I am using a large health dataset as a part of a research project (N = ~18 000). My colleagues and I are investigating whether smoking predicts the presence or absence of a mental illness.
We are using logistic regression in Stata 13 for our analyses. We are weighting our regression model per the instructions of the user documentation for the dataset. However, with the inclusion of these weighting variables, Stata is providing robust standard errors (HC1) in lieu of standard errors. To the best of my knowledge, there is no way to disable this feature.
Initially, I found the inclusion of robust standard errors surprising as homoscedasticity is NOT an underlying assumption for logistic regression. However, when I investigated this in the Stata documentation, the rationale for the provision of robust standard errors was not exclusively for the correction of heteroscedasticity. From what I was reading, it was noted that in non-simple, non-random sampling, estimates of covariance were artificially inflated because persons from the same primary sampling unit were being selected. In other words, systematic error was being introduced into the model. This systematic error variance was being mitigated through the usage of robust standard errors.
My question is this:
Will using robust standard errors in logistic regression, produce erroneous results?
