**Date:** Thu, 24 Aug 2006 07:27:46 -0700
**Reply-To:** "docsms@gmail.com" <docsms@GMAIL.COM>
**Sender:** "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
**From:** "docsms@gmail.com" <docsms@GMAIL.COM>
**Organization:** http://groups.google.com
**Subject:** Logistic Regression for Grouped & Ungrouped Data
**Content-Type:** text/plain; charset="iso-8859-1"
Hello All,

I realize that my posting was dissappear, and I don't know why. Maybe,
I added another question, which replace the original posting. I am
sorry if I cause any inconvenience to you.

Hello. I am examining whether analysts' certain types of forecast
issuing given firm characteristics, let's say analysts' negative
recommendation.

Dependent variable (Y_j) is an individual analyst j's issuing of
negative recommendation (0 or 1, binormial) and independent variable is

firm size (X1), positive earnings (X2), for example. And assume that
the number of analysts following the firm i is N, and the total number
of analysts' negative recommendation for firm i is SUM_Y.

I can use logistic regression model with Grouped data;

Proc GENMOD Data=AAA;
Model SUM_Y/N = X1 X2 / link=logist d=binomial;
Run;

My question is, can I take logistic regression model with ungrouped
data approach? That means,

Proc GENMOD Data=AAA;
Model Y_j = X1 X2 /link=logit d=binomial;
Run;

Q1. If I take the second approach, the X1 and X2 variables will be
constant for a firm. But, they are varing with firm level. Will it
cause any problem?

Q2. My data is very unbalanced. That means, N is varing seriously from

1 - 300. Even with this seriously unbalnaced data, can I use logistic
model with grouped data? I mean, for example, 1 negative
recommendation with 3 analysts is 33%, but only 10% with 10 analysts.
Is this causing any problem?

Q3. In the case of using grouped approach, should I use N as weight
too?

Thank you your help in advance.

Minsup