topic Re: OLS regression model in Statistical Procedures

OLS regression model

uzma03505621 — Wed, 02 Jun 2021 00:24:09 GMT

Hi everyone,

I am using proc reg for the analysis of my study data.

Dependent variable= mcs score

independent variable= cat2 cat3 age income

where cat2 and cat3 are both categorical variables. My reference group is category 1..I have created dummy variables.

My code is as follows:

ods graphics on;
proc reg data=dummyfinal plots(maxpoints=none);
model mcs42=cat2 cat3;
output out=new P=YHAT RSTUDENT=RESID L95M=LOW U95M=HIGH;
run;
quit;
ods graphics off;

after getting the predicted value (YHAT) of the dependent variable, I have to obtain the mean mcs scores across the 3 categories (cat1 cat2 cat3) along with the confidence intervals and do multiple comparison tests( eg: Tukey kramer).

can anyone please help me with the SAS codes that I should run to obtain the following results.

My results should look like this:

means and SE
	MCS
	Mean (SE)	p value
cat1( reference group)	40.45 (0.94) ^∗∗∗	<0.001
cat2	43.76 (0.71) ^∗∗∗	<0.001
cat3	46.96 (0.78) ^∗∗∗	<0.001

Re: OLS regression model

PaigeMiller — Wed, 02 Jun 2021 11:40:30 GMT

@uzma03505621 wrote:

Hi everyone,

I am using proc reg for the analysis of my study data.

Dependent variable= mcs score

independent variable= cat2 cat3 age income

where cat2 and cat3 are both categorical variables. My reference group is category 1..I have created dummy variables.

My code is as follows:

ods graphics on;
proc reg data=dummyfinal plots(maxpoints=none);
model mcs42=cat2 cat3;
output out=new P=YHAT RSTUDENT=RESID L95M=LOW U95M=HIGH;
run;
quit;
ods graphics off;

after getting the predicted value (YHAT) of the dependent variable, I have to obtain the mean mcs scores across the 3 categories (cat1 cat2 cat3) along with the confidence intervals and do multiple comparison tests( eg: Tukey kramer).

can anyone please help me with the SAS codes that I should run to obtain the following results.

My results should look like this:

means and SE

MCS

Mean (SE) p value

cat1( reference group) 40.45 (0.94) ^∗∗∗ <0.001

cat2 43.76 (0.71) ^∗∗∗ <0.001

cat3 46.96 (0.78) ^∗∗∗ <0.001

There are some things that really aren't clear, such as you say age and time are independent variables, but these are not in your model. In addition, you talk about a reference category of category1, even though you haven't put category1 into the model, and I assume these are three levels of a single variable.

So anyway, here is how to handle a categorical variable with 3 levels, which I have named CAT.

To get means in this case, you can use PROC GLM, and you don't have to create the dummy variables yourself.

proc glm data=dummyfinal;
class cat(ref='1');
model mcs42=cat;
means cat/t;
quit;

Re: OLS regression model

uzma03505621 — Wed, 02 Jun 2021 12:18:59 GMT

I'm sorry for the typing error, the correct code is :

ods graphics on;
proc reg data=dummyfinal plots(maxpoints=none);
model mcs42=cat2 cat3 age income;
output out=new P=YHAT RSTUDENT=RESID L95M=LOW U95M=HIGH;
run;
quit;
ods graphics off;

My original variable is called category with three values (1,2,3). I created dummy variables as follows:
If category=2 then cat2=1 else 0
If category=3 then cat3=1 else 0
So when I run the proc reg program, category=1 will be used as reference by-default? ( please correct me if I am wrong)

Thanks for proc glm code, I have used this before. I am being told to use proc reg only, that's why I created dummy variables, I need help with codes to compare the new regression adjusted mcs MEANS (with confidence interval) between these 3 categories.

I appreciate your time and consideration.

Thank you.

Re: OLS regression model

PaigeMiller — Wed, 02 Jun 2021 12:38:24 GMT

@uzma03505621 wrote:
I'm sorry for the typing error, the correct code is :

ods graphics on;
proc reg data=dummyfinal plots(maxpoints=none);
model mcs42=cat2 cat3 age income;
output out=new P=YHAT RSTUDENT=RESID L95M=LOW U95M=HIGH;
run;
quit;
ods graphics off;

My original variable is called category with three values (1,2,3). I created dummy variables as follows:
If category=2 then cat2=1 else 0
If category=3 then cat3=1 else 0
So when I run the proc reg program, category=1 will be used as reference by-default? ( please correct me if I am wrong)

Thanks for proc glm code, I have used this before. I am being told to use proc reg only, that's why I created dummy variables, I need help with codes to compare the new regression adjusted mcs MEANS (with confidence interval) between these 3 categories.

I appreciate your time and consideration.

Thank you.

I don't know how to get the means that you are asking for using PROC REG only. As you can see, it's very easy to get the means from PROC GLM.

Re: OLS regression model

uzma03505621 — Wed, 02 Jun 2021 15:06:28 GMT

I used dummy coding:

data dummyfinal;
set finalfile;
if category=2 then cat2=1; else cat2=0;
if category=3 then cat3=1; else cat3=0;

run;

Then I did proc reg (unadjusted model without income and age) as follows:

/*unadjusted model*/
ods graphics on;
proc reg data=dummyfinal plots(maxpoints=none);
model mcs42=cat2 cat3;
output out=new P=YHAT RSTUDENT=RESID L95M=LOW U95M=HIGH;
run;
quit;
ods graphics off;

My results look this like

Parameter Estimates
Variable	Label	DF	Parameter Estimate	Standard Error	t Value	Pr > \|t\|
Intercept	Intercept	1	41.48793	0.26104	158.93	<.0001
cat2		1	-4.81016	0.43795	-10.98	<.0001
cat3		1	11.55753	0.28816	40.11	<.0001

Now Yhat is my predicted dependent variable, I want to compare the statistically significant difference in the means of my predicted dependent variable across my three category (independent) variable.

So should I do ANOVA+post hoc of this predicted Yhat? or is there any other method to get an output as below:

Unadjusted means and SE
	PCS		MCS
	Mean (SE)	p value	Mean (SE)	p value
Category 1 (reference)	37.27 (0.96) ^∗∗∗	<0.001	40.45 (0.94) ^∗∗∗	<0.001
Category2	37.02 (0.97) ^∗∗∗	<0.001	43.76 (0.71) ^∗∗∗	<0.001
category3	38.38 (1.04) ^∗	0.016	46.96 (0.78) ^∗∗∗	<0.001

Re: OLS regression model

PaigeMiller — Wed, 02 Jun 2021 15:24:04 GMT

I think the real problem here is whoever told you that PROC REG has to be used, this is bad advice, when PROC GLM makes this simple.

Nevertheless, I still don't know how to do this with PROC REG, specifically I'm not sure how you get the CORRECT standard errors, and so I cannot advise further if PROC REG has to be used.