STA261H1S Term Test 1 (Wed Evening)
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STA261H1S Term Test 1 (Wed Evening)
2022
Question 1
Let’s assume that X1, X2 , ..., Xn ∼ f (x; θ), where the density is given by,
f (x; θ) = (e)se ; for x > 0
(a) [6 Marks] Derive the maximum likelihood estimator of θ . Assuming that the true value of θ is
θ0 , calculate the Fisher Information.
Solution:
L(θ) = θ−n exp (− )
ℓ(θ) = −n ln(θ) −
S(θ) = − + ⇒ θˆ =
Now the second derivative,
ℓ′′(θ) = − 2
E (ℓ′′(θ))|θ=θ0 = E ( − 2 )
n
= −
Thus, the Fisher’s information = −E(ℓ′′(θ))|θ=θ0 =
(b) [3 Marks] Show that the E(S(θ))|θ=θ0 = 0, where S(θ) is the score function from (a).
Solution:
E(S(θ))|θ=θ0 = E (− + )|θ=θ0
n nθ0
θ0 θ0(2)
n n
θ0 θ0
(c) [1 Marks] What are the mean and the variance of the asymptotic distribution of the MLE? Solution: as n → ∞ we have θˆ ∼ N (θ0 , θ0(2)/n)
Question 2
Let X1, X2 , . . . , X25 N (µ, σ1(2) = 9) and independently of X1, X2 , . . . , X25 , let Y1, Y2 , . . . , Y20 N (µ, σ1(2) = 16) where µ ∈ R is unknown. Let = ∑i(2)5(1) Xi and = ∑i(2) Yi . We consider estimating µ by T = a + (1 − a), where a ∈ (0, 1).
(a) [5 Marks] Is T an unbiased estimator of µ? If yes, prove your answer. If no, calculate the bias. Solutions: E(T) = E(a +(1− a)) = aE +(1− a)E = aµ +(1− a)µ = µ(a +1− a) = µ and so T is an unbiased estimator of µ . ■
(b) [5 Marks] Find the value of a that minimizes the variance of T (i.e. Var(T)).
Note: You don’t have to do any second derivative tests.
Solutions: V (T) = V (a + (1 − a)) inp a2 V () + (1 − a)2 V () = a2 × + (1 − a)2 × = 2a × − 2(1 − a) ×
Setting this equal to zero and solving for a
=⇒ a = = 0.6896552 ■
Question 3
Suppose we have a finite population Π and a measurement X : Π → {0, 1}, where |Π| = 20 and |{π : X(π) = 0}| = 13.
(a) [2 Marks] Determine fX (0) and fX (1).
Solutions: Here, fX (0) = 13/20 and fX (1) = 7/20.
(b) [4 Marks] For an i.i.d. sample of size 5 (this means sampling with replacement), determine the probability that 5fˆX (1) = 4.
Solution
Note that 5fˆX (1) is the number of 1’s in the sample of size 5 and under iid sampling 5fˆX (1) ∼ Bin(n = 5, θ = 7/20) and so P (5fˆX (1) = 4) = (4(5)) ( )4 (1 − )1 = 0.04877031 ■
(c) [4 Marks] For simple random sample of size 5 (without replacement), determine the probability that 5fˆX (1) = 2.
Solution
Let X be the number the number of elements with measurement 1 in a sample of size 5. Then, P (5fˆX (1) = 2) ⇒ P (fˆX (1) = 2/5). That is we need to find P (X = 2) with replacement. Implying hypergeometric rule we get,
(2(5))( 3(15))
(5(20))
■
Question 4
X1, X2 , ..., Xn are random variables from N (0, θ) distribution. Here, θ = σ 2 > 0 is the variance of the distribution, and is an unknown parameter.
(a) [4 Marks] Find the sufficient statistic for θ
Solutions: Here we have,
L(θ) = ( ) − exp (− xi(2)) = gθ(T (s), h(s))
here, h(s) = 1 and T (s) = ∑ xi(2) and thus, ∑ xi(2) is a sufficient statistic for θ .
(b) [6 Marks] Find the MLE of θ (no need to calculate the second derivative). Is the obtained MLE a consistent estimator?
Solution: Here,
L(θ) = ( ) −
exp (− xi(2))
ℓ(θ) = − log(θ) − log(2π) − xi(2)
= − + xi(2) = 0
⇒θˆ = xi(2)
Since, the mean parameter µ = 0, here and thus, V (X) = E(X2) = θ . Due to the WLLN
∑ xi(2) E(X2), and thus θˆ is a consistent estimator.
Question 5
Suppose that a statistical model is given by the family of Bernoulli (θ) distributions where θ ∈ Ω = [0, 1].
(a) [4 Marks] If the interest is in making inference about the probability that two independent observations from this model are both equal to zero, then determine ψ(θ)
Solution
ψ(θ) = P (X1 = 0, X2 = 0) = (1 − θ)2 ■
(b) [6 Marks] If {X1, X2 , . . . , Xn} are iid random variables from this model, find the maximum
likelihood estimator for the aforementioned ψ(θ) and calculate the bias in your estimator.
Solution
is the MLE of θ and this ψ(θ) a is a 1-1 function of θ in Ω = (0, 1) and so T = (1 − )2 is the MLE of ψ(θ)
ET = E((1 − )2 ) = E((1 − ))2 + V ((1 − )) = (1 − θ)2 + = ψ(θ) + =⇒ Bias(T) = ET − ψ(θ) = ■
2022-04-12