Skip to document
This is a Premium Document. Some documents on Studocu are Premium. Upgrade to Premium to unlock it.

Principles of Statistics 2016-2017 Example Sheet 1

Module

Principles of Statistics

5 Documents
Students shared 5 documents in this course
Academic year: 2016/2017
Uploaded by:
Anonymous Student
This document has been uploaded by a student, just like you, who decided to remain anonymous.
Newcastle University

Comments

Please sign in or register to post comments.

Preview text

PRINCIPLES OF STATISTICS – EXAMPLES 1/

Part II, Michaelmas 2016, Quentin Berthet (email: q@statslab.cam.ac)

Throughout, the abbreviations ‘i.i.’, ‘pdf/pmf’ and ‘MLE’ stand for ‘independent and iden- tically distributed’, ‘probability density/mass function’ and ‘maximum likelihood estimator’, re- spectively. A normal distribution inRdwith mean vectorμand covariance matrix Σ is denoted byNd(μ,Σ), andN(μ, σ 2 ) corresponds to the univariate cased= 1.

  1. Consider an i.i. sampleX 1 ,... , Xnof random variables. For each of the following parametric models of pmf/pdf’s, find the MLE of the unknown parameter,the score equation and the Fisher information.

a)Xi∼i.i.d(θ), θ∈[0,1], b)Xi∼i.i.d(θ,1), θ∈R, c)Xi∼i.i.d(0, θ), θ∈(0,∞), d)Xi∼i.i.d(μ, σ 2 ), θ= (μ, σ 2 )T∈R×(0,∞), e)Xi∼i.i.d oisson(θ),θ∈(0,∞), f)Xi∼i.i.d model{f(·, θ) :θ∈(0,∞)}with pdff(x, θ) = (1/θ)e−x/θ, x≥0. g)Xi∼i.i.d model{f(·, θ) :θ∈(0,∞)}with pdff(x, θ) =θe−θx, x≥0. 2 which of the examples of the previous exercise is the MLE unbiased(i., does one have Eθθˆ=θfor allθ∈Θ)? When unbiased, deduce whether the variance of the MLE attains the Cram`er-Rao lower bound or not.

  1. LetX 1 ,... , Xnbe i.i. Poisson random variables with parameterθ >0, and letX ̄n= (1/n)

∑n i=1Xi, S 2 n= (n−1) − 1 ∑n i=1(Xi−X ̄n) 2. Show thatV ar(X ̄n)≤V ar(S 2 n). 4. Find the MLE for an i.i. sampleX 1 ,... , Xnarising from the models a)N(θ,1) where θ∈Θ = [0,∞) and b)N(θ, θ) whereθ∈Θ = (0,∞).

5 an i.i. sampleX 1 ,... , Xnarising from the model

{f(·, θ) :θ∈R}, f(x, θ) =

1

2

e−|x−θ|, x∈R,

ofLaplace distributions. Assumingnto be odd for simplicity, show that the MLE is equal to the sample median. Discuss what happens whennis even. Can you calculate the Fisher information?

  1. Consider observing ann×1 random vectorY ∼N(Xθ, I) whereXis a non-stochastic n×pmatrix of full column rank, whereθ∈Θ =Rpforp≤n, and whereIis then×nidentity matrix. Compute the MLE and find its distribution. Calculate the Fisher information for this model and compare it to the variance of the MLE. Deduce, as a special case,the form of the MLE and Fisher information in the case whenp=nandX=I.

7 (X, Xn:n∈N) be random vectors inRk. a) Prove thatXn→PXasn→ ∞if and only if each vector componentXn,j, j= 1,... , k, ofXnconverges in probability to the corresponding vector componentXj ofXasn→ ∞. Formulate and prove an analogous result for random symmetrick×k-matrices.

b) SupposeE‖Xn−X‖ →0 asn→ ∞where‖·‖is the Euclidean norm onRk. Deduce thatXn→PXasn→∞.

c) Show that the converse in b) is false, that is, give an example of real random variables Xn→PXasn→∞butE|Xn−X|6→0.

1

8 1 ,... , Xni.i. random variables such thatEX 1 = 0, EX 12 ∈(0,∞), theStudent t-statisticis given by

tn=

nX ̄n Sn

, X ̄n=

1

n

∑n

i=

Xi, Sn 2 =

1

n− 1

∑n

i=

(Xi−X ̄n) 2.

Show thattn→dN(0,1) asn→ ∞. Assuming nowEX 1 =μ∈R, deduce an asymptotic level 1 −αconfidence interval forEX 1.

