STAT 714

LINEAR STATISTICAL MODELS

Fall, 2010

Lecture Notes

Joshua M. Tebbs
Department of Statistics
The University of South Carolina
TABLE OF CONTENTS STAT 714, J. TEBBS

Contents

1 Examples of the General Linear Model 1

2 The Linear Least Squares Problem 13

2.1 Least squares estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.2 Geometric considerations . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

2.3 Reparameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 25

3 Estimability and Least Squares Estimators 28

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2 Estimability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28

3.2.1 One-way ANOVA . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

3.2.2 Two-way crossed ANOVA with no interaction . . . . . . . . . . . 37

3.2.3 Two-way crossed ANOVA with interaction . . . . . . . . . . . . . 39

3.3 Reparameterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40

3.4 Forcing least squares solutions using linear constraints . . . . . . . . . . . 46

4 The Gauss-Markov Model 54

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

4.2 The Gauss-Markov Theorem . . . . . . . . . . . . . . . . . . . . . . . . . 55

4.3 Estimation of σ 2 in the GM model . . . . . . . . . . . . . . . . . . . . . 57

4.4 Implications of model selection . . . . . . . . . . . . . . . . . . . . . . . . 60

4.4.1 Underfitting (Misspecification) . . . . . . . . . . . . . . . . . . . . 60

4.4.2 Overfitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61

4.5 The Aitken model and generalized least squares . . . . . . . . . . . . . . 63

5 Distributional Theory 68

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68

i
TABLE OF CONTENTS STAT 714, J. TEBBS

5.2 Multivariate normal distribution . . . . . . . . . . . . . . . . . . . . . . . 69

5.2.1 Probability density function . . . . . . . . . . . . . . . . . . . . . 69

5.2.2 Moment generating functions . . . . . . . . . . . . . . . . . . . . 70

5.2.3 Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 72

5.2.4 Less-than-full-rank normal distributions . . . . . . . . . . . . . . 73

5.2.5 Independence results . . . . . . . . . . . . . . . . . . . . . . . . . 74

5.2.6 Conditional distributions . . . . . . . . . . . . . . . . . . . . . . . 76

5.3 Noncentral χ2 distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 77

5.4 Noncentral F distribution . . . . . . . . . . . . . . . . . . . . . . . . . . 79

5.5 Distributions of quadratic forms . . . . . . . . . . . . . . . . . . . . . . . 81

5.6 Independence of quadratic forms . . . . . . . . . . . . . . . . . . . . . . . 85

5.7 Cochran’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

6 Statistical Inference 95

6.1 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95

6.2 Testing models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97

6.3 Testing linear parametric functions . . . . . . . . . . . . . . . . . . . . . 103

6.4 Testing models versus testing linear parametric functions . . . . . . . . . 107

6.5 Likelihood ratio tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109

6.5.1 Constrained estimation . . . . . . . . . . . . . . . . . . . . . . . . 109

6.5.2 Testing procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . 111

6.6 Confidence intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.6.1 Single intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . . 113

6.6.2 Multiple intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . 115

7 Appendix 118

7.1 Matrix algebra: Basic ideas . . . . . . . . . . . . . . . . . . . . . . . . . 118

7.2 Linear independence and rank . . . . . . . . . . . . . . . . . . . . . . . . 120

ii
TABLE OF CONTENTS STAT 714, J. TEBBS

7.3 Vector spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

7.4 Systems of equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 128

7.5 Perpendicular projection matrices . . . . . . . . . . . . . . . . . . . . . . 134

7.6 Trace, determinant, and eigenproblems . . . . . . . . . . . . . . . . . . . 136

7.7 Random vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 142

iii
CHAPTER 1 STAT 714, J. TEBBS

1 Examples of the General Linear Model

Complementary reading from Monahan: Chapter 1.

INTRODUCTION : Linear models are models that are linear in their parameters. The
general form of a linear model is given by

Y = Xβ + ,

where Y is an n × 1 vector of observed responses, X is an n × p (design) matrix of fixed

constants, β is a p × 1 vector of fixed but unknown parameters, and  is an n × 1 vector
of (unobserved) random errors. The model is called a linear model because the mean of
the response vector Y is linear in the unknown parameter β.

SCOPE : Several models commonly used in statistics are examples of the general linear
model Y = Xβ + . These include, but are not limited to, linear regression models and
analysis of variance (ANOVA) models. Regression models generally refer to those for
which X is full rank, while ANOVA models refer to those for which X consists of zeros
and ones.

GENERAL CLASSES OF LINEAR MODELS :

• Model I: Least squares model : Y = Xβ + . This model makes no assumptions

on . The parameter space is Θ = {β : β ∈ Rp }.

• Model II: Gauss Markov model : Y = Xβ + , where E() = 0 and cov() = σ 2 I.

The parameter space is Θ = {(β, σ 2 ) : (β, σ 2 ) ∈ Rp × R+ }.

• Model III: Aitken model : Y = Xβ + , where E() = 0 and cov() = σ 2 V, V

known. The parameter space is Θ = {(β, σ 2 ) : (β, σ 2 ) ∈ Rp × R+ }.

• Model IV: General linear mixed model : Y = Xβ + , where E() = 0 and

cov() = Σ ≡ Σ(θ). The parameter space is Θ = {(β, θ) : (β, θ) ∈ Rp × Ω},
where Ω is the set of all values of θ for which Σ(θ) is positive definite.

PAGE 1
CHAPTER 1 STAT 714, J. TEBBS

GAUSS MARKOV MODEL: Consider the linear model Y = Xβ + , where E() = 0

and cov() = σ 2 I. This model is treated extensively in Chapter 4. We now highlight
special cases of this model.

Example 1.1. One-sample problem. Suppose that Y1 , Y2 , ..., Yn is an iid sample with
mean µ and variance σ 2 > 0. If 1 , 2 , ..., n are iid with mean E(i ) = 0 and common
variance σ 2 , we can write the GM model

Y = Xβ + ,

where
     
Y1 1 1
     
     
 Y2   1   2 
Yn×1 =
 ..
,
 Xn×1 =
 ..
,
 β 1×1 = µ, n×1 =
 ..
.

 .   .   . 
     
Yn 1 n
Note that E() = 0 and cov() = σ 2 I. 

