LESSON 2 (3hrs)

INFERENCE ON THE MEAN VECTOR  WHEN  IS KNOWN

2.1 Introduction
In this Lesson we will consider inference on the mean vector  when  is known
Key terms
• Multivariate normal random vector: a vector whose probability density
function follows multivariate normal distribution
2.2 Lesson Learning Outcomes
By the end of this Lesson, you shall be able to:
2.2.1 Estimate the mean vector and the dispersion matrix from N p (  , )
2.2.2 Test hypothesis on the Mean Vector  when the dispersion matrix  is
known
2.2.3 Test for difference between two mean vectors when the common variance-
covariance matrix  is known

1.2.1. Estimation of the Mean Vector and the Dispersion Matrix from N p (  , )
Suppose a random vector X = ( X1 , X 2 , ..., X p ) N p ( , ) where  is a vector of
constants and  is a symmetric positive definite matrix of constants. Suppose that a
random sample of size n is taken from this population. Then the observation matrix(or
data matrix) is given by

 x11 x12 ... x1n 

 
 x21 x22 ... x2 n 
X = . . ... .  =  X1 X 2 ... X n 
 
 . . ... . 
x xp2 ... x pn 
 p1

Where

 X1 j 
 
X2 j 
X j = .  , j = 1, 2, ..., n .
 
. 
 
 X pj 
The sample mean vector is defined by

1
  X1 
 
X2  n
X = .  , where X i = X
1
is the mean of the observation of the i − th component of
  n j =1
ij

. 
 
 X p 
X.

X
1
We note that X = j .
n j =1

We define the sample dispersion matrix to be the matrix of sample variances and
covariances of X1 , X 2 , ..., X p .

That is

 s11 s12 ... s1 p 

   n n

 s21 s22 ... s2 p  
1 n
 X X ik jk 
S = .

. ... .  where sij =  X ik X jk −
 n  k =1
 k =1
n
k =1  , i, j = 1, 2, ..., p

 . . ... .   
s sp2 ... s pp   
 p1

So that we can write S as

(X
1
S= j − X )( X j − X ) which is a symmetric positive definite matrix. It can be shown
n j =1

that the maximum likelihood estimators(M.L.E.’s) of  and  are respectively

ˆ = X and ˆ = S .

Sampling Distribution of X when  is Known

If X 1 X 2 ... X n are iid N p (  , ) then it can be shown that the sample mean vector
1
X N p (  , ) .
n
We know that ( X −  )−1 ( X −  )  2p
So that

n( X −  )−1 ( X −  )  p2

2
2.2.2 Testing for the Mean Vector  when  is Known
Let X 1 X 2 ... X n be iid N p (  , )
Suppose we wish to test

H 0 :  = 0 against H a :   0 (  0 is a specified vector and  is specified)

at  level of significance. Then under H 0 the statistic

Q = n( X − 0 )−1 ( X − 0 )  p2

The critical region of the test is given by

n( X − 0 )−1 ( X − 0 )   2p, 

Where  2p, is such that pr (  p2   p2,  ) = 

Thus the test is to compute Q = n( X − 0 )−1 ( X − 0 ) and reject H 0 at  level whenever
Q   2p, 
In two-dimensional space Q is an ellipse and for p  2 Q is an ellipsoid.
The confidence set for  is given by the ellipsoid

C = { : n( X −  )−1 ( X −  )   2p,  }

That is,

pr {   C} = 1 − 

Compare this to Confidence Interval for  in 1-dimensional case. i.e.

 
C = { : X − z 2    X + z 2 }
n n

Example 2.1: In Example 1.1 in Lesson 1 suppose that we assume that thorax length( X 1 )
and elytra length ( X 2 ) have a joint bivariate normal distribution with mean vector  and
known variance-covariance matrix

 300 225 
= 
 225 350 
The sample mean vector was obtained as

X = [194.9 263.4] . Test at  = 0.05 level

3
 H 0 :  = 0 against H a :   0

Where 0 = [195 262] . Give a 95% confidence set.

Solution: Here n = 10

Q = n( X − 0 )−1 ( X − 0 ) 22 under H 0

Now

1  350 −225 
 −1 =  
54375  −225 300 

So that

10  350 −225  −0.1

Q= ( −0.1 1.4 )   
54375  −225 300  1.4 

10  654.5
= =0.12
54375

Now

2,2 0.05 = 5.99 . Since Q  2,2 0.05 we fail to reject H 0

The 95% confidence set is given by

C = { :10( X −  )−1 ( X −  )  5.99} i.e. all vectors  that fall inside the ellipsoid

10
C = { : [350( X1 − 1 )2 + 300( X 2 − 2 )2 − 450( X1 − 1 )( X 2 − 2 )  5.99}
54375

2.2.3. Testing for difference between two mean vectors when the common variance-
covariance matrix  is known

Suppose X 1 X 2 ... X n are iid N p ( 1 , )

Y1 Y2 ... Ym are iid N p ( 2 , )
Where X and Y samples are assumed independent.
Then

4
 n

X
1 1 n
X= j N p ( 1 , ) , Y = 1 Y j N p ( 2 ,
1
)
n j =1 n m j =1 m

Let Z = X − Y , then

E(Z ) = E( X − Y )
= 1 − 2
and
var(Z ) = var( X ) + var(Y ))

1 1 m+n
= + = 
n m mn

Since Z = X − Y is a linear combination of normal vectors it is also normal.

