[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: Mlogit, zero cells, and constraints

From   James Shaw <>
Subject   st: Mlogit, zero cells, and constraints
Date   Mon, 30 Jul 2007 17:26:37 -0700 (PDT)

Dear Stata Listservers:

I have a question regarding the use of constraints
with mlogit.  I am fitting multinomial logit models to
a sparse data set and am running into problems with
categorical independent variables with zero
observations in some equations.  In cases where this
occurs, I am unable to derive the associated parameter
and standard error estimates.  The estimates go to
negative infinity, and the standard errors are

Is it valid to simply constrain the parameter
estimates to equal zero in such cases?  That is,
constrain the coefficient to be zero in equations with
zero obs but free it to be estimated in other
equations.  I observed changes in other parameter
estimates after applying such constraints, which I
would not expect.  

Also, other than collapsing adjacent categories, what
can be done when one is unable to derive parameter and
standard error estimates for a categorical independent
variable (dummy variable) in one or more equations due
to having too few, but not zero, observations?  I
would not think it safe to constrain the estimate to
be zero in such cases.

Thank you in advance for your help.


James W. Shaw, Ph.D., Pharm.D., M.P.H.
Assistant Professor
Director of Outcomes Research, Jefferson Headache Center
Department of Neurology
Jefferson Medical College
Thomas Jefferson University
Office:  215-955-2097
Cellular:  215-852-3045
Fax:  215-955-1960

Got a little couch potato? 
Check out fun summer activities for kids. 
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index