10-15-2015 03:49 PM
I am trying to estimate a model using 3sls (because of cross-equation correlation) that will account for survey weights and heteroskedasticity. I am using the following basic syntax:
proc model data=mydata;
fit equation 1...equation7 / 3sls hccme=1;
I have tried different options for hccme, and none of them appear to be giving me appropriate standard errors (everything's significant, which I know is not the case). I originally tried estimating the model using proc syslin since the models are indeed linear; however, I did not see an option for obtaining robust standard errors with that procedure. Any help regarding correcting the proc model syntax or on other procedures that I can look into would be much appreciated.