for Structural Modeling with Polytomous and Continuous Variables.
Berkovits, Ilona; Hancock, Gregory R.
A simulation study compared three methods of estimating parameters within structural equation models (SEM) with polytomous variables. These methods appear in three SEM computer software packages: (1) LISREL (Joreskog and Sorbom, 1996) with PRELIS (Joreskog and Sorbom); (2) EQS (Bentler, 1995); and (3) the new Mplus (Muthen and Muthen, 1998). The differences in parameter estimation accuracy were evaluated by manipulating sample size, exogenous and endogenous factors with polytomous indicator variables, number of polytomous variables per factor, number of categories per polytomous variable, and different thresholds for the underlying continuum to create the categories. Results show some severe parameter estimation problems in LISREL (LISREL results are not discussed in this paper). EQS had problems with convergence, but the parameter estimates did not appear to be affected, allowing comparison of EQS results with Mplus results. Mplus displayed a very slight estimation difference depending on whether the scaling variable on the endogenous factor was polytomous. Mplus did appear to estimate the path coefficients closer to the true values than EQS for all conditions, with Mplus results having less variability. Mplus is faster and converges more often than EQS, and thus appears preferable for the estimation of SEM models with polytomous variables. (Contains 8 tables, 4 figures, and 19 references.) (Author/SLD)
Publication Type: Reports - Research; Speeches/Meeting Papers
Education Level: N/A
Sponsor: Educational Testing Service, Princeton, NJ.
Authoring Institution: N/A