st: clogit, initial values not feasible

[Glen Waddell <glen.waddell@gmail.com>]

I am attempting a conditional logit model with obs in the 410,000 
range.  There are 70 groups spanning 9 years.  I have four RHS variables 
of interest.  Upon executing the clogit, "note: multiple positive 
outcomes within groups encountered" appears.  This is expected, given 
the underlying data.

If I take a random sample (of approx 1 percent) the model runs 
successfully.  However, the following error is returned when I attempt 
the full sample (or anything above a 10 percent sample).

initial values not feasible
r(1400)

This is returned whether or not I feed it initial values (e.g., from a 
comparable OLS specification). 

The return code and my experience with the small sample has me believing 
that size matters, in some sense.  yet, I am finding this on a machine 
that allocates 15GB to memory in Stata.  With the dimension above, what 
should I expect from clogit?

Thoughts?

--


Glen Waddell
Department of Economics  |  University of Oregon
http://www.uoregon.edu/~waddell



*
*   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/