Work

The performance of cross-validation indices used to select among competing covariance structure models

Year: 2003
Type: dissertation
Abstract: When testing structural equation models, researchers attempt to establish a \nmodel that will generalize to other samples from the same population. Unfortunately, \nresearchers tend to test and respecify models during this attempt, capitalizing on the \ncharacteristics inherent within the sample data in which the model is being developed. \nSeveral measures of model fit exist to aid researchers when trying to select a model \nthat fits the sample data well. However, these measures fail to consider the predictive \nvalidity of a model, or how well it will generalize to other samples from the same \npopulation. \nIn 1983, Cudeck and Browne proposed using cross-validation as a model \nselection technique. They recommended cross-validating several plausible models \nand selecting the model with the most predictive validity. Several cross-validation \nindices have been proposed in the past twenty years, including the single-sample AIC, \nCAIC, and BCI; the multiple-sample C; the two-sample CVI; and the “pseudo” \nsingle-sample C* and C *. Previous studies have investigated the performance of \nthese various indices, but have been limited with respect to their study design \ncharacteristics. The purpose of this study is to extend the literature in this area by \nexamining the performance of the previously mentioned cross-validation indices \nunder additional study design characteristics, such as nonnormality and crossvalidation \ndesign. Factor loading, sample size, and model misspecification conditions \nwere also manipulated. \nThe performance of each cross-validation index was measured in terms of \nhow many times out of 1,000 replications they selected the correct confirmatory \nfactor model. The results indicated that the performance of the cross-validation \nindices tended to improve as factor loading and sample size increased. The double \ncross-validated indices outperformed their simple cross-validated counterparts in \ncertain conditions. The performance of the cross-validation indices tended to \ndecrease as nonnormality increased. Recommendations are provided as to which \ncross-validation methods would optimally perform in a given condition. It is hoped \nthat this study provides researchers with useful information concerning the use of \ncross-validation as a model selection technique and that researchers will begin to \nfocus on the predictive validity of a structural equation model in addition to overall \nmodel fit.
Language: English
Open Access status: green
OpenAlex ID: W1937329879
Cites: 48
Cited by: 1
Related to: 10