Tutorial 8

The University of Hong Kong

1 Recap

Estimator properties

1. Confidence Intervals

2 This week
1. Confidence Intervals

2.1 Interval Estimation

- An interval estimator of θ is a random interval [L(X), U (X)], where L(X) := L (X1 , . . . , Xn ) and
U (X) := U (X1 , . . . , Xn ) are two statistics such that L(X) ≤ U (X) with probability 1 .

- If X = x is observed, [L(x), U (x)] is the interval estimate of θ.

- For an interval estimator [L(X), U (X)] of θ, the confidence coefficient [L(X), U (X)], denoted by
(1 − α) is the infimum of the coverage probabilities P(θ ∈ [L(x), U (x)]), that is,

1 − α = inf P(θ ∈ [L(x), U (x)]).

θ

- A random variable Q(X, θ) = Q (X1 , . . . , Xn , θ) is a pivotal quantity if the distribution of Q(X, θ)

is free of θ. That is regardless of distribution of X, Q(X, θ) has the same distribution of all values
of θ.

2.2 Confidence Intervals for Means

Let X = {X1 , . . . , Xn } be an independent random sample from the population N µ, σ 2 . Define

2
X̄ = n1 ni=1 Xi and S 2 = n1 ni=1 Xi − X̄ .
P P

- If σ 2 is known, a 1 − α confident interval of µ is

 
σ σ
x̄ − zα/2 √ , x̄ + zα/2 √
n n

given that the observed value of X̄ is x̄.

1
- If σ 2 is unknown, a 1 − α confident interval of µ is
 
s s
x̄ − tα/2,df =n−1 √ , x̄ + tα/2,df =n−1 √
n−1 n−1

given that the observed values of X̄ and S are x̄ and s respectively.

2.3 Distribution Theory

Let X = {X1 , . . . , Xn } be an independent random sample from the population N µ, σ 2 . Define

2
X̄ = n1 ni=1 Xi and S 2 = n1 ni=1 Xi − X̄ .
P P

- X̄ and S 2 are independent.

Pk 2
- If Z1 , . . . , Zk are k independent N (0, 1) random variables, then i=1 Zi ∼ χ2k .
Pn 2
nS 2 (Xi −X̄ )
- σ2
= i=1
σ2
∼ χ2n−1 .

- If Z ∼ N (0, 1), U ∼ χ2k and Z and U are independent, then T = √Z ∼ tk .

U/k

X̄−µ
- √
S/ n−1
∼ tn−1 .

- When n → ∞, the distribution function of tn tends to the distribution function of N (0, 1).

3 Exercise

1. A publisher wants to estimate the mean length of time (in minutes) all adults spend on reading
newspaper. To determine this estimate, the publisher takes a random sample of 15 people and
obtains the following results

11, 9, 8, 10, 10, 9, 7, 11, 11, 7, 6, 9, 10, 8, 10

Assume that the population of reading times is normally distributed.

a. Construct a 90% confidence interval for the population mean if the publisher assumes that σ is
1.5 minutes from past studies.

b. Construct a 90% confidence interval for the population mean if the publisher assumes that σ is
unknown.

Solution:

a. Normal population with known population variance. Let µ be the population, mean length

2
 of time all adults spent on reading news paper.
 
σ
CI90% (µ) = x̄ ± z0.05 √ Note: x̄ = 9.0667, z0.05 = 1.645
n
 
1.5
= 9.0667 ± 1.645 √
15
= [8.4296, 9.7038]

b. Normal population with unknown population variance.

 
s
CI90% (µ) = x̄ ± t0.05,n−1 √ Note: s ≈ 1.5261, t0.05,14 = 1.761
n−1
 
s
= x̄ ± t0.05,14 √
14
 
1.5261
= 9.0667 ± 1.761 × √
14
≈ [8.3484, 9.7850]

2. An education officer wants to use the mean of a random sample of 16 students to estimate the
mean score (µ) that all students of the city would get if they took a certain MAF-test. Based on
experience, the officer knows that the population has a normal distribution with σ = 35.

a. What is the width of the 95% confidence interval?

b. Suppose that sample data are:

137, 125, 184, 116, 98, 163, 150, 145, 178, 163, 151, 103, 75, 86, 193, 141

Construct a 90% confidence interval for µ.

c. If the education officer wants to obtain a 99% confidence interval for with a width not exceeding
20 points, how many students must be included for sample test?