Example 1.2. Simple linear regression. Consider the model where a response variable
Y is linearly related to an independent variable x via

Yi = β0 + β1 xi + i ,

for i = 1, 2, ..., n, where the i are uncorrelated random variables with mean 0 and
common variance σ 2 > 0. If x1 , x2 , ..., xn are fixed constants, measured without error,
then this is a GM model Y = Xβ +  with
     
Y1 1 x1 1
       
     
 Y2   1 x2  β0  2 
Yn×1 =  .  , Xn×2 =  . . 
  
, β 2×1 =  , n×1 = ..
.
 ..   .. .. 
 
β1  . 
     
Yn 1 xn n
Note that E() = 0 and cov() = σ 2 I. 

Example 1.3. Multiple linear regression. Suppose that a response variable Y is linearly
related to several independent variables, say, x1 , x2 , ..., xk via

Yi = β0 + β1 xi1 + β2 xi2 + · · · + βk xik + i ,

PAGE 2
CHAPTER 1 STAT 714, J. TEBBS

for i = 1, 2, ..., n, where i are uncorrelated random variables with mean 0 and common
variance σ 2 > 0. If the independent variables are fixed constants, measured without
error, then this model is a special GM model Y = Xβ +  where
 
    β0  
Y 1 x11 x12 · · · x1k   1
 1    
 β1

  
 1 x21 x22 · · · x2k 
     
 Y2     2 
Y =  .  , Xn×p =  . . , β p×1 =  ,  =  ,
 
  
. . .
 β2 ..
 ..   .. .. .. .. .. 
  

 ..

  . 
     .   
Yn 1 xn1 xn2 · · · xnk   n
βk

and p = k + 1. Note that E() = 0 and cov() = σ 2 I. 

Example 1.4. One-way ANOVA. Consider an experiment that is performed to compare

a ≥ 2 treatments. For the ith treatment level, suppose that ni experimental units are
selected at random and assigned to the ith treatment. Consider the model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where the random errors ij are uncorrelated random
variables with zero mean and common variance σ 2 > 0. If the a treatment effects
α1 , α2 , ..., αa are best regarded as fixed constants, then this model is a special case of the
GM model Y = Xβ + . To see this, note that with n = ai=1 ni ,
P

 
    µ
Y11 1n1 1n1 0n1 · · · 0n1  
α
     
 1 
0n2 1n2 · · · 0n2 
   
 Y12   1  
Yn×1 =   , Xn×p =  n2  , β p×1 =   α2  ,

 ..   .. .. .. ... .. 
 .   . . . .   .. 
 
     . 
Yana 1na 0na 0na · · · 1na  
αa

where p = a + 1 and n×1 = (11 , 12 , ..., ana )0 , and where 1ni is an ni × 1 vector of ones
and 0ni is an ni × 1 vector of zeros. Note that E() = 0 and cov() = σ 2 I.

NOTE : In Example 1.4, note that the first column of X is the sum of the last a columns;
i.e., there is a linear dependence in the columns of X. From results in linear algebra,
we know that X is not of full column rank. In fact, the rank of X is r = a, one less

PAGE 3
CHAPTER 1 STAT 714, J. TEBBS

than the number of columns p = a + 1. This is a common characteristic of ANOVA

models; namely, their X matrices are not of full column rank. On the other hand,
(linear) regression models are models of the form Y = Xβ + , where X is of full column
rank; see Examples 1.2 and 1.3. 

Example 1.5. Two-way nested ANOVA. Consider an experiment with two factors,
where one factor, say, Factor B, is nested within Factor A. In other words, every level
of B appears with exactly one level of Factor A. A statistical model for this situation is

Yijk = µ + αi + βij + ijk ,

for i = 1, 2, ..., a, j = 1, 2, ..., bi , and k = 1, 2, ..., nij . In this model, µ denotes the overall
mean, αi represents the effect due to the ith level of A, and βij represents the effect
of the jth level of B, nested within the ith level of A. If all parameters are fixed, and
the random errors ijk are uncorrelated random variables with zero mean and constant
unknown variance σ 2 > 0, then this is a special GM model Y = Xβ + . For example,
with a = 3, b = 2, and nij = 2, we have
   
Y 1 1 0 0 1 0 0 0 0 0
 111     
 Y112   1 1 0 0 1 0 0 0 0 0  µ
   
     
 Y121   1 1 0 0 0 1 0 0 0 0   α1
     

     
 Y122   1 1 0 0 0 1 0 0 0 0   α2
     

     
 Y211   1 0 1 0 0 0 1 0 0 0  α3
     
 
     
     
 Y212   1 0 1 0 0 0 1 0 0 0   β
 , β =  11

Y= 
, X = 
   
,

 Y221   1 0 1 0 0 0 0 1 0 0   β12 
     
     
 Y222   1 0 1 0 0 0 0 1 0 0   β21 
     
     
 Y311   1 0 0 1 0 0 0 0 1 0   β22 
     
     
 Y312   1 0 0 1 0 0 0 0 1 0 
  β31 
    
   
 Y321   1 0 0 1 0 0 0 0 0 1  β32
   
Y322 1 0 0 1 0 0 0 0 0 1

and  = (111 , 112 , ..., 322 )0 . Note that E() = 0 and cov() = σ 2 I. The X matrix is not
of full column rank. The rank of X is r = 6 and there are p = 10 columns. 

PAGE 4
CHAPTER 1 STAT 714, J. TEBBS

Example 1.6. Two-way crossed ANOVA with interaction. Consider an experiment with
two factors (A and B), where Factor A has a levels and Factor B has b levels. In general,
we say that factors A and B are crossed if every level of A occurs in combination with
every level of B. Consider the two-factor (crossed) ANOVA model given by

Yijk = µ + αi + βj + γij + ijk ,

for i = 1, 2, ..., a, j = 1, 2, ..., b, and k = 1, 2, ..., nij , where the random errors ij are
uncorrelated random variables with zero mean and constant unknown variance σ 2 > 0.
If all the parameters are fixed, this is a special GM model Y = Xβ + . For example,
with a = 3, b = 2, and nij = 3,
   
