topic Re: Force a variable's coefficient during logistic regression ... in Statistical Procedures

Force a variable's coefficient during logistic regression ...

jzhang332002 — Fri, 26 Sep 2014 16:38:21 GMT

I want to do a logistic regression y (0 or 1) over a list of variables x1, x2, x3, x4. Now I want to fix x1's coefficient to be 2.5 and regress against x2-x4. How can I do it?

proc genmod data = mmm descending;

model y = x1 x2 x3 x4 /d=b;

run;

Now I want to fix x1's coefficient to be 2.5 such as:

proc genmod data = mmm descending;

model y = x1(coefficient to be fixed as 2.5) x2 x3 x4 /d=b;

run;

Is that doable? Thanks.

Jian

Re: Force a variable's coefficient during logistic regression ...

stat_sas — Fri, 26 Sep 2014 18:40:20 GMT

Please check estimation method for proc genmod.

Re: Force a variable's coefficient during logistic regression ...

Community_Help — Fri, 26 Sep 2014 20:18:09 GMT

Hi there - I moved your post from the general support community space to the SAS Statistical procedures community so users can see your question there. Thanks

Re: Force a variable's coefficient during logistic regression ...

lvm — Fri, 26 Sep 2014 20:52:13 GMT

Use the offset option. This forces the coefficient to be 1. So, you first create a variable that is 2.5*x1, and use this.

data mmm; set mmm;

x1_25 = x1*2.5;

proc genmod data=mmm descending;

model y = x2 x3 x4 x5 / offset=x1_25 dist=b;

run;

Re: Force a variable's coefficient during logistic regression ...

SteveDenham — Fri, 03 Oct 2014 18:08:38 GMT

Oh, that is sweet. I would have had some sort of data step where I subtracted off some horrible looking backtransformed from the logit value from the dependent.

Steve Denham

Re: Force a variable's coefficient during logistic regression ...

Rick_SAS — Fri, 03 Oct 2014 19:32:15 GMT

I agree. The drawback of the OFFSET= hack is that it only permits one variable, so isn't as versatile as the RESTRICT statement in PROC REG. Regardless. it's sooo clever!

Re: Force a variable's coefficient during logistic regression ...

lvm — Fri, 03 Oct 2014 19:59:59 GMT

If you want b1=2.5 and b2=5.0, one could do:

data mmm; set mmm;

offset = 2.5*x1 + 5*x2;

run;

proc genmod data=mmm descending;

model y = x3 x4 x5 / offset=offset dist=b;

run;

And so on...