Dan Mungas posted on Saturday, May 22, 2004 - 8:25 pm
I am running two different confirmatory factor analyses using ordinal indicators. In one analyses, there are three possible values for 22 items, in the second there are five values for 6 items. Exploratory factor analysis with categorical indicators for both indicates a very large first eigenvalue, and intercorrelations of items are high. A one dimensional CFA model yields very acceptable CFI, TLI, and SRMR values, and residuals of the correlation matrix are small (generally<.10). RMSEA values are in the .11-.13 range. This isn't substantially changed by recoding to increase cell frequencies. I am not sure how to interpret the model fit in light of the different information from RMSEA and the other fit indices. Additionally, in one analysis, but not the other, using the WLS estimator instead of the default WLSMV results in RMSEA < .05. Any help would be appreciated.
bmuthen posted on Saturday, May 22, 2004 - 9:00 pm
Simulation studies for CFA with categorical outcomes point to CFI as a trustworthy fit index, with the requirement of values of at least 0.96. See the Yu dissertation posted on the Mplus home page for a comparison of fit indices.