9 the examples from Exercise 1, derive directly (without using the general asymptotic theory for MLEs) the asymptotic distribution of

n(θˆMLE−θ) asn→∞. 10. Suppose one observesone random vectorX = (X 1 , X 2 )T from a bivariate normal distributionN 2 (θ,Σ) whereθ= (θ 1 , θ 2 )Tand where Σ is an arbitrary butknown 2 ×2 positive definite covariance matrix.

i) Compute the Cram`er-Rao lower bound for estimating the first coefficientθ 1 if a)θ 2 is known and b) ifθ 2 is unknown.

ii) Show that the two bounds in i) coincide when Σ is a diagonal matrix. iii) Show that the bound in i)a) is always less than or equal to the boundin i)b), and give an information-theoretic interpretation of this result.

2

Was this document helpful?
This is a Premium Document. Some documents on Studocu are Premium. Upgrade to Premium to unlock it.

Principles of Statistics 2016-2017 Example Sheet 1

Module: Principles of Statistics

5 Documents
Students shared 5 documents in this course
Was this document helpful?

This is a preview

Do you want full access? Go Premium and unlock all 2 pages
  • Access to all documents

  • Get Unlimited Downloads

  • Improve your grades

Upload

Share your documents to unlock

Already Premium?
PRINCIPLES OF STATISTICS EXAMPLES 1/4
Part II, Michaelmas 2016, Quentin Berthet (email: q.berthet@statslab.cam.ac.uk)
Throughout, the abbreviations ‘i.i.d.’, ‘pdf/pmf and ‘MLE’ stand for ‘independent and iden-
tically distributed’, ‘probability density/mass function’ and ‘maximum likelihood estimator’, re-
spectively. A normal distribution in Rdwith mean vector µand covariance matrix Σ is denoted
by Nd(µ, Σ), and N(µ, σ2) corresponds to the univariate case d= 1.
1. Consider an i.i.d. sample X1, . . . , Xnof random variables. For each of the following
parametric models of pmf/pdf’s, find the MLE of the unknown parameter, the score equation
and the Fisher information.
a) Xii.i.d. Bernoulli(θ), θ [0,1],
b) Xii.i.d. N(θ, 1), θ R,
c) Xii.i.d. N(0, θ), θ (0,),
d) Xii.i.d. N(µ, σ2), θ = (µ, σ2)TR×(0,),
e) Xii.i.d. P oisson(θ), θ(0,),
f) Xii.i.d. from model {f(·, θ) : θ(0,)}with pdf f(x, θ) = (1)ex/θ , x 0.
g) Xii.i.d. from model {f(·, θ) : θ(0,)}with pdf f(x, θ) = θeθx, x 0.
2. In which of the examples of the previous exercise is the MLE unbiased (i.e., does one have
Eθˆ
θ=θfor all θΘ)? When unbiased, deduce whether the variance of the MLE attains the
Cram`er-Rao lower bound or not.
3. Let X1, . . . , Xnbe i.i.d. Poisson random variables with parameter θ > 0, and let ¯
Xn=
(1/n)Pn
i=1 Xi, S2
n= (n1)1Pn
i=1(Xi¯
Xn)2. Show that V ar(¯
Xn)V ar(S2
n).
4. Find the MLE for an i.i.d. sample X1, . . . , Xnarising from the models a) N(θ, 1) where
θΘ = [0,) and b) N(θ, θ) where θΘ = (0,).
5. Consider an i.i.d. sample X1, . . . , Xnarising from the model
{f(·, θ) : θR}, f(x, θ) = 1
2e−|xθ|, x R,
of Laplace distributions. Assuming nto be odd for simplicity, show that the MLE is equal to the
sample median. Discuss what happens when nis even. Can you calculate the Fisher information?
6. Consider observing an n×1 random vector YN(Xθ, I) where Xis a non-stochastic
n×pmatrix of full column rank, where θΘ = Rpfor pn, and where Iis the n×nidentity
matrix. Compute the MLE and find its distribution. Calculate the Fisher information for this
model and compare it to the variance of the MLE. Deduce, as a special case, the form of the
MLE and Fisher information in the case when p=nand X=I.
7. Let (X, Xn:nN) be random vectors in Rk.
a) Prove that XnPXas n if and only if each vector component Xn,j , j = 1, . . . , k,
of Xnconverges in probability to the corresponding vector component Xjof Xas n .
Formulate and prove an analogous result for random symmetric k×k-matrices.
b) Suppose EkXnXk 0 as n where k · k is the Euclidean norm on Rk. Deduce
that XnPXas n .
c) Show that the converse in b) is false, that is, give an example of real random variables
XnPXas n but E|XnX| 6→ 0.
1

Why is this page out of focus?

This is a Premium document. Become Premium to read the whole document.