Dear All
I have a number of questions re: the multinomial logit model. Replies on
any of the questions raised will be much appreciated.
Here is a brief description of what I'm doing:
I have a model of mobility of sportsmen using (annual) career data for
approx. 300 players (effectively this is a panel data set
but I'm looking at overall mobility)
Dependent var = 0 (stays with same team)
= 1 (moves to new club)
= 2 (retires from sport)
Indep vars = (age,experience, player performance stats, team performance
stats and various qualitative dummy vars.)
My questions are as follows:-
1. Can I compare results directly with a binomial logit (0,1) model?
a) Is comparison of pseudo R-squared from the two models valid?
b) Signs are good that (0,1,2) is a better spec, given lots of significant
p-values in categories 1 and 2 (11 in each category),
but are there formal tests I should be using to compare the two models?
2. I've investigated problems of collinearity but how can I test/account
for heteroscedasticity (or serial correlation)?
e.g. I know that logit specs can give 'robust' standard errors but
this doesn't appear to be the case for multinomial logit
3. Is there some testing I should be doing on the residuals? (I believe
these are less informative than residuals from OLS for
example).
Hope someone out there can help.
Thanks
Ron Dorsey
*
* 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/
