I am running a twolevel CFA (4 factors, 22 variables; ordinal data, WLSMV estimator) examining model fit for an unconstrained (Ho) and constrained (Ha) model (factor loading invariance across levels).
All my reading indicates using the DIFFTEST procedure to compare nested models when using the WLSMV estimator.
However, placing the DIFFTEST option in the SAVEDATA command when estimating the Ha model generates an error message:
"The DIFFTEST option is not available for TYPE=TWOLEVEL. Request for DIFFTEST will be ignored."
I am following the 2-step process describe in the MPLUS manual (p.625; 2012 edition) for applying DIFFTEST with the WLSMV estimator .
I assume from the message I cannot use a DIFFTEST option method because it is a TWOLEVEL model, and I will need to compute the test statistic manually. Is there a reference for how use the output from my Ho & Ha models to do this?
It is not possible to compute DIFFTEST manually. Luckily you don't need to for your task. Use Model Test instead, see page 711-713 in the manual.
Es Maths posted on Tuesday, October 15, 2019 - 4:23 am
Is it possible to use 'MODEL TEST' for a two-level CFA using WLSMV and categorical data when comparing different models instead of DIFFTEST option? I need to compare for instance one-factor model at the higher level model vs two-factor model at the higher level.
Would it be OK to make the comparisons based on the model fit indices e.g. RMSEA and others?
I have been using WLSMV from the beginning of my analyses, I wonder if it is ok to keep my analyses with WLSMV in the paper but do the statistical comparison by using ML? Hope this question is not silly.