Chap 5.2

Result 5.
1: Let X , X , …, X be a random vectors from N (μ, Σ)
population, then the test statistic of H0:μ = u v.s. H1:μ ≠ u
/
for the L.R.T. is Λ = (1 + ) .
Pf:
Let the (p+1)*(p+1) matrix A be defined by
(X − X)(X − X)′ √n(X − μ ) A A

A= =
A A
√n X − μ ′ −1
By exercise 4.8, |A| = |A | ∗ |A −A A A |
= |A | ∗ |A −A A A |
From which we obtain
(-1) ∑ (X − X X − X + n X − μ X−μ |
=∑ (X − X X − X | ∗ {−1 − n X − μ ∑ (X −
X X −X ] X−μ }
Since ∑ X −μ X −μ ′=∑ X −X X −X ′+
n X−μ X−μ , the equality above involving
determinants can be written as
(-1) | ∑ X −μ X −μ ′|
T
= X − X X − X ′ ∗ (−1)(1 + )
n−1
or nΣ = |nΣ| ∗ (1 + ).
/ | |
Thus Λ = = (1 + ) .
Generalized Likelihood Ratio Test:
Let θ be a vector consisting of all the unknow parameters, and
let L(θ) be the likelihood function obtained by evaluating the
joint density of X , X , …, X at their observed values
x , x , …, x .
The parameter vector θ takes its values in the parameter set Θ.
Ex: for a p-dim. Normal case,
θ = (μ , μ , … , μ , σ , σ ,…,σ ,σ ,…,σ , …,σ )′ and Θ
consists of the union of the p-dim. space. where -∞<μ <∞, …,

( )
-∞<μ <∞, and the -dim. space of variance and cov. such
( )
that Σ is a p.d. matrix. Therefore, Θ has dim. v = p + .
Under the null hypo. H0:μ = u , we have
θ ={μ =μ ,…, μ =μ ; σ , …,σ , …,σ with p.d. Σ}, so

( )
Θ has dim v = .
H0:θ = θ  Θ
H1:θ ≠ θ  Θ
Θ ∪ Θ = Θ  parameter space can be divided by 2 parts.
A likelihood ratio test of H0:θ ∈ Θ rejects H in favor of
∈ ( )
H0:θ ∉ Θ if Λ = < .
( )
Result 5.2: When the simple size n is large, the function
−2 ln(∧) ∼ χ ( ). Here the degrees of freedom are
v − v =(dim. of Θ) – (dim. of Θ ).
5.4 Confidence Region and Simultaneous Comparisons of
Component Means
Let θ be a vector of unknown population parameters and Θ
the set of all possible values of θ. A confidence region is
determined by the data and denoted by R(X), where
X=[X , X , …, X ] is the data matrix. The region R(X) is said to
be a 100(1-α)% confidence region of data matrix X if

Pr[ R(X) will cover the true θ ] =1-α before the sample is
collected. This prob. is calculated under the true, but unknown
value of θ.
The confidence region before the sample is selected for the
mean μ of a p-dimension Normal population is to satisfy
( − 1)
Pr{n(X − ) (X − ) ≤ , ( )} = 1 −
−
whatever the values of the unknown μ and Σ. In other words,
( ) /
X will be within ( , )) of with probability
1 − α provided distance is defined in terms of ( ) .
A 100(1-α)% C.R. for the mean of p-dim. normal population is
the set determined by all such that
( − 1)
(x − )′ (x − ) ≤ , ( )
−
where x and are the sample mean vector and sample
covariance matrix.
Let X~ N μ, Σ and form the L.C.
Z = ℓ X +⋯+ℓ X = ℓ X
From above, we have μ = ℓ μ and σ = ℓ Σℓ. Moreover,
Z~N(ℓ μ, ℓ Σℓ).
If a random sample from the N μ, Σ is available, a
corresponding sample of z’s can be created by taking
z = ℓ x + ⋯ + ℓ x = ℓ x j = 1,2, … , n
The sample mean and variance of this sample are z = ℓ x and
s = ℓ Sℓ , where x and S are the sample mean vector and
variance matrix of the original data X respectively. For ℓ
fixed and σ unknown, a 100(1-α)% C.I. for μ is based on
z−μ √n(ℓ x − ℓ μ)
t= = ~t
s /√ ℓ Sℓ
and leads to the statement
s α s α
z− t ≤μ ≤z+ t ( )
√n 2 √n 2
or
ℓ Sℓ α ℓ Sℓ α
ℓ x− t ≤ℓ μ≤ℓ x+ t ( )
√n 2 √n 2
Above inequality can be interpreted as a statement about the
components of the mean vector μ. For example, ℓ μ = μ
with ℓ = (1,0,0, … ,0), it becomes the usual C.I. for a normal
population mean. We could make several confidence statements

