Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit

MATH2831 - Assignment 2 - 2023 T3

Question 2 [20 marks] - must be submitted as a group work on Moodle

You can work on this question as a group of up to four students.  Your answers can be typed (e.g. as a Latex

or Word document) or handwritten, then converted (printed) into a pdf file and uploaded on Moodle submission link. Only one solution per group must be submitted. Please make sure that you list names of all group

members and their zIDs on the cover page of your submitted document.

For full marks, answers need to be mathematically correct and written clearly using plain English.

In the box below, list names and zIDs of all group members.

Part A. [7 marks] Consider the general linear model

y = Xβ + ε,   (1)

where as usual y is an n × 1 vector of responses, X is an n × p design matrix, β is a p × 1 vector of parameters, and ε is a vector of uncorrelated errors with zero mean and variance σ2 .

Consider the following transformation of ε:

q = β + Lε,

where L is a real p × n matrix.

A1. Determine E(q), the mean vector of q, and Var(q), the covariance matrix of q.

Give your answers as expressions with only  β , σ 2 and L (you must justify your answers).

(Hint: Covariance matrix of a random vector Z is dened as Var(Z) = E[(Z − E(Z))(Z − E(Z)) ]).

A2. Determine E(q q).

Give your answer as an expression with only β , σ 2 and L (you must justify your answer).

(Hint: you may use the formula for the expectation of quadratic forms).

A3. Assume that the error term in model (1) is multivariate normal, ε N(0, σ 2I), where I denotes the n × n identity matrix.

Determine the distribution of the vector q (you must justify your answer).

Part B. [13 marks] Consider the general linear model (1) from part A, and assume that X has a full rank. Denote by b the least squares estimator of β .

B1. Show that we can write b in the following form:

b = β + Lε,

where L = (XX)−1X .

B2. Show that the residual sum of squares for model (1) can be written as

SSres = ε (I XL)ε,

where I is n × n identity matrix.

B3. Determine

E[ε (I XL)ε].

(Hint: use the formula forE(εAε).)

B4. Apply the results from previous parts to show that the usual estimator of σ2

model (1) is given by

2 = ε (I XL)ε

n p

and then show that 2 is an unbiased estimator of σ2 .

B5. Assume that ε ∼ N(0, σ 2I). Derive the following formula for the 100(1 − α)% confidence interval for βj, j = 0, 1, . . . ,k:

bj± tα/2,n−p ,

where lj(⊤) is the jth row of martix L.

You must show all steps. You may use the result (n p)2/σ2 χn(2) −p without providing proof.