You are on page 1of 6

Given a data set X , X , … , X and a particular ℓ, the

confidence t-interval in ℓ μ is the set of ℓ μ values for which


√ ℓ ℓ
|t| = | |≤t ( )
ℓ ℓ

or equivalently

ℓ ℓ ℓ( )
t = = ≤t ( )
ℓ ℓ ℓ ℓ

It seems reasonable to expect that the constant t ( ) is above

equality be replaced by a large value c , when statements are

developed for many choices of ℓ. Considering the values of ℓ

for which t ≤ c , we are natually led to the determinant of

ℓ( )
ℓ t = ℓ ℓ ℓ

Using the maximization in (2-50) with x = ℓ, d = (X − μ) and

B=S, we get

n ℓ X−μ ℓ X−μ
ℓ =n ℓ
ℓ Sℓ ℓ Sℓ

= n X−μ S X−μ = T

with the maximum occurring for ℓ proportional to S X−μ .


Result 5.3 : Let X , X , … , X be a random sample from

N(μ, Σ) with Σ positive definite. Then, simultaneously for all ℓ,

the interval
( ) ℓ ℓ / ( ) ℓ ℓ /
( ℓ X − [ F , (α) ] , ℓ X + [ F , (α) ] )

will contain ℓ μ with probability 1-α.

Pf.: Because T = n X − μ S X − μ ≤ c , it also


implies that ≤c for every ℓ.
ℓ ℓ
(n−1)p
Choosing c = Fp,n−p (α) gives interval that will contain ℓ μ for
n−p

all ℓ with probability 1-α ≤ pr(T ≤ c ).

Above intervals are called T −intervals. If we choose

ℓ = (1,0, … ,0) and so on through ℓ = (0, … ,0,1), the

T −intervals allow us to conclude

(n − 1)p s (n − 1)p s
x − F , (α) ≤μ ≤x + F , (α)
n−p n n−p n

(n − 1)p s (n − 1)p s
x − F , (α) ≤μ ≤x + F , (α)
n−p n n−p n

all hold simultaneously with confidence coefficient 1-α. We can

also make statements about the difference μ − μ by choosing

ℓ = (0, … ,0,1,0, . . ,0,1,0, … ,0).


In this case, ℓ′ Sℓ = s − 2s + s and the two end-points of

the statement are

(n−1)p
((x − x ) ± Fp,n−p (α) )
n−p n

See example 5.4.

The Bonferroni’s Method of Multiple Comparisons

Suppose prior to the collection of data, confidence statements

about linear combinations ℓ ′μ, ℓ ′μ,…, ℓ ′μ are required for

small m. Let c denote a confidence statement about the value

of ℓ ′μ with Pr(c true) = 1 − α , i = 1,2, … , m.

Now, Pr(all c true)=1-Pr(at least one c false)

≥1−∑ pr( c false)

= 1− (1 − pr(c true))

= 1 − (α + α +. . . +α ) (see example 5.6)

Above inequality is a special case of the Bonferroni’s inequality.

Bonferroni inequality: Pr(A ∪ A ∪ … ∪ A ) ≤ ∑ pr(A ).


Let us develop simultaneous interval estimates for the

components μ of μ. We consider the individual t-intervals

x ± t ( ) , i=1,2,…,m
n

with α = .

Since pr x ± t contains μ = 1 − , i=1,…,m.


n

We have pr x ± t contains μ , all i


n

α α α
≥ 1−( + + ⋯+ ) = 1 − α
m m m

Therefore, with an overall confidence level greater than or equal

to 1 − α, we make the m=p statements

α s α s
x − tn−1 ≤ μ ≤ x + tn−1
2p n 2p n

α s α s
x − tn−1 ≤ μ ≤ x + tn−1
2p n 2p n

Ex.: Let p=3, n=20, and α = 0.05.

We construct the simultaneous 95% Bonforroni C.I. for


.
μ , μ and μ corresponding to the choice α = , i=1,2,3
s 2.879
x − t19(0.0083) ≤ μ ≤ 4.64 + 2.625
n 20

s 199.798
x − t 19 (0.0083) ≤ μ ≤ 45.4 + 2.625
n 20

s 3.628
x − t19 (0.0083) n ≤ μ ≤ 9.965 + 2.625 20

5.5

Result 5.4 : Let X , X , … , X be a random sample from a

population with μ and positive de inite Σ. When n-p is large,

then H : μ = μ is rejected in favor of H : μ ≠ μ at a

significant α if

n X − μ S X−μ > χ (α)

Result 5.5 : Let X , X , … , X be a random sample from a

population with μ and positive de inite Σ. If n-p is large, then

ℓ ℓ
ℓ′x ± χ (α) will contain ℓ′μ for every ℓ with

probability approximately 1 − α.
Midterm
7 5 9 6 5 4
1. Let the Normal data matrix X ∗ = 68 70 72 69 66 69 and α =
80 90 95 90 93 92
0.05.

(1) Find the deviation vectors d , d and d .

(2) Determine the matrix S from (1).


X +X
(3) Find the sample covariance matrix for Y = X .
X −X
(4) Find the value of Wilk’s lambda for this data under
H : μ = (8,65,85) v. s. H : not H .
(5) What is your conclusion for (3) by using the Hotelling T2 distribution?
(6) Determine the T2 simultaneous C.I. for μ , μ and μ .
(7) Determine the Bonferroni’s simultaneous C.I. for μ , μ and μ .
μ 7
(8) Test the hypothesis H : μ − μ = and make your conclusion.
0
(9) Find the Hotelling’s T2-value for (3).

2. Let X , … , X be a random sample from N(μ, σ ). Prove that X and s =

∑ (X − X) are independent.

You might also like