topic Re: How to control for Covariates in PROC NPAR1WAY in Statistical Procedures

How to control for Covariates in PROC NPAR1WAY

Jep — Tue, 17 Sep 2019 14:39:32 GMT

Hello All,

Is there a way to control for covariates using the NPAR1WAY procedure? I have an outcome which is continuous but not normally distributed and a predictor which is categorical with 3 levels. I also have a covariate which is continuous but also not normally distributed that I would like to control for. Does anybody have an idea of how I can do this?

Re: How to control for Covariates in PROC NPAR1WAY

PaigeMiller — Tue, 17 Sep 2019 14:55:10 GMT

@Jep wrote:

Hello All,

Is there a way to control for covariates using the NPAR1WAY procedure? I have an outcome which is continuous but not normally distributed and a predictor which is categorical with 3 levels. I also have a covariate which is continuous but also not normally distributed that I would like to control for. Does anybody have an idea of how I can do this?

There is no such capability in PROC NPAR1WAY.

Your response variables don't have to be normally distributed in order to fit a model using PROC GLIMMIX, but you do have to figure out an approximate distribution of the responses; and you can use PROC GLM to fit a model if your response variable errors (not the variable itself) are normally distributed.

Your covariate does not have to be normally distributed.

Re: How to control for Covariates in PROC NPAR1WAY

PGStats — Tue, 17 Sep 2019 17:26:48 GMT

AFAIK, given the limited information that you provided, the best way to handle this is with PROC QUANTREG, which by default fits the median response. You can define the covariate effect as a polynomial or spline with the EFFECT statement, and test the main effect with nonparametric statistics in the TEST statement.