Analysis of Variance and Design of

Experiments
Experimental Design Models
:::
Lecture 16
Tukey’s Test for Non‐additivity

Shalabh
Department of Mathematics and Statistics
Indian Institute of Technology Kanpur

Slides can be downloaded from http://home.iitk.ac.in/~shalab/sp1

Tukey’s test for non‐additivity:
Consider the set up of two way classification with one observation
per cell and interaction as

yij    i   j   ij   ij , i  1, 2..., I , j  1, 2,..., J

I J
with 
i 1
i  0,   j  0.
j 1

The distribution of degrees of freedom in this case is as follows:

2
Tukey’s test for non‐additivity:
Source Degrees of freedom
A I‐ 1
B J‐ 1
AB(interaction) (I – 1) (J – 1)
Error 0
_______________________________________________
Total (IJ – 1)
_______________________________________________

There is no degree of freedom for error.

3
Tukey’s test for non‐additivity:
The problem is that the two factor interaction effect and
random error component are subsumed together and cannot
be separated out.
There is no estimate for  2 .

If no interaction exists, then H 0 :  ij  0 for all i, j is accepted

and the additive model
yij     i   j   ij

is well enough to test the hypothesis H 0 :  i  0 and H 0 :  j  0

with error having ( I  1)( J  1) degrees of freedom.
4
Tukey’s test for non‐additivity:
If interaction exists, then H 0 :  ij  0 is rejected.
In such a case, if we assume that the structure of the interaction
effect is such that it is proportional to the product of individual
effects, i.e.,  ij   i  j then a test for testing H 0 :   0 can be
constructed.

Such a test will serve as a test for non‐additivity.

It will help in knowing the effect of the presence of interact effect

and whether the interaction enters into the model additively.

5
Tukey’s test for non‐additivity:
Such a test is given by Tukey’s test for non‐additivity which requires
one degree of freedom leaving  I -1 J -1 -1 degrees of freedom
for error.

Let us assume that departure from additivity can be specified by

introducing a product term and writing the model as

E ( yij )     i   j   i  j ; i  1, 2,..., I , j  1, 2,..., J

I J
with 
i 1
i  0, 
j 1
j  0.

When   0, the model becomes a nonlinear model and the least‐

squares theory for linear models is not applicable.
6
Tukey’s test for non‐additivity:
Note that using , we have

1 I J
1 I J
yoo 
IJ

i 1 j 1
yij 
IJ
    
i 1 j 1
i   j   i  j   ij 

1 I 1 J  I J
     i    j  (  j )(  j )   oo
I i 1 J j 1 IJ i 1 j 1

    oo

E ( yoo )  

 ˆ  yoo .

7
Tukey’s test for non‐additivity:
Next
1 J 1 J
yio   yij       i   j   i  j   ij 
J j 1 J j 1
1 J 1 J
    i    j   i   j   io
J j 1 J j 1
    i   io

E ( yio )     i
 ˆi  yio  ˆ  yio  yoo .
Similarly
yoj     j
 ˆ j  yoj  ˆ  yoj  yoo . 8
Tukey’s test for non‐additivity:
Thus ˆ , ˆ i and ̂ j remain the unbiased estimators of  ,  i
and  j , respectively irrespective of whether   0 or not.
Also
E  yij  yio  yoj  yoo    i  j
or
E ( yij  yoo )  ( yio  yoo )  ( yoj  yoo )    i  j .

9
Tukey’s test for non‐additivity:
Consider the estimation of  ,  i ,  j and  based on the
minimization of
S    ( yij     i   j   i  j ) 2
i j

   Sij2 .
i j

Use least squares and observe the outcome.

10
Tukey’s test for nonadditivity:
The normal equations are solved as
S I J
 0   Sij  0  ˆ  yoo
 i 1 j 1

S J
 0   (1   j ) Sij  0
 i j 1

S I
 0   (1   i ) Sij  0
 j i 1

S I J
 0    i  j Sij  0
 i 1 j 1
I J
or   ( y
i 1 j 1
i j ij     i   j   i  j )  0

I J

  i

j yij
  (say) which can be estimated provided  i
i 1 j 1
or  
 I 2  J 2 
 i     j  and  j are assumed to be known.
 i 1   j 1 
11
Tukey’s test for non‐additivity:
Since  i and  j can be estimated by ̂ i  yio  yoo and
̂ j  yoj  yoo irrespective of whether   0 or not, so we can
substitute them in place of  i and  j in  which gives
I J I J

 ˆi ˆ j yij ( IJ ) ˆ i ˆ j yij

ˆ  i 1 j 1

i 1 j 1

 I 2  J ˆ2   I 2  J ˆ2 
  ˆ i     j   J  ˆ i   I   j 
 i 1   j 1   i 1   j 1 
I J
IJ  ( yio  yoo )( yoj  yoo ) yij
i 1 j 1

