Date:         Fri, 7 Jan 2011 18:44:57 -0600
Reply-To:     Robin R High <rhigh@UNMC.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Robin R High <rhigh@UNMC.EDU>
Subject:      Re: Logistic Regression Intercept Pr>Chisq
Comments: To: Tanmoy Mukherjee <tkmcornell@yahoo.com>
In-Reply-To:  <79137.24240.qm@web30202.mail.mud.yahoo.com>
Content-Type: text/plain; charset="US-ASCII"

Tanmoy,

The pvalue for the intercept that is not-significant implies that it is
close to 0 (on the logit scale?).  If you set all the other explanatory
data to 0, the intercept becomes a reference value so to speak, to predict
the probability of a success, which implies a 50/50 occurrence of the
success:

prob(y=1) = EXP(0) / (1 + EXP(0))
          = 1/2
          = .5

I would keep it in the model (and I assume by backward selection you
selected variables essentially on your own and not the code generated
version?)

Robin High
UNMC






From:
Tanmoy Mukherjee <tkmcornell@YAHOO.COM>
To:
SAS-L@LISTSERV.UGA.EDU
Date:
01/07/2011 06:17 PM
Subject:
Logistic Regression Intercept Pr>Chisq
Sent by:
"SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>



I am running a Binomial Logistic Regression and I am getting some weird
result with regards to the Intercept values. I will appreciate if someone
can shed some light on the interpretaion of the same.

Data set :
Total Number of observations 351877
Response =1 9511

Modeling probability of response=1

Using Backward selection method to compute the best model

Intercept
Wald Chi Square 0.38
Pr>ChiSquare      0.5376

Based on this does this mean that we should reject the Intercept.

I will appreciate if someone can please explain this.

Thanks and Regards,
Tanmoy




Tanmoy Kumar Mukherjee
3 Perrine Court,
East Brunswick, NJ 08816
Phone: 9173994540
Email: tkmcornell@yahoo.com