Standard errors for heteroscedasticity and cluster

[rfrancis]

Hello, I'm searching for a method (preferably a SAS Proc) that will generate standard errors for a cross-sectional OLS model that are robust to heteroscedasticity as well as clustering in the cross-sections. There is no time-series dimension, hence PROC MODEL will not work. PROC REG will address the heteroscedasticity but will not address clustering.  PROC SURVEYREG will address the clustering, but does address heteroscedasticity. I'm curious if demeaning the data based on the cluster variable (which is SIC or industry) will address the clustering issue. If so, then it is possible to use the HCCME options with PROC REG to address the heteroscedasticity. I grateful for any insight.  Rick

[SteveDenham]

Hi Rick,

Seems like I recommend it for everything, but PROC GLIMMIX may be an option.  Heteroscedasticity can be specified for RANDOM effects (which is what I would assume describes your clusters), and cross-sectional models can be fit using the RESIDUAL option in a second RANDOM statement.  Take a look at the documentation, and see if it might be serviceable.

SteveDenham

[rfrancis]

Thank you Steve!  I will definitely follow your advice!  Rick