Colleagues,

I've been working with a series of hierarchical models using the gllamm
procedure.  I'm trying to estimate ordered logit parameters with a random
intercept on a data set with 33 countries and around 36,000 observations.
I have six level-1 variables, seven level-two variables, and several
cross-level interactions, with most of interesting being focused on the
cross-level interaction.

Here is my problem -- my understanding of the "best" practice for getting
parameters is to use adaptive quadrature.  That works pretty well for some
specifications of the model, but not when the entire set of variables are
used.  In that case, the maximization routine gets stuck at a particular
spot on the likelihood function and at each iteration just "back ups" to
the that spot.  I can get parameter estimates using "regular" quadature,
but I'm concerned about using those without a better understanding of why I
can get results with one and not the other.

I've re-run the other, partly specified models with both routines and get
similar (but not the same) results.  I can live with noting that the point
estimates are exactly the same, noting that its a source of uncertainty.
But, I can't be the first person who has run into the problem and I'm
interested in hearing your advice/thoughts on the best way to approach this.

Best,

scott

-- 
-- 

Scott D. McClurg
Chairperson, Political Science
Director of Graduate Studies, Political Science
Professor of Political Science and Sociology
PN-L Moderator
Department of Political Science
Southern Illinois University
Mailcode 4501
Carbondale, IL 62901

(e) [log in to unmask]
(p) 618.453.3179

**********************************************************
             Political Methodology E-Mail List
   Editors: Ethan Porter        <[log in to unmask]>
            Gregory Whitfield   <[log in to unmask]>
**********************************************************
        Send messages to [log in to unmask]
  To join the list, cancel your subscription, or modify
           your subscription settings visit:

          http://polmeth.wustl.edu/polmeth.php

**********************************************************