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: ∙ We tend to focus on three kinds of rejection rules for a properly chosen constant c , called a critical value : T c T c  T  c 21 ∙ But others are possible, such as an asymmetric rejection rule in place of  T  c : reject H if T c L or T c U . ∙ Asymmetric rules are sometimes preferred when testing hypotheses about a parameter that must be strictly postive: 0. 22 EXAMPLE : Suppose in a Normal ,1 population, we want to test H : ≤ 0 against H 1 : 0. We can obtain a random sample of size of size n . We have an unbiased estimator of , X ̄ . It seems reasonable to base the rejection rule on how large X ̄ is. ∙ Let x ̄ denote the average for the actual data we observe. ∙ If we get x ̄ ≤ 0, that is no evidence against H : ≤ 0. 23 ∙ The harder question is what to do if x ̄ 0. Should we reject H if, say, x ̄ .02? What about x ̄ 1.30? We need to determine a rule once we have chosen the size of the test. To do that, we need to choose a statistic whose distribution we can compute under the null hypothesis. ∙ Using sd X ̄ 1/ n , we can use the standardized sample average: T X ̄ / sd X ̄ X ̄ / 1/ n n X ̄ 24 ∙ Note how the sample size n enters T directly, along with the sample average. For a positive outcome on X ̄ , T increases with n . ∙ Because T has the same sign as X ̄ , negative values of T are no evidence against H : ≤ 0. We only will reject in favor of H 1 : 0 if T is positive and “sufficiently large.” 25 ∙ In other words, we choose a rejection rule of the form T c for a constant c to be chosen by us. ∙ To find a rejection rule that allows us to control the size of the test, note that if 0, T has a standard normal distribution: T Normal 0,1 . ∙ If 0 then E T 0. Therefore, of all the null values, 0 is the least favorable case because it is the null value we are most likely to reject when the null is true. 26 ∙ Suppose we designate .05, that is, we want to make a Type I error in only 5% of the random samples we draw. (It is very common to choose a size of 5%, but it is not written in stone.) Then we need to find c such that P Z c .05 where Z Normal 0,1 . (Remember, T has a standard normal distribution if 0.) 27 ∙ From the standard normal distribution we can find that c ≈ 1.65. ∙ So the rejection rule that gives a 5% (or .05) size is: “Reject H in favor of H 1 if T 1.65.” 28 area = .05 .1 .2 .3 .4 phi(z) c = 1.65 OneSided Rejection Rule, alpha = .05 29 ∙ A test against a onesided alternative is also called a one tailed test ....
View
Full Document
 Fall '12
 Jeff
 Normal Distribution, Null hypothesis, Statistical hypothesis testing, alternative hypotheses

Click to edit the document details