@larameix wrote:
Hey everybody!
I´m using procedures for complex survey data in SAS (proc surveyreg). I´ve detected heteroscedasticity in my regression models.
I found this link, which claims that the procedure for complex samples controls for heteroscedasticity by default (by using an asymptotically-consistent covariance matrix):
https://wrds-www.wharton.upenn.edu/pages/support/applications/programming-examples-and-other-topics/...
I found that the ACOV Option can be used in the proc reg statement, in order to display the estimated asymptotic covariance matrix of the estimates under the hypothesis of heteroscedasticity. But I didn´t find detailed information on how this works using the procedure for complex samples. Does the Taylor expansion theory have anything to do with this?
I am thankful for any little hint! Thank you in advance!
@larameix wrote:
Hey everybody!
I´m using procedures for complex survey data in SAS (proc surveyreg). I´ve detected heteroscedasticity in my regression models.
I found this link, which claims that the procedure for complex samples controls for heteroscedasticity by default (by using an asymptotically-consistent covariance matrix):
https://wrds-www.wharton.upenn.edu/pages/support/applications/programming-examples-and-other-topics/...
I found that the ACOV Option can be used in the proc reg statement, in order to display the estimated asymptotic covariance matrix of the estimates under the hypothesis of heteroscedasticity. But I didn´t find detailed information on how this works using the procedure for complex samples. Does the Taylor expansion theory have anything to do with this?
I am thankful for any little hint! Thank you in advance!
I'm assuming that SAS uses a robust* sandwich estimator for producing the variance-covariance matrix of regression coefficients, such as done by SUDAAN. Taylor series may be used to estimate the variance-covariance matrix terms but Taylor series doesn't have anything to do with the notion of robustness.
*Robust means the model is not that sensitive to departures from the assumed variance-covariance structure.
SAS is using a sandwich estimator of the variance-covariance matrix by default but I can't tell if it's a robust estimator. I'd contact SAS tech support.
