This preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full DocumentThis preview has intentionally blurred sections. Sign up to view the full version.
View Full Document
Unformatted text preview: n X ̄ n . If 0 then, by the CLT, T n d → Normal 0,1 . ∙ If 0 then T n n X ̄ n − n O p 1 n p → − because n X ̄ n − d → Normal 0,1 and 0 so n → − . It follows that the asymptotic size of the test for 0 is zero. 8 ∙ Therefore, just as in the finite sample case, we can focus on the size of the test under H : 0. ∙ Because of the CLT, we can use the same approach to choosing critical values that we did in the case of a normal population: given asymptotic size , we choose critical value c such that P Z c where Z Normal 0,1 . 9 ∙ Theoretically, the important difference is that (when 0) we can only conclude P T n c → as n → and not P T n c for all n . This is why we now call c an asymptotic critical value , and is the asymptotic size. 10 ∙ In the same way, we compute pvalues using the standard normal distribution. So, if (say) r n is the outcome of the test statistic (and r n 0) and Z denotes a standard normal random variable, we compute P Z r n and report this as the asymptotic p value . ∙ The language of hypothesis testing is the same as before, except we often use the qualifier “asymptotic” or “approximate.” ∙ The approach to twosided tests is exactly the same as with finitesample analysis. 11 ∙ What happens if we consider the general population assumption X , 2 and use the usual t statistic, T n n X ̄ n S n , where S n is the sample standard deviation? Generally, we cannot find the distribution of T n for any n because we cannot even determine the distribution of X ̄ n , S n in general. 12 ∙ We do know that, under H : 0, T n n X ̄ n 1 S n − 1 n X ̄ n n X ̄ n o p 1 O p 1 n X ̄ n o p 1 13 ∙ Therefore, by the asymptotic equivalence lemma, T n n X ̄ n S n d → Normal 0,1 when 0. ∙ As an exercise, you can show that it still holds that T n p → − if 0. 14 ∙ This means that we can use the standard normal critical values for testing hypotheses when the variance is generally unknown; and we can use it to compute asymptotic pvalues. ∙ Asymptotically, replacing with S n has no effect on the limiting distribution, and so, when conducting asymptotic tests, it is as if we know . 15 ∙ What about using the t n − 1 critical values rather than the standard normal critical values? Because the t n − 1 crtical values converge to those of the standard normal as n → , there is nothing wrong with it. In fact, it is usually preferred (for n 120 or so) because it includes exact inference for the normal population as a special case....
View
Full Document
 Fall '12
 Jeff
 Normal Distribution, Statistical hypothesis testing, Tn, local power

Click to edit the document details