ABSTRACT
We briefly describe several model fit indices. First we consider four sample-based indices: the Bentler-Bonett normed fit index, a df -adjusted version of it, and two corresponding goodness-of-fit indices initially implemented in LISREL. Then we examine briefly a philosophically different index, Akaike’s information criterion, and a related index, Cudeck and Browne’s estimated cross-validation index. After that, we discuss a number of population-based indices and then revisit examining residuals to extend our discussion from Chapter 2. At the end of the appendix, we provide a summary table of the indices (Table D-3). (Additional information may be found in Notes to Chapter 2.)
Bentler and Bonett’s normed fit index: NFI
Bentler and Bonett (1980) suggested that the goodness of fit of a particular model may be usefully assessed relative to the fit of some baseline “null model.” Such a null model would be an arbitrary, highly restricted model-say, that all correlations are zero, or that all correlations are equal, or some such-which would represent a baseline level that any realistic model would be expected to exceed. The index would then represent the point at which the model being evaluated falls on a scale running from this baseline model to perfect fit.
