Scribd
Upload
7 views3 pages

Tut4 5

This document covers key concepts in multivariate statistics, including random vectors, mean vectors, covariance matrices, and the multivariate normal distribution. It also discusses the likelihood function, maximum likelihood estimation (MLE), and the likelihood ratio test. Additionally, it includes exercises related to calculating sample means, covariance matrices, and conducting statistical tests using provided data.

Uploaded by

Chan Hufflepuff
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
7 views3 pages

Tut4 5

This document covers key concepts in multivariate statistics, including random vectors, mean vectors, covariance matrices, and the multivariate normal distribution. It also discusses the likelihood function, maximum likelihood estimation (MLE), and the likelihood ratio test. Additionally, it includes exercises related to calculating sample means, covariance matrices, and conducting statistical tests using provided data.

Uploaded by

Chan Hufflepuff
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

STAT3613

Tutorial 4.5 2024/10/29

Review
Multivariate Statistics

• Random vector
○ The elements of a vector are random variables

= ⋮

○ y1, …, yn can be dependent or independent

• Mean vector
○ The elements of the vector are the expected value of the elements of the random
vector
( )
= ( )= ⋮
( )

• Covariance matrix
○ X an m  1 random vector and Y an n  1 random vector
○ Variance-covariance matrix / covariance matrix is defined as

 Cov X, Y   E X  μ X Y  μ Y T 
○ Properties
 Cov(X,Y) = (ij) where ij = Cov(xi,yj)
 Dimensions of Cov(X,Y) = m  n
 Cov X, Y   E XY T   E X E Y T 
 
 Σ  Var Y   E Y  μ Y Y  μ Y T  E YY T   E Y E Y T 
○ Correlation
, = =

○ Correlation matrix
=
⬚ 0
= diag( ) = ⬚ ⬚
0 ⬚

• Linear combinations
○ Let A and B are constant matrices, b and c are constant vectors and X and Y are
two n  1 random vectors with Var(Y) = , then
 E(AY + B X + b) = AE(Y) + BE(X) + b
 Cov(AY + b,BY + c)=AB’

1
 Var(AY + b)= AA’

• Multivariate normal distribution


○ When Yp1 follows a multivariate normal distribution, Y ~ N (  , ), the
probability density function of Y is
1  1 
 Y   exp  Y  μ  Σ 1 Y  μ 
T

2  Σ
p/2 1/ 2
 2 
Then
 E(Y) = , Var(Y) =  = (ij)
 E(yi) = i, Var(yi) = ii, Cov(yi yj) = ij
 yi ~ N (i,ii)

○ For a sample of size n, y1, …, yn, the likelihood function and log-likelihood
function are
1  1 
Lμ, Σ | y1 ,, y n   i 1 y i   i 1 exp  y i  μ  Σ 1 y i  μ 
n n T

2  Σ
p/2 1/ 2
 2 
n n 1 n
  log L   log 2   log Σ   y i  μ  Σ 1 y i  μ 
p T

2 2 2 i 1

○ MLE
 = = ∑ = ~ ,
 = ∑ ( − )( − ) = ( − ) ( − )

○ For constant matrix A and constant vector c,


AY + c ~ N (A+ c, AA’)

○ Let Y partition into Y1 and Y2


Y   μ   Σ Σ12  
Y   1  ~ N   1 ,  11  
Y
 2 μ Σ
  2   21 Σ 22  

 11 = Var(Y1), 22 = Var(Y2), 12 = Cov(Y1,Y2)


 Y1 and Y2 independent iff 12 = 0
 ( | )= + ( − )
 ( | )= −

• Likelihood ratio test


○ H0 vs H1 (H0  H1)

−2 ~

 (asymptotically)
 v = number of free parameters under H1
 v0 = number of free parameters under H0

2
Exercises
• Consider variables YrHgt (inches), FtFrBody (pounds), PrctFFB (percentages),
BkFat (inches), SaleHt (inches) and SalseWt (pounds) in "TUT402.csv".
a) Calculate the sample means of the variables.
b) Calculate the sample covariance matrix.
c) Calculate the sample correlation matrix.
d) Assuming multivariate normality of the 6 variables, find MLE for
([ , , ]|[ , , ] = [45,1000,80])
e) Calculate the sample means and covariance matrix of the variables when they are
measured in (feet, pounds, percentages, feet, inches, pounds)
f) Calculate the maximum log-likelihood under multivariate normality.
g) Calculate the maximum log-likelihood under multivariate normality and the mean
= [55,1200,68,0.20,50,1200].
h) Conduct a likelihood ratio test for : = [55,1200,68,0.20,50,1200] at
5%.

You might also like