The KS test uses the statistic $$ D_n=\sup_x |\hat{F}_n(x)-F_0(x)| $$ where $F_0(x)$ is the distribution to be tested and $\hat{F}_n(x)$ the empirical distribution. Under the null hypothesis $D_n$ is distributed according to the Kolmogorov distribution and the KS-test is performed testing $D_n$ against the Kolmogorov distribution.

My question concerns the bootstrapped version , i.e., $$ D^*_n=\sup_x |\hat{F}_n^*(x)-F_0(x)| $$ where $\hat{F}_n^*(x)$ is obtained through bootstrapping from $\hat{F}_n(x)$. Given the data, which is the distribution of $D^*_n$ generated by bootstrap? Which is the relationship between this distribution and the kolmogorov distribution for $n\rightarrow \infty$?

Can I derive a KS-test based on the distribution $D^*_n$ given the data? Is there some known result about that?

Thank you for any help.

  • $\begingroup$ I am curious as to why you are interested in this. Generally speaking, the bootstrap is about constructing an approximate sampling distribution with which to do inference. Here, the distribution of $D_n$ is fixed (invariant to the underlying data) and known under quite general hypotheses. It can be computed pretty readily as well. $\endgroup$
    – cardinal
    Apr 10 '13 at 11:58
  • $\begingroup$ I'm trying to implement a KS test based on the Dirichlet process and I ended with this problem. To compare the classic KS test and mine, I need to understand the relationship between the distributions of $D_n$ and $D_n^*$. $\endgroup$
    – vatna
    Apr 10 '13 at 12:57

Your Answer

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

Browse other questions tagged or ask your own question.