RE: st: Differences between LISREL and SEM in Stata 12 that might cause the same model to give different results??

[Cameron McIntosh <cnm100@hotmail.com>]

Brent,

Yes, as Richard requested, tell us exactly what the differences are. But with only three observed items, a one-factor model is just-identified, so I'm pretty sure that the only difference could be in the standard errors. You have used the cluster option so unless you also invoke LISREL's options for complex survey designs they will be different... and in LISREL the only estimator for a CFA/SEM model with a complex survey design is FIML, and it will use all available information (complete and partially complete cases). -sem- in Stata 12 has a comparable FIML procedure so you should get the same results if you do it this way across programs. Note, however, that a three-item, one-factor model has no chance of failing to fit the data, so you won;t learn anything by estimating it, unless this is purely a software comparison exercise.  

And as for the comment on rotation -- this is confirmatory not exploratory factor analysis, so no rotation is involved. The only type of CFA modeling that permits rotation is exploratory structural equation modeling (ESEM), which is currently available only in the Mplus package:

Asparouhov, T. & Muthén, B. (2009). Exploratory structural equation modeling. Structural Equation Modeling, 16, 397-438.http://www.statmodel.com/download/SEM-Asparouhov2009.pdf ;
Marsh, H.W., Muthén, B., Asparouhov, A., Lüdtke, O., Robitzsch, A., Morin, A.J.S., & Trautwein, U. (2009). Exploratory structural equation modeling, integrating CFA and EFA: Application to students’ evaluations of university teaching. Structural Equation Modeling, 16, 439-476.http://www.statmodel.com/download/SEM-Marsh2009.pdf
Marsh, H. W., Lüdtke, O., Muthén, B., Asparouhov, T., Morin, A. J. S., Trautwein, U. & Nagengast, B. (2010). A new look at the big-five factor structure through exploratory structural equation modeling. Psychological Assessment, 22(3), 471-491. http://www.statmodel.com/download/BIG5%20ESEM%20MS.pdf  ;

Cam

> From: richards.868@buckeyemail.osu.edu
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Differences between LISREL and SEM in Stata 12 that might cause the same model to give different results??
> Date: Mon, 31 Oct 2011 16:13:40 +0000
> 
> Hey everybody-
> 
> I have been trying to replicate other researchers' factor analysis 
> measurement model (estimating the relationship between three observed 
> variables and one latent factor). They ran the model using LISREL. I have tried running the same model, using the same data, using the structural equation modeling feature of Stata 12. 
> But I get different results. 
> 
> Here's the code I'm using:
> 
> sem (LNC5 -> perprescmc) (LNC5 -> travalonmc) (LNC5 -> chspklngmc) ///
> (perprescmc <- _cons@0) (travalonmc <- _cons@0) (chspklngmc <- _cons@0), ///
> covstructure(e.perprescmc e.travalonmc e.chspklngmc, zero) nocons ///
> trace iterate(400) latent(LNC5) vce(cluster clustid2) ///
> var(e.chspklngmc, init(0.125)) ///
> var(e.travalonmc, init(0.8)) ///
> var(e.perprescmc, init(4))
> 
> I emailed the authors, and they suggested the differing results may
> be due to the the possibility that a) Stata 12 calculates its standard 
> errors differently from LISREL and b) LISREL does not use the delta 
> method, while Stata 12 might use it. Another friend suggested it might
> due to the different rotation methods used by the two programs.
> 
> Does this sound correct - could these differences between LISREL 
> and Stata 12 be what's causing me to get different results? If not, are 
> there other differences between the two programs that might be causing 
> this?
> 
> Thank you for your consideration.
> 
> Brent
> 
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
 		 	   		  
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/