topic Re: Graph Multiple Regression with 2 lines in Graphics Programming

Graph Multiple Regression with 2 lines

LGrau — Thu, 05 Oct 2017 17:59:36 GMT

Hi everyone,

I am trying to create a graph for a multiple regression (used PROC GLM to create) with 9 covariates. I want to make a graph with two lines (druguse=0 and druguse=1), plugging in for 'typical' values of the other covariates (ie, age50, race=white, etc).

Is that even possible? I am using SAS Studio.

Here's what I've done so far:

PROC SGPLOT DATA=sample.name;

SCATTER X=BASELINE Y=CHANGE/GROUP=DRUGUSE;

REG X=BASELINE Y=CHANGE/CLI CLM NOMARKERS;

RUN;

However, this REG doesn't include all my covariates! Not sure what to do!

Re: Graph Multiple Regression with 2 lines

DanH_sas — Thu, 05 Oct 2017 18:01:22 GMT

Add GROUP=DRUGUSE to the REG statement as well.

Re: Graph Multiple Regression with 2 lines

WarrenKuhfeld — Thu, 05 Oct 2017 18:02:47 GMT

Typically a regression plot with multiple lines is only made when there is a CLASS variable (although there could be more than one) and a single continuous variable--the kind of graph you made with PROC SGPLOT. What is it you want? How do the other variables enter in?

Re: Graph Multiple Regression with 2 lines

LGrau — Thu, 05 Oct 2017 18:05:07 GMT

It gets really crazy ! Plus, doing that doesn't include my full model.

The full model has drug use, age, baseline score, bmi, income, high school, marijuana use, etc.

Re: Graph Multiple Regression with 2 lines

LGrau — Thu, 05 Oct 2017 18:11:20 GMT

My Model is this:

PROC GLM DATA=BIOS6623.AGG_MENT;

CLASS INCOME2;

MODEL DELTAAGGMENT=AGG_MENT0 HARD_DRUGS AGE BMI MARI ALCOHOL CSMOKER INCOME2 HS ADHY2_ NHW/SOLUTION CLPARM;

TITLE "MODEL 1: FULL MODEL FOR CHANGE IN AGGREGATE MENTAL SCORE FOR ALL VARIABLES";

QUIT; TITLE;

I want to create a graph with two lines (Hard_Drugs=1 and Hard_Drugs=0), with age=50 race=NHW Mari=0 alcohol=0 HS=0 ADHY2=1 CSMOKER=0, etc.

This is the code I have so far, but it is not happening.

PROC SGPLOT DATA=BIOS6623.AGG_MENT;
TITLE "CHANGE IN AGGREGATE MENTAL SCORE";
SCATTER X=AGG_MENT0 Y=DELTAAGGMENT/GROUP=HARD_DRUGS;
REG X=AGG_MENT0 Y=DELTAAGGMENT/GROUP=HARD_DRUGS CLI CLM NOMARKERS;
RUN;TITLE;

Re: Graph Multiple Regression with 2 lines

WarrenKuhfeld — Thu, 05 Oct 2017 18:30:11 GMT

It might be that the EFFECT and EFFECTPLOT statements in many of the modeling procedures (but not GLM) can do what you want. Sorry, but I don't have time today to play with this.

http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_introcom_sect015.htm

http://support.sas.com/documentation/cdl/en/statug/68162/HTML/default/viewer.htm#statug_introcom_sect025.htm

Re: Graph Multiple Regression with 2 lines

WarrenKuhfeld — Thu, 05 Oct 2017 21:10:27 GMT

ods graphics on;
proc orthoreg data=sashelp.class;
   class sex;
   effectplot slicefit;
   model weight = height age sex;
run;

See if this gets you going in the right direction. I hope this helps.

Re: Graph Multiple Regression with 2 lines

ballardw — Thu, 05 Oct 2017 22:03:13 GMT

@LGrau wrote:

It gets really crazy ! Plus, doing that doesn't include my full model.

The full model has drug use, age, baseline score, bmi, income, high school, marijuana use, etc.