That is

m+n
Z N p ( , ) where  = 1 − 2
mn

We then have

mn
(Z −  )−1 ( Z −  )  2p
m+n

To test

H 0 : 1 = 2 against H a : 1  2  H 0 : 1 − 2 = 0 against H a : 1 − 2  0

The test is to compute the value of the test statistic

mn
Q= Z −1Z
m+n

mn
= ( X − Y )−1 ( X − Y )  p2 under H 0
m+n

and reject H 0 at  level whenever the computed value Qc   p2,  (tabular value)
the 100(1 −  )% confidence set for  = 1 − 2 is given by

5
 mn
C = {1 − 2 : (Z −  )−1 (Z −  )   p2,  }
m+n

Example 2.2:

Let X1 , X 2 , ..., X 10 be iid N2 ( 1 , )

Y1 , Y2 , ..., Y12 be iid N 2 ( 2 , )

 1 0.8 
Where  =   is the known common dispersion matrix.
 0.8 2 

Given the sample mean vectors are X = [99 127] , Y = [105 119] respectively

Test H 0 : 1 = 2 against H a : 1  2
at  = 0.05 level and give 95% confidence set.

Solution:
Here we wish to test H 0 : 1 − 2 = 0 against H a : 1 − 2  0

mn
We have Q = ( X − Y )−1 ( X − Y )  22 under H 0
m+n

Here m = 12 , n = 10 , p = 2 so that we get the computed value of Q as

(12)(10)  1   2 −0.8  −6 
Qc =   (−6, 8)   
12 + 10  1.36   −0.8 1  8 

 120  212.8 
=   =853.48
 22  1.36 

Comparing this to 2,2 0.05 = 5.99 we have Qc  2,2 0.05 we reject H 0

The 95% confidence set for 1 − 2 is given by

120
C = {1 − 2 : [2(6 + 1 )2 + (8 − 2 )2 + (1.6)(6 + 1 )(8 − 2 )]  5.99}
22

2.4 Assessment Questions

6
2.4.1 . Write down the pgf of the multivariate normal distribution
2.4.2 . What is the distribution of the quadratic form in the exponent of the pdf of p-
variate normal distribution? How is this fact used is testing a hypothesis on the mean
vector of the p-variate normal distribution?

2.5 Exercise
2.5.1 A random sample of size n=10 is obtained from a N 2 (  , ) distribution where

 4 4.2 
 =  
 4 .2 9 
To test the null hypothesis H 0 :  = (8, 7) ' against the alternative H1 :   (8, 7) '
the sample mean vector is found to be X = (8.8, 7.2) '
a) Carry out the test at  = 0.01 level of significance.
b) Give a 95% confidence region for the mean vector 

2.5.2 If random sample of size n=101 is obtained from a N 2 (  , ) distribution where

the dispersion matrix is

 2 1
= 
 1 1
To test the null hypothesis H 0 :  = (5.0 , 3.5) ' against the alternative
H 1 :   (5.0 , 3.5) the sample mean vector is found to be X = (5.5., 3.4) Carry out
' '

the test at  = 0.05 level of significance.

Summary
In this Lesson we have considered inference on the mean vector  when  is known. In
particular we have:
1. Estimated the mean vector and the dispersion matrix from N p (  , )
2. Tested hypothesis on the Mean Vector  when the dispersion matrix  is known
3. Tested for difference between two mean vectors when the common variance-
covariance matrix  is known

7
 References

1. Manly, B.F.J.(2004). Multivariate Statistical Methods: A Primer, 3rd Edition. Chapman&

Hall/HRC. ISBN-1584884142, ISBN-13: 978-1583883149.
2. Morrison, D. F. (2004). Multivariate Statistical Methods; 4th Edition; McGraw Hill; ISBN:
07-043185.
3. Krzanowski, W. J. (2000). Principles of Multivariate Analysis; 2nd Edition; Oxford
University Press; ISBN: 0198507089, 97801198507086.
4. https://www.worldcat.org/title/applied-multivariate-analysis/oclc/1035710263