Convergence problem-categorical modeling
Message/Author
 Yung Chung posted on Wednesday, June 28, 2006 - 4:44 pm
Hi, Drs. Muthen:

I am running a three wave, 2 latent variables cross-lagged model. Each latent variable has three indicators. The indicators for the latent variable "COMM" are on ordinal level, whose responses are anchored on a Likert-type scale (1, 2, and 3). Below is my syntax:

COMM1 by BYS36AR@1 BYS36BR BYS36CR;
COMM2 BY F1S105AR@1 F1S105BR F1S105CR;
COMM3 BY F2S99AR@1 F2S99BR F2S99CR;

BYS36AR WITH F1S105AR;
F1S105AR WITH F2S99AR;
BYS36AR WITH F2S99AR;

BYS36BR WITH F1S105BR;
F1S105BR WITH F2S99BR;
BYS36BR WITH F2S99BR;

BYS36CR WITH F1S105CR;
F1S105CR WITH F2S99CR;
BYS36CR WITH F2S99CR;

BY2XRTH WITH F12XRTH;
F12XRTH WITH F22XRTH;
BY2XRTH WITH F22XRTH;
BY2XMTH WITH F12XMTH;
F12XMTH WITH F22XMTH;
BY2XMTH WITH F22XMTH;
BY2XSTH WITH F12XSTH;
F12XSTH WITH F22XSTH;
BY2XSTH WITH F22XSTH;

COMM2 ON COMM1;
COMM3 ON COMM2;

Output:
standardized sampstat modindices (4)TECH4;

When I used WLSMV to model the ordinal varaibles, Mplus failed to converge (even so after trying different starting values for these indicators). However, when I used ML assuming that they are interval, the model fit the data very nicely (with good fit indices).

I am wondering if I want to use categorical modeling, do you have any suggestion as to how I can make the program converge?
 Bengt O. Muthen posted on Sunday, July 02, 2006 - 5:33 am
The only way to answer this is for you to send your input, data, output, and license number to support@statmodel.com.
 Shuai Wang posted on Friday, September 20, 2013 - 11:07 am
Hi, Drs. Muthen:

I have an another convegence problem for my analysis. The EFA result is very good, then I go on to fit its CFA model without cross loadings. But the fit is not very good. Thus I began to add paths into the CFA model to allow cross loadings, thinking it might improve the fit. The paths added are those paths with loadings >0.3 in the EFA model. But whether I added them all at once or added them one by one, either standard error could not be computed or there is no convergence. Do you have any suggestion for me to improve the CFA fit?