topic Re: Using linear regression with categorical and continuous covariates in SAS Procedures

Using linear regression with categorical and continuous covariates

cassifields — Sat, 11 Apr 2015 13:13:06 GMT

I am trying to do a linear regression to computer whether body weight can predict blood pressure after controlling for age, healthstatus, and physical activity. Here, my DV (blood pressure) is continuous and my IV(body weight) is continuous. Also, I have several covariate IVs; age is continous, healthstatus is categorical (dummy coded 1-5), and physical activity is continous (# hours doing physical activity). To do this, should I use proc glm or proc reg?

Re: Using linear regression with categorical and continuous covariates

gergely_batho — Sat, 11 Apr 2015 23:53:43 GMT

I recommend proc glm. There you can use the class statement to dummy code the categorical variables.

You can use also proc reg, but then first you need to pre-process your data (creating dummy variables "manually").

Re: Using linear regression with categorical and continuous covariates

cassifields — Sun, 12 Apr 2015 21:41:56 GMT

Can you check multicollinearity using proc glm?

Re: Using linear regression with categorical and continuous covariates

Ksharp — Mon, 13 Apr 2015 12:06:47 GMT

You can use STEPWISE BACKWISW ... option of MODEL to get rid of multicollinearity and get the best model .

Re: Using linear regression with categorical and continuous covariates

SteveDenham — Mon, 13 Apr 2015 14:10:17 GMT

"Best models" with categorical variables are messy, and those methods in PROC REG really won't help. The categorical dummy variables are by definition collinear.

I would approach this with PROC GLM or MIXED, and look at the chapter on Analysis of Covariance in SAS for Mixed Models, 2nd ed.

Steve Denham

Re: Using linear regression with categorical and continuous covariates

Ksharp — Mon, 13 Apr 2015 14:42:40 GMT

Doc Steve,

Sorry to make you confused . My statistical background was not as good as you were .