I'm a bit confused about power analysis and the non-central t-distribution.

If I am in a situation where $H_0$ is $\mu = 0$, and $H_1$ is $\mu > 0$, I calculate for what value $\mu$ would have a probability of 95% or less of occurring. I do this by saying that, if $\mu$ really were 0, then $\frac{\bar{X}}{s(X)}$ has a t-distribution.

Now for a power analysis, I suppose that $\mu$ is something like 1.5 and then I calculate the probability of correctly rejecting $H_0$. It seems then that $\frac{\bar{X} - 1.5}{s(X)}$ should be a t-distribution also. I can then calculate the probability of exceeding the critical value (that would be obtained by assuming $H_0$) I have using this distribution.

Why would the distribution of the test-statistic change based on what I consider $H_0$ to be? I must be missing something.

Update: It looks like the answer may be that (the way I'm being taught) is that you use the same test statistic as you would under ${H_0}$, but it now has a non-central t-distribution because of the new assumed value of $\mu$. However, instead of using the non-central t distribution -- you could do what I described in my third paragraph. Is that correct?


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.