Date: Wed, 3 Nov 2010 05:32:05 -0400
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Peter Flom <peterflomconsulting@MINDSPRING.COM>
Subject: Re: Multinomial Logit-Insignificant pvalues
Content-Type: text/plain; charset="us-ascii"
Pam Shin wrote
I am building a MNL model that has three outcomes. However for some of the
covariates, the pvalue for chi-sq statistics comes up as insignificant for
one of the outcomes. Is that a problem? Does this imply that this
variable should not be part of model?
Any insight, input would be very helpful.
First, you shouldn't really be deciding on covariates based on p-values
(although many people do)
Second, could you provide some context?
Third, for the particular case where a variable's effects are large at one
comparison and small at others, I'd look at the actual values and the actual
comparisons and see what made substantive sense. One possibility is to do
TWO separate logistics, each dichotomous.
But, with all that said, and taken as cautions, in general, I'm inclined to
include variables that make a difference.
Peter Flom PhD.
Peter Flom Consulting LLC
5 Penn Plaza, Ste 2342
NY NY 10001