topic Re: Surveyreg to obtain robust standard errors in Statistical Procedures

Surveyreg to obtain robust standard errors

Callam1 — Mon, 25 Dec 2023 17:15:37 GMT

Is the code below correct to obtain robust standard errors for a regression where the outcome variable is binary (0,1) and the independent variables are categorical and binaries? The dataset is at individual level: one unique person (identified by ID) per row. Is it right to use ‘cluster id’ to obtain robust S.E.? Is it ok to use individual weights together with the ‘cluster id’ option as done below?

proc surveyreg data=matched_dataset;
cluster id;
class is_treated(ref='0') categorical_covariates;
model outcome_variable = is_treated covariate1 covariate2 …/ CLPARM solution;
weight _MATCHWGT_;
run;

Re: Surveyreg to obtain robust standard errors

sbxkoenk — Tue, 26 Dec 2023 10:32:31 GMT

Hello,

If you only want robust standard errors, you can indeed specify the cluster variable to be the identifier variable.

Like here:

proc surveyreg data=yourdata;
cluster id;
model y = x1 x2 x3 / covb;
run; QUIT;

And, of course, you can specify a WEIGHT statement (if the weights are sampling weights).

If you do not specify a WEIGHT statement or a REPWEIGHTS statement, PROC SURVEYREG assigns all observations a weight of one.

But are you sure you want PROC SURVEYREG??
Keep in mind that your outcome variable is binary (0,1).
PROC SURVEYLOGISTIC seems more appropriate then anyway.

Koen

Re: Surveyreg to obtain robust standard errors

Callam1 — Tue, 26 Dec 2023 21:48:46 GMT

Thank you