Master’s Written Examination and Solution

Option: Statistics and Probability Spring 2017

Full points may be obtained for correct answers to eight questions. Each
numbered question (which may have several parts) is worth the same num-
ber of points. All answers will be graded, but the score for the examination
will be the sum of the scores of your best eight solutions.

Use separate answer sheets for each question. DO NOT PUT YOUR
NAME ON YOUR ANSWER SHEETS. When you have finished,
insert all your answer sheets into the envelope provided, then seal it.

1
Problem 1 – Stat 401

Consider a random experiment that consists of drawing at random one

chip from a bowl containing 10 chips. Each chip has an ordered pair of
numbers on it: one with (1, 1), one with (2, 1), two with (3, 1), one with
(1, 2), two with (2, 2), and three with (3, 2). Let the random variables X1
and X2 be defined as the respective first and second values of the ordered
pair.

1. Find the joint probability mass function of (X1 , X2 ).

2. Find the marginal probability mass function of X1 and X2 .
3. Let g(x1 , x2 ) = x1 x2 . Compute E g(X1 , X2 ).

Solution:
1. The joint probability mass function of (X1 , X2 ) is given by the follow.
X2 = 1 X2 = 2
1 1
X1 = 1 10 10
1 2
X1 = 2 10 10
2 3
X1 = 3 10 10

2. The marginal probability mass function of X1 is

X1 = 1 X1 = 2 X1 = 3
2 3 5
10 10 10
The marginal probability mass function of X2 is
X2 = 1 X2 = 2
4 6
10 10

3.
1 1 1 2
Eg(X1 , X2 ) =1 × 1 × +1×2× +2×1× +2×2×
10 10 10 10
2 3
+3×1× +3×2× .
10 10

2
Problem 2 – Stat 401

A group of friends plan to go hiking on a summer weekend. In recent years,

the probability that it rains on a summer day is 4%. Unfortunately, the
weatherman has predicted rain for the day of the hiking. When it actually
rains, the weatherman correctly forecasts rain 90% of the time. When it
doesn’t rain, he incorrectly forecasts rain 10% of the time. What is the
probability that it will rain on the day of the hiking?

Solution:
Let

R = {it rains on a day}

N R = {it does Not rain on a day}
W = {the weatherman predicts it will rain on a day}

According to the problem, we have

P(R) = 4%, P(N R) = 96%, P(W |R) = 90%, P(W |N R) = 10%.

We are asked to compute

P(R|W ).
By the Bayes’ Theorem, we have
P(W |R)P(R)
P(R|W ) =
P(W |R)P(R) + P(W |N R)P(N R)
90% × 4%
= .
90% × 4% + 10% × 96%

Problem 3 – Stat 401

Let (X, Y ) be a random vector with joint probability density function

given by
 −(x+y)
2e ; 0 < x < y < ∞,
f (x, y) =
0; otherwise.
1. Find the conditional probability density function of Y given X = x.
2. Compute E(Y |X = x).

3
Solution:
1. The probability density function fX (x) of X is given by
Z ∞ Z ∞
fX (x) = f (x, y)dy = 2e−x e−y dy = 2e−2x ; 0 < x < ∞.
x x
Therefore, the conditional probability density function of Y given
X = x is:
f (x, y)
fY |X (y|x) = = ex−y ; 0 < x < y < ∞.
fX (x)
2. The conditional expectation is computed by the conditional density
as follows.
Z ∞
E(Y |X = x) = y · ex−y dy = x + 1.
x

Problem 4 – Stat 401

Let X have a Poisson distribution with parameter m. If m is an experi-
mental value of a random variable having a gamma distribution with α = 2
and β = 1, compute P (X ≤ 2).
Solution:
Let M be the random variable with an experimental value m. The joint
distribution of X and M can be written as f (x|m)g(m), where f (x|m) is
the conditional density function of X given M and g(m) is the density
function of M . Thus the joint distribution of X and M is
e−m mx −m
f (x, m) = me
x!
1
= mx+1 e−2m .
x!
The marginal distribution of X is
Z ∞
f (x) = f (x, m)dm
0
Z ∞
1 x+1 −2m
= m e dm
0 x!
Γ(x + 2)( 21 )x+2 ∞
Z
1 x+1 −2m
= 1 x+2 m e dm
x! 0 Γ(x + 2)( 2 )
 x+2
1
= (x + 1) , x = 0, 1, 2, . . . ,
2
4
Thus
P (X ≤ 2) = P (X = 0) + P (X = 1) + P (X = 2)
 2  3  4
1 1 1
= +2× +3×
2 2 2
11
= .
16

Problem 5 – Stat 411

Let X1 ,X2 ,...,Xn be a random sample from a Poisson distribution with

parameter θ > 0.

