Tutorial Week 2: Solutions

Statistical Inference: STAT3013/4027/6027

1. a. The sample space for 𝑈 is 𝑆𝑈 = {0, 1, 2, 3, 4}:

𝑝𝑈 (0) = 𝑃 (𝑈 = 0) = 𝑃 (𝑋 = 0, 𝑌 = 0) = 0.1
𝑝𝑈 (1) = 𝑃 (𝑈 = 1) = 𝑃 (𝑋 = 0, 𝑌 = 1) + 𝑃 (𝑋 = 1, 𝑌 = 0) = 0.1 + 0.25 = 0.35
𝑝𝑈 (2) = 𝑃 (𝑈 = 2) = 𝑃 (𝑋 = 0, 𝑌 = 2) + 𝑃 (𝑋 = 1, 𝑌 = 1) + 𝑃 (𝑋 = 2, 𝑌 = 0) = 0.2 +
0 + 0.05 = 0.25
𝑝𝑈 (3) = 𝑃 (𝑈 = 3) = 𝑃 (𝑋 = 1, 𝑌 = 2) + 𝑃 (𝑋 = 2, 𝑌 = 1) = 0.2 + 0.05 = 0.25
𝑝𝑈 (4) = 𝑃 (𝑈 = 4) = 𝑃 (𝑋 = 2, 𝑌 = 2) = 0.05
b. The marginal distribution for 𝑋 is:

𝑝𝑋 (0) = 𝑃 (𝑋 = 0) = 𝑃 (𝑋 = 0, 0 ≤ 𝑌 ≤ 2) = 0.1 + 0.1 + 0.2 = 0.4

𝑝𝑋 (1) = 𝑃 (𝑋 = 1) = 𝑃 (𝑋 = 1, 0 ≤ 𝑌 ≤ 2) = 0.25 + 0 + 0.2 = 0.45
𝑝𝑋 (2) = 𝑃 (𝑋 = 2) = 𝑃 (𝑋 = 2, 0 ≤ 𝑌 ≤ 2) = 0.05 + 0.05 + 0.05 = 0.15

The marginal distribution for 𝑌 is:

𝑝𝑌 (0) = 𝑃 (𝑌 = 0) = 𝑃 (0 ≤ 𝑋 ≤ 2, 𝑌 = 0) = 0.1 + 0.25 + 0.05 = 0.4

𝑝𝑌 (1) = 𝑃 (𝑌 = 1) = 𝑃 (0 ≤ 𝑋 ≤ 2, 𝑌 = 1) = 0.1 + 0 + 0.05 = 0.15
𝑝𝑌 (2) = 𝑃 (𝑌 = 2) = 𝑃 (0 ≤ 𝑋 ≤ 2, 𝑌 = 2) = 0.2 + 0.2 + 0.05 = 0.45

Clearly, 𝑋 and 𝑌 are not independent as 𝑃 (𝑋 = 0, 𝑌 = 0) = 0.1 ≠ 𝑃 (𝑋 = 0) × 𝑃 (𝑌 =

0) = 0.4 × 0.4 = 0.16.

1
 c. Using the mutliplication rule for independent random variables we have:

Values of 𝑋1 Values of 𝑌1
0 1 2
0 0.16 0.06 0.18
1 0.18 0.0675 0.2025
2 0.06 0.0225 0.0675

d. Similar to part (a), the pmf of 𝑈1 is calculated as:

𝑝𝑈1 (0) = 𝑃 (𝑈1 = 0) = 𝑃 (𝑋1 = 0, 𝑌1 = 0) = 0.16

𝑝𝑈1 (1) = 𝑃 (𝑈1 = 1) = 𝑃 (𝑋1 = 0, 𝑌1 = 1) + 𝑃 (𝑋1 = 1, 𝑌1 = 0) = 0.06 + 0.18 = 0.24
𝑝𝑈1 (2) = 𝑃 (𝑈1 = 2) = 𝑃 (𝑋1 = 0, 𝑌1 = 2) + 𝑃 (𝑋1 = 1, 𝑌1 = 1) + 𝑃 (𝑋1 = 2, 𝑌1 = 0) = 0.18 +
0.0675 + 0.06 = 0.3075
𝑝𝑈1 (3) = 𝑃 (𝑈1 = 3) = 𝑃 (𝑋1 = 1, 𝑌1 = 2) + 𝑃 (𝑋1 = 2, 𝑌1 = 1)0.2025 + 0.0225 = 0.225
𝑝𝑈1 (4) = 𝑃 (𝑈1 = 4) = 𝑃 (𝑋1 = 2, 𝑌1 = 2) = 0.0675