Y111 1 1 0 0 1 0 1 0 0 0 0 0
   
 Y112   1 1 0 0 1 0 1 0 0 0 0 0 
   
   
 Y113   1 1 0 0 1 0 1 0 0 0 0 0 
   
     
 Y121   1 1 0 0 0 1 0 1 0 0 0 0  µ
   
     
 Y122   1 1 0 0 0 1 0 1 0 0 0 0   α1
     

     
 Y123   1 1 0 0 0 1 0 1 0 0 0 0   α2
     

     
 Y211   1 0 1 0 1 0 0 0 1 0 0 0   α3
     

     
 Y212   1 0 1 0 1 0 0 0 1 0 0 0   β1
     

     
     
 Y213   1 0 1 0 1 0 0 0 1 0 0 0   β
, β =  2

Y= 
, X = 
   
,

 Y221   1 0 1 0 0 1 0 0 0 1 0 0   γ11 
     
     
 Y222   1 0 1 0 0 1 0 0 0 1 0 0   γ12 
     
     
 Y223   1 0 1 0 0 1 0 0 0 1 0 0   γ21 
     
     
 Y311   1 0 0 1 1 0 0 0 0 0 1 0   γ22 
     
     
 Y312   1 0 0 1 1 0 0 0 0 0 1 0   γ31 
     
   
 Y313   1 0 0 1 1 0 0 0 0 0 1 0 
 γ32
  
   
 Y321   1 0 0 1 0 1 0 0 0 0 0 1 
   
   
 Y322   1 0 0 1 0 1 0 0 0 0 0 1 
   
Y323 1 0 0 1 0 1 0 0 0 0 0 1

and  = (111 , 112 , ..., 323 )0 . Note that E() = 0 and cov() = σ 2 I. The X matrix is not
of full column rank. The rank of X is r = 6 and there are p = 12 columns. 

PAGE 5
CHAPTER 1 STAT 714, J. TEBBS

Example 1.7. Two-way crossed ANOVA without interaction. Consider an experiment

with two factors (A and B), where Factor A has a levels and Factor B has b levels. The
two-way crossed model without interaction is given by

Yijk = µ + αi + βj + ijk ,

for i = 1, 2, ..., a, j = 1, 2, ..., b, and k = 1, 2, ..., nij , where the random errors ij are
uncorrelated random variables with zero mean and common variance σ 2 > 0. Note that
no-interaction model is a special case of the interaction model in Example 1.6 when
H0 : γ11 = γ12 = · · · = γ32 = 0 is true. That is, the no-interaction model is a reduced
version of the interaction model. With a = 3, b = 2, and nij = 3 as before, we have
   
Y111 1 1 0 0 1 0
   
 Y112   1 1 0 0 1 0 
   
   
 Y113   1 1 0 0 1 0 
   
   
 Y121   1 1 0 0 0 1 
   
   
 Y122   1 1 0 0 0 1 
   
   
 Y123   1 1 0 0 0 1 
   
     
 Y211   1 0 1 0 1 0  µ
   
     
 Y212   1 0 1 0 1 0   α1 
     
     
     
 Y213   1 0 1 0 1 0   α2 
Y= 
, X = 
 
, β = 
 
,

 Y221   1 0 1 0 0 1   α3 
     
     
 Y222   1 0 1 0 0 1   β1 
     
   
 Y223   1 0 1 0 0 1  β2
   
   
 Y311   1 0 0 1 1 0 
   
   
 Y312   1 0 0 1 1 0 
   
   
 Y313   1 0 0 1 1 0 
   
   
 Y321   1 0 0 1 0 1 
   
   
 Y322   1 0 0 1 0 1 
   
Y323 1 0 0 1 0 1

and  = (111 , 112 , ..., 323 )0 . Note that E() = 0 and cov() = σ 2 I. The X matrix is not
of full column rank. The rank of X is r = 4 and there are p = 6 columns. Also note that

PAGE 6
CHAPTER 1 STAT 714, J. TEBBS

the design matrix for the no-interaction model is the same as the design matrix for the
interaction model, except that the last 6 columns are removed. 

Example 1.8. Analysis of covariance. Consider an experiment to compare a ≥ 2

treatments after adjusting for the effects of a covariate x. A model for the analysis of
covariance is given by
Yij = µ + αi + βi xij + ij ,

for i = 1, 2, ..., a, j = 1, 2, ..., ni , where the random errors ij are uncorrelated random
variables with zero mean and common variance σ 2 > 0. In this model, µ represents the
overall mean, αi represents the (fixed) effect of receiving the ith treatment (disregarding
the covariates), and βi denotes the slope of the line that relates Y to x for the ith
treatment. Note that this model allows the treatment slopes to be different. The xij ’s
are assumed to be fixed values measured without error.

NOTE : The analysis of covariance (ANCOVA) model is a special GM model Y = Xβ+.

For example, with a = 3 and n1 = n2 = n3 = 3, we have
     
Y11 1 1 0 0 x11 0 0 11
       
 Y12   1 1 0 0 x12 0 0  µ 12
     
 
       
 Y13   1 1 0 0 x13 0 0   α1  13
       
 
       
 Y21   1 0 1 0 0 x21 0   α2  21
       
 
       
Y =  Y22  , X =  1 0 1 0 0 x22 0  , β =  α3  ,  =  .
       
22
       
       
 Y23   1 0 1 0 0 x23 0 β 23
 1 
    
     
       
 Y31   1 0 0 1 0 0 x31   β2   31 
       
     
 Y32   1 0 0 1 0 0 x32  β3  32 
     
Y33 1 0 0 1 0 0 x33 33

Note that E() = 0 and cov() = σ 2 I. The X matrix is not of full column rank. If there
are no linear dependencies among the last 3 columns, the rank of X is r = 6 and there
are p = 7 columns.

REDUCED MODEL: Consider the ANCOVA model in Example 1.8 which allows for
unequal slopes. If β1 = β2 = · · · = βa ; that is, all slopes are equal, then the ANCOVA

PAGE 7
CHAPTER 1 STAT 714, J. TEBBS

model reduces to
Yij = µ + αi + βxij + ij .

That is, the common-slopes ANCOVA model is a reduced version of the model that
allows for different slopes. Assuming the same error structure, this reduced ANCOVA
model is also a special GM model Y = Xβ + . With a = 3 and n1 = n2 = n3 = 3, as
before, we have
     
Y 1 1 0 0 x11 11
 11     
 Y12 1 1 0 0 x12 12
     
    
       
 Y13 1 1 0 0 x13 µ 13
     
    
       
 Y21 1 0 1 0 x21  α1  21
       
    
       
Y =  Y22 , X =   , β =  α2  ,  =  .
       
1 0 1 0 x22 22
       
       
 Y23   1 0 1 0 x23   α3   23 
       
     
 Y31   1 0 0 1 x31  β  31 
     
     
 Y32   1 0 0 1 x32   32 
     
Y33 1 0 0 1 x33 33

As long as at least one of the xij ’s is different, the rank of X is r = 4 and there are p = 5
columns. 

GOAL: We now provide examples of linear models of the form Y = Xβ +  that are not
GM models.

TERMINOLOGY : A factor of classification is said to be random if it has an infinitely

large number of levels and the levels included in the experiment can be viewed as a
random sample from the population of possible levels.

Example 1.9. One-way random effects ANOVA. Consider the model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where the treatment effects α1 , α2 , ..., αa are best
regarded as random; e.g., the a levels of the factor of interest are drawn from a large
population of possible levels, and the random errors ij are uncorrelated random variables

PAGE 8
CHAPTER 1 STAT 714, J. TEBBS

with zero mean and common variance σ 2 > 0. For concreteness, let a = 4 and nij = 3.
The model Y = Xβ +  looks like
   
Y11 11
   
 Y12  12
   
 
   
 Y13  13
   
 
   
 Y21  21
   
 
      
 Y22  13 03 03 03 α1 22
   
 
      
      
 Y23   03 13 03 03   α2   23 
Y=  
 = 112 µ + 



+
 


 Y31   03 03 13 03   α3   31 
      
   
 Y32  03 03 03 13 α4  32 
  | {z } | {z }  
   
 Y33  = Z1 = 1  33 
   
   
 Y41   41 
   
   
 Y42   42 
   
Y43 43
| {z }
= 2
= Xβ + Z1 1 + 2 ,

where we identify X = 112 , β = µ, and  = Z1 1 + 2 . This is not a GM model because

cov() = cov(Z1 1 + 2 ) = Z1 cov(1 )Z01 + cov(2 ) = Z1 cov(1 )Z01 + σ 2 I,

provided that the αi ’s and the errors ij are uncorrelated. Note that cov() 6= σ 2 I. 

Example 1.10. Two-factor mixed model. Consider an experiment with two factors (A
and B), where Factor A is fixed and has a levels and Factor B is random with b levels.
A statistical model for this situation is given by

Yijk = µ + αi + βj + ijk ,

for i = 1, 2, ..., a, j = 1, 2, ..., b, and k = 1, 2, ..., nij . The αi ’s are best regarded as fixed
and the βj ’s are best regarded as random. This model assumes no interaction.

APPLICATION : In a randomized block experiment, b blocks may have been selected

randomly from a large collection of available blocks. If the goal is to make a statement

PAGE 9
CHAPTER 1 STAT 714, J. TEBBS

about the large population of blocks (and not those b blocks in the experiment), then
blocks are considered as random. The treatment effects α1 , α2 , ..., αa are regarded as
fixed constants if the a treatments are the only ones of interest.

NOTE : For concreteness, suppose that a = 2, b = 4, and nij = 1. We can write the
model above as
   
Y 11
 11   
 Y12 12
   
  
     
 Y13 β 13
     
 1
  


 

 µ 

 
 


 Y14 14 14 04    I4  
 β 14
 2
   
Y= = + + .

  α1
     
 Y21  14 04 14   I4  β3   21 



 α2   



 Y22  | {z } β4  22 
= Xβ
  | {z }  
   
 Y23  = Z1 1  23 
   
Y24 24
| {z }
= 2

NOTE : If the αi ’s are best regarded as random as well, then we have

   
Y 11
 11   
 Y12  12 
   

     
 Y13  β 13 
   
 1  

      
     
 Y14  14 04 α1 I4  β2   14 
Y= 
 = 18 µ + 

 + 

+
 
.

 Y21  04 14 α2 I4  β3   21 
  | {z }    
   
 Y22  = Z1 1 β4  22 
  | {z }  
   
 Y23  = Z2 2  23 
   
Y24 24
| {z }
= 3

This model is also known as a random effects or variance component model. 

GENERAL FORM : A linear mixed model can be expressed generally as

Y = Xβ + Z1 1 + Z2 2 + · · · + Zk k ,

where Z1 , Z2 , ..., Zk are known matrices (typically Zk = Ik ) and 1 , 2 , ..., k are uncorre-
lated random vectors with uncorrelated components.

PAGE 10
CHAPTER 1 STAT 714, J. TEBBS

Example 1.11. Time series models. When measurements are taken over time, the GM
model may not be appropriate because observations are likely correlated. A linear model
of the form Y = Xβ + , where E() = 0 and cov() = σ 2 V, V known, may be more
appropriate. The general form of V is chosen to model the correlation of the observed
responses. For example, consider the statistical model

Yt = β0 + β1 t + t ,

for t = 1, 2, ..., n, where t = ρt−1 + at , at ∼ iid N (0, σ 2 ), and |ρ| < 1 (this is a
stationarity condition). This is called a simple linear trend model where the error
process {t : t = 1, 2, ..., n} follows an autoregressive model of order 1, AR(1). It is easy
to show that E(t ) = 0, for all t, and that cov(t , s ) = σ 2 ρ|t−s| , for all t and s. Therefore,
if n = 5,  
2 3 4
1 ρ ρ ρ ρ
 
2
 ρ 1 ρ ρ ρ 
 3 
 
V = σ 2  ρ2 ρ 1 ρ ρ 2  . 
 
 
 3 2 
 ρ ρ ρ 1 ρ 
 
ρ4 ρ3 ρ2 ρ 1

Example 1.12. Random coefficient models. Suppose that t measurements are taken
(over time) on n individuals and consider the model

Yij = x0ij β i + ij ,

for i = 1, 2, ..., n and j = 1, 2, ..., t; that is, the different p × 1 regression parameters β i
are “subject-specific.” If the individuals are considered to be a random sample, then we
can treat β 1 , β 2 , ..., β n as iid random vectors with mean β and p × p covariance matrix
Σββ , say. We can write this model as

Yij = x0ij β i + ij

= x0ij β + x0ij (β i − β) + ij .

|{z} | {z }
fixed random

If the β i ’s are independent of the ij ’s, note that

var(Yij ) = x0ij Σββ xij + σ 2 6= σ 2 . 

PAGE 11
CHAPTER 1 STAT 714, J. TEBBS

Example 1.13. Measurement error models. Consider the statistical model

Yi = β0 + β1 Xi + i ,

where i ∼ iid N (0, σ2 ). The Xi ’s are not observed exactly; instead, they are measured
with non-negligible error so that

Wi = Xi + Ui ,

where Ui ∼ iid N (0, σU2 ). Here,

Observed data: (Yi , Wi )

Not observed: (Xi , i , Ui )

Unknown parameters: (β0 , β1 , σ2 , σU2 ).

As a frame of reference, suppose that Y is a continuous measurement of lung function

in small children and that X denotes the long-term exposure to NO2 . It is unlikely that
X can be measured exactly; instead, the surrogate W , the amount of NO2 recorded at a
clinic visit, is more likely to be observed. Note that the model above can be rewritten as

Yi = β0 + β1 (Wi − Ui ) + i

= β0 + β1 Wi + (i − β1 Ui ) .
| {z }
= ∗i

Because the Wi ’s are not fixed in advance, we would at least need E(∗i |Wi ) = 0 for this
to be a GM linear model. However, note that

E(∗i |Wi ) = E(i − β1 Ui |Xi + Ui )

= E(i |Xi + Ui ) − β1 E(Ui |Xi + Ui ).

The first term is zero if i is independent of both Xi and Ui . The second term generally
is not zero (unless β1 = 0, of course) because Ui and Xi + Ui are correlated. Therefore,
this can not be a GM model. 

PAGE 12
CHAPTER 2 STAT 714, J. TEBBS

2 The Linear Least Squares Problem

Complementary reading from Monahan: Chapter 2 (except Section 2.4).

INTRODUCTION : Consider the general linear model

Y = Xβ + ,

where Y is an n × 1 vector of observed responses, X is an n × p matrix of fixed constants,

β is a p × 1 vector of fixed but unknown parameters, and  is an n × 1 vector of random
errors. If E() = 0, then
E(Y) = E(Xβ + ) = Xβ.

Since β is unknown, all we really know is that E(Y) = Xβ ∈ C(X). To estimate E(Y),
it seems natural to take the vector in C(X) that is closest to Y.

2.1 Least squares estimation

DEFINITION : An estimate β
b is a least squares estimate of β if Xβ
b is the vector in

C(X) that is closest to Y. In other words, β

b is a least squares estimate of β if

b = arg min (Y − Xβ)0 (Y − Xβ).

β
β∈R p

LEAST SQUARES : Let β = (β1 , β2 , ..., βp )0 and define the error sum of squares

Q(β) = (Y − Xβ)0 (Y − Xβ),

the squared distance from Y to Xβ. The point where Q(β) is minimized satisfies
   
∂Q(β)
0
 ∂β1   
 ∂Q(β)   
∂Q(β)  ∂β2   0 
= 0, or, in other words, 
 ..  =  ..  .
  
∂β  .   . 
   
∂Q(β)
∂βp
0

This minimization problem can be tackled either algebraically or geometrically.

PAGE 13
CHAPTER 2 STAT 714, J. TEBBS

Result 2.1. Let a and b be p × 1 vectors and A be a p × p matrix of constants. Then

∂a0 b ∂b0 Ab
=a and = (A + A0 )b.
∂b ∂b

Proof. See Monahan, pp 14. 

NOTE : In Result 2.1, note that

∂b0 Ab
= 2Ab
∂b

if A is symmetric.

NORMAL EQUATIONS : Simple calculations show that

Q(β) = (Y − Xβ)0 (Y − Xβ)

= Y0 Y − 2Y0 Xβ + β 0 X0 Xβ.

Using Result 2.1, we have

∂Q(β)
= −2X0 Y + 2X0 Xβ,
∂β

because X0 X is symmetric. Setting this expression equal to 0 and rearranging gives

X0 Xβ = X0 Y.

These are the normal equations. If X0 X is nonsingular, then the unique least squares
estimator of β is
b = (X0 X)−1 X0 Y.
β

When X0 X is singular, which can happen in ANOVA models (see Chapter 1), there can
be multiple solutions to the normal equations. Having already proved algebraically that
the normal equations are consistent, we know that the general form of the least squares
solution is
b = (X0 X)− X0 Y + [I − (X0 X)− X0 X]z,
β

for z ∈ Rp , where (X0 X)− is a generalized inverse of X0 X.

PAGE 14
CHAPTER 2 STAT 714, J. TEBBS

2.2 Geometric considerations

CONSISTENCY : Recall that a linear system Ax = c is consistent if there exists an x∗

such that Ax∗ = c; that is, if c ∈ C(A). Applying this definition to

X0 Xβ = X0 Y,

the normal equations are consistent if X0 Y ∈ C(X0 X). Clearly, X0 Y ∈ C(X0 ). Thus, we’ll
be able to establish consistency (geometrically) if we can show that C(X0 X) = C(X0 ).

Result 2.2. N (X0 X) = N (X).

Proof. Suppose that w ∈ N (X). Then Xw = 0 and X0 Xw = 0 so that w ∈ N (X0 X).
Suppose that w ∈ N (X0 X). Then X0 Xw = 0 and w0 X0 Xw = 0. Thus, ||Xw||2 = 0
which implies that Xw = 0; i.e., w ∈ N (X). 

Result 2.3. Suppose that S1 and T1 are orthogonal complements, as well as S2 and T2 .
If S1 ⊆ S2 , then T2 ⊆ T1 .
Proof. See Monahan, pp 244. 

CONSISTENCY : We use the previous two results to show that C(X0 X) = C(X0 ). Take
S1 = N (X0 X), T1 = C(X0 X), S2 = N (X), and T2 = C(X0 ). We know that S1 and
T1 (S2 and T2 ) are orthogonal complements. Because N (X0 X) ⊆ N (X), the last result
guarantees C(X0 ) ⊆ C(X0 X). But, C(X0 X) ⊆ C(X0 ) trivially, so we’re done. Note also

C(X0 X) = C(X0 ) =⇒ r(X0 X) = r(X0 ) = r(X). 

NOTE : We now state a result that characterizes all solutions to the normal equations.

Result 2.4. Q(β) = (Y − Xβ)0 (Y − Xβ) is minimized at β

b if and only if β
b is a solution

to the normal equations.

Proof. (⇐=) Suppose that β
b is a solution to the normal equations. Then,

Q(β) = (Y − Xβ)0 (Y − Xβ)

= (Y − Xβ b − Xβ)0 (Y − Xβ
b + Xβ b − Xβ)
b + Xβ

b 0 (Y − Xβ)
= (Y − Xβ) b − Xβ)0 (Xβ
b + (Xβ b − Xβ),

PAGE 15
CHAPTER 2 STAT 714, J. TEBBS

b − Xβ)0 (Y − Xβ)
since the cross product term 2(Xβ b = 0; verify this using the fact that

β b + z0 z, where
b solves the normal equations. Thus, we have shown that Q(β) = Q(β)
b − Xβ. Therefore, Q(β) ≥ Q(β)
z = Xβ b for all β and, hence, β
b minimizes Q(β). (=⇒)

Now, suppose that β e ≥ Q(β),

e minimizes Q(β). We already know that Q(β) b where
b = (X0 X)− X0 Y, by assumption, but also Q(β)
β e ≤ Q(β)
b because β
e minimizes Q(β).

Thus, Q(β)
e = Q(β).
b But because Q(β) b + z0 z, where z = Xβ
e = Q(β) b − Xβ,
e it must be
b − Xβ
true that z = Xβ e = 0; that is, Xβ
b = Xβ.
e Thus,

X 0 Xβ
e = X 0 Xβ
b = X0 Y,

since β
b is a solution to the normal equations. This shows that β
e is also solution to the

normal equations. 

INVARIANCE : In proving the last result, we have discovered a very important fact;
namely, if β
b and β
e both solve the normal equations, then Xβ
b = Xβ.
e In other words,

Xβ
b is invariant to the choice of β.
b

NOTE : The following result ties least squares estimation to the notion of a perpendicular
projection matrix. It also produces a general formula for the matrix.

Result 2.5. An estimate β

b is a least squares estimate if and only if Xβ
b = MY, where

M is the perpendicular projection matrix onto C(X).

Proof. We will show that

(Y − Xβ)0 (Y − Xβ) = (Y − MY)0 (Y − MY) + (MY − Xβ)0 (MY − Xβ).

Both terms on the right hand side are nonnegative, and the first term does not involve
β. Thus, (Y − Xβ)0 (Y − Xβ) is minimized by minimizing (MY − Xβ)0 (MY − Xβ), the
squared distance between MY and Xβ. This distance is zero if and only if MY = Xβ,
which proves the result. Now to show the above equation:

(Y − Xβ)0 (Y − Xβ) = (Y − MY + MY − Xβ)0 (Y − MY + MY − Xβ)

= (Y − MY)0 (Y − MY) + (Y − MY)0 (MY − Xβ)

| {z }
(∗)

+ (MY − Xβ) (Y − MY) +(MY − Xβ)0 (MY − Xβ).

0
| {z }
(∗∗)

PAGE 16
CHAPTER 2 STAT 714, J. TEBBS

It suffices to show that (∗) and (∗∗) are zero. To show that (∗) is zero, note that

(Y − MY)0 (MY − Xβ) = Y0 (I − M)(MY − Xβ) = [(I − M)Y]0 (MY − Xβ) = 0,

because (I − M)Y ∈ N (X0 ) and MY − Xβ ∈ C(X). Similarly, (∗∗) = 0 as well. 

Result 2.6. The perpendicular projection matrix onto C(X) is given by

M = X(X0 X)− X0 .

b = (X0 X)− X0 Y is a solution to the normal equations, so it is a

Proof. We know that β
least squares estimate. But, by Result 2.5, we know Xβ
b = MY. Because perpendicular

projection matrices are unique, M = X(X0 X)− X0 as claimed. 

NOTATION : Monahan uses PX to denote the perpendicular projection matrix onto

C(X). We will henceforth do the same; that is,

PX = X(X0 X)− X0 .

PROPERTIES : Let PX denote the perpendicular projection matrix onto C(X). Then

(a) PX is idempotent

(b) PX projects onto C(X)

(c) PX is invariant to the choice of (X0 X)−

(d) PX is symmetric

(e) PX is unique.

We have already proven (a), (b), (d), and (e); see Matrix Algebra Review 5. Part (c) must
be true; otherwise, part (e) would not hold. However, we can prove (c) more rigorously.

Result 2.7. If (X0 X)− 0 − 0

1 and (X X)2 are generalized inverses of X X, then

1. X(X0 X)− 0 0 − 0
1 X X = X(X X)2 X X = X

2. X(X0 X)− 0 0 − 0
1 X = X(X X)2 X .

PAGE 17
CHAPTER 2 STAT 714, J. TEBBS

Proof. For v ∈ Rn , let v = v1 + v2 , where v1 ∈ C(X) and v2 ⊥C(X). Since v1 ∈ C(X),

we know that v1 = Xd, for some vector d. Then,

v0 X(X0 X)− 0 0 0 − 0 0 0 0 − 0 0 0 0
1 X X = v1 X(X X)1 X X = d X X(X X)1 X X = d X X = v X,

since v2 ⊥C(X). Since v and (X0 X)−

1 were arbitrary, we have shown the first part. To

show the second part, note that

X(X0 X)− 0 0 − 0 0 − 0 0 − 0
1 X v = X(X X)1 X Xd = X(X X)2 X Xd = X(X X)2 X v.

Since v is arbitrary, the second part follows as well. 

Result 2.8. Suppose X is n × p with rank r ≤ p, and let PX be the perpendicular

projection matrix onto C(X). Then r(PX ) = r(X) = r and r(I − PX ) = n − r.
Proof. Note that PX is n × n. We know that C(PX ) = C(X), so the first part is obvious.
To show the second part, recall that I − PX is the perpendicular projection matrix onto
N (X0 ), so it is idempotent. Thus,

r(I − PX ) = tr(I − PX ) = tr(I) − tr(PX ) = n − r(PX ) = n − r,

because the trace operator is linear and because PX is idempotent as well. 

SUMMARY : Consider the linear model Y = Xβ + , where E() = 0; in what follows,

the cov() = σ 2 I assumption is not needed. We have shown that a least squares estimate
of β is given by
b = (X0 X)− X0 Y.
β

This solution is not unique (unless X0 X is nonsingular). However,

b≡Y
PX Y = X β b

is unique. We call Y b is the point in C(X)

b the vector of fitted values. Geometrically, Y

that is closest to Y. Now, recall that I − PX is the perpendicular projection matrix onto
N (X0 ). Note that
(I − PX )Y = Y − PX Y = Y − Y
b ≡b
e.

PAGE 18
CHAPTER 2 STAT 714, J. TEBBS

We call b e ∈ N (X0 ). Because C(X) and N (X0 ) are

e the vector of residuals. Note that b
orthogonal complements, we know that Y can be uniquely decomposed as

Y=Y
b +b
e.

We also know that Y

b and b
e are orthogonal vectors. Finally, note that

Y0 Y = Y0 IY = Y0 (PX + I − PX )Y

= Y0 PX Y + Y0 (I − PX )Y

= Y0 PX PX Y + Y0 (I − PX )(I − PX )Y
b 0Y
= Y e0 b
b +b e,

since PX and I−PX are both symmetric and idempotent; i.e., they are both perpendicular
projection matrices (but onto orthogonal spaces). This orthogonal decomposition of Y0 Y
is often given in a tabular display called an analysis of variance (ANOVA) table.

ANOVA TABLE : Suppose that Y is n × 1, X is n × p with rank r ≤ p, β is p × 1, and

 is n × 1. An ANOVA table looks like

Source df SS
Model r b 0Y
Y b = Y 0 PX Y

Residual n − r e0 b
b e = Y0 (I − PX )Y
Total n Y0 Y = Y0 IY

It is interesting to note that the sum of squares column, abbreviated “SS,” catalogues
3 quadratic forms, Y0 PX Y, Y0 (I − PX Y), and Y0 IY. The degrees of freedom column,
abbreviated “df,” catalogues the ranks of the associated quadratic form matrices; i.e.,

r(PX ) = r

r(I − PX ) = n − r

r(I) = n.

The quantity Y0 PX Y is called the (uncorrected) model sum of squares, Y0 (I − PX )Y

is called the residual sum of squares, and Y0 Y is called the (uncorrected) total sum of
squares.

PAGE 19
CHAPTER 2 STAT 714, J. TEBBS

NOTE : The following “visualization” analogy is taken liberally from Christensen (2002).

VISUALIZATION : One can think about the geometry of least squares estimation in
three dimensions (i.e., when n = 3). Consider your kitchen table and take one corner of
the table to be the origin. Take C(X) as the two dimensional subspace determined by the
surface of the table, and let Y be any vector originating at the origin; i.e., any point in
R3 . The linear model says that E(Y) = Xβ, which just says that E(Y) is somewhere on
the table. The least squares estimate Y
b = Xβ
b = PX Y is the perpendicular projection

e = (I − PX )Y is the vector
of Y onto the surface of the table. The residual vector b
starting at the origin, perpendicular to the surface of the table, that reaches the same
height as Y. Another way to think of the residual vector is to first connect Y and
PX Y with a line segment (that is perpendicular to the surface of the table). Then,
shift the line segment along the surface (keeping it perpendicular) until the line segment
has one end at the origin. The residual vector b
e is the perpendicular projection of Y
onto C(I − PX ) = N (X0 ); that is, the projection onto the orthogonal complement of the
table surface. The orthogonal complement C(I − PX ) is the one-dimensional space in
the vertical direction that goes through the origin. Once you have these vectors in place,
sums of squares arise from Pythagorean’s Theorem.

A SIMPLE PPM : Suppose Y1 , Y2 , ..., Yn are iid with mean E(Yi ) = µ. In terms of the
general linear model, we can write Y = Xβ + , where
     
Y 1 1
 1     
     
 Y2   1   2 
Y=  .. 
 , X = 1 =   , β = µ,
 ..  =
 ..
.

 .   .   . 
     
Yn 1 n

The perpendicular projection matrix onto C(X) is given by

P1 = 1(10 1)− 10 = n−1 110 = n−1 J,

where J is the n × n matrix of ones. Note that

P1 Y = n−1 JY = Y 1,

PAGE 20
CHAPTER 2 STAT 714, J. TEBBS

Pn
where Y = n−1 i=1 Yi . The perpendicular projection matrix P1 projects Y onto the
space
C(P1 ) = {z ∈ Rn : z = (a, a, ..., a)0 ; a ∈ R}.

Note that r(P1 ) = 1. Note also that

 
Y −Y
 1 
 Y2 − Y
 

(I − P1 )Y = Y − P1 Y = Y − Y 1 = 
 ..
,

 . 
 
Yn − Y

the vector which contains the deviations from the mean. The perpendicular projection
matrix I − P1 projects Y onto
( n
)
X
C(I − P1 ) = z ∈ Rn : z = (a1 , a2 , ..., an )0 ; ai ∈ R, ai = 0 .
i=1

Note that r(I − P1 ) = n − 1.

REMARK : The matrix P1 plays an important role in linear models, and here is why.
Most linear models, when written out in non-matrix notation, contain an intercept
term. For example, in simple linear regression,

Yi = β0 + β1 xi + i ,

or in ANOVA-type models like

Yijk = µ + αi + βj + γij + ijk ,

the intercept terms are β0 and µ, respectively. In the corresponding design matrices, the
first column of X is 1. If we discard the “other” terms like β1 xi and αi + βj + γij in the
models above, then we have a reduced model of the form Yi = µ + i ; that is, a model
that relates Yi to its overall mean, or, in matrix notation Y = 1µ + . The perpendicular
projection matrix onto C(1) is P1 and

2
Y0 P1 Y = Y0 P1 P1 Y = (P1 Y)0 (P1 Y) = nY .

PAGE 21
CHAPTER 2 STAT 714, J. TEBBS

This is the model sum of squares for the model Yi = µ + i ; that is, Y0 P1 Y is the sum of
squares that arises from fitting the overall mean µ. Now, consider a general linear model
of the form Y = Xβ + , where E() = 0, and suppose that the first column of X is 1.
In general, we know that

Y0 Y = Y0 IY = Y0 PX Y + Y0 (I − PX )Y.

Subtracting Y0 P1 Y from both sides, we get

Y0 (I − P1 )Y = Y0 (PX − P1 )Y + Y0 (I − PX )Y.

