Re: st: Multinomial logit - not significant coefficents -

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Sun, 13/6/10, Paolina Medina wrote:
> I'm running a multinomial logit with 10 possible outcomes.
> Some of the variables i am including are significant for
> the 6 first outcomes, but not for the last outcomes.
> Should i just keep the variables that are significant for
> all the possible outcomes and run the mlogit again?
> Is there anyway to use different regressor for the
> different possible outcomes? Would it be conceptually right?

Excluding variables because they are insignificant is always
conceptually wrong. If you think a variable is interesting 
enough to introduce it in your model, then it is still 
interesting to show that it is insignificant. If you think of 
these variables as control variables, see:
<http://www.stata.com/support/faqs/stat/stepwise.html>

However, it is even worse in -mlogit-, the coefficients do 
not correspond with one outcome but two: the outcome mentioned
in the equation and reference category. Unless you have a 
very specific hypothesis that that ratio of relative risks
equals zero, you should not impose that constraint.

Hope this helps,
Maarten

--------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany

http://www.maartenbuis.nl
--------------------------


      

*
*   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/