RE: st: Mlogit with constraints

[Richard Williams <Richard.A.Williams.5@ND.edu>]

At 05:04 AM 12/12/2005, Maarten Buis wrote:
> Edlira:
>
> Just because you prefer one set of coefficients, doesn't mean that 
> that is the best way of describing the relationship. By constraining 
> all parameters (including the constant) to be the same you have, in 
> essence, done a single logit of categories 2 till 5 against category 
> 1, which is a severe constraint which needs to be justified. 
> Deciding whether this constraint is justified is what we do when 
> testing constraints. So doing a log likelihood ratio test of the 
> unconstrained model and the constrained model (-help lrtest-) seems 
> in order before reporting the constrained model.
>
> Note that if you constrain all the parameters (including the 
> constant) to be the same than the probabilities of group membership 
> for an average individual will also be the same for all groups. A 
> more sensible constraint would be to let the constant be different 
> across groups and constrain the effect of time to be the same across groups.
I agree with Maarten.  Also, one thing I find peculiar here is that 
the equality constraint on the constants isn't working, i.e. the 
constant in the first 3 equations is 0 while in the last it is 
non-zero.  I would have expected a non-zero and identical value for 
the constant in all 4 equations. I am not sure why this is but in any 
event I don't think you would want the constants constrained to be 
equal anyway.


-------------------------------------------
Richard Williams, Notre Dame Dept of Sociology
OFFICE: (574)631-6668, (574)631-6463
FAX:    (574)288-4373
HOME:   (574)289-5227
EMAIL:  Richard.A.Williams.5@ND.Edu
WWW (personal):    http://www.nd.edu/~rwilliam
WWW (department):    http://www.nd.edu/~soc 
*
*   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/