Anonymous posted on Thursday, October 07, 2004 - 5:21 am
I want to check the construct-validity of an expected two factor model (one factor with 7 observed variables and one factor with 4 variables). I treat the 11 variables as categorical. Making a CFA with Analysis: Type=General and Output: Standardized I get a CFI 0.93. How is it possible to see if or which items should be taken out of the model, to get a better CFI? How is it possible to see that maybe some items are highly correlated with the items of the other factor? How is it possible to compute residual covariances? Some residual variances are quite high. Which is the best estimator for such a check of construct-validity? If i look to the model in an EFA i can see in a varimax rotation that some Items load high on the other factor. But in an EFA there is no CFI. Do i have to change between EFA and CFA to make the construct-validity? Thank you.
The way to see which items may not behave the way you expect them to is to do an EFA as you have done. You can also do an EFA in a CFA framework where you can obtain standard errors, other fit measures like CFI, and modification indices.
Anonymous posted on Wednesday, October 13, 2004 - 12:27 am
Thank you very much. But what does "EFA in a CFA framework" mean? If I make a CFA with for example: Type is General, and Model: f1 by y1 y2 etc. and Output: Standardized. I get standard errors and see significant unstandardized loadings. The standardized loadings are nearly conform to those estimated (unrotated) in an EFA. I can see in this CFA also RMSEA and SRMR. But this seems not to be that what you understand with "EFA in a CFA-Framework". The other thing is which "Estimator" is a "good" one for this check of the construct validity?
bmuthen posted on Wednesday, October 13, 2004 - 4:53 pm
For EFA in a CFA framework, please see the handout for Day 1 of our short courses - see the Mplus web site under Web Training and Handouts.