st: MLOGIT CLUSTER and XTLOGIT

[John Litfiba <cariboupad@gmx.fr>]

Dear all,

I need your help and expertise about LOGISTIC regression with Stata
and in particular with the MLOGIT and XTLOGIT functions...

There are some threads about these topics, but it seems that the
questions is still unanswered so maybe someone can help me :
I have a very large panel dataset with more than 100000 individuals
and 6 million observations (each individual has to make a choice at
different dates) and I wanted to fit a logit model...

I have two big problems

1) What exactly means the Cluster option with the MLOGIT stata
function? I know that it is the famous sandwith/robust variance
estimator (see FAQ by stata) but then, how the multinomial logit model
writes?
The random terms Ui's only depend on individual-cluster i, like a
fixed effect model? I do not know simply how to write the logit
equation estimated by mlogit when cluster() is used

2) The second point is that, knowing that, I have tried to estimate a
fixed effect logit (or a random effect)  on my database using
xtlogit... without success : either the log likelihood is infinitely
small, either the hessian is not demi definite positive :
does it comes from the large size of my panel data ? In other word :
is there any hope to fit a fixed effect logit on such a large dataset?


Many thanks in advance,
Best,

John
*
*   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/