I am running a multiple group two factor CFA with four and 24 continuous indicators for each factor (many variables are quite nonnormal).
When I ran the CFA, I got the following error message: THE STANDARD ERRORS OF THE MODEL PARAMETER ESTIMATES COULD NOT BE COMPUTED. THE MODEL MAY NOT BE IDENTIFIED. CHECK YOUR MODEL. PROBLEM INVOLVING PARAMETER 62.
Parameter 62 is the factor correlation.
In order to deal with this identification problem, I tried the bootstrapping option with 1000 iterations. The output looked fine, giving me model fit indices, etc. However, the standard errors estimated were all given as 999.000.
Are these standard errors due to the non-normality of the data, or do they indicate some other problem? Are the model fit indices still trustworthy so that I can use them for group invariance tests? Is there perhaps a better way to deal with this identification issue? Thanks a lot for your reply.
Here is the code I have used:
Analysis: estimator=ml; bootstrap=100; MODEL: F1 BY DW1 DW2 WCt1 - WCt6 Rt1 - Rt13 Lt1 - Lt5 ; F2 BY St1 St2 St3 S14; F1 WITH F2 ;
Model new: F1 BY DW1 DW2 WCt1 - WCt6 Rt1 - Rt13 Lt1 - Lt5 ; F2 BY St1 St2 St3 S14;
Hello, As is well known, when we use bootstrap method to generate bootstrap samples, the model parameter estimates of some samples may be improper. Are the standard errors of the model parameter estimates which mplus gives based on all bootstrap samples or the bootstrap samples that have the proper solution? For example, if I set bootstrap=1000(generating 1000 bootstrap samples) and the model parameter estimates of 20 bootstrap samples are improper, are the standard errors of the model parameter estimates that mplus gives are based on 1000 bootstrap samples or 980 bootstrap samples£¨only having proper bootstrap samples) £¿
Hello, In other word, standard errors of the model parameter estimates that mplus gives are based on 1000 bootstrap samples and not 980 bootstrap samples. Is it(that I understand above) right? Thank you.