This is different from the pmf of 𝑈 , despite the equality of the marginal distributions of
the components of 𝑈1 and 𝑈 . Thus, the joint distributions is important in determining
the distribution of the sum (or any multi-variable function) of random variables.
e. Let’s compute the following:

𝐸[𝑋] = ∑ possibility × probability

𝑥∈𝑋
2
= ∑ 𝑥𝑃 (𝑋 = 𝑥)
𝑥=0
= (0 × 0.4) + (1 × 0.45) + (2 × 0.15) = 0.75

𝐸[𝑌 ] = ∑ possibility × probability

𝑦∈𝑌
2
= ∑ 𝑦𝑃 (𝑌 = 𝑦)
𝑦=0

= (0 × 0.4) + (1 × 0.15) + (2 × 0.45) = 1.05

2
 2
𝐸[𝑌 2 ] = ∑ 𝑦2 𝑃 (𝑌 = 𝑦)
𝑦=0

= (02 × 0.4) + (12 × 0.15) + (22 × 0.45) = 1.95

2
𝑉 [𝑌 ] = 𝐸[𝑌 2 ] − [𝐸[𝑌 ]]
= 1.95 − 1.052 = 0.8475
2
f. We note that 𝐸(𝑋|𝑌 = 𝑦) = ∑𝑥=0 𝑥𝑃 (𝑋 = 𝑥|𝑦) where 𝑃 (𝑋 = 𝑥|𝑦) is the
conditional pmf of 𝑋 given 𝑌 = 𝑦 which we can calculate for all possible pairs
(𝑥, 𝑦) as:

𝑃 (𝑋 = 0, 𝑌 = 0) 0.1
𝑃 (𝑋 = 0|𝑌 = 0) = = = 0.25
𝑃 (𝑌 = 0) 0.4
𝑃 (𝑋 = 1, 𝑌 = 0) 0.25
𝑃 (𝑋 = 1|𝑌 = 0) = = = 0.625
𝑃 (𝑌 = 0) 0.4
𝑃 (𝑋 = 2, 𝑌 = 0) 0.05
𝑃 (𝑋 = 2|𝑌 = 0) = = = 0.125
𝑃 (𝑌 = 0) 0.4

𝑃 (𝑋 = 0, 𝑌 = 1) 0.1
𝑃 (𝑋 = 0|𝑌 = 1) = = = 0.667
𝑃 (𝑌 = 1) 0.1
𝑃 (𝑋 = 1, 𝑌 = 1) 0
𝑃 (𝑋 = 1|𝑌 = 1) = = =0
𝑃 (𝑌 = 1) 0.15
𝑃 (𝑋 = 2, 𝑌 = 1) 0.05
𝑃 (𝑋 = 2|𝑌 = 1) = = = 0.333
𝑃 (𝑌 = 1) 0.15

𝑃 (𝑋 = 0, 𝑌 = 2) 0.2
𝑃 (𝑋 = 0|𝑌 = 2) = = = 0.444
𝑃 (𝑌 = 2) 0.45
𝑃 (𝑋 = 1, 𝑌 = 2) 0.2
𝑃 (𝑋 = 1|𝑌 = 2) = = = 0.444
𝑃 (𝑌 = 2) 0.45
𝑃 (𝑋 = 2, 𝑌 = 2) 0.05
𝑃 (𝑋 = 2|𝑌 = 2) = = = 0.111
𝑃 (𝑌 = 2) 0.45

3
 𝐸(𝑋|𝑌 = 0) = (0 × 0.25) + (1 × 0.625) + (2 × 0.125) = 0.875
𝐸(𝑋|𝑌 = 1) = (0 × 0.667) + (1 × 0) + (2 × 0.333) = 0.667
𝐸(𝑋|𝑌 = 2) = (0 × 0.444) + (1 × 0.444) + (2 × 0.111) = 0.667

We can see:

