Scribd
Upload
36 views2 pages

18.443 Statistics For Applications: Mit Opencourseware

This document is a problem set for the course 18.443 Statistics for Applications from MIT OpenCourseWare. It contains 5 problems related to statistical hypothesis testing, decision theory, Bayesian estimation, and properties of common probability distributions.

Uploaded by

Lionel Carlos
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
36 views2 pages

18.443 Statistics For Applications: Mit Opencourseware

This document is a problem set for the course 18.443 Statistics for Applications from MIT OpenCourseWare. It contains 5 problems related to statistical hypothesis testing, decision theory, Bayesian estimation, and properties of common probability distributions.

Uploaded by

Lionel Carlos
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 2

MIT OpenCourseWare

http://ocw.mit.edu

18.443 Statistics for Applications


Spring 2009

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.

18.443 Problem Set 10


1. A statistic X is measured in a medical test. Suppose that for a certain disease, D,
the hypothesis H0 is that the person being tested does not have D and in that case the
distribution of X is N (3, 1). The alternative hypothesis H1 is that the person does have
D, although they may not have any symptoms yet. Under H1 , X has distribution N (5, 1).
(a) Show that for any c, X c is a best critical region for testing H0 vs. H1 , in the
sense that it has largest power for given size (recall the Neyman-Pearson lemma).
(b) Suppose it costs $40 each time X is measured for one patient. Suppose this will
be done for people in a risk group in which a priori, the probability of having D is 0.08.
If the test is positive (H0 is rejected), a test based on a dierent statistic costing $1000
will be done which will give a correct result in essentially all cases. If the original $40 test
is negative (H0 is not rejected) then no further test or treatment will be done. In that
case, if a patient has D, assume a loss of $1,000,000. How should c be chosen to minimize
the expected loss?
(c) In a general population where the a priori probability of having D is 105 , is it
cost-eective to do any such test procedure (for any c)? Hint: would it be, even if an
initial $40 test always gave the correct result?
2. Suppose Xj are i.i.d. with distribution U [0, 1] under H0 and U [0.2, 1.2] under H1 . Each
observation costs $4 to make. The prior probabilities are p0 = p1 = 1/2. Choosing the
wrong Hj will cost $50.
(a) Is it better to ip a coin to choose an Hj without observations or to take observa
tions until H0 or H1 can be conrmed as correct?
(b) What is the probability, under H0 or H1 , that after 20 observations, we still cannot
conrm which hypothesis is true?
3. If a binomial probability p has a beta(a, b) prior and X successes are observed in n

trials, what is the Bayes estimate of the function g(p) = p2 ? Recall that for n = 2, we

found early on that unbiased estimation gave a strange answer when X = 1. For the U [0, 1]

prior and n = 2, what are the Bayes estimates of p2 for X = 0, 1, and 2?

4. It was stated without proof in PS9 that a standard Cauchy distribution is the same as

a t1 distribution (t distribution with one degree of freedom). Show this in two stages:

(a) If Y and Z are i.i.d. N (0, 1), show that Y /Z has a standard Cauchy distribution. Hint:

Y /Z x is equivalent to Y xZ if Z > 0 but not if Z < 0. Consider two cases.

(b) Do the same for Y /|Z| (as a corollary of part (a)) and relate this to the t1 distribution.

5. Show that each of the following parametric families is exponential, and give explicit

representations for each.

(a) Gamma family (a, ), a > 0, > 0;

(b) Beta family beta(a, b), a > 0, b > 0;

(c) Geometric family, geometric (p), 0 < p < 1.

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy