STAT 632-600 Final Exam: Statistical Inference - Prof. Jeffrey Hart, Exams of Statistics

The instructions and questions for a final exam in a statistical inference course. The exam covers topics such as decision rules, james-stein theory, bayesian inference, and markov chain monte carlo (mcmc). Students are required to answer each question specifically and circle the correct answer for multiple choice questions.

Typology: Exams

Pre 2010

Uploaded on 02/10/2009

koofers-user-iz0
koofers-user-iz0 🇺🇸

5

(1)

10 documents

1 / 5

Toggle sidebar

This page cannot be seen from the preview

Don't miss anything!

bg1
STAT 632-600
December 13, 2005
Final Exam
Instructions: Answer each question on the test paper provided. Circle the correct answer
on multiple choice questions. Be specific on short-answer questions. Each multiple choice
question is worth four points, and each of the other questions is worth 12 points. Have a
happy holiday!
1. A decision rule δis inadmissible if
(a) its Bayes risk is larger than that of the Bayes rule.
(b) there exist at least two decision rules having uniformly smaller risk than that of δ.
(c) there exists a rule δ1for which R(θ, δ1)R(θ, δ) for all θwith strict inequality for at
least one θ.
(d) there exists a rule δ1such that supθΘR(θ, δ1)<supθΘR(θ, δ).
(e) it is unable to stand up throughout an entire Aggie home game.
2. James-Stein theory is concerned mainly with
(a) admissibility of the sample mean as an estimator of a multivariate normal population
mean.
(b) admissibility of the sample variance as an estimator of population variance.
(c) large sample properties of maximum likelihood estimators.
(d) large sample properties of the mean of a posterior distribution.
(e) how to properly pour Pils vom Faβinto a German beer stein.
3. The observations Y1, . . . , Ynare a random sample from a normal distribution with mean
µand variance σ2, both of which are unknown. The prior distribution is
π(µ, σ) = 1
σφµµ
σ·2φ(σ)I(0,)(σ),
where φis the standard normal density and I(0,)is an indicator function. In this particular
problem, the prior πis
(a) a noninformative prior.
(b) a conjugate prior.
(c) an improper prior.
(d) none of the above.
(e) saddened at the passing of his buddy Richard Pryor.
1
pf3
pf4
pf5

Partial preview of the text

Download STAT 632-600 Final Exam: Statistical Inference - Prof. Jeffrey Hart and more Exams Statistics in PDF only on Docsity!

STAT 632-

December 13, 2005 Final Exam

Instructions: Answer each question on the test paper provided. Circle the correct answer on multiple choice questions. Be specific on short-answer questions. Each multiple choice question is worth four points, and each of the other questions is worth 12 points. Have a happy holiday!

  1. A decision rule δ is inadmissible if (a) its Bayes risk is larger than that of the Bayes rule. (b) there exist at least two decision rules having uniformly smaller risk than that of δ. (c) there exists a rule δ 1 for which R(θ, δ 1 ) ≤ R(θ, δ) for all θ with strict inequality for at least one θ. (d) there exists a rule δ 1 such that supθ∈Θ R(θ, δ 1 ) < supθ∈Θ R(θ, δ). (e) it is unable to stand up throughout an entire Aggie home game.
  2. James-Stein theory is concerned mainly with (a) admissibility of the sample mean as an estimator of a multivariate normal population mean. (b) admissibility of the sample variance as an estimator of population variance. (c) large sample properties of maximum likelihood estimators. (d) large sample properties of the mean of a posterior distribution. (e) how to properly pour Pils vom Faβ into a German beer stein.
  3. The observations Y 1 ,... , Yn are a random sample from a normal distribution with mean μ and variance σ^2 , both of which are unknown. The prior distribution is

π(μ, σ) = σ^1 φ

( (^) μ σ

) · 2 φ(σ)I(0,∞)(σ),

where φ is the standard normal density and I(0,∞) is an indicator function. In this particular problem, the prior π is (a) a noninformative prior. (b) a conjugate prior. (c) an improper prior. (d) none of the above. (e) saddened at the passing of his buddy Richard Pryor.

  1. In a binomial experiment with unknown proportion θ and prior

π(θ) ∝ θa−^1 (1 − θ)b−^1 I(0,1)(θ),

if a > 1 and b > 1 then the posterior mean is (a) equal to the posterior mode. (b) equal to the usual frequentist estimate of θ. (c) a weighted average of the means of two beta distributions. (d) all of the above. (e) often confused with the more popular Bayes point estimate known as the “rump me- dian.”

  1. Gibbs sampling is (a) a special case of the single component Metropolis-Hastings algorithm. (b) equivalent to an independence sampler. (c) always based on a symmetric proposal distribution. (d) a special case of the random walk Metropolis algorithm. (e) a little known form of wine tasting perfected by the brothers Gibb (better known as the Bee Gees).
  2. A certain statistical decision problem has parameter space Θ = [0, 1]. The decision rules δ 1 , δ 2 , δ 3 have the following risk functions: R(θ, δ 1 ) = 2θ, 0 ≤ θ ≤ 1 , R(θ, δ 2 ) = 3

( θ − (^12)

) 2 , 0 ≤ θ ≤ 1 R(θ, δ 3 ) = 7

∣∣ ∣∣θ − 1 2

∣∣ ∣∣^3 , 0 ≤ θ ≤ 1.

If a uniform prior is used to compute the Bayes risk for each rule, then the respective minimax and Bayes rules are (a) δ 3 and δ 2. (b) δ 1 and δ 3. (c) δ 2 and δ 3. (d) δ 2 and δ 1. (e) impossible to determine given the amount of sleep I had last night.

  1. Lindley’s paradox (a) is concerned with paradoxical differences between Bayes and frequentist point estimates. (b) is a problem due to using noninformative priors. (c) occurs when the posterior probability of the null hypothesis is very small but a frequen- tist P -value is not. (d) occurs when a frequentist P -value is very small but the posterior probability of the null hypothesis is not. (e) is a band that cites the Velvet Underground as one of its major influences.
  2. Define the term highest posterior density region and describe how you would determine such a region given a posterior density π( · |data).
  3. A set of data y is assumed to have come from one of the three models M 1 , M 2 and M 3. The respective likelihoods for the three models are f 1 (y|θ), f 2 (y|θ) and f 3 (y|θ), where the parameter space is Θ in each case. The respective conditional priors for θ are πi(θ), i = 1, 2 , 3, and the prior probabilities for the three models are

P (M 1 ) =^12 , P (M 2 ) =^14 , and P (M 3 ) =^14.

Write down an expression for P (M 1 |y) in terms of the quantities defined above.

  1. Define the terms Bayes factor and posterior odds ratio.
  2. Discuss the essential difference between frequentist and Bayesian inference. In doing so, define the terms initial precision and final precision.
  3. Describe a physical setting in which it would be natural to use a hierarchical model. In so doing, be sure to convince me that you know what a hierarchical model is.