I assume you want to know how the starting values of the parameter estimates are computed. The starting values are displayed by using the ITPRINT option in the MODEL statement. The starting values are computed as discussed in the McCullagh and Nelder reference (section 2.5 in the 2nd edition). Basically, you can get them using PROC REG by regressing the logit of the response, y, on the predictors with weight ny(1-y). For example, using the remission data in the stepwise example in the PROC LOGISTIC documentation:
data a; set remission;
y=.1*(remiss=0)+.9*(remiss=1);
logit=log(y/(1-y));
wt=y*(1-y);
run;
proc reg;
model logit=smear cell;
weight wt;
run;
Note that for binary response data (rather than binomial, events/trials, data), n=1 and zeros are replaced by 0.1 and ones are replaced by 0.9. For binomial data, y=events/trials with the same adjustment if y=0 or 1.
... View more