about the components of μ each with associated different
vector ℓ . However, the confidence associated with all of the
statements together is NOT 1-α. Intuitively, it would be
desirable to associate a “collective” confidence coefficient of
1-α with the confidence intervals that can be generated by all
choices of ℓ . The method is to find a wider interval than the
above t-interval for ℓ μ for a specific choice of ℓ.

Chap 5.2

Uploaded by

Copyright:

Available Formats

Chap 5.2

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Chap 5.2

Uploaded by

Copyright:

Available Formats

Result 5.

1: Let X , X , …, X be a random vectors from N (μ, Σ)

population, then the test statistic of H0:μ = u v.s. H1:μ ≠ u

Let the (p+1)*(p+1) matrix A be defined by

(X − X)(X − X)′ √n(X − μ ) A A

By exercise 4.8, |A| = |A | ∗ |A −A A A |

From which we obtain

n X−μ X−μ , the equality above involving

determinants can be written as

Generalized Likelihood Ratio Test:

Let θ be a vector consisting of all the unknow parameters, and

let L(θ) be the likelihood function obtained by evaluating the

joint density of X , X , …, X at their observed values

The parameter vector θ takes its values in the parameter set Θ.

Ex: for a p-dim. Normal case,

θ = (μ , μ , … , μ , σ , σ ,…,σ ,σ ,…,σ , …,σ )′ and Θ

consists of the union of the p-dim. space. where -∞<μ <∞, …,

Under the null hypo. H0:μ = u , we have

θ ={μ =μ ,…, μ =μ ; σ , …,σ , …,σ with p.d. Σ}, so

Θ ∪ Θ = Θ  parameter space can be divided by 2 parts.

A likelihood ratio test of H0:θ ∈ Θ rejects H in favor of

Result 5.2: When the simple size n is large, the function

−2 ln(∧) ∼ χ ( ). Here the degrees of freedom are

5.4 Confidence Region and Simultaneous Comparisons of

Let θ be a vector of unknown population parameters and Θ

the set of all possible values of θ. A confidence region is

determined by the data and denoted by R(X), where

X=[X , X , …, X ] is the data matrix. The region R(X) is said to

be a 100(1-α)% confidence region of data matrix X if

collected. This prob. is calculated under the true, but unknown

The confidence region before the sample is selected for the

mean μ of a p-dimension Normal population is to satisfy

1 − α provided distance is defined in terms of ( ) .

A 100(1-α)% C.R. for the mean of p-dim. normal population is

the set determined by all such that

where x and are the sample mean vector and sample

Let X~ N μ, Σ and form the L.C.

If a random sample from the N μ, Σ is available, a

corresponding sample of z’s can be created by taking

The sample mean and variance of this sample are z = ℓ x and

s = ℓ Sℓ , where x and S are the sample mean vector and

variance matrix of the original data X respectively. For ℓ

fixed and σ unknown, a 100(1-α)% C.I. for μ is based on

components of the mean vector μ. For example, ℓ μ = μ

with ℓ = (1,0,0, … ,0), it becomes the usual C.I. for a normal

population mean. We could make several confidence statements

vector ℓ . However, the confidence associated with all of the

statements together is NOT 1-α. Intuitively, it would be

desirable to associate a “collective” confidence coefficient of

1-α with the confidence intervals that can be generated by all

choices of ℓ . The method is to find a wider interval than the

above t-interval for ℓ μ for a specific choice of ℓ.

You might also like