(i) Find the complete sufficient statistics for θ.

(ii) Find the MVUE of θ2 e−θ .

(iii) Find the MLE of θ2 e−θ .

(iv) Derive the asymptotic distribution of the MLE.

Solution:
(i) The probability mass function of Xi can be written as

exp{Xi ln θ − ln Xi − θ}, Xi = 0, 1, 2, 3, . . . ,

Since the support does not depend on θ and ln θ is nontrivial contin-

uous
Pn function of θ, this is a regular exponential class. Consequently,
i=1 Xi is a complete sufficient statistics for θ.

θ2 e−θ and ni=1 Xi is a complete sufficient statis-

P
(ii) Since 2P (X1 = 2) = P
tics, the 2P (X1 = 2| ni=1 Xi ) is the P
MVUE of θ2 e−θ . Next we should
derive the expression of 2P (X1 = 2| ni=1 Xi ).

5
 n
P (X1 = 2, ni=1 Xi )
X P
2P (X1 = 2| Xi ) = 2
P ( ni=1 Xi )
P
i=1
−θ 2

 −(n−1)θ Pn
X −2
Pn 
e (θ) /2 e ((n − 1)θ) i=1 i
/( i=1 Xi − 2)!
=2 Pn
e−nθ (nθ) i=1 Xi /( ni=1 Xi )!
P
Pn
(n − 1) i=1 Xi −2 ( ni=1 Xi )!
P
= Pn X Pn
n i=1 i ( i=1 Xi − 2)!2!
Pn   2  Pni=1 Xi −2
i=1 Xi 1 1
=2 1−
2 n n

(iii) The log likelihood function can be written as

n n
!
X Y
l(θ) = −nθ + Xi ln θ − ln Xi ! .
i=1 i=1

The first derivative of l(θ) is

Pn
i=1 Xi
l0 (θ) = −n + .
θ
Solve l0 (θ) = 0, we have θ̂ = X̄. Here θ̂ is the MLE of θ. Thus the
MLE of θ2 e−θ is X̄ 2 e−X̄ .
(iv) We first derive the Fisher information matrix I(θ).
 2 
∂ l(θ)
I(θ) = −E
∂θ2
 Pn 
X i
= −E − i=12
θ
n
= .
θ
∂θ2 e−θ
Since ∂θ = (2θ − θ2 )e−θ , by the asymptotic theory of MLE,
√  2 −X̄ 2 −θ

n X̄ e − θ e ∼ N (0, θ(2θ − θ2 )2 e−2θ ).

6
Problem 6 – Stat 411

Let X1 ,X2 ,...,Xn be a random sample from a distribution with density

function f (x; θ) = 1θ e−x/θ , 0 < x < ∞.

(i) Are ni=1 Xi and minni=1 Xi / ni=1 Xi independent? Justify your an-
P P

swer.
Pn
(ii) Compute E(minni=1 Xi / i=1 Xi ).

Solution:
(i) The pdf of X can be written as exp(−x/θ − ln θ), where Pn 0 < x < ∞.
Thus it is a regular exponential class. Consequently i=1P Xi is a com-
plete sufficient statistics. On the other hand, mini=1 Xi / ni=1 Xi is aP
n

scale invariance function and Pf (x; θ) is scale function, thus minni=1 Xi / ni=1 Xi
is ancillary statistics. Thus ni=1 Xi and minni=1 Xi / ni=1 Xi are in-
P
dependent.
(ii) First we have Z ∞
x −x/θ
E(Xi ) = e dx = θ.
0 θ
So we E( ni=1 Xi ) = nθ. The PDF of minni=1 Xi is given by
P

n − x
n−1 − x n nx
f1 (x) = e θ e θ = e− θ .
θ θ
Thus Z ∞
n nx − nx θ
E(min Xi ) = e θ dx = .
i=1 0 θ n
By the conclusion in (i), we have
n
! ! n
!
n
minni=1 Xi
 
