slides_13_inferasymptotic

# 39 we can use the same approach for two sided

This preview shows pages 39–48. Sign up to view the full content.

39

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
We can use the same approach for two-sided alternatives. Now, when we write H n 1 : n / n , can be negative as well as positive. We get the same limiting distribution for T n : T n d T Normal / ,1 where can be negative or positive. 40
Choose a 5% asymptotic size, so c 1.96. Then we need to compute P | T | 1.96 1 P 1.96 T 1.96 1 P 1.96 / T / 1.96 / 1 1.96 / 1.96 /  1.96 / 1.96 / The shape of the local power function is identical to that for the exact power function under normality. 41

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
.05 .2 .4 .6 .8 1 local power -5 0 5 delta Local Power: Asymptotic Size = .05, sigma = 1 42
Recall that the asymptotic distribution of the t statistic under the null hypothesis is the same when we replace with S n , the sample standard deviation. This continues to be true under local alternatives: T n n X ̄ n S n n X ̄ n n n n S n n X ̄ n n S n S n d Normal 0,1 Normal / ,1 43

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
The derivation uses n X ̄ n n S n n X ̄ n n o p 1 d Normal 0,1 S n o p 1 A consequence is that for local power analysis, we can act as if S n is . This is not true for exact power analysis, where we move from the normal distribution (using ) to a t distribution (using S n ). 44
When is unknown, the best we can do is estimate the local power function. For example, against H n 1 : n / n the estimator of the power function is ̂ n / S n 1.65 The same local power analysis goes through if we replace with n where n 0 and S n n o p 1 This is why we usually only model the drift for the parameters that we are testing under H 0 – the mean in this case – but not the other parameters (the variance). 45

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
None of the preceding analysis relies on the null being of the form H 0 : 0. In general, if the null is H 0 : 0 and H n 1 : n 0 / n , the same sort of local power calculations work. Suppose that, in the earlier Poisson example, the null is H 0 : 2. The three statistics are T n 1 n X ̄ n 2 2 , T n 2 n X ̄ n 2 X ̄ n , T n 3 n X ̄ n 2 S n Not only do these all have limiting Normal 0,1 distributions under H 0 , they all have limiting Normal / 2 ,1 distributions under H n 1 . 46
We can show this by using more general results. Let n be the mean of the random sample X ni : i 1,2,..., n and n 2 Var X ni . We take as given that the CLT holds, that is, n X ̄ n n n d Normal 0,1 .

This preview has intentionally blurred sections. Sign up to view the full version.

View Full Document
This is the end of the preview. Sign up to access the rest of the document.

{[ snackBarMessage ]}

### What students are saying

• As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

Kiran Temple University Fox School of Business ‘17, Course Hero Intern

• I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

Dana University of Pennsylvania ‘17, Course Hero Intern

• The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

Jill Tulane University ‘16, Course Hero Intern