st: -xtlogit- vs -xtmelogit- and predicted random effects

[Jeph Herrin <stata@spandrel.net>]

I have a random effects logit model which I would like to bootstrap in order
to get confidence intervals on the random effect terms. The model has a single
random intercept, and I can estimate it two different ways:

  xtset groupid
  xtlogit depvar indvar1 indvar2

or

  xtmelogit depvar indvar1 indvar2 || groupid:


These two models give nearly identical results, which is good, but while the
first model takes 6 minutes to run, the second takes 123 minutes (!). Since I
am bootstrapping 1000 samples, only -xtlogit- is practical.

However, despite StataCorp's assurance 4 years ago that it would look into
providing predicted random effects after estimation using -xtlogit-:

  http://www.stata.com/statalist/archive/2007-03/msg00135.html

this feature is still not available. My questions:

 1) I cannot find an estimation option for -xtmelogit- to match the adaptive
    quadrature of -xtlogit-, which would presumably speed it up. Is there one
    that I am missing?
 2) Or, has someone else looked under the hood of -xtlogit-, as it were, and
    written some code for getting the predicted random effects?
 3) This is for StataCorp: any progress on adding this postestimation option
    to -xtlogit-, per the above posting?


Much to my chagrin - this was for a revise and resubmit on a manuscript, so we had
a very short time to work - I gave this problem to a SAS-using colleague, who ran
it in GLIMMIX in the course of day. GLIMMIX uses pseudo ML, which I know doesn't do as
good a job of estimating the random effect terms, but in a pinch one has to do what
one can - however, I would be very pleased to have a Stata solution (that does not
take 3 months to run).

thanks,
Jeph
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/