I was working on the following problem:

Consider two probability density functions on $[0,1]: f_0(x) = 1$, and $f_1(x) = 2x$. Among all tests of the null hypothesis $H_0: X \sim f_0(x)$ versus the alternative $X \sim f_1(x)$, with significance level $\alpha = 0.1$, how large can the power possibly be?

I think we need to begin by looking at an arbitrary test with significance level $\alpha = 0.1$ but I am having a hard time doing this. I am not even sure this is the direction we want to head in so I was hoping to get some hints regarding this problem.


Let $\{\Gamma_0, \Gamma_1\}$ be a partition of $[0,1]$ into (measurable) sets and the decision rule be that hypothesis $H_i$ is true if the observation $X$ belongs to $\Gamma_i$. Then, the false alarm probability is $$\alpha = \int_{\Gamma_1} f_0(x)\,\mathrm dx = 0.1$$ while the power of the test is $$\beta = \int_{\Gamma_1} f_1(x)\,\mathrm dx.$$ Now,since $f_0(x)$ is the uniform density, we know that the total "length" of the set $\Gamma_1$ is $0.1$ and so the question is

What should we choose $\Gamma_1$ to be so as to maximize $\displaystyle \int_{\Gamma_1} f_1(x)\,\mathrm dx$?

Sketching the shape of the density $f_1(x)$ might reveal the answer to you in a flash, and then you can write down a more formal way of getting to it if you like.


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.