2

I asked before what is the intuition behind sandwich estimators. I must still missing something because I don't understand why sandwich estimators are not always applied to OLS residuals.

Can you explain what sandwich estimators do in English and why we don't see them always applied to OLS models?

Community
  • 1
Robert Kubrick
  • 4,528
  • 2
    In short, they're less efficient than the classic SEs, given that the assumptions behind these hold. – abaumann Oct 24 '14 at 15:09
  • @abaumann Less efficient in what? – Robert Kubrick Oct 24 '14 at 15:21
  • If you know for sure that the underlying model is linear with Gaussian residuals then what the estimator may see as outliers is real data that shouldn't be discarded. – Arthur B. Oct 24 '14 at 15:52
  • 2
    @RobertKubrick http://en.wikipedia.org/wiki/Efficiency_%28statistics%29 – abaumann Oct 24 '14 at 15:59
  • @abaumann So can we say that for large sets (say, more than 5,000 observations) sandwich estimators are always preferable? Why to use the standard diagonal std errors? Then again, unless the residuals are extremely heavy-tailed, the larger the sample the less likely the standard errors bias... – Robert Kubrick Oct 24 '14 at 17:52

0 Answers0