Analysis 

of Variance and Design of 
Experiments
General Linear Hypothesis and Analysis of Variance 
:::
Lecture 4
ANOVA models  and Least Squares Estimation of 
Parameters

Shalabh 
Department of Mathematics and Statistics
Indian Institute of Technology Kanpur

Slides can be downloaded from http://home.iitk.ac.in/~shalab/sp1
Regression model for the general linear hypothesis:
Let Y1 , Y2 ,..., Yn be a sequence of n independent random variables
associated with responses. Then we can write it as
p
Yi    j xij   i , i  1, 2,..., n; j  1, 2,..., p
j 1

where 1 ,  2 ,...,  p are the unknown parameters and xij’s are the
known values of independent covariates X 1 , X 2 ,..., X p , i ’s are
i.i.d. random error component with

E ( i )  0 , V ar (  i )   2 , C ov (  i ,  j )  0( i  j ).

2
Relationship between the regression and ANOVA models:
Now consider the case of one‐way model and try to understand
its interpretation in terms of the multiple regression model.

The covariate X is now measured at different levels, e.g., if X is

the quantity of fertilizer then suppose there are p possible values,
say 1 Kg., 2 Kg.,..., p Kg. then X 1 , X 2 ,..., X p denotes these p values
in the following way.

3
Relationship between the regression and ANOVA models:
The linear model now can be expressed as
Y   0   1 X 1   2 X 2  ...   p X p  

by defining

1 if effect of 1 Kg. fertilizer is present

X1  
0 if effect of 1 Kg. fertilizer is absent

1 if effect of 2 Kg. fertilizer is present

X2  
0 if effect of 2 Kg. fertilizer is absent

1 if effect of p Kg. fertilizer is present
Xp  
0 if effect of p Kg. fertilizer is absent.
4
Relationship between the regression and ANOVA models:
If the effect of 1 Kg. of fertilizer is present, then other effects will
obviously be absent and the linear model is expressible as
Y   0   1 ( X 1  1)   2 ( X 2  0)  ...   p ( X p  0)  
  0  1  

If the effect of 2 Kg. of fertilizer is present then

Y   0  1 ( X 1  0)   2 ( X 2  1)  ...   p ( X p  0)  
 0  2  

If the effect of p Kg. of fertilizer is present then

Y   0  1 ( X 1  0)   2 ( X 2  0)  ...   p ( X p  1)  
 0   p  

and so on.
5
Relationship between the regression and ANOVA models:
If the experiment with 1 Kg. of fertilizer is repeated n1 number of
times then n1 observation on response variables are recorded
which can be represented as

Y11   0  1.1   2 .0  ...   p .0  11

Y12   0  1.1   2 .0  ...   p .0  12


Y1n1   0  1.1   2 .0  ...   p .0  1n1

6
Relationship between the regression and ANOVA models:
If X2 = 1 is repeated n2 times, then on the same lines n2
number of times then n2 observation on response variables are
recorded which can be represented as

Y21   0  1.0   2 .1  ...   p .0   21

Y22   0  1.0   2 .1  ...   p .0   22


Y2 n2   0  1.0   2 .1  ...   p .0   2 n2

7
Relationship between the regression and ANOVA models:
The experiment is continued and if Xp = 1 is repeated np times,
then on the same lines
Yp1   0  1.0   2 .0  ...   p .1   P1

Yp 2   0  1.0   2 .0  ...   p .1   P 2

Ypn p   0  1.0   2 .0  ...   p .1   pn p

8
Relationship between the regression and ANOVA models:
All these n1 , n 2 ,.., n p observations can be represented as
y   
 11
  1 1 0 0  0 0   11 
 y12     12 
 1 1 0 0  0 0 
   
                         
 y1n     1n 
 1 1 0 0  0 0 
 1
  1 
 y21  1 0 1 00 0    21 
y    
 0  
   1 0 1 0  0 0     22 
22

                                                      1                  
       
y
 2 n2   1 0 1 0  0 0  
   
 2 n2 
                       p
 
     
 y p1  1 0 0 00 1    p1 
     
 1 0 0 00 1  
 p2 
y  p2 
                     
  
1 0 0 00 1    
 y pn      pn 
 p  p
or
     Y  X    . 9
Relationship between the regression and ANOVA models:
In the two‐way analysis of  variance model, there are two 
covariates and the linear model is expressible as  

