Re: st: Start values for MLE with xtmixed

[Arne Risa Hole <arnehole@gmail.com>]

This won't help the original poster but you can supply your own
starting values and impose constraints with -mixlogit- (available from
SSC) as this uses -ml- to maximise the simulated likelihood function.
See Figure 1 in
http://www.shef.ac.uk/polopoly_fs/1.105377!/file/2011011.pdf for an
example of how this feature can be used to study empirical
identification as mentioned by Jay.

Arne

On 24 October 2012 13:48, JVerkuilen (Gmail) <jvverkuilen@gmail.com> wrote:
> On Wed, Oct 24, 2012 at 3:03 AM, Maarten Buis <maartenlbuis@gmail.com> wrote:
>> In the case of starting
>> values, you go to the link called maximize_options, and than you look
>> for an option called -from()-. No such option is listed, so you cannot
>> provide starting values.
>
> I looked for that too and came to the same conclusion. -xtmixed- is
> using the EM algorithm to generate starting values. For the vast
> majority of situations it's what you'd want and is thus the
> appropriate option.
>
> However, there are a number of circumstances---examples include random
> effects meta-analysis with a relatively complex design, profiling a
> random effect's parameter(s), or studying empirical
> identification---where it would be highly useful to be able to assign
> parameters specific starting values and/or fix them in place with no
> iteration. To my knowledge (and based on prior statements by Vince
> Wiggins here: http://www.stata.com/statalist/archive/2012-09/msg00518.html),
> -xtmixed-, -xtmelogit-, and -xtmepoisson- can't do that at present. I
> think you could probably do that with -gllamm-.
>
> Jay
> *
> *   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/
*
*   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/