X mini=1 Xi X
E Xi Pn =E Xi E Pn .
i=1 i=1 X i i=1 i=1 Xi

On the other hand,

n
! !
minni=1 Xi
  
X n
E Xi Pn = E min Xi .
i=1 i=1 Xi i=1

Pn
Thus E(minni=1 Xi / i=1 Xi ) = 1
n2

7
Problem 7 – Stat 411

Let X1 , X2 , . . . , X25 denote a random sample of size 25 from a normal

distribution N (0, θ), where 0 < θ < ∞.

(a) Find a most powerful critical region of size α = 0.05 for testing H0 :
θ = 1 against H1 : θ = 3 .
For your reference, P (Y ≥ 37.652) = 0.05 if Y ∼ χ2 (25).
(b) Determine the power of the test you obtained in (a) .
Note that it’s enough to conclude something like “the power is P (Y ≥
37.652) with Y ∼ χ2 (25)”.
(c) Find a uniformly most powerful critical region of size α = 0.05 for
testing H0 : θ = 1 against H1 : θ > 1 .
(d) Is there a uniformly most powerful test of H0 : θ = 1 against H1 : θ 6=
1? Why?

Solution:

(a) Let x = (x1 , . . . , x25 )T stand

P25for 2the data. The likelihood function
−n/2 −n/2
L(θ; x) = (2π) θ exp{− i=1 xi /(2θ)}. Then the likelihood ratio
statistic
( 25
) 25
L(θ = 1; x) n/2 1 X
2
X
= 3 · exp − xi ≤ k ⇐⇒ x2i ≥ c
L(θ = 3; x) 3 i=1 i=1

where k and c = −3 log(k · 3−n/2 ) are some positive constants. P

According to the Neyman-Pearson theorem, C = {(x1 , . . . , x25 )T | 25 2
i=1 xi ≥
c} is a most powerful
hP critical region,
i where the value of c can be determined
25 2
by 0.05 = Pθ=1 i=1 Xi ≥ c .
If θ = 1, then 25 2 2
P
i=1 Xi ∼ χ (25). Thus c = 37.652 in this case and a most
powerful critical region of size α = 0.05 is
25
X
T
{(x1 , . . . , x25 ) | x2i ≥ 37.652}.
i=1
P25 2
(b) Under H1 , θ = 3 and thus i=1 Xi /3 ∼ χ2 (25). Then the power of
the test

8
 " 25
# " 25
#
X X
Pθ=3 Xi2 ≥ 37.652 = Pθ=3 Xi2 /3 ≥ 12.551 = P [Y ≥ 12.551]
i=1 i=1

if Y ∼ χ2 (25).
(c) Repeating the procedure in (a) for H0 : θ = 1 against H1 : θ = θ1 , we
can
P25always get the same most powerful critical region C = {(x1 , . . . , x25 )T |
2
i=1 xi ≥ 37.652}, as long as θ1 > 1. Therefore, C is a uniformly most
powerful critical region for testing H0 : θ = 1 against H1 : θ > 1 .
(d) From (c), we know that a uniformly most powerful critical region
for testing H0 : P θ = 1 against H1 : θ > 1 has to take the form of
{(x1 , . . . , x25 ) | 25
T 2
i=1 xi ≥ c} for some constant c. Similarly, we know
that a uniformly most powerful critical region for testing P H0 : θ = 1
against H1 : θ < 1 has to take the form of {(x1 , . . . , x25 ) | 25
T 2
i=1 xi ≤ d}
for some constant d. Therefore, there is no uniformly most powerful test
for H0 : θ = 1 against H1 : θ 6= 1.

Problem 8 – Stat 411

Consider a random sample X1 , X2 , . . . , Xn from a distribution with pdf

f (x; θ) = θ(1 − x)θ−1 , 0 < x < 1, zero elsewhere, where θ > 0.

(a) What’s the name of the distribution?

(b) Find the likelihood ratio statistic Λ for testing H0 : θ = 1 against
H1 : θ 6= 1.
(c) Write
Qn the critical region for the likelihood ratio test in (b) in terms
of i=1 (1 − xi ).

