Analysis of Variance and

Design
g of Experiments
Experiments-
p -I
MODULE – II
LECTURE - 9

GENERAL LINEAR HYPOTHESIS

AND ANALYSIS OF VARIANCE
Dr. Shalabh
Department of Mathematics and Statistics
Indian Institute of Technology Kanpur
 2

Case of rejection of H 0

If F > F1−α ( p − 1, n − p), then H 0 : β1 = β 2 = ... = β p is rejected. This means that at least one βi is different from other effects
which is responsible for the rejection. So objective is to investigate and find out such βi and divide the population into
groups such that the means of populations within the group are same. This can be done by pairwise testing of β ′s.
Test H 0 : βi = βk (i ≠ k ) against H1 : βi ≠ βk .
This can be tested using following t-statistic

Yio − Yko
t=
⎛1 1⎞
s2 ⎜ + ⎟
⎝ ni nk ⎠

q2
which follows the t distribution with (n − p ) degrees of freedom under H 0 and s =
2
.
n− p

Thus the decision rule is to reject H 0 at level α if the observed difference

⎛1 1⎞
( yioi − yko
k ) >t α s2 ⎜ + ⎟
1− , n − p
2 ⎝ ni nk ⎠

⎛1 1⎞
The quantity t α s 2 ⎜ + ⎟ is called the critical difference.
1− , n − p
2 ⎝ ni nk ⎠
 3

Thus following steps are followed:

1. Compute all possible critical differences arising out of all possible pairs ( βi , β k ), i ≠ k = 1, 2,..., p.

2. Compare them with their observed differences.

3 Divide the p populations into different groups such that the populations in the same group have same means
3. means.

The computation are simplified if ni = n for all i. In such a case , the common critical difference (CCD) is

2s 2
CCD = t α
1− , n − p n
2

and the observed difference ( yio − y ko ),

) i ≠ k are compared with CCD
CCD.

If yio − yko > CCD

then the corresponding effects/means y io and y ko are coming from populations with the different means.
 4

Note: In general we say that if there are three effects β1 , β2 , β3 then

if H 01 : β1 = β 2 (≡ denote as event A) is accepted

and if H 02 : β 2 = β3 (≡ denote as event B ) is accepted
then H 03 : β1 = β 2 (≡ denote as event C ) will be accepted.

The question arises here that in what sense do we conclude such statement about the acceptance of H 03 .

The reason is as follows:

Since event A ∩ B ⊂ C , so

P ( A ∩ B ) ≤ P (C )

In this sense if the probability of an event is higher than the intersection of the events, i.e., the probability that H 03 is
accepted is higher than the probability of acceptance of H 01 and H02 both, so we conclude, in general , that the
acceptance of H 01 and H02 imply the acceptance of H 03 .
 5

Multiple
p comparison
p tests

One interest in the analysis of variance is to decide whether population means are equal or not. If the hypothesis of

equal means is rejected then one would like to divide the populations into subgroups such that all populations with

same means come to the same subgroup. This can be achieved by the multiple comparison tests.

A multiple comparison test procedure conducts the test of hypothesis for all the pairs of effects and compare them at a

significance level α , i.e., it works on per comparison basis.

t statistic If we want to ensure that the significance level α simultaneously for all group
This is based mainly on the t-statistic.

comparison of interest, the approximate multiple test procedure is one that controls the error rate per experiment

basis.

There are various available multiple comparison tests. We will discuss some of them in the context of one way

classification. In two way or higher classification, they can be used on similar lines.
 6

1. Studentized range test

It is assumed in the Studentized range test that the p samples, each of size n, have been drawn from p normal
populations. Let their sample means be y1o , y 2 o ,..., y po These means are ranked and arranged in an ascending order
as y1* , y2* ,..., y *p where y1* = Min yio and y *p = Max yio , i = 1, 2,..., p.
i i

Find the range.

R = y *p − y1* .

The Studentized range is defined as

R n
q p, n− p =
s

where q α , p , γ is the upper 100α % point of Studentized range when γ = n − p. The tables for q α , p ,γ are available.

The testing procedure involves the comparison of q p , γ with q α , p ,γ in the usual way as follows:
• if q p , n − p < q α , p , n − p then conclude that β1 = β 2 = ... = β p ,
• if q p , n − p > q α , p , n − p then all β ′s in the group are not the same.
 7

2. Studentized - Newman - Keuls test

The Student-Newman-Keuls test is similar to Studentized range test in the sense that the range is compared with 100α %
points on critical Studentized range W p given by
.
s2
W p = qα , p , γ .
n

The observed range R = y *p − y1* is now compared with W p .

• If R < W p then stop the process of comparison and conclude that β1 = β 2 = ... = β p .
• if R > W p then
* * *
i. divide the ranked means y1 , y2 ,..., y p into two subgroups containing

( y *p , y *p −1 ,..., y 2* ) and ( y *p −1 , y *p − 2 ,..., y1* ) .

ii. Compute the ranges R1 = y *p − y2* and R2 = y *p −1 − y1* . Then compare the ranges R1 and R2 with W p −1.

• If either range R1 or R2 is smaller than W p −1, then means (or β i ’s) in each of the groups are equal.
• If R1 and / or R2 are greater then W p −1 , then the ( p − 1) means (or β i ’ s) in the group concerned are divided
into two groups of (p – 2) means each and compare the range of the two groups with W p − 2 .

Continue with this procedure until a group of i means (or β i ’s) is found whose range does not exceed Wi .

By this method, the difference between any two means under test is significant when the range of the observed means
of each and every subgroup containing the two means under test is significant according to the Studentized critical
range. This procedure can be easily understood by the following flow chart.
 8

Arrange yio ' s in increasing order

y1* ≤ y2* ≤ ... ≤ y p*

Compute R = y *p − y1*
s2
Compare with W p = qα , p ,γ
n

If R < W p If R > W p
Stop and conclude continue
β1 = β 2 = ... = β p

Divide ranked mean in 2 groups

( y ,.*p ,..., y2* ) and ( y *p −1 ,..., y1* )

Compute R1 = y *p − y2*
R2 = y *p −1 − y1*
Compare R1 and R2 with Wp −1

4 possibilities
ibiliti off R1 and
d R 2 with
ith W p −1
 9

R1 < W p −1 R1 > W p −1
R1 < Wp −1 R1 > W p −1
R2 > W p −1 R2 < W p −1
R2 < W p −1 R2 > W p −1
⇒ β 2 = β3 = ... = β p
⇒ β 2 = β3 = ... = β p ⇒ β1 = β 2 = ... = β p −1
and
and at least one and at least one
β1 = β 2 = ... = β p −1 Divide ranked means
β i ≠ β j , i ≠ j = 1, 2,..., p − 1 β i ≠ β j , i ≠ j = 2,3,..., p
in four groups
which is β1 which is β p
⇒ β1 = β 2 = ... = β p 1. y *p ,..., y3*
⇒ one subgroup is ⇒ One subgroup has
( β 2 , β 3 ,..., β p ) 2. y *p −1 ,..., y2*
( β1 , β 2 ,..., β p −1 )
and another group has only β1 and another has only β p 3. y *p −1 ,..., y2* (same as in 2)
4. y *p − 2 ,..., y1*

Compute
R3 = y *p − y3*
R4 = y *p −1 − y2*
R5 = y *p − 2 − y1*
and compare
with Wp − 2

Continue till we get subgroups with same β ' s