CFA with ordinal data PreviousNext
Mplus Discussion > Confirmatory Factor Analysis >
Message/Author
 Thomas Carpenter posted on Friday, November 30, 2012 - 3:31 pm
Hi there,

I am trying to replicate an analysis done in LISREL. Because I have ordinal indicators (7-point likert), I used polychoric correlations and an asymptotic covariance matrix in lisrel with diagonally weighted least squares. I am trying to do the same analyses in MPlus.

To do this, I used (1) categorical are... to make the indicators categorical. I also (2) used the WLSMV estimator. However, fit is much worse using this method than the lisrel procedure above. I found that ULSMV gave me a better approximation to the lisrel results, but still they are giving me a different pattern of results. Any idea how to use (1) polychoric, (2) asymptotic covariance matrix, and (3) diagonally weighted least squares in mplus?

Note: I am open to other procedures than the above, but at the very least I would like to be able to replicate the first analysis.
 Linda K. Muthen posted on Friday, November 30, 2012 - 7:34 pm
You are doing this correctly in Mplus. We know that the weight matrices differ between Mplus and LISREL. It seems odd the LISREL results are closer to unweighted least squares.

How large is your sample? Does it have a threshold structure or only correlations. Is it a single or multiple group analysis?
 Thomas Carpenter posted on Monday, December 03, 2012 - 1:42 pm
Thanks Linda for the quick reply.

The same size is small, n = 146, so I was treating the data with caution anyway. However, I had expected the two software packages to converge at least a little more. It is a single-group analysis. I am analyzing raw data, not correlations.

Could it be the small sample size?

The fit good to fair in LISREL, but poor in MPlus. Note that we did the analyses in R and got results similar to MPlus's WLS results.
 Bengt O. Muthen posted on Tuesday, December 04, 2012 - 2:35 am
It could be due to the small sample size. Our simulations show that the chi-square performance is reasonable down to about n=200. I haven't seen LISREL simulations on this.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: