What is the proper way to estimate a logistic model

bigbigben — Wed, 03 Jun 2015 19:09:37 GMT

Assuming I have a set of data where the response variable y only takes the value of 0 and 1, and x are the covariate.

Follow the logic of the logistic function, define:

p=Probability(y=1) and we know the underlying function is that

p= 1/(1+exp(-beta*x))

How should I write the SAS code for it?

Is the following correct:

proc genmod data=XXXX

model y= x/dist=binomial

I am not sure if SAS actually maximize the likelihood function according to the aforementioned specification or SAS simply takes

g(y)=logit(y)=x*beta and run a OLS type regression to get beta.

If that is case, I do not think , and what should be the correct setup?

Thank you very much.

Re: What is the proper way to estimate a logistic model

lvm — Thu, 04 Jun 2015 15:38:06 GMT

Both GENMOD and LOGISTIC (and other PROCs) use MLE for a binomial distribution, definitely not OLS. Check out LOGISTIC for great graphics.

Reeza — Thu, 04 Jun 2015 15:41:27 GMT

Use proc logistic