Assume that $\pi$ is the coupling of probability measures $\mu$ and $\nu$ on $[0,1]$. The hypothesis test for independence is that $$ H_0: \pi=\mu\times \nu \, , H_a: \pi\neq \mu\times \nu $$ The test statistic is $W(\hat{\pi}^N)$.

My question is how to get the power: $$\text{power}=1−P(\text{type II error})=1−P(H_0 \text{ accept}|H_0 \text{ false})$$ Can I take two dependent random data sets and for this one means conditional probability on $H_0$ false?

Also, I am confused about the meaning of this Corollary. Does it mean the $$P(\text{reject } H_0|H_0 \text{ is true})$$, which is the Type-1 error? But in the simulation result, the author calculates the power and assume the significance level $\alpha=0.1$.


1 Answer 1


In the simulations, the author computes the power function of the test, which, in general, is the probability of a test $T$ rejecting the null as a function of the distribution. Thus, it is a little different from the power. I.e., if the parameter for the distribution is $\theta$, then the power function $\beta_T(\theta)$ is given as $\beta_T(\theta) = p(T=1|\theta)$. In particular, it is defined for all distributions, those in $H_0$ and those in $H_1$.

The corollary just says that there is a constant $C(\nu)$ such that $W$ would asymptotically make sense as a test statistic for independence.


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.