Y   0 + 1 X 1   2 X 2  ...   p X p   1 Z 1   2 Z 2  ...   q Z q  

X 1 , X 2 ,..., X p
where                           denotes, e.g., the p levels of the quantity of 
Z 1 , Z 2 ,..., Z q
fertilizer, say 1 Kg., 2 Kg.,...,p Kg. and                          denotes, e.g., 
the q levels of level of irrigation, say 10 Cms., 20 Cms.,…100  Cms. 
etc.  The levels                                                  are the counter variable 
X 1 , X 2 ,..., X p Z 1 , Z 2 ,..., Z q
indicating the presence or absence of the effect as in the earlier 
case.  
10
Relationship between the regression and ANOVA models:
If the effect of X1 and Z1 are present, i.e., 1 Kg of fertilizer and 10
Cms. of irrigation is used then the linear model is written as
Y   0  1.1   2 .0  ...   p .0   1.1   2 .0  ...   p .0  
  0  1   1   .

If X2 = 1 and Z2 = 1 is used, then the model is

Y  0   2   2   .

The design matrix can be written accordingly as in the one‐way

analysis of variance case.

In the three‐way analysis of variance model

Y    1 X 1  ...   p X p   1Z1  ...   q Z q  1W1  ...   rWr  
11
Relationship between the regression and ANOVA models:
The regression parameters  ' s can be fixed or random.

• If all  ' s are unknown constants, they are called as parameters

of the model and the model is called as a fixed effect model or
model I.
The objective, in this case, is to make inferences about the
parameters and the error variance  2 .

• If for some j, xij = 1 , for all i = 1,2,…n then  j is termed an

additive constant. In this case,  j occurs with every observation
and so it is also called a general mean effect. 12
Relationship between the regression and ANOVA models:
• If all  ' s are observable random variables except the additive
constant, then the linear model is termed as random effect
model, model II or variance components model.

The objective, in this case, is to make inferences about the

variances of  ' s , i.e.,  21 ,  22 ,...,  2 p and error variance and/or
certain functions of them.

13
Relationship between the regression and ANOVA models:

• If some parameters are fixed and some are random variables,

then the model is called a mixed effect model or model III. In
the mixed effect model, at least one  j is constant and at least
one  j is a random variable.

The objective is to make inference about the fixed effect

parameters, variance of random effects and error variance  2 .

14
Analysis of variance:
Analysis of variance is a body of statistical methods of analyzing
the measurements assumed to be structured as

y i   1 xi1   2 xi 2  ...   p xip   i , i  1, 2,..., n

where xij are integers, generally 0 or 1 indicating usually the

absence or presence of effects  j ; and i ’s are assumed to be
identically and independently distributed with mean 0 and
variance  2.
It may be noted that the i ’s can be assumed additionally to
follow a normal distribution N (0,  2 ) .
15
Analysis of variance:
It is needed for the maximum likelihood estimation of parameters
from the beginning of the analysis, but in the least‐squares
estimation, it is needed only when conducting the tests of
hypothesis and the confidence interval estimation of parameters.

The least‐squares method does not require any knowledge of

distribution like normal up to the stage of estimation of
parameters.

We need some basic concepts to develop tools.

16
Least squares estimate of 
Let y1 , y2 ,..., yn be a sample of observations on Y1, Y2 ,..., Yn .
The least‐squares estimate of  is the values ˆ of  for which the
sum of squares due to errors, i.e.,
n
S 2    i2   '   ( y  X  )( y  X  )
i 1

 yy  2 X ' y    X X 

is minimum where y  ( y1 , y2 ,..., yn ).

17
Least squares estimate of  
Differentiating S2 with respect to  and substituting it to be
zero, the normal equations are obtained as

dS 2
 2 X X   2 X y  0
d

or X X   X y

If X has full rank p, then ( X X ) has a unique inverse and the

unique least squares estimate (LSE) of  is

ˆ  ( X X ) 1 X y

18
Least squares estimate of  
ˆ (LSE) is the best linear unbiased estimator of  in the sense of
having minimum variance in the class of linear and unbiased
estimator.

If the rank of X is not full, then generalized inverse is used for

finding the inverse of (X’X).

If L  is a linear parametric function where L  (  1 ,  2 ,...,  p ) is

a non‐null vector, then the least‐squares estimate of L  is Lˆ .

19
Least squares estimate of  
A question arises that what are the conditions under which a
linear parametric function L  admits a unique least‐squares
estimate in the general case.

The concept of estimable function is needed to find such

conditions.

20