G14FOS-E1

The University of Nottingham

SCHOOL OF MATHEMATICAL SCIENCES

A LEVEL 4 MODULE, AUTUMN 2013–2014

FUNDAMENTALS OF STATISTICS

Time allowed THREE Hours

Candidates may complete the front cover of their answer book and sign their desk card but must
NOT write anything else until the start of the examination period is announced.

This paper contains FIVE questions which carry equal marks.

Full marks may be obtained for FOUR complete answers.
Credit will be given for the best FOUR answers.

An indication is given of the approximate weighting of each section of a question by means of a figure
enclosed by square brackets, eg [12], immediately following that section.

Only silent, self-contained calculators with a Single-line Display

or Dual-line Display are permitted in this examination.

Dictionaries are not allowed with one exception. Those whose first language is not
English may use a standard translation dictionary to translate between that language and
English provided that neither language is the subject of this examination. Subject specific
translation dictionaries are not permitted.
No electronic devices capable of storing and retrieving text, including electronic
dictionaries, may be used.

Do NOT turn examination paper over until instructed to do so.

ADDITIONAL MATERIAL: Neave Tables

G14FOS-E1 Turn over

 ∼ 2 ∼ G14FOS-E1

1 (a) Consider a probability density function of the form

pX (x|θ) = exp{A(θ)B(x) + C(θ) + D(x)}. (1.1)

(i) Write down the name of the class of distribution of type (1.1).
(ii) State the Neyman factorisation theorem.
(iii) Suppose that X = (X1 ,..,Xn ) is an independent and identically distributed (IID)
∑
n
sample from (1.1). Show that T (X) = n−1 B(Xi ) is sufficient for θ.
i=1
(iv) By considering the ratio

pX (x|θ) ∏ n ∏ n

′ , where p (x|θ) = p (xi |θ) and p (x ′

|θ) = pX (x′i |θ), (1.2)
pX (x |θ) X
i=1
X X
i=1

show that T (X) is minimal sufficient for θ.

[Note: No marks will be given for quoting the result that a complete sufficient statistic
is minimal sufficient; you should prove minimal sufficiency directly using the ratio
(1.2).]
(v) Find an expression for g(θ) = E[T (X)], the expectation of T (X), in terms of A(θ)
and C(θ). State but do not prove any results that you use.
(vi) Show that T (X), viewed as an estimator of g(θ), attains the Cramér-Rao Lower
Bound (CRLB). Again, state but do not prove any results that you use. [19]

(b) Suppose X has distribution N (θ, c2 θ) where c > 0 is a known constant, θ > 0 is an un-
known parameter and N (µ, σ 2 ) is the normal distribution with mean µ and variance σ 2 .

(i) Write the probability density function of X in the form (1.1).

(ii) Suppose now that X = (X1 , .., Xn ) is an IID sample from N (θ, c2 θ). Find a statistic
T (X) that is minimal sufficient for θ.
(iii) Determine g(θ) = E[T (X)], the expectation of T (X), in terms of θ and c.
(iv) Determine the CRLB for unbiased estimators of g(θ). [11]

G14FOS-E1
 ∼ 3 ∼ G14FOS-E1

2 (a) (i) Explain what is meant by a randomised hypothesis test. When might we wish to use
such a test?
(ii) Define (I) a simple hypothesis and (II) a composite hypothesis.
(iii) State the Neyman-Pearson Lemma.
(iv) Define what is meant by a Uniformly Most Powerful (UMP) test of size α for testing
H0 : θ = θ0 against a composite hypothesis H1 . [10]

(b) Consider a sample of size 1 (i.e. n = 1) from the distribution with probability mass function
pX (x|θ) = (1 − θ)θx , x = 0, 1, 2, ..., θ ∈ (0, 1).
(i) Consider testing the null hypothesis H0 : θ = 12 against the alternative hypothesis
H1 : θ > 12 .
Determine the UMP test of H0 against H1 of size α = 0.05.
(ii) Now consider the alternative hypothesis H2 : θ < 21 .
Determine the UMP test of H0 against H2 of size α = 0.05
(iii) Discuss, giving justification, whether or not a UMP test of H0 : θ = θ0 against
H3 : θ ̸= θ0 exists, where θ ̸= θ0 means θ < θ0 or θ > θ0 . [10]

(c) Suppose that θ̂ = (θ̂1 , .., θ̂p )T is an estimator of θ = (θ1 , .., θp )T such that, for large n,
√
n(θ̂ − θ) ≈ Np (0p , V ),
where 0p is the p-vector of zeros and V is a p × p covariance matrix.
( ) 12
∑p
(i) Consider the function d(θ) = ∥θ∥ = θj2 .
j=1

Stating any results that you use, show that

√ ( ) ( )
n d(θ̂) − d(θ) ≈ N 0, {θ T V θ/d(θ)2 } .

