Figure 8.11 plots these power functions for various degrees of freedom and noncentrality parameters. We could use Fig.8.11 to find the power of the test in Example 8.5.4 when ,that is when . It appears to be about 0.45. (The actual power is 0.438.)
P – Values for two – sided tests. The p – values for hypotheses (8.5.6) when U= u can be computed as follows. The test rejects when . The d.f. of when is . So the p – values is . For instance, in Example 8.5.4, the p – value is . Note that this is twice the p – value when the hypotheses were (8.5.1). For t tests, if the p – value for testing hypotheses (8.5.1) or (8.5.5) is p, then the p – value for hypotheses (8.5.6) is either 2p or .
Figure 8.11 The power functions of two – sided level 0.05 and level 0.05 and level 0.01 t tests with various degrees of freedom for various values of the noncentrality parameter
Likelihood Ratio Tests
A very popular form of hypothesis test is the likelihood ratio test. This is a generalization of the optimal test for simple null and alternative hypotheses that was developed in Section 8.2. It is based on the likelihood function .The likelihood function tends to be highest near the true values of . Indeed, this is why maximum likelihood estimation works well in so many cases. Now, suppose that we wish to test the hypotheses
,
. (8.5.8)
In order to compare these two hypotheses, we might wish to see whether the likelihood function is higher on or on . If we restrict attention to , for example, then we can compute the largest value of the likelihood . Similarly, for , we can compute . The ratio of these two values can then be used for testing the hypotheses (8.5.8). Define the likelihood ratio statistic
. (8.5.9)
Then, a likelihood ratio test of hypotheses (8.5.8) is to reject if for some constant k. That is, we shall reject if the likelihood function on is sufficiently large compared to the likelihood function on . Generally, k is chosen so that the test has a desired level , if that is possible.
Figure 8.11 plots these power functions for various degrees of freedom and noncentrality parameters. We could use Fig.8.11 to find the power of the test in Example 8.5.4 when ,that is when . It appears to be about 0.45. (The actual power is 0.438.)
P – Values for two – sided tests. The p – values for hypotheses (8.5.6) when U= u can be computed as follows. The test rejects when . The d.f. of when is . So the p – values is . For instance, in Example 8.5.4, the p – value is . Note that this is twice the p – value when the hypotheses were (8.5.1). For t tests, if the p – value for testing hypotheses (8.5.1) or (8.5.5) is p, then the p – value for hypotheses (8.5.6) is either 2p or .
Figure 8.11 The power functions of two – sided level 0.05 and level 0.05 and level 0.01 t tests with various degrees of freedom for various values of the noncentrality parameter
Likelihood Ratio Tests
A very popular form of hypothesis test is the likelihood ratio test. This is a generalization of the optimal test for simple null and alternative hypotheses that was developed in Section 8.2. It is based on the likelihood function .The likelihood function tends to be highest near the true values of . Indeed, this is why maximum likelihood estimation works well in so many cases. Now, suppose that we wish to test the hypotheses
,
. (8.5.8)
In order to compare these two hypotheses, we might wish to see whether the likelihood function is higher on or on . If we restrict attention to , for example, then we can compute the largest value of the likelihood . Similarly, for , we can compute . The ratio of these two values can then be used for testing the hypotheses (8.5.8). Define the likelihood ratio statistic
. (8.5.9)
Then, a likelihood ratio test of hypotheses (8.5.8) is to reject if for some constant k. That is, we shall reject if the likelihood function on is sufficiently large compared to the likelihood function on . Generally, k is chosen so that the test has a desired level , if that is possible.
การแปล กรุณารอสักครู่..
