Hi, I needed to rerun an analysis originally conducted in version 5.21 (to conduct a DIFFTEST we didn't do back then) so just ran it in version 6.11. I used the exact same input file and data file and the parameter estimates are exactly the same (CFA using categorical indicators, loadings were identical as were indicator thresholds) but the fit indices differed for all but WRMR. Which values for the fit indices should we trust? If from the new version, does that mean that all analyses done using previous versions should be re-run? Thanks! Rick Zinbarg
In Version 6 we changed to a new method for second-order chi-square adjustment for WLSMV, ULSMV, and MLMV resulting in the usual degrees of freedom. It is an alternative to what we had been using. Both can be trusted. We changed to avoid the confusion regarding the degrees of freedom with WLSMV.
thanks for the very speedy reply Linda. Given that the two analyses produced different values of CFI, TLI, and RMSEA (the WRMR values are the same), I am wondering which ones I should report? I guess I am hoping you will say it is ok to report the values from version 5.21 as otherwise it will mean having to rerun all of our analyses so that the values we are reporting are comparable across models (and the only results from the version 6.11 analyses we are running now will be the difftest results). If, however, the version 6.11 values are more accurate then we will bite the bullet and rerun all of our analyses.
You can report either one. There is no difference in accuracy. They are asymptotically equivalent. The difference is that the degrees of freedom in the new version are as expected. In both cases, the important value is the p-value.