STAT 741 TAKEHOME EXAM
Hello, dear friend, you can consult us at any time if you have any questions, add WeChat: daixieit
STAT 741 TAKEHOME EXAM
Spring 2023
Problem 1
Consider the vector of measurements Yi = (yi1 . yi2 . 口口口. yini)\ taken on subject i, and assume the following model,
Yi = Xi8 + Zi bi + ∈i . (1)
where Xi and Zi are nonrandom ni × p and ni × g design or covariate ma- trices, respectively, 8 is a p–dimensional parameter of fixed effects common to all subjects, bi is g–dimensional subject–specific (nonrandom) parame- ter considered nuisance, and ∈i ∼ Nni(0. g2 I) is the ni –dimensional vector representing noise or measurement error. Notice that the covariance matrix g I2 = g I2ni is ni × ni and its elements do not depend on i. Denote by fi (yi |bi . 8 . g2 ) the density of Yi .
1. Show that ZYi is sufficient for bi , and obtain the conditional pdf fi (yi |Z y i . 8 . g2 ).
2. Argue that one may try to get inference about 8 . g2 is by maximizing the conditional likelihood fi (yi |Z y i . 8 . g2 ) but that this may not be a good idea.
3. Assume that the bi are random, each having the same distribution o(·). Explain how to get inference about 8 . g2 and o.
4. A simpler alternative assumes that the random effects are normally distributed in which case model (1) is referred to as linear mixed– effects model, where Xi , Zi , and 8 are as above, but bi ∼ Nq (0, D), and ∈i ∼ Nni(0. Σ) is the ni –dimensional error. Here Σ = Σi is ni × ni covariance matrix whose components do not depend on i. It is assumed that the ∈i and bi are independent. Obtain inference about all the
parameters, and show how to test H0 : C8 = 80 versus H0 : C8 80 .
Problem 2
Let x ∼ b(n. T), and consider the distribution of x|x > 0. x is said to have a truncated binomial distribution.
1. Argue that the distribution of x|x > 0 is a tilted or distorted distribu- tion, and obtain its form.
2. Data were collected on the number of girls and boys in 28 families in the form of pairs (No. girls, No. boys):
10 10 11 11 11 11 11 11 11 11 11 20 20 20 21 21 21 21 12 12 30 31 31
13 13 40 41 14
Since there is at least one girl in each family, do the data suggest the number of girls follow a truncated binomial distribution x|x > 0, where x ∼ b(n. 1/2)?
Problem 3
Explain how the GLM theory works for binomial (not Bernoulli) data using a general link. Then appply your results to any suitable data set of your choice using the canonical link.
2023-05-05