Math 362, Problem Set 6

Due 3/23/11

1. (6.1.2) Let X1 , X2 , . . . , Xn be a random sample from a Γ(α = 3, β = θ)

distribution, 0 < θ < ∞. Determine the mle of θ.
Answer
We have

1 Y P
L(θ) = (Xi2 )e− Xi /θ
θ3n 2n X X
`(θ) = −3n log(θ) − n log(2) + 2 log(Xi ) − Xi /θ
P
−3n Xi
`0 (θ) = + 2
.
θ θ
Solving `0 (θ) = 0, we have
1
θ̂ = X̄.
3
2. (6.1.5) Suppose X1 , . . . , Xn are iid with pdf f (x; θ) = 2x/θ2 , 0 < x ≤ θ,
zero elsewhere. Find
(a) The mle θ̂ for θ.
(b) The constant c so that E[cθ̂] = θ.
(c) The mle for the median of the distribution.
Answer:
We have that
2n
Q
(Xi )

θ 2n max Xi < θ
L(θ) =
0 max(Xi ) > θ.
n Q
Since 2 θ2n(Xi ) is a decreasing function of θ, L(θ) is maximized when θ =
max(Xi ). That is θ̂ = max(Xi ) = Yn .
2
Using the pdf of the order statistics, and the fact that FX (x) = n xθ2 .
θ
x2n
Z
2n
E[θ̂] = 2n = θ.
0 θ2n 2n + 1

1
 2n+1
Thus we take c = 2n .
For (b), we compute that the median of X’s distribution is Q2 so that
Q2
Q2
Z
1 2x
= 2
= 22 .
2 0 θ θ

Thus Q2 = √θ .
2
Yn
Thus Q̂2 = √
2
.

3. (6.1.9) Suppose X1 , . . . , Xn are iid with pdf f (x; θ) = (1/θ)e−x/θ . Find

the mle of P(X ≤ 2).
Answer:
For this distribution we have that
X
`(θ) = −n log(θ) − Xi /θ
P
−n Xi
`0 (θ) = + 2
.
θ θ

Solving `0 (θ) = 0, we see that θ̂ = X̄.

Since Z 2
P(X ≤ 2) = (1/θ)e−x/θ = 1 − e−2/θ .
0

Thus the mle of P(X ≤ 2) is

1 − e−2/X̄ .

4. (6.1.10) If X1 , X2 , . . . Xn be a random sample from a Bernoulli distribution

with parameter p. If p is restricted so that we know that 21 ≤ p ≤ 1, find
the mle of this parameter.
Answer:
We have that
P P
L(θ) = p Xi (1 − p)n− Xi
X X
`(θ) = (Xi ) log(p) + (n − Xi ) log(1 − p)
P P
Xi n − Xi
`0 (θ) = + .
p 1−p

Setting `0 (θ) = 0, and solving we find θ = X̄. On the other hand since
`0 (θ) > 0 for θ < X̄ and `0 (θ) < 0 for θ < X̄. We see that if X̄ < 12 , we
have `(θ) is maximized at θ = 12 . Thus θ̂ = max{ 12 , X̄}.

2
5. (6.2.1) Prove that X̄, the mean of a random sample of size n from a distri-
bution that is N (θ, σ 2 ) is, for every known σ 2 > 0, an efficient estimator
of θ.
Answer:
We compute
1 1
(x − θ)2


f (x; θ) = √ exp −
2πσ 2 2σ 2
1 1
log(f (x; θ)) = − log(2πσ 2 ) − 2 (x − θ)2
2 2σ
∂ log(f (x; θ)) x−θ
=− 2
∂θ σ
∂ 2 log(f (x; θ)) 1
= − 2.
∂θ σ
Thus
1 1
I(θ) = −E[− 2
] = 2.
σ σ
2
On the other hand, Var(X̄) = σn . We have that k(θ) = E[X̄] = θ, so that
2
k 0 (θ) = 1. Thus the Rao-Cramer bound is σn which matches the variance
of X̄. Thus X̄ is an efficient estimator of θ.
6. (6.2.7’) Let X have a gamma distribution with α = 3 and β = θ > 0.
(a) Find the Fisher information I(θ).
(b) If X1 , . . . , Xn is a random sample from this distribution, show that
the mle of θ is an efficient estimator of θ.
√
(c) What is the asymptotic distribution of n(θ̂ − θ)?
Note: I changed α = 4 in the original problem to α = 3 since you
computed the mle for θ in this case above.
Answer:
We have that

1 2 −x/θ
f (x; θ) = x e
2θ3
x
log(f (x; θ)) = − log(2) − 3 log(θ) + 2 log(x) −
θ
∂ log(f (x; θ)) −3 x
= + 2
∂θ θ θ
∂ 2 log(f (x; θ)) 3 2x
= 2 − 3.
∂θ θ θ

3
 3
Since E[X] = 3θ, we see that I(θ) = θ2 .

We found the mle for θ in problem one to be X̄/3. Note that Var(θ̂) =
1 θ2 θ2
9n Var(Xi ) = 3n . The Rao-Cramer bound is 3n as well, so θ̂ is efficient.
√ 1
Note that n(θ̂ − θ) → N (0, I(θ) ).