I'm working with the Nadaraya Watson estimator and calculate the optimal bandwidth h with leave one out cross validation.

Now I'd like to get the variance of h (not the variance of the NW estimator). Is this possible? Can anyone recommend literature?


1 Answer 1


h is bandwidth and a single value(for uni-variate case). How it is possible to calculate a variance of a single value? Even for multivariate case, bandwidth is single for every input variable. Suppose you have 5 variable so you got 5 bandwidths but every bandwidth is for each variable.

  • $\begingroup$ Meanwhile i found a Bayesian approach where you use the Metropolis Hastings algorithm. With this approach you can get a variance (and intervals). $\endgroup$
    – To Mate
    Sep 23, 2018 at 6:22
  • $\begingroup$ Bayesian approach for what? To calculate bandwidth or in NW..........??? $\endgroup$
    – Angel
    Sep 23, 2018 at 6:25
  • 1
    $\begingroup$ Food for thought: The sample mean is also only a single value yet people find a variance for it all the time $\endgroup$
    – einar
    Dec 8, 2020 at 10:12
  • $\begingroup$ @einar that is mean of different samples. $\endgroup$
    – Angel
    Dec 14, 2020 at 5:05

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.