Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
st: meologit - multilevel ordered logit - converging problems
From
<[email protected]>
To
<[email protected]>
Subject
st: meologit - multilevel ordered logit - converging problems
Date
Sun, 22 Sep 2013 22:49:57 +0000
Julian Garritzmann <[email protected]> reports a convergence problem using
the -meologit- command. My guess is that his model is too complicated for the data he has.
Julian remarks that "I can, of course, include the macro level variables with the other micro
variables, but then of course the SEs are incorrect"
I will assert that the SEs would be incorrect even if the model converged. The reason is that there are only 17 level-2 observations. SEs with nice properties for the country parameters require a "large" number of level-2 obs.
For more about this, see http://econpapers.repec.org/paper/bocusug13/04.htm and, for more gory detail, https://www.iser.essex.ac.uk/publications/working-papers/iser/2013-14 and references therein. The working paper also refers to regression methods aside from multilevel (mixed) models that one might apply to data with the same structure as Julian's.
Stephen
------------------
Stephen P. Jenkins <[email protected]>
> ------------------------------
>
> Date: Sat, 21 Sep 2013 11:44:11 -0400
> From: Julian Garritzmann <[email protected]>
> Subject: st: meologit - multilevel ordered logit - converging problems
>
> Dear List-users,
>
> I have a problem implementing multilevel ordered logit models in Stata
> (with the new meologit command).
>
> Back in Stata 12 I used the GLLAMM-commands but had the usual troubles
> (complex to implement and long computational time). So I was very much
> looking forward to Stata 13's new meologit-command. This command works
> fine as long as I only have variables on one level, but as soon as I
> include a second level, I constanlty receive error messages such as:
>
> nearby values are missing
> Iteration 12: log likelihood = -22475.38 (backed up)
> could not calculate numerical derivatives -- discontinuous region with
> missing values encountered
> could not calculate numerical derivatives -- discontinuous region with
> missing values encountered
> r(430);
>
> I am using survey data (18000 observations nested in 17 countries). The
> DV is given in five categories, so meologit seems ideal. As soon as only
> variables from one level are included, it works fine, so for example the
> follwing model works fine:
> meologit eduspendord income female educ age agesq i.empstat
> HHmax2persons govcutspend || country:
>
> However, as soon as I try to include variables on the second level, the
> models do not converge anymore. So, for example, the following models
> won't work (same model as above, but "open" included as a macro variable
> on the country level):
> meologit eduspendord income female educ age agesq i.empstat
> HHmax2persons govcutspend || country: open
> meologit eduspendord income female educ age agesq i.empstat
> HHmax2persons govcutspend || country: open inequality
>
> I can, of course, include the macro level variables with the other micro
> variables, but then of course the SEs are incorrect:
> meologit eduspendord income female educ age agesq i.empstat
> HHmax2persons govcutspend open || country:
>
> In general, I program in the following way:
> meologit DV IVs_individual_level || country: IVs_country_level
>
> I tried the following solutions:
> - - included only variables without missing values
> - - I also tried defining starting values or starting grids, but was not
> successful.
>
> What am I doing wrong? How could I solve this?
>
> Thank you so much!
>
> - ---
> Julian L. Garritzmann, M.A.
>
> Visiting Fellow at the Graduate School of Arts and Sciences
> Harvard University
>
> Wissenschaftlicher Mitarbeiter
> Lehrstuhl für Politikwissenschaft,
> insb. Policy-Analyse und politische Theorie
> Prof. Dr. Marius R. Busemeyer
> Universität Konstanz
Please access the attached hyperlink for an important electronic communications disclaimer: http://lse.ac.uk/emailDisclaimer
*
* 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/