NAME:

SEMESTER I 2009/2010 ST2132

MATHEMATICAL STATISTICS

MID-TERM TEST

• Show your work to get full credit.

• Write neatly and clearly.

• Partial credit may be given for wrong or incomplete answers.

• The test comprises three problems and the maximum score is 50 points.

• Answer all questions.

• Good luck!

Question Marks scored Max. Marks

1 16

2 17

3 17

Total 50

1
1. Let X1 , X2 , . . . , X10 be an independent and identically distributed sample from
a normal distribution with mean µ = 0 and variance σ 2 = 3.

(a) Give an example of a statistic, based on X1 , X2 , . . . , X10 , which has the

χ210 distribution. [4 points]
√
Solution Observe that Xi / 3 ∼ N (0, 1) Then

X12 X22 X2
U= + + · · · + 10 ∼ χ210.
3 3 3

(b) Give an example of a statistic, based on X1 , X2 , . . . , X10 , which has the

F4,5 distribution. [4 points]
Solution Let
4 9
X Xi2 X Xi2
V = , W = .
i=1 3 i=5 3

Then (V/4)/(W/5) ∼ F4,5.

(c) Using only X1 , . . . , X10, is it possible to construct a statistic which has the
t10 distribution? Why? [4 points]
Solution No, because any statistic q T having the t10 distribution must
necessarily be of the form T = Z/ U/10 with Z ∼ N (0, 1) independent
of U . But this is not possible here as Z must be a function of X1 , . . . , X10.
(d) Let
10
1 X
X̄ = Xi ,
10 i=1
10
2 1X
S = (Xi − X̄)2 .
9 i=1

If we are not told that the i.i.d. sample X1 , . . . , X10 come from a nor-
mal distribution, can we say that X̄ and S 2 are statistically independent?
Explain. [4 points]
Solution It is proved in class that X̄ and S 2 are independent if the popu-
lation is normal. This is not true in general as X̄ and S 2 are both functions
of X1 , . . . , X10 .

2
2. Let X1 , X2 , . . . , Xn be an i.i.d. sample from the following probability density
function with parameter θ > 0:
(
x −x2 /(2θ2 )
θ2
e , if x > 0,
f (x|θ) =
0, otherwise.

(a) Let θ̂ denote the method of moments estimate of θ. Find θ̂. Hint: Think
of a density of a normal distribution. [5 points]
Solution
Z ∞ x2 −x2 /(2θ2 )
µ1 = E(X1 ) = 2
e dx
√0 θZ
2π ∞ x2 −x2 /(2θ2 )
= √ e dx
θ 0 2πθ
√ Z
2π ∞ x2 −x2 /(2θ2 )
= √ e dx
2θ −∞ 2πθ
√
2π 2
= θ
√2θ
πθ
= √ .
2
Thus the method of moments estimate of θ is
√ √ n
2 2 X
µ̂ = √ µ̂1 = √ Xi .
π πn i=1

(b) Let θ̃ denote the maximum likelihood estimate of θ. Find θ̃. [4 points]
Solution The loglikelihood function is
n
Y xi −x2 /(2θ2 )
l(θ) = log[ 2
e i ]
i=1 θ
n n
X X x2i
= log(xi ) − 2n log(θ) − .
i=1 i=1 2θ2

Hence n
d 2n X x2i
l(θ) = − + 3
.
dθ θ i=1 θ

Equating l0(θ) to zero and solving for θ, we obtain the mle of θ to be

sP
n
i=1Xi2
θ̃ = .
2n

3
(c) Find the asymptotic variance of θ̃. Hint: Use the Gamma function
Z ∞
Γ(z) = tz−1 e−t dt.
0
[4 points]
Solution
∂2
I(θ) = −E[ log f (X1 |θ)]
∂θ2
∂2 X2
= −E{ 2 [log(X1 ) − 2 log(θ) − 12 ]}
∂θ 2θ
2 3X12
= −E[ 2 − 4 ]
θ θ
2 3E(X12 )
= − 2+ .
θ θ4
q
Now letting y = x2/(2θ2 ) and hence x = y/(2θ2 ) and dx = (θ2 /x)dy, we
get
Z ∞ x3 −x2 /(2θ2 )
E(X12 ) = e dx
0 θ2
Z ∞ (2θ2 )3/2y 3/2 −y θ
= e √ dy
0 θ2 2y
Z ∞
= 2θ2 ye−y dy
0
= 2θ2 Γ(2)
= 2θ2 ,
since Γ(2) = 1! = 1. Thus the aymptotic variance of θ̃ is
1 θ2
= .
nI(θ) 4n
(d) Find a sufficient statistic for θ. [4 points]
Solution The joint pdf is
n Qn Pn
Y i=1 xi − x2i /(2θ2 )
f (xi |θ) = e i=1

i=1 θ2n
= g(t, θ)h(x1, . . . , xn ),
where
n
X
t = x2i ,
i+1
2
g(t, θ) = θ−2n e−t/(2θ ) ,
n
Y
h(x1, . . . , xn ) = xi .
i=1

4
 Pn
It follows from the factorization theorem that T = i=1 Xi2 is sufficient for
θ.

3. Let X1 , X2 be a simple random sample (without replacement) of size n = 2

from a finite population of size N = 5. Let µ denote the population mean and
we are interested in estimating µ using an estimator of the form

µ̂c = c1 X1 + c2 X2 ,

where c1 , c2 are constants.

(a) Compute E(µ̂c ) in terms of c1 and c2? [4 points]

Solution
E((µ̂c ) = c1 E(X1 ) + c2 E(X2 ) = (c1 + c2 )µ.
(b) Find a condition on c1 and c2 such that µ̂c is an unbiased estimator of µ.
[4 points]
Solution For µ̂c to be an unbiased estimator of µ, the condition is

c1 + c2 = 1.

(c) What is the variance of µ̂c in terms of c1 and c2? [4 points]

Solution Let σ 2 denote the population variance. Then (using Lemma A
and Lemma B of pages 205 and 207 of the textbook)

Var(µ̂c ) = Var(c1 X1 ) + Var(c2 X2 ) + 2Cov(c1 X1 , c2X2 )

= c21 σ 2 + c22 σ 2 + 2c1 c2 Cov(X1 , X2 )
σ2
= c21 σ 2 + c22 σ 2 − 2c1 c2
N −1
c1 c2
= σ 2(c21 + c22 − ).
2

(d) Find the values of c1 and c2 such that µ̂c is an unbiased estimator of µ and
has the smallest standard deviation. [5 points]
Solution Easiest way is to write c2 = 1 − c1 (from part (b)) and minimize

c1 c2 c1(1 − c1)
c21 + c22 − = c21 + (1 − c1 )2 −
2 2
with respect to c1 . This gives us c1 = c2 = 1/2.
Another way isto use the method of Lagrange multipliers.