S ASB
I I
where S A  J  ˆ  J  ( yio  yoo ) 2
i
2

i 1 i 1
J J
S B  I  ˆ  I  ( yoj  yoo ) 2 .
2
j
j 1 j 1 12
Tukey’s test for non‐additivity:
Assuming  i and  j to be known
2
 
   I J 
  1     i  j Var ( yij )  0 
Var ( )  I 2 2

 J
2   i 1 j 1
   i
2
  j 

 i 1 j  1 
 I
  J
2
  i     j 
2 2

 i 1   j 1  2
 
 I 2  J 2 
2 2
 2  2
 i     j 
I J

 i     j   i 1   j 1 
 i 1   j 1 
using
Var ( yij )   2 , Cov( yij , y jk )  0 for all i  k .
13
Tukey’s test for non‐additivity:
When  i and  j are estimated by ˆ i and ˆ j then substitute
them back in the expression of Var () and treating it as Var (ˆ)
gives
2
Var (ˆ ) 
 I 2  J ˆ2 
  ˆi     j 
 i 1   j 1 
IJ  2

S ASB

for given ˆ i and ˆ j .

14
Tukey’s test for non‐additivity:
Note that if   0, then
 I J 
  i j ij 
  y
E  ˆ / ˆ i , ˆ j for all i, j   E  I 
i 1 j 1

 J
2 
 i j 
 
2

 i 1 j 1 
 I J 
  i j   (    i   j  0   )
ij 
0
 E i 1 j 1
   0.
 I J
 I J

 (   i
2
)(   2
j )   i  j)
(  2
) (  2

 i 1 j 1  i 1 j 1

As ̂ i and ̂ j remains valid irrespective of   0, or not, in

this sense ̂ is a function of yij and hence normally distributed
as
ˆ  IJ  2

 ~ N  0, . 15

 S ASB 
Tukey’s test for non‐additivity:
Thus the statistic
2 2
 I J   I J 
IJ   ˆi  j yij 
ˆ IJ   ( yio  yoo )( yoj  yoo ) yij 
(ˆ ) 2
 i 1 j 1   i 1 j 1 
 
Var (ˆ )  2 S ASB  2 S ASB
2
 I J 
IJ   ( yio  yoo )( yoj  yoo )( yij  yio  yoj  yoo ) 
  
i 1 j 1 SN

 2 S ASB 2
follows a  2 ‐ distribution with one degree of freedom where
2
 I J 
IJ   ( yio  yoo )( yoj  yoo )( yij  yio  yoj  yoo ) 
SN   
i 1 j 1

S ASB

is the sum of squares due to non‐additivity. 16

Tukey’s test for non‐additivity:
Note that I J

S AB
  ij io oj oo
( y 
i 1 j 1
y  y  y )2


2 2
 S N SAB 
Follows  2 (( I  1)( J  1)), so  2  2  is nonnegative and
  
follows  2  ( I 1)( J 1) 1
The reason for this is as follows:
yij     i   j  non additivity   ij
and so
TSS  SSA  SSB  S N  SSE
 SSE  TSS  SSA  SSB  S N
has degrees of freedom
( IJ  1)  ( I  1)  ( J  1)  1  ( I  1)( J  1)  1 17
Tukey’s test for non‐additivity:
We need to ensure that SSE > 0. So using the result
“If Q , Q1 and Q2 are quadratic forms such that
Q  Q1  Q2 with Q ~  2 ( a ), Q2 ~  2 (b ) and Q2 is non‐negative, then
Q1 ~  2 ( a  b )"
SN SAB
ensures that the difference  is nonnegative.
2 2

Moreover SN (SS due to non‐additivity) and SSE are orthogonal.

18
Tukey’s test for non‐additivity:
Thus the F‐test for non‐additivity is
 SN /  2 
 
 1 
F
 SSE /  2 
 
 ( I  1)( J  1)  1 
SSN
  ( I  1)( J  1)  1
SSE
~ F 1, ( I  1)( J  1)  1 under H 0 .

So the decision rule is

Reject H 0 :   0 whenever
F  F1 1,( I  1)( J  1)  1 .

19
Tukey’s test for non‐additivity:
The analysis of variance table for the model including a term for
non‐additivity is as follows:
ANOVA Table
________________________________________________________________________
Source of Degrees Sum of Mean F‐value
variation of freedom squares squares
________________________________________________________________________
Due to Factor A I–1 SA MSA

Due to Factor B J–1 SB MSB

MSN
Due to 1 SN MSN = SN F
Non‐additivity N MSE

Error (I – 1)(J – 1) ‐1SSE MSE

(By subtraction)
__________________________________________________________________
Total IJ – 1 TSS
20