topic Re: Proc logistic effect selection methods in SAS Programming

Proc logistic effect selection methods

eabc0351 — Wed, 28 Mar 2018 23:01:07 GMT

Hello. I am tying to force variables into my proc logisitic model in a blockwise fashion. See example below for blocks. I have not been able to figure out syntax to force in blocks, only individual variables. Any help would be appreciated. Thank you.

block one: zses zach

block two: zses zach zach2 zses2 zach3 zses3

block three: zses zach zach2 zses2 zach3 zses3 zach*zses

block four: zses zach zach2 zses2 zach3 zses3 zach*zses zach2*zses zses2*zach zach2*zses2

Re: Proc logistic effect selection methods

ballardw — Wed, 28 Mar 2018 23:15:32 GMT

What would "blockwise fashion" actually entail? What impact would your "block" have on your data?

If you mean that you want to force specific variables to stay in the model when using a selection method you pretty much have abandoned a lot of control of which variables are in your final model.

Re: Proc logistic effect selection methods

eabc0351 — Thu, 29 Mar 2018 00:13:34 GMT

No, I'm not talking about forcing variables to stay in the final model. I'm talking about forcing variables into the model in order to calculate the change in the -2 Log L between models. What I want to know is does the entry of the additional block of variables produce a significant pvalue for the change in -2 Log L between that model and the previous model. If it does, I would keep the block of variables in the model. If not, they come out.
This is not stepwise regression. It's heirarchal ordered entry. I can figure out the syntax in proc reg but cannot not figure it out in proc logisitc.

Re: Proc logistic effect selection methods

PGStats — Thu, 29 Mar 2018 03:30:36 GMT

What do you want proc logistic to do beyond fitting the four models and giving you some fit statistics?