[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: xtlogit vs gllamm6
I have been having problems receiving messages from the list for the past one week and I am sorry if this message has already come through and someone has already replied to it. Since I desparately need an answer to this I am posting it again.
I have run the same model using gllamm6 as well as xtlogit.
Here is the the summary of results I have got:
command issued: gllamm dp time....., family(binomial) link(logit) i(uni)
no of level 1 units=22321
no of level 2 units=19
level 2 var=0.12045214
xtlogit: (this is the RE)
command issued: xtlogit dp time..., re i(uni)
number of obs=22321
number of groups=19
log likelihood = -2797.01
comparison model last reported lok lik=-2973.4769
fitted full model rho=0.0 log lik = -2814.1954
sigma u = 0.3358706
Note: 1 compeltely determined panel
My questions are:
(1) why are the two log lik different (parameter estimates are different too)?
(2) under xtlogit, why is the simple logit lik different to the starting one with rho=0.0?
(3) Why am I getting the "Note: 1 completely determined panel" the comment at the end of the xtlogit set of results?
(4) Does this mean that xtlogit has dropped some observations from the estimation. This is the obvious reason why the log likelihood value has gone up. But then why does it still give me the same numebr of obs and groups as what I was expecting to get - same as gllamm?
Any help in solving this mystery would be very much appreciated.
Professor Wiji Arulampalam,
Department of Economics,
University of Warwick,
Tel: +44 (24) 7652 3471
Sec. Tel: +44 (24) 7652 3202
Fax: +44 (24) 7652 3032
* For searches and help try: