STAT4CI3/6CI3 Computational Methods for Inference
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STAT4CI3/6CI3 Computational Methods for Inference
STATISTICS 4CI3/6CI3
FALL 2022
FINAL EXAM
Q. 1 a) [10 marks]
Write an efficient R function to generate a random sample from the distribution with probability mass function
f (x) = { 0.2}x尸0
Verify that your function does indeed generate a sample from this distribution.
b) [10 marks]
Consider the following R function
rLaplace <- function(N , rate=1) 桂
if (rate <= 0) stop("rate must be positive")
Z <- -log(runif(N))/rate
X <- ifelse(runif(N)<0.5 , -Z, Z)
X
|
Prove (analytically) that this function will return a random sample from the Laplace distribution with probability density function
f (x; λ) =
e一λ|北 | for x 〕IR
where the λ is the rate parameter of the distribution.
Hint: Find P(Xi ≤ x; λ) considering x < 0 and x > 0 separately and show that derivative of this does give the required density.
Q. 2 a) [10 marks]
Euler’s constant arises in number theory and is defined as
&
γ = 尸 log(x)e一北 dx
Write an R function to estimate γ and give its standard error using Monte Carlo simulation and give the results of running the function with a Monte Carlo simulation size of N = 1, 000, 000.
b) [10 marks]
Write another function that uses antithetic variables to estimate γ and show that, for the same number of randam variables generated, the standard error of the antithetic variable estimator is much less that of the estimator in (a).
Hint: Recall that U 一 Uniform(0, 1) ~ X = 尸 log(U) 一 exponential(1).
Q. 3 Suppose that X1 , . . . , Xn is a random sample from an exponential distribution with mean µ . An approximate 95% confidence interval for µ which is commonly used in practice is given
by
┌ ┐
where
a(n) = 1 尸
and b(n) = 1 +
a) [10 marks]
Show that if X 一 exponential(µ) then Z = X/µ 一 exponential(1) and that the coverage probability of the above interval can be written as
P ╱a(n)X < µ < b(n)X、 = P ╱a(n)Z < 1 < b(n)Z、 = P ╱
< Z <
、
Comment on the relevance of the above result for a simulation study to estimate the coverage of the interval.
b) [10 marks]
Describe and implement a simulation study to assess how the coverage probability of this interval changes with n. Ensure that you give standard errors for any estimates of the coverage.
Q. 4 Suppose that X1 , . . . , Xn are a sample from a normal distribution with mean 0 and unknown standard deviation σ . We wish to conduct Bayesian inference for σ with the prior distribution that σ exponential(1). We will do this using an independence Metropolis-Hastings algorithm with candidate distribution equal to the prior for σ .
Use the following data for this question and also for Question 5.
-4.46 -4.09 -3.66 -2.82 -2.45 -2.16 -2.12 -1.51 -1.41 -1.32
-1.26 -0.97 -0.53 -0.46 -0.29 0.55 0.69 1.29 3.52 4.07
a) [5 marks]
Show that the posterior distribution for σ is of the form
π(σ } x1 , . . . , xn ) 体
exp {尸σ 尸
} σ > 0
b) [5 marks]
Suppose that σ is a candidate drawn from the prior distribution and that the current state of the chain is σt一1 . Show that the probability of moving at iteration t is equal to
ρt = min { L
t一(σ)
1 }
c) [10 marks]
Implement this independence Metropolis-Hastings algorithm to get a sample from the posterior distribution of σ and estimate the posterior mean, posterior standard deviation and an equitailed posterior 95% credible interval for σ .
Q. 5 In this question we shall do a frequentist analysis of the same data and model as in Question 4 using the bootstrap.
a) [10 marks]
Assuming the population distribution is Normal(0, σ) find the maximum likelihood estimator for σ and use a parametric bootstrap to estimate the bias and standard error of the maximum likelihood estimator. Also construct a 95% confidence interval for σ using the basic bootstrap method.
b) [10 marks]
Now suppose that the only assumption we make about the population distribution is that it has a finite standard deviation σ which we wish to estimate based on the same data. Find the plug-in estimator of σ and use the non-parametric bootstrap to estimate the bias and standard error of this estimator and again use the basic bootstrap method to construct a 95% confidence interval for σ based on the non-parametric bootstrap.
2023-01-02