linear regression: vce robust standard errors with svy prefix

[Sara Somaruga]

Hello there,
I am estimating linear regression models with weighted data using the 'svy' prefix. Since the dependent variable is dichotomous (I am deliberately estimating linear models anyway), I wanted to estimate with robust standard errors, which is often recommended. However, I've realized that it's not possible to estimate robust standard errors with the 'svy' prefix.

In my research, I came across the following explanation:
'The option vce(robust) is not compatible with the svy prefix because svy already performs a form of robust variance estimation. The svy prefix uses Taylor linearization to calculate the variance of the estimators, taking into account the sample design. This is a specialized form of robust variance estimation suitable for survey data. When you use the svy prefix, a robust variance estimation that takes into account the complex sample design is already being performed. The additional use of vce(robust) would be redundant and potentially misleading in this context, as it does not consider the additional aspects of the sample design covered by svy.'

Can someone with deeper insights into the underlying estimation methods tell me if this information is correct?

Thanks in advance, Sara

[Carlo Lazzaro]

Sara:
the -svy estimation- entry, Stata .pdf manual confirms what you report (without full reference, though. See the FAQ. Thanks).