[R] sem package and AMOS

[John Fox]

Dear Anthony,

sem() does FIML estimation, not 2SLS, and so it's hard to understand
why you're getting "nearly identical" parameter estimates but very
different coefficient standard errors and model chi-squares. In fact,
unless the problem is very ill-conditioned, the parameter estimates
should be the same within rounding error, as should the model
chi-square. There is some room for small differences in the standard
errors -- sem() uses a numerical Hessian and I'm not sure what AMOS
does -- but large differences are indicative of some problem. 

I suspect that you're not fitting quite the same model in sem() and
AMOS.

I hope this helps,
 John

On Mon, 02 Feb 2009 14:50:26 -0600
 Anthony Dick <adick at uchicago.edu> wrote:
> Hello-
> 
> I am using R to build my initial models, but need to use AMOS to
> compare the models of two groups (adults vs. kids). The problem is I
> am getting different results with R and AMOS for the initial models
> of the separate groups (and the R results make more sense).
> 
> The parameter estimates (path coefficients and variances) from both
> programs are nearly identical, but the model chi-squares (and
> significance estimates of the parameter estimates) are different. I
> am using Maximum Likelihood in AMOS. R I think defaults to
> two-stage-least squares estimate, and AMOS 16 does not implement
> 2SLS.
> 
> I am using fMRI data, so the error variances are likely correlated,
> and the data non-normal to varying degrees. Is 2SLS the better way to
> go for these kinds of data?
> 
> Is there a way to change the default method for R sem? I couldn't
> find this in the ?help. I note I have run some of the AMOS examples
> in R and have gotten identical results across platforms, so I believe
> the problem is not in specifying things incorrectly across platforms.
> Also, the dfs are identical for both analyses. I must use AMOS to do
> model comparison (and thus maximum likelihood), but would like to
> achieve similar results across platforms for the basic models before
> I proceed, and would like to track down the reason for the
> difference.
> 
> Thanks,
> 
> Anthony
> 
> -- 
> Anthony Steven Dick, Ph.D.
> Post-Doctoral Fellow
> Human Neuroscience Laboratory
> Department of Neurology
> The University of Chicago
> 5841 S. Maryland Ave. MC-2030
> Chicago, IL 60637
> Phone: (773)-834-7770
> Email: adick at uchicago.edu
> Web: http://home.uchicago.edu/~adick/
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
> http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

--------------------------------
John Fox, Professor
Department of Sociology
McMaster University
Hamilton, Ontario, Canada
http://socserv.mcmaster.ca/jfox/