2
𝐸 [𝐸[𝑋|𝑌 ]] = ∑ 𝐸[𝑋|𝑌 = 𝑦]𝑃 (𝑌 = 𝑦)
𝑦=0

= (0.875 × 0.4) + (0.667 × 0.15) + (0.667 × 0.45) = 0.75

= 𝐸[𝑋]

4
 2. We calculate the moment genrating function as follows:

∞
1 1
𝑀𝑋 (𝑡) = 𝐸[exp(𝑋𝑡)] = ∫ exp(𝑡𝑥) √ exp {− 2 (𝑥 − 𝜇)2 } 𝑑𝑥
−∞ 𝜎 2𝜋 2𝜎
∞
1 1
= ∫ √ exp {− 2 (𝑥2 − 2𝜇𝑥 + 𝜇2 − 2𝜎2 𝑡𝑥)} 𝑑𝑥
−∞ 𝜎 2𝜋 2𝜎
∞
1 1
= ∫ √ exp {− 2 (𝑥2 − 2(𝜇 + 𝜎2 𝑡)𝑥 + 𝜇2 )} 𝑑𝑥
−∞ 𝜎 2𝜋 2𝜎
∞
1 1
= ∫ √ exp {− 2 (𝑥2 − 2(𝜇 + 𝜎2 𝑡)𝑥 + 𝜇2 + (𝜇 + 𝜎2 𝑡)2 − (𝜇 + 𝜎2 𝑡)2 )} 𝑑𝑥
−∞ 𝜎 2𝜋 2𝜎
∞
1 1
= ∫ √ exp {− 2 (𝑥2 − 2(𝜇 + 𝜎2 𝑡)𝑥 + (𝜇 + 𝜎2 𝑡)2 − 2𝜇𝜎2 𝑡 − 𝜎4 𝑡2 )} 𝑑𝑥
−∞ 𝜎 2𝜋 2𝜎
∞
1 2 4 2 1 1
= exp {− 2 (−2𝜇𝜎 𝑡 − 𝜎 𝑡 )} ∫ √ exp {− 2 (𝑥2 − 2(𝜇 + 𝜎2 𝑡)𝑥 + (𝜇 + 𝜎2 𝑡)2 )} 𝑑𝑥
2𝜎 −∞ 𝜎 2𝜋 2𝜎
∞
1 1 1 2
= exp {− 2 (−2𝜇𝜎2 𝑡 − 𝜎4 𝑡2 )} ∫ √ exp {− 2 (𝑥 − (𝜇 + 𝜎2 𝑡)) } 𝑑𝑥
2𝜎 −∞ 𝜎 2𝜋 2𝜎
1
= exp {− 2 (−2𝜇𝜎2 𝑡 − 𝜎4 𝑡2 )} × 1
2𝜎
1
= exp {(𝜇𝑡 + 𝜎2 𝑡2 )}
2

5
 √ 𝑑𝑥
3. First, we note that 𝑌 = 𝑋 2 implies that 𝑋 = 𝑌 and 𝑑𝑦
= 0.5𝑌 −1/2 . So, using
the change of variable formula, we have (for 𝑦 > 0):
a.

√ 𝑑𝑥
𝑓𝑌 (𝑦) = 𝑓𝑋 (𝑥 = 𝑦)∣ ∣
𝑑𝑦
2 1 √
= √ exp {− ( 𝑦)2 } 0.5𝑦−1/2
𝜋 2
1 1
= √ exp {− 𝑦}
2𝜋𝑦 2

√
b. Note: Γ(1/2) = 𝜋, so we have:

1 1 1
𝑓𝑌 (𝑦) = √ exp {− 𝑦}
Γ(1/2) 2𝑦 2
1 1
= 1/2 𝑦1/2−1 exp {− 𝑦}
2 Γ(1/2) 2

This is a 𝜒2 distribution with 1 degree of freedom.

• Note: since the square of a standard normal random variable has a chi-squared
distribution with 1 degree of freedom, 𝑍 2 = |𝑍|2 , we see that 𝑋 = |𝑍|. Thus, the
name of the distribution of 𝑋 arises from the fact that it is the absolute value of a
standard normal random variable or, more colorfully, 𝑋 has a normal distribution
which is “folded” over at the origin.