Which would make that a 7+ dimension graph. We don't really have good technology for displaying 3 dimensions and you want to display more than 4 more? In a single plot?

Re: Graph Multiple Regression with 2 lines

Rick_SAS — Fri, 06 Oct 2017 14:23:55 GMT

For a discussion and examples of using the EFFECTPLOT statement to slice and dice high-dimensional regression models, see "Use the EFFECTPLOT statement to visualize regression models in SAS."

Re: Graph Multiple Regression with 2 lines

LGrau — Fri, 06 Oct 2017 19:55:30 GMT

This worked! However, it fits de line using the means of the covariates. Is there a way to fit with me specifying which values to use?

Re: Graph Multiple Regression with 2 lines

Rick_SAS — Fri, 06 Oct 2017 20:32:10 GMT

Yes. My blog post links to the documentation. You might want to look at the AT= option.

You can specify statistics (mean, min, max, midrange) or specify particular values:

effectplot / at(x=min max x=2 5 7);

Re: Graph Multiple Regression with 2 lines

LGrau — Fri, 06 Oct 2017 20:35:44 GMT

That gives me the error "VARIABLE CONFIGURATION IS NOT CURRENTLY SUPPORTED BY THE EFFECTPLOT STATEMENT"

Re: Graph Multiple Regression with 2 lines

Rick_SAS — Fri, 06 Oct 2017 20:39:14 GMT

Please post your code and specify what version of SAS you are running.

Re: Graph Multiple Regression with 2 lines

LGrau — Fri, 06 Oct 2017 20:40:31 GMT

I'm using SAS Studio!

ods graphics on;
proc orthoreg data=BIOS6623.AGG_MENT;
CLASS HARD_DRUGS INCOME2;
effectplot slicefit/ cli clm;
model DELTAAGGMENT=AGG_MENT0 HARD_DRUGS AGE BMI MARI ALCOHOL CSMOKER INCOME2 HS ADHY2_ NHW;
run;

proc orthoreg data=BIOS6623.AGG_PHYS;
CLASS HARD_DRUGS INCOME2;
effectplot slicefit/ cli clm;
model DELTAAGGPHYS=AGG_PHYS0 HARD_DRUGS AGE BMI MARI ALCOHOL CSMOKER INCOME2 HS ADHY2_ NHW;
run;

proc orthoreg data=BIOS6623.LEU3N;
CLASS HARD_DRUGS INCOME2;
effectplot slicefit/ cli clm;
model DELTALEU3N=LEU3N0 HARD_DRUGS AGE BMI MARI ALCOHOL CSMOKER INCOME2 HS ADHY2_ NHW ;
run;

proc orthoreg data=BIOS6623.LOG10VLOAD;
CLASS HARD_DRUGS INCOME2;
effectplot slicefit/ cli clm;
model LOGVLOAD= LOGVLOAD0 HARD_DRUGS AGE BMI MARI ALCOHOL CSMOKER INCOME2 HS ADHY2_ NHW ;
run;

Re: Graph Multiple Regression with 2 lines

Rick_SAS — Sat, 07 Oct 2017 11:45:39 GMT

SAS Studio is an interface. It does not tell me what version of SAS you are using by running

%put SYSVLONG = &SYSVLONG;

The examples you posted do not use the AT= option. Please post the code that is giving the error, along with the portion of the SAS log that shows the error messages.

Re: Graph Multiple Regression with 2 lines

Rick_SAS — Mon, 09 Oct 2017 14:03:11 GMT

I do not know the valid values of your continuous variables, but try something like this:

CLASS HARD_DRUGS INCOME2;
effectplot slicefit/ at(

AGE = 30

BMI = 25

MARI = 0

ALCOHOL = 1

CSMOKER = 1

HS = 1

ADHY2_ NHW = 0;

);

Whatever value you use on the right-hand side will determine the value at which the model is evaluated for the effectplot. The first continuous variable MODEL statement (AGG_MENT0 ) determines the continuous axis for the effect plot.