Solution:

a. Normal population with known population variance.

 
σ
CI95% (µ) = x̄ ± z0.025 √
n

Width of CI = 2z0.025 √σn = 2 × 1.96 × √35

16
= 34.3

b. Normal population with known population variance.

 
σ
CI90% (µ) = x̄ ± z0.05 √
n
 
35
= 138 ± 1.645 × √ ≈ [123.6063; 152.3938] x̄ = 138, z0.05 = 1.645.
16

3
 c.
σ
CI = 2z0.005 √ ⩽ 20
n
2z0.005 σ 2
 
⇒n⩾ ≈ 81.2883
20
∴ n = 82

3. Consider a random sample T1 , T2 , . . . , Tn drawn from an exponential distribution with rate λ > 0.
P
a. Show that 2λ Ti has a Chi-squared distribution with 2n degrees of freedom. (Hint: Chi-
squared distribution with n degrees of freedom is the same as Gamma (0.5n, 0.5), the Gamma
distribution with the shape parameter equal to 0.5n and the rate parameter equal to 0.5 . The pdf
of Gamma(α, β) is f (x) = β α xα−1 e−βx /Γ(α).)

b. Find an equal-tailed confidence interval for λ with α level.

Solution:

a.
iid
Ti ∼ Exp(λ), λ > 0 (1)
λ
∴ MTi (t) = for t < λ (2)
λ−t
X n
Let Y = 2λ Ti (3)
i=1

Let’s prove the mgf of Y is the same as the mgf of Γ 2n 1 2

2 , 2 , i.e., χ2n .
   
MY (t) = E etY = E et2λΣTi = E e2λtT1 e2λtT2 · · · e2λtTn


     
ind.
= E e2λtT1 E e2λtT2 · · · eE e2λtTn
identical
= [MT i (2λt)]n
n  n !n
 1
λ 1
= = = 1 2
λ − 2λt 1 − 2t 2 −t
   
1 2n 1
∴ Y ∼ Γ n, ≡Γ , ≡ χ22n
2 2 2

4
 b.
n
X
∵ Y = 2λ Ti ∼ χ22n
i=1
n
!
X
∴1−α=P χ21− α ,2n < 2λ 2
Ti < χ α ,2n
2 2
i=1
χ21− α ,2n χ2α ,2n
!
2
=P < λ < P2n
2 ni=1 Ti
P
2 i=1 Ti
" 2
χ2α ,2n
#
χ1− α ,2n
2 2
∴ CI(1−α) (λ) = , P
2 ni=1 Ti 2 ni=1 Ti
P

4. Suppose that we obtained a single observation Y from an exponential distribution with mean
θ, f (y) = 1θ e−y/θ I(y ≥ 0). Use Y to form a 90% confidence interval for θ.

Solution:

First of all, let’s find the cdf of Y .

Z y Z y
1 −t 1 h −t iy
FY (y) = fY (t)dt = e θ dt = e θ · (−θ) (4)
0 0 θ θ 0
y
= 1 − e− θ for y > 0. (5)
Now, let’s find [a, b] s.t. p(Y < a) = 0.05 and P (Y < b = 0.95).
a a
0.05 = P (Y < a) = FY (a) = 1 − e− θ → e− θ = 0.95 (6)
a
→ − = ln 0.95 → a = −θ ln 0.95. (7)
θ
b b
0.95 = p(Y < b) = FY (b) = 1 − e− θ → e− θ = 0.05 (8)
b
→ − = ln 0.05 → b = −θ ln 0.05. (9)
θ

∴ 0.90 = P (a < Y < b) (10)

= P (θ ln 0.95 < Y < −θ ln 0.05) (11)
Y Y
= P( <θ< ) (12)
− ln 0.05 − ln 0.95
Y Y
= P( <θ< ). (13)
2.9957 0.0513

5
 y y
∴ CI90% (θ) = [ 2.9957 , 0.0513 ], where y is the single observation of Y.