Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: inflated SEs with FIML in sem


From   "JVerkuilen (Gmail)" <jvverkuilen@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: inflated SEs with FIML in sem
Date   Mon, 31 Dec 2012 09:42:43 -0500

On Sun, Dec 30, 2012 at 11:46 PM, Eric Reither <eric.reither@usu.edu> wrote:
> Hello everyone.
>
> I have a question about the estimation of SEs with complex survey data in Stata, when using the FIML option in the new sem command.  But first, a small bit of background.

<snip>

> Now the question(s): Has anyone else had a similar experience with sem in Stata?  Any thoughts about the cause of the problem, or about possible solutions?>

I haven't run both FIML and svy at the same time, so I'm just going to
have to speculate a bit and hopefully other folks will chime in. FIML
partitions the multivariate normal likelihood to deal with the missing
data (in effect making missing parts fit perfectly). This is briefly
described on the bottom of p. 211 of the SEM manual. Somewhere the svy
weights and the partitioning aren't working correctly together,
possibly by creating a block of constant data.

You might want to output as many summary statistics as possible and
look for sources of trouble, such as extraordinarily large standard
errors or a singular Hessian. Are the iterations throwing the
non-concave log-likelihood error?

In many respects I think MI is probably better than FIML because it
doesn't assume multivariate normality.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index