关键词 > STATS310/732
STATS 310/732 Introduction to Statistical Inference SEMESTER 1, 2019
发布时间:2022-06-02
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STATS 310/732
SEMESTER 1, 2019
STATISTICS
Introduction to Statistical Inference
Foundations of Statistical Inference
1. Let X1 ,X2 have the joint probability density function
f(x1 ,x2 ) = 8x1 x2 , 0 < x1 < x2 < 1.
Let Y1 = X1 /X2 and Y2 = X2 .
(a) Find E(X1 X2 ).
[6 marks]
(b) Show that the joint probability density function of Y1 and Y2 is fY1,Y2 (y1 ,y2 ) = 8y1y2(3) ,
and state the support of the distribution.
[6 marks]
(c) Show that Y2 |Y1 = y1 has the probability density function
f2 (y2 |y1 ) = 4y2(3) ,
and state the support of the distribution.
[6 marks]
(d) Are Y1 and Y2 independent? Justify your answer.
[2 marks] [20 marks]
2. Explain in sucient detail why the weak law of large numbers does not hold for the mean of a random sample independently drawn from the Cauchy distribution which has density function
f(x) = , x ∈ R.
[5 marks]
3. Let X ∼ χν(2), i.e., the chi-square distribution with ν degrees of freedom, which has E(X) = ν and Var(X) = 2ν . Explain using the central limit theorem why
X − ν
converges in distribution to a standard normal random variable as ν → ∞ .
[5 marks]
Answer: Write X = P Xi , where Xi χν(2)0 and ν = nν0 (if ν is an integer, we can simply set ν0 = 1 and n = ν). Therefore, Var(Xi ) = 2ν0 is nite, and
Y = = =
=
to which the CLT can be applied.
X − ν
√2nν0 nX − nν0
4. Let X1 ,X2 , . . . ,Xn be a random sample drawn independently from the geometric distribution with probability function
f(x;θ) = θ(1 − θ)x , x = 0, 1, 2, . . . , 0 < θ < 1, which has mean (1 − θ)/θ and variance (1 − θ)/θ2 .
(a) Write down the likelihood and log-likelihood functions.
[4 marks]
(b) Find the maximum likelihood estimator
(c) Show that the Fisher information I(θ) = n/{θ2 (1 − θ)}.
[4 marks]
(d) Is θ an MVUE of θ? Justify your answer. [4 marks]
95% con dence interval for θ .
[4 marks] [20 marks]
5. Consider the multinomial distribution with 3 categories, where the random variables X1 ,X2 and X3 have the joint probability function
f(x;θ) = n!x1!x2!x3!θ 1(x)1 θ2(x)2 (1 − θ 1 − θ2 )x3 , x1 ,x2 ,x3 ≥ 0,
where x = (x1 ,x2 ,x3 )T , θ = (θ1 ,θ2 )T , n = x1 + x2 + x3 , θ 1 > 0, θ2 > 0 and
θ3 ≡ 1 − θ 1 − θ2 > 0.
(a) Write down expressions for the log-likelihood function ℓ(θ) and the score statis- tic vector U(x;θ).
[5 marks]
(b) Find the maximum likelihood estimator θ of θ . (You do not need to perform
(c) Show that the Fisher information is
I(θ) = n 1θ1 1θ3
1θ2 1θ3 .
[5 marks]
(d) Consider using −2 log(LR) to test the null hypothesis H0 : θ 1 = θ2 = θ3 against
[5 marks] [20 marks]
6. Let X have a distribution with density function
f(x;θ) = θxθ −1 , 0 < x < 1, θ > 0.
One is interested in testing
H0 : θ = 1 vs. H1 : θ > 1,
based on a random sample x1 , . . . ,xn independently drawn from the distribution.
(a) Write down the sample space S, the parameter space and the null parameter space 0 of the test.
[2 marks]
(b) Let n = 1 and C = {x : x > aα } be the critical region of some test at the signi cance level of α . Find the value of a0.05 .
[4 marks] (c) Let n ≥ 1. Show that the uniformly most powerful (UMP) test has the form
n
Reject H0 , if xi > bα ,
i=1
where bα is so chosen that the test has a signi cance level of α .
[4 marks] [10 marks]
8. Consider the subject of decision theory.
(a) For a decision-making problem, the following table gives R(di ,θj ), the risk of each candidate decision rule di for each state of nature θj .
θ1 θ2
d1 10 40
d2 22 36
d3 14 30
d4 26 26
d5 14 34
d6 26 30
d7 18 24
d8 30 20
Find all admissible rules.
[4 marks]
(b) Still with the table given in part (a), nd the minimax rule(s). Justify your answer.
[4 marks]
(c) If a Bayesian has a prior belief that P( = θ 1 ) = 0.3 and P( = θ2 ) = 0.7, will the Bayesian prefer d3 to d4 ? Justify your answer.
[4 marks]
(d) Show that the beta distribution is a conjugate prior to the negative binomial distribution.
Note: The Beta(α,β) and NegBinomial(k,p) distributions have PDFs given by, respectively,
f(x) = 1B(α,β)xα −1(1 − x)β −1 , 0 < x < 1, α > 0, β > 0,
where B(α,β) is the Beta function, and
f(x) = k +x(x) − 1 pk (1 − p)x , x = 0, 1, 2, . . . , k = 1, 2, . . . , 0 < p < 1.
[4 marks]
(e) Explain in sucient detail how a Bayes rule can be conveniently determined without examining the entire set of decision rules.
[4 marks] [20 marks]