You are on page 1of 2

Stat 111 Homework 7, Spring 2015

Due Friday, April 10 at 1:11pm, handed into one of the 3rd floor HW submission box for Stat 111,
outside the 3rd floor suite of Stat offices in the Science Center (at the end of the hall near Littauer).
Please write your name on your homework. Show your work and make sure to give clear, convincing,
and succinct explanations.
Collaboration policy (for this and all future homeworks): You are encouraged to discuss the
problems with other students, but you must write up your solutions yourself and in your own words.
Copying someone elses solution, or just making trivial changes is not acceptable. And make sure that
you write-up any R-code in your own words.
Problem 1. 1. Show that the asymptotic likelihood ratio test statistic, 2LRT = 2 log(), to test
H0 : = 0 vs. HA : 6= 0 is truly 2df =1 distributed for i.i.d. Y1 , ..., Yn N (, 2 ) when 2 is
known and H0 is true. That is, the exact distribution is 2 ; its not just an asymptotic result to test
for the Normal distribution.
Problem 2. Likelihood Ratio Test for Exponential Family of Distributions: The exponential family
of distributions, both discrete and continuous, based on a parameter is defined by:
f (x|) = c(x)d() exp [a()b(x)]
*Note: is a scalar...there is a generalization for the multi-parameter case.
(a) Show that Pois() is a special case of this family. That is, determine the values for the functions
a, b, c, and d.
(b) Suppose that X1 , ..., Xn are i.i.d. according to a general distribution in the exponential family.
Consider a hypothesis test of H0 : = 0 vs. HA : = 1 , where 0 < 1 . Derive an expression for the
likelihood ratio test statistic (not the asymptotic one).
P
(c) For a general distribution in the exponential family, let T = [b(xi )]. Show that T is a sufficient
statistic for .
Problem 3. Let X1 , ..., Xn Bern(p).
(a) What is the likelihood function, L(p), for p? What is the log-likelihood function, l(p)?
(b) What is the score function, U (p), for p?
(c) Calculate the expected Fishers information matrix, In (p), for p.
Wed like to test H0 : p = p0 vs. HA : p 6= p0 at the = 0.05 level.
(d) Calculate the asymptotic likelihood ratio test statistic, 2LRT = 2 log(), for these hypotheses.
(e) Calculate the score test statistic, S, for these hypotheses.
(f) Calculate the Wald test statistic, W , for these hypotheses.
(g) In 2-3 sentences, compare the three different likelihood-based asymptotic tests for this setting.
Problem 4. A survey of Harvard students was taken and the respondents were asked Are you
1

vegetarian? Of the n = 73 responses, x = 7 said they were vegetarian. The Vegetarian Resource
Group (VRG) ran a poll that said 5% of all Americans are vegetarian: <http://www.peta.org/
living/food/2011-vegetarian-vegan-stats/>. Using the Harvard survey as a simple random
sample, wed like to determine whether or not Harvards population has a different percentage of
vegetarians than the US as a whole [assuming VRGs survey is truth for the US population].
(a) For this dataset, perform the three asymptotic tests (LR, Score, and Wald) in problem 3 above.
Be sure to state the hypotheses, calculate the test statistic, the asymptotic critical values and pvalues, and state the conclusion to determine whether Harvards population has a different percentage
of vegetarians than the US as a whole. The R commands qchisq(p,df) and pchisq(x,df) may be
very useful.
(b) Perform an exact test (based on the exact sampling distribution of or X) to determine whether
Harvards population has a different percentage of vegetarians than the US as a whole. Be sure to state
your hypotheses, the critical value, the p-value, and state the conclusion in context of the problem.
The R commands qbinom(p,size,prob) and 2*(1-pbinom(x,size,prob)) may be very useful.
(c) Calculate the 95% Confidence Intervals based on each of the 4 tests above.
(d) Write a paragraph comparing the results above. Do the tests all come to the same conclusion? Which test statistic has the smallest/greatest p-value? Why? Which confidence interval is
widest/narrowest? Why?

You might also like