Solution:

(a) It’s beta(1,θ) distribution.

θ−1
(b) The likelihood function is L(θ) = ni=1 θ(1−xi )θ−1 =Qθn [ ni=1 (1 − xi )] .
Q Q

The log-likelihood function is l(θ) = n log θ + (θ − 1) log ni=1 (1 − xi ). Solv-

ing l0 (θ) = n/θ + log ni=1 (1 − xi ) = 0, we get θ’s mle
Q

n
θ̂ = Qn .
− log i=1 (1 − xi )

9
Then ni=1 (1 − xi ) = e−n/θ̂ and the likelihood ratio statistic for testing
Q

H0 : θ = 1 against H1 : θ 6= 1 is
" n
#1−θ̂
L(1) Y
Λ= = (θ̂)−n (1 − xi ) = (θ̂)−n · e−n/θ̂ · en
L(θ̂) i=1
Qn
(c) Denote Y = i=1 (1 − xi ) ∈ (0, 1). Then θ̂ = −n/(log Y ) ∈ (0, ∞).
From (b), Λ = Λ(θ̂) = en · (θ̂)−n e−n/θ̂ . Then log Λ = n − n log θ̂ − n/θ̂ and

∂ log Λ n n n
=− + = (1 − θ̂)
∂ θ̂ θ̂ (θ̂)2 (θ̂)2
As a function of θ̂, Λ or log Λ increases before θ̂ = 1 and decreases after
θ̂ = 1. Therefore,

Λ ≤ λ ⇐⇒ θ̂ ≤ c1 or θ̂ ≥ c2 ⇐⇒ Y ≤ d1 or Y ≥ d2
with λ = Λ(c1 ) = Λ(c2 ) providing critical values d1 = e−n/c1 and d2 =
e−n/c2 for Y . That is, the critical region for the likelihood ratio test in (b)
takes the form of
n
Y n
Y
{(x1 , . . . , xn ) | (1 − xi ) ≤ d1 or (1 − xi ) ≥ d2 }.
i=1 i=1

Problem 9 – Stat 481

Let X̄ be the mean of a random sample of size n = 16, from a normal

distribution with mean µ and standard deviation σ = 8. To test the hy-
potheses H0 : µ ≤ 35 against H1 : µ > 35. Suppose we reject the null
hypotheses H0 when X̄ ≥ 36.5 and retain H0 otherwise.
(1). Determine the significant level α of this test;
(2). The probability of type I error at µ = 34;
(3). The type II error and its power at µ = 36.

Solution:


It is known that X̄ ∼ N µ, σ 2 /n .

10
(1). The significant level is the maximum type I error which can be reached
at the boundary point µ = 35.
 

X̄ − µ 36.5 − 35
α = P X̄ ≥ 36.5 µ = 35 = P √ ≥ = 1 − Φ (0.75) .
σ/ n 8/4
(2). Type I error probability at µ = 34


P X̄ ≥ 36.5 µ = 34 = 1 − Φ (1.25)

(3). Type II error probability at µ = 36



P X̄ < 36.5 µ = 36 = Φ (0.25) ,

Power at µ = 36


P X̄ ≥ 36.5 µ = 36 = 1 − Φ (0.25) .

Problem 10 – Stat 481

A multiple linear regression
model Yi = β0 + β1 xi1 + β2 xi2 + εi , i =

i.i.d. 2
1, ...16, εi ∼ N 0, σ . It is calculated that SST O = 200, SSR = 154.
(1). Construct the ANOVA table.
(2). Calculate and interpret the coefficient of the determination R2 .
(3). Test H0 : β1 = β2 = 0 at the α = 0.05 significance level. [
F (0.05, 2, 13) = 3.8, F (0.05, 1, 14) = 4.6]
(4). Find the statistic for an individual test H0 : β1 = 0 against H1 : β1 6=
0. Derive its sampling distribution.

