Dear Drs. Muthén, is it possible in MPlus to obtain the SRMR fit index for a measurement model with ordinal categorical variables and the WLSM estimator? If not, is it at all possible to obtain or calculate the SRMR for ordinal data? Thanks!
I believe we do SRMR for ordinal variables. It is not done for models with thresholds or covariates however. The default is to include thresholds. You can add MODEL = NOMEANSTRUCTURE to the ANALYSIS command to avoid having thresholds in the model.
Ryan Veal posted on Sunday, October 27, 2019 - 8:15 pm
Hello Professor Muthen,
I ran a series of EFA and ESEM for ordered categorical variables, and as expected, had identical factor loadings and parameters, as well as identical model fit indices. The one exception was SRMR, which had values of a slightly lower magnitude for the ESEM models.
Can you propose a reason why this would be the SRMR values would be lower in the ESEM compared to the EFA models?
The denominator is different. In EFA there is no mean structure. In ESEM, you can have mean structure, even though you didn't. Since in ESEM mean structure is included in the SRMR definition, and is fitted perfectly you get lower value (same S, bigger denominator).
Ryan Veal posted on Monday, October 28, 2019 - 3:21 pm