Before an election, combining the results of 12,625 polls with 14,491,635 samples in total, it shows that 6,413,959 responders (44.3%) say they will vote for the first candidate and 6,134,272 responders (42.3%) say they will vote for the other candidate. Assume a binomial model Binomial(n,p) of the polls for the first and second candidates, where p is the percentage of the votes to the first candidate and n is the total number of votes to the first candidate or the second candidate. Suppose we are interested in whether the first candidate wins more than half of the votes to the first and second candidates:
H0: p = 0.5 v.s. H1: p > 0.5
(a) Compute the test statistics of the generalized likelihood ratio test. Is this test a uniformly most powerful test?
(b) Use Wilks' theorem to compute the critical value of the generalized likelihood ratio test under α = 0.05 level. Make a decision.
(c) Another test has test statistics p - po/√po(1 - po)/n, where po = 0.5. Compute the p-value of this test using the central limit theorem and make a decision. Assume the significance level α = 0.05.
(d) If the second candidate wins the election, comment on possible problems in this statistical analysis.