(ii) Given a suitable estimator V̂ of V , determine an approximate 95% confidence internal

for d(θ), with the limits of the confidence interval expressed in terms of θ̂ and V̂ . [10]

G14FOS-E1 Turn over

 ∼ 4 ∼ G14FOS-E1

3 (a) (i) Explain what is meant by a family of conjugate prior distributions for a given probability
model pX (x|θ).
(ii) Consider a single observation from a distribution with probability density function
{ 1 3 −4 −β/x
β x e x>0
f (x|β) = 2
0 x ≤ 0.
Determine a conjugate family of prior distributions.
(iii) Calculate the Bayes factor for comparing the following models:
1
M0 : f (x|β = 1), π(M0 ) = ;
2
−β 1
M1 : f (x|β), π(β) = e (β > 0), π(M1 ) = .
2
(iv) Show that model M0 is always preferred when x = 31 and explain why this is a sensible
finding. [14]

(b) (i) Let x = (x1 , .., xn ) denote a random sample from the Uniform(0, θ) distribution, i.e.
the uniform distribution on the interval (0, θ), where θ > 0. Show, providing careful
justification, that the likelihood L(θ|x) is given by
1
L(θ|x) = I{θ: m<θ} (θ)
θn
where m = max1≤i≤n xi .
[Note: For a set A, the indicator function IA (θ) = 1 if θ ∈ A and IA (θ) = 0 otherwise.]
(ii) Suppose that θ in (b)(i) has prior density
{
(1 − c)θ−c θ ∈ (0, 1)
π(θ) =
0 θ∈/ (0, 1),
where c ∈ (0, 1) is a constant. Determine the posterior density π(θ|x).
(iii) For each α ∈ (0, 1), the posterior α−quantile of θ is defined by
∫ θα
π(θ|x)dθ = α.
m

Show that
m
θα = .
{1 − c + cmn+c−1 }1/(n+c−1)

(iv) Derive the 100(1 − α)% Highest Posterior Credible Region for θ. [16]

G14FOS-E1
 ∼ 5 ∼ G14FOS-E1

4 The normal-inverse-gamma distribution N IG(a, d, m, v) has bivariate probability distribution

function (pdf )
( a )d/2 [ { }]
( ) −(d+3)/2 −1 (µ − m) 2
p(µ, σ ) = ( d2) √
2
σ 2
× exp +a , (4.1)
Γ 2 2πv 2σ 2 v

where µ ∈ (−∞, ∞), σ 2 > 0, a > 0, d > 0, γ > 0 and m ∈ (−∞, ∞).
(a) (i) The inverse gamma distribution IG(a, d) is the marginal distribution of σ 2 in the
N IG(a, d, m, v) distribution. Determine the pdf of IG(a, d).
(ii) Find an expression for p(µ), the marginal pdf of µ in the N IG(a, d, m, v) distribution.
[10]
(b) Let x1 , .., xn denote a random sample from the N (µ, σ 2 ) distribution, and suppose that
(µ, σ 2 ) has prior distribution N IG(a, d, m, v). Show that the posterior distribution of
(µ, σ 2 ) is of the form N IG(a1 , d1 , m1 , v1 ), where the quantities d1 , m1 , and v1 , should be
expressed in terms of a, d, m, v and the sample x1 , .., xn .
[Note: You are not required to find a1 .] [10]
2 2
(c) Now suppose y = (y1 , .., yn ) is a random sample from the N (µ, σ0 ) distribution where σ0
is a known quantity. Suppose that µ has an N (δ, τ 2 ) prior where δ and τ 2 > 0 are known
constants. If z is a new observation from the N (µ, σ02 ) distribution, find the predictive
distribution f (z|y). [10]

G14FOS-E1 Turn over

 ∼ 6 ∼ G14FOS-E1

5 Let pX (x|θ) and L(d, θ) denote, respectively, a probability model and a loss function, where
x ∈ X , the sample space, θ ∈ Θ, the parameter space, and d ∈ D, the decision space. Let
δ : X → D denote a decision rule.
(a) (i) Define the risk R(δ, θ) of the decision rule δ.
(ii) Define what is meant by an admissible decision rule.
(iii) Given a prior distribution π(θ), define the Bayes risk R̄(δ, π).
(iv) Prove that the Bayes risk R̄(δ, π) is minimised over the set of decision rules by
δπ (x) = d∗ (x) where, for each x ∈ X , d∗ (x) minimises the expected posterior loss
∫
L(d, θ)π(θ|x)dθ. [14]
θ∈Θ

(b) (i) Let L(d, θ) = (d − θ)2 denote the quadratic loss function. Show that in this case the
posterior expected loss is minimised by the posterior mean of θ.
(ii) Let x = (x1 , .., xn ) denote a random sample from the Poisson (θ) distribution given by
θx e−θ
P (X = x|θ) = , x = 0, 1, 2, ... , θ > 0.
x!
Find the posterior distribution π(θ|x) when the Jeffreys prior is used for θ.

(iii) Find the estimator of θ that minimises the posterior expected quadratic loss.
(iv) Discuss whether the estimator of θ you found in (b)(iii) admissable. [16]

G14FOS-E1 End