Is there a name for the following process?

Say I have an absolutely continuous probability density function $f$ with compact support, and I take $k$ independent samples $x_1,\dots,x_k$ from $f$. Then, I let $x^*$ be the sample for which $f(x_i)$ is the largest, i.e. $x^* = \arg \max_i f(x_i)$. My question is, is there a name for this process, and can anything be said about the distribution of $x^*$? This basically has the effect of emphasizing the modes of $f$ more aggressively.

  • $\begingroup$ "Subsampled estimate of the mode." But in your problem is $f$ known? Otherwise, how would you compute $x^*$? $\endgroup$ – David G. Stork Apr 8 '16 at 23:39
  • $\begingroup$ Yes, $f$ is known. I'm interested in describing the distribution of $x^*$ in terms of $f$. $\endgroup$ – Tom Solberg Apr 8 '16 at 23:51
  • 1
    $\begingroup$ As $k\to\infty$, the distribution of $x^*$ will obviously be concentrated near the maximum of $f$ (assuming for simplicity that $f$ has a unique global max), as you already pointed out yourself. What else do you hope to be able to say in this generality? $\endgroup$ – Christian Remling Apr 9 '16 at 0:22

If your $k$ samples are taken with replacement, then your method is a Bootstrap estimate of the mode. In the limit of a large number of such samples, the Bootstrap estimate coincides with the true mode.

Bootstrapping is useful when seeking the variance of a statistic such as the mode, as in your case.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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