Proc RobustReg - standard errors

[VioletaB]

Hi all,

I want to use proc robust in my analysis for treatment effect; however, i can not find out how to output standard errors into table. Specially when proc robustreg used during multiple imputation process. Can anyone help?

Thank you

[PaigeMiller]

Example:

proc robustreg data=sashelp.class;
    ods output parameterestimates = paramestimates;
    model weight=height;
run;

[SAS_Rob]

The example below shows the syntax necessary to combine results from ROBUSTREG in MIANALYZE.

 
/*Assume the imputation is already done*/
data a (drop=i);
do _imputation_=1 to 25;
do i=1 to 1000;
do trt=1 to 3;

x1=rannor(1234);
x2=rannor(1234);
e=rannor(1234);

      if i > 900 then y=100 + e;
      else y=10 + 5*x1 + 3*x2 + .5 * e;
output;
end;
end;
end;
run;

proc robustreg data=a method=m(wf=huber);
by _imputation_;
class trt;
model y = x1 x2 trt;
ods output parameterestimates=parms;
run;

proc mianalyze parms(classvar=level)=parms;
class trt;
modeleffects intercept x1 x2 trt;
run;

 

 

[VioletaB]

Thanks for this. This helps, however, this is first time i use robustreg.  The code provide only estimate for one of treatment arm; is were a way to get those for both treatment arms and then a comparison like you would get using proc mixed?

[SAS_Rob]

It would be helpful if you could post your ROBUSTREG code.  

[VioletaB]

Pro sort data=imp_24;by _imputation_;run;

proc robustreg data=imp_24 method=m (wf=huber);

by _imputation_;

class trt;

model ratio=base trt;

ods output parameterstimates=parms;

run;

proc mianalyze [arms (classvar=level)=parms;

class trt;

modeleffects intercept base trt;

ods output parameterstomates=mi;

run;

Ratio and base are continuous variables and trt has 2 levels: 1,2.

it provides only information for one treatment arm. Can we get from proc robustreg similar results as we get from proc mixed: LSmeans for each treatment arm and then a comparison.

[SAS_Rob]

ROBUSTREG does not supply LSMeans.

Since it is using GLM coding and you only have one CLASS variable and a single covariate, the estimate for the intercept is the adjusted mean value for the second treatment arm.