Address correspondence to: Stephen A. Sivo, College of Education, University of Central
Florida, 4000 Central Florida Parkway, Orlando, FL 32816-1250. E-mail:
ssivo@mail.ucf.edu
The Journal of Experimental Education, 2006, 74(3), 267–288
Copyright © 2006 Heldref Publications
267
The Search for “Optimal” Cutoff
Properties: Fit Index Criteria in
Structural Equation Modeling
STEPHEN A. SIVO
University of Central Florida
XITAO FAN
University of Virginia
E. LEA WITTA
University of Central Florida
JOHN T. WILLSE
University of North Carolina–Greensboro
ABSTRACT. This study is a partial replication of L. Hu and P. M. Bentler’s (1999)
fit criteria work. The purpose of this study was twofold: (a) to determine whether
cut-off values vary according to which model is the true population model for a
dataset and (b) to identify which of 13 fit indexes behave optimally by retaining all
of the correct models while simultaneously rejecting all of the misspecified models in
a manner invariant across sample size and data distribution. The authors found that
for most indexes the results do not vary depending on which model serves as the cor-
rect model. Furthermore, the search for an optimal cut-off value led to a new dis-
covery about the nature of McDonald’s measure of centrality and the root mean
square error of approximation. Unlike all other indexes considered in this study, the
cut-off value of both indexes actually decreases for incorrect models as sample size
increases. This may suggest that power calculations are more likely to be optimal
when based on those indices.
Key words: cut-off values, fit indexes, model fit, structural equation modeling