topic Re: How to estimate spline function in a multi-variable logistic regression in Statistical Procedures

How to estimate spline function in a multi-variable logistic regression

Chung-Li — Mon, 22 May 2017 07:52:32 GMT

Hi all,

When researchers want to conduct the relationship between a binary dependent variable and independent variables, one way they can do is as follows:

PROC LOGISTIC DATA = MyDtSt;
    MODEL Y(event='1') = Main X, confounders;
RUN;

This one is easy!

It simply use main x to construct a model for Y, also with the adjustment of confounders.

However, my question is that what if I want to use spline(X) to model Y, and at the same time I want to control for linear confounding effects?

Is there any way allows me to model binary Y with spline function of X, and also enables me to take into account confounding effects?

Hope you guys can provide me with some suggestions.

Thanks in advance!

Re: How to estimate spline function in a multi-variable logistic regression

Ksharp — Mon, 22 May 2017 12:15:07 GMT

Calling @Rick_SAS

Re: How to estimate spline function in a multi-variable logistic regression

PaigeMiller — Mon, 22 May 2017 12:25:03 GMT

The EFFECT statement in PROC LOGISTIC (and many other PROCs) allows you to use a spline fit through X in the model. This example, which uses PROC GLIMMIX, ought to function the same in PROC LOGISTIC: http://documentation.sas.com/?cdcId=statcdc&cdcVersion=14.2&docsetId=statug&docsetTarget=statug_introcom_sect015.htm&locale=en

Re: How to estimate spline function in a multi-variable logistic regression

Rick_SAS — Mon, 22 May 2017 12:35:52 GMT

You can use the EFFECT statement in PROC LOGISTIC to define a spline effect.

Simple example with lots of links to papers and doc: "Nonsmooth models and spline effects"

The EFFECT statement supports several kinds of splines, so read the doc for how to specify the basis functions. For an example that uses restricted cubic splines, see "Regression with restricted cubic splines in SAS". The article demonstrates linear regression, but you can use the same ideas and syntax in PROC LOGISTIC.

Re: How to estimate spline function in a multi-variable logistic regression

Chung-Li — Tue, 23 May 2017 00:09:17 GMT

Ksharp,

Thank you for doing this!

Re: How to estimate spline function in a multi-variable logistic regression

Chung-Li — Tue, 23 May 2017 00:11:45 GMT

Paige Miller,

Thank you for this information!

I'll definitely check "effect" statement.

Re: How to estimate spline function in a multi-variable logistic regression

Chung-Li — Tue, 23 May 2017 00:15:10 GMT

Rick,

This is exactly what I'm looking for!

With this information, I think I can achieve what I want.

Thank you!