cchien posted on Thursday, July 01, 2010 - 1:51 pm
When I used Mplus 5 previously, the output of EFA contained the factor determinacy, but recently I found this part does not appear in the output by Mplus 6. Was the factor determinacy removed in version 6? Or say...do I need to add any additional syntax to make it show in the output? There is no clue in the latest user's guide.
I ran an ESEM and extracted 3 factors with the following factor determinacy: F1 = 0.658; F2 = 0.423; F3 = 0.954 I know that the quality of the factor scores of F1 and F2 do not look great, but the items mostly loaded on to the factors as expected and the model fit was acceptable: RMSEA = 0.061 [0.057 0.065]; CFI = 0.857; SRMR =0.049. My chi-square was SG but I did not take it into account as my N=401. My question: Is it OK to use the 3 factors (though F3 determinacy is pretty low) in the structural part of my ESEM where I test the relationship of these factors with some predictors? I noticed that we cannot ignore any factor in the model specified. Are there specific criteria of when determinacy is too low? Thanking you in advance for your response.
I wouldn't worry so much about factor determinacy, but model fit. The CFI is poor and a significant chi-square can't really claim oversensitivity due to a large sample when you have only n=401.
The factor determinacy is mostly relevant if you work with estimated factor scores. It is true that a lower determinacy tends to give larger SEs for structural coefficients, but if you have signficant structural coefficients it may not hurt.
1) OK. So the chi-square and the CFI are not good. Do you think this model (given the RMSEA and SRMR) is defendable based on substantive grounds: the items are loading as expected though there are cross loadings, which are explainable.
2) In the structural part of my model, the factors (meaning factor scores?) were regressed on predictors. Should I be worried here about the low factor determinancy?