Solution:

(1). ANOVA Table

Source S.S. DF M S F
Regression 154 2 77 21.76
Error 46 13 3.54 Total

Total 200 15

11
(2). The coefficient of the determination
SSR
R2 = = 77%
SST O
It means that there are 77% of variation of the response explained by the
linear regression model.
(3) The F -statistic in the ANOVA table F = 21.76 > F (0.05, 2, 13) =
3.80. Therefore, we will reject the null hypothesis at level 0.05, i.e. at
least one predictor contribute significantly to explain the variation in the
response.
(4). The regression coefficient β1 is the effect on mean response for a unit
increase in predictor variable X1 , while holding the other predictor X2
constant.
Use t-test for individual test H0 : β1 = 0 and the statistic is
β̂1
tβ̂1 =  
s β̂1
 
where β̂1 and s β̂1 are the least square estimate and its standard error
of linear coefficient β1 . Under H0 : β1 = 0, the test statistic tβ̂1 ∼ t (13) .
Problem 11 – Stat 481
(1). In the context of one-way ANOVA ( fixed effect with k levels), write
down the model, the asummptions, the null hypothesis and the expression
for (I).Total Sum of Squares (SSTO), (II) Treatment Sum of Squares
(SSTR), (III) Error Sum of Squares (SSE).
(2). Establish that SST O = SST R + SSE .
(3). Under the null hypothesis show that
SSE 2 SST R
2
∼ χ (N − k) , 2
∼ χ2 (k − 1)
σ σ
and SSE is independent of SST R.
Solution:


(1). ANOVA model: Yij = µ + τi + εij , εij ∼i.i.d. N 0, σ 2 , i = 1, ..., k; j =
1, ...n. Null hypothesis H0 : τ1 = τ2 = τ3 = 0, the sum of squares
k X
n k k X
n
X
2 X
2 X
2
SST O = Yij − Ȳ·· , SST R = n Ȳi· − Ȳ·· , SSE = Yij − Ȳi· .
i=1 j=1 i=1 i=1 j=1

12
(2).
k X
X n k X
X n



Yij − Ȳi· Ȳi· − Ȳ·· = (εij − ε̄i· ) (ε̄i· − ε̄·· )
i=1 j=1 i=1 j=1
k
X n
X
= (ε̄i· − ε̄·· ) (εij − ε̄i· ) = 0
i=1 j=1
k X
n
X
2
SST O = Yij − Ȳi· + Ȳi· − Ȳ··
i=1 j=1
k k X
n
X
2 X
2
=n Ȳi· − Ȳ·· + Yij − Ȳi· = SST R + SSE
i=1 i=1 j=1

2
(3). It is know that (n − 1) s2i /σ 2 = nj=1 Yij − Ȳi· /σ 2 ∼ χ2 (n − 1) .
P

All observations are independent, based on the additivity of the Chisquare

distribution
k
SSE X (n − 1) s2i
2
= 2
∼ χ2 (N − k) , N = kn.
σ i=1
σ


Note that under H0 : τi
= 0, i = 1, 2, 3, Yij = µ + εij , Yij ∼i.i.d. N µ, σ 2 ,
and Ȳi· ∼i.i.d. N µ, σ 2 /n . Based on the Student’s Theorem
Pk
2
SST R Ȳ i· − Ȳ··
2
= i=1 2 ∼ χ2 (k − 1) .
σ σ /n
In Student’s Theorem, the sample mean Ȳi· is independent of the sample
variance s2i . Hence SSE is independent of SST R.
Problem 12 – Stat 481
Following is the partially completed ANOVA table from a two-factor 3 × 4
factorial experiment with n = 3 observations in each cell:
Source SS df MS F
A 20.12
B 15.21
AB 15.52
Error
Total 102.00

13
(a) For a fixed-effects model, complete the ANOVA table.
(No need to calculate the p-values.)
(b) What assumptions do you need for the fixed-effect model?
Briefly describe how to check your assumptions.
(c) For a random-effects model, complete the ANOVA table.
(No need to calculate the p-values.)

Solution:

(a)

Source SS df MS F
A 20.12 2 10.06 4.72
B 15.21 3 5.07 2.38
AB 15.52 6 2.59 1.21
Error 51.15 24 2.13
Total 102.00 35

(b) The random errors are assumed to be iid normal. After fitting the
regression model and obtaining the residuals, we may use the residual plot
to check the model assumption.
(c)

Source SS df MS F
A 20.12 2 10.06 3.884*
B 15.21 3 5.07 1.958*
AB 15.52 6 2.59 1.21
Error 51.15 24 2.13
Total 102.00 35

14