Re: st: "not concave"

["JVerkuilen (Gmail)" <jvverkuilen@gmail.com>]

On Fri, Mar 1, 2013 at 7:55 AM, Wahideh Achbari
<wahideh.achbari@vub.ac.be> wrote:
> Hi everyone,
>
> Here's a query about a very simple confirmatory factor analysis with 4
> factors. I have been trying to estimate this model, but unfortunately
> the model is "not concave" running a Maximum Liklihood algorithm. What
> surprises me most is that I have been running the same model in AMOS
> and MPLUS without any difficulties. I would like to know what can be
> done to remedy this. Your help is greatly appreciated.

As William Buchanan noted, you have to check to see whether you really
are running the same model. You may not be. You're also using the
missing data option, which is a likely substantial source of
difference across programs.

First thing I'd do is to use listwise deletion to fit the model on
complete data in all three programs. If you get the same answer (up to
identification constraints and the usual level of numerical
imprecisions) then you know it's the missing data that's causing the
trouble.

Also this model is right at the level of identification, so you may
have empirical unidentification issues, which can really differ across
programs in how they're handled. Your model may run but have ludicrous
standard errors in one program and crash in another.

Study the observed correlations to see if there are any really extreme
differences, say one that's .9 and others that are -.1. Factor
analysis doesn't like that.
*
*   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/