Re: st: Multilevel modelling: convergence, value of the log odds ratio, and dummy variable issues.

[Maarten Buis <maartenlbuis@gmail.com>]

It is very hard (impossible) for us to answer that question. Basically
we would need access to the data and the exact .do-file to look at
what might be the cause of that. The fact that you estimated your
models step-by-step does not guarantee that you can estimate your
model. In fact, it will do exactly nothing to help in that sense. It
does help you pin-down where the problem might be. So if you could
estimate your model with only the main effects, but the model with
interactions won't converge, then the problem is with the
interactions. Try not to estimate all interactions at once, but add
them one at the time. Now you can try to pin-point which interaction
is the problem. Now, you can look at lots of cross tabulations, and
see if you understand why there is a problem with that interaction.
Once you figured that out, you can make an informed decision on what
to do next. As you can see, it is impossible for us to do that and
diagnose the problem remotely.

A log odds ratio of 27 for an indicator (dummy) variable is extremely
high, 27 for the odds ratio is possible depending on the exact
circumstances. The reason why you get such numbers critically depends
on your data and model, so again it is not possible for use to
diagnose that problem over the internet.

With Stata 11 I would tend to use factor variables for the categorical
variables and their interactions rather than create the indicator
variables yourself.

-- Maarten (not Marteen)


On Fri, Nov 8, 2013 at 3:32 PM, liliana <liliana.andriano@gmail.com> wrote:
> Hello,
>
> I am a new statalist user. I have registered here because I need some help.
>
> I am writing my Master's thesis and as a part of it I am analyzing the data
> from the Demographic and Health Surveys (DHS).
> I am investigating the individual- and community-level determinants of child
> mortality. I have generated the variables and started to run the
> regressions. As Marteen suggests in many of his posts, I have run started
> from the null model, then added the individual-level variables, then the
> community-level variables, and then the interaction terms
> (individual*community).
>
> However, Stata (I hold the version 11.2)  tells me that the convergence is
> not achieved. I have included the iter(20) and the difficult options in the
> xtmelogit expression. Why?
>
> Moreover, in the last model (with the interactions) the dummy variable which
> assumes value 1 if the mother of the baby has a higher education has got a
> log odds ratio equal to 27. Can you imagine why?
>
> Also, all the variables in my model are dummies. Are dummies better than
> categorical variables for these estimations?
>
> Many thanks in advance.
>
> I look forward to hearing from you soon.
>
> Liliana
>
>
>
>
> --
> View this message in context: http://statalist.1588530.n2.nabble.com/Multilevel-modelling-convergence-value-of-the-log-odds-ratio-and-dummy-variable-issues-tp7580450.html
> Sent from the Statalist mailing list archive at Nabble.com.
> *
> *   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/



-- 
---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------
*
*   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/