Stanley Watt <[email protected]> asks:
> I'm sorry to bother you, but I was wondering whether anyone had ever run
> into this problem. This is a problem for me both on Stata 7 and Stata 8.
> I am running a clogit regression where the dependent variable can be 1
> multiple times for observations within the same group.
> When I run a clogit regression against an independent variable which is a
> characteristic of the choice (and which is the same across different
> groups), I get the following results:
> . clogit indicator rgdpch, group(id_number)
> note: multiple positive outcomes within groups encountered.
> Iteration 0: log likelihood = -41843.543
> Iteration 1: log likelihood = -38224.759
> Iteration 2: log likelihood = -37944.648
> Iteration 3: log likelihood = -37831.499
[estimation results follow with log-likelihood, standard error, etc. missing]
This is quite curious considering that, for both examples, -clogit- apparently
iterated towards a solution only to switch to a missing log-likelihood once
it got there.
As a first step, I recommend that Stanley make sure his Stata 8 is fully
up-to-date. During the lifetime of Stata 8 (July 2004), -clogit- was changed
from an internal command to one implemented in ado and optimized using -ml-
rather than Stata's internal optimizer. This change is 99% transparent, but
very slight differences can result for some datasets and, when they do, we
firmly believe the new results to be superior.
If Stanley's Stata 8 is already current, or if updating doesn't fix the
problem, I would be glad to look into this further if Stanley can email me his
data. Naturally, I would report back to the list as appropriate.
--Bobby
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