The quantity Y0 (I−P1 )Y is called the corrected total sum of squares and the quantity
Y0 (PX − P1 )Y is called the corrected model sum of squares. The term “corrected”
is understood to mean that we have removed the effects of “fitting the mean.” This is
important because this is the sum of squares breakdown that is commonly used; i.e.,

Source df SS
Model (Corrected) r−1 Y0 (PX − P1 )Y
Residual n−r Y0 (I − PX )Y
Total (Corrected) n−1 Y0 (I − P1 )Y

In ANOVA models, the corrected model sum of squares Y0 (PX − P1 )Y is often broken
down further into smaller components which correspond to different parts; e.g., orthog-
onal contrasts, main effects, interaction terms, etc. Finally, the degrees of freedom are
simply the corresponding ranks of PX − P1 , I − PX , and I − P1 .

NOTE : In the general linear model Y = Xβ + , the residual vector from the least
e = (I − PX )Y ∈ N (X0 ), so b
squares fit b e0 X = 0; that is, the residuals in a least squares
fit are orthogonal to the columns of X, since the columns of X are in C(X). Note that if
1 ∈ C(X), which is true of all linear models with an intercept term, then
n
X
e0 1 =
b ebi = 0,
i=1

that is, the sum of the residuals from a least squares fit is zero. This is not necessarily
true of models for which 1 ∈
/ C(X).

PAGE 22
CHAPTER 2 STAT 714, J. TEBBS

Result 2.9. If C(W) ⊂ C(X), then PX − PW is the perpendicular projection matrix

onto C[(I − PW )X].
Proof. It suffices to show that (a) PX − PW is symmetric and idempotent and that (b)
C(PX − PW ) = C[(I − PW )X]. First note that PX PW = PW because the columns of
PW are in C(W) ⊂ C(X). By symmetry, PW PX = PW . Now,

(PX − PW )(PX − PW ) = P2X − PX PW − PW PX + P2W

= PX − PW − PW + PW = PX − PW .

Thus, PX − PW is idempotent. Also, (PX − PW )0 = P0X − P0W = PX − PW , so PX − PW

is symmetric. Thus, PX − PW is a perpendicular projection matrix onto C(PX − PW ).
Suppose that v ∈ C(PX − PW ); i.e., v = (PX − PW )d, for some d. Write d = d1 + d2 ,
where d1 ∈ C(X) and d2 ∈ N (X0 ); that is, d1 = Xa, for some a, and X0 d2 = 0. Then,

v = (PX − PW )(d1 + d2 )

= (PX − PW )(Xa + d2 )

= PX Xa + PX d2 − PW Xa − PW d2

= Xa + 0 − PW Xa − 0

= (I − PW )Xa ∈ C[(I − PW )X].

Thus, C(PX − PW ) ⊆ C[(I − PW )X]. Now, suppose that w ∈ C[(I − PW )X]. Then
w = (I − PW )Xc, for some c. Thus,

w = Xc − PW Xc = PX Xc − PW Xc = (PX − PW )Xc ∈ C(PX − PW ).

This shows that C[(I − PW )X] ⊆ C(PX − PW ). 

TERMINOLOGY : Suppose that V is a vector space and that S is a subspace of V; i.e.,

S ⊂ V. The subspace
SV⊥ = {z ∈ V : z⊥S}

is called the orthogonal complement of S with respect to V. If V = Rn , then SV⊥ = S ⊥

is simply referred to as the orthogonal complement of S.

PAGE 23
CHAPTER 2 STAT 714, J. TEBBS

Result 2.10. If C(W) ⊂ C(X), then C(PX − PW ) = C[(I − PW )X] is the orthogonal
complement of C(PW ) with respect to C(PX ); that is,

C(PX − PW ) = C(PW )⊥
C(PX ) .

Proof. C(PX − PW )⊥C(PW ) because (PX − PW )PW = PX PW − P2W = PW − PW = 0.

Because C(PX − PW ) ⊂ C(PX ), C(PX − PW ) is contained in the orthogonal complement
of C(PW ) with respect to C(PX ). Now suppose that v ∈ C(PX ) and v⊥C(PW ). Then,

v = PX v = (PX − PW )v + PW v = (PX − PW )v ∈ C(PX − PW ),

showing that the orthogonal complement of C(PW ) with respect to C(PX ) is contained
in C(PX − PW ). 

REMARK : The preceding two results are important for hypothesis testing in linear
models. Consider the linear models

Y = Xβ +  and Y = Wγ + ,

where C(W) ⊂ C(X). As we will learn later, the condition C(W) ⊂ C(X) implies that
Y = Wγ +  is a reduced model when compared to Y = Xβ + , sometimes called
the full model. If E() = 0, then, if the full model is correct,

E(PX Y) = PX E(Y) = PX Xβ = Xβ ∈ C(X).

Similarly, if the reduced model is correct, E(PW Y) = Wγ ∈ C(W). Note that if

the reduced model Y = Wγ +  is correct, then the full model Y = Xβ +  is also
correct since C(W) ⊂ C(X). Thus, if the reduced model is correct, PX Y and PW Y
are attempting to estimate the same thing and their difference (PX − PW )Y should be
small. On the other hand, if the reduced model is not correct, but the full model is, then
PX Y and PW Y are estimating different things and one would expect (PX − PW )Y to
be large. The question about whether or not to “accept” the reduced model as plausible
thus hinges on deciding whether or not (PX − PW )Y, the (perpendicular) projection of
Y onto C(PX − PW ) = C(PW )⊥
C(PX ) , is large or small.

PAGE 24
CHAPTER 2 STAT 714, J. TEBBS

2.3 Reparameterization

REMARK : For estimation in the general linear model Y = Xβ + , where E() = 0,

we can only learn about β through Xβ ∈ C(X). Thus, the crucial item needed is
PX , the perpendicular projection matrix onto C(X). For convenience, we call C(X) the
estimation space. PX is the perpendicular projection matrix onto the estimation space.
We call N (X0 ) the error space. I − PX is the perpendicular projection matrix onto the
error space.

IMPORTANT : Any two linear models with the same estimation space are really the
same model; the models are said to be reparameterizations of each other. Any two
such models will give the same predicted values, the same residuals, the same ANOVA
table, etc. In particular, suppose that we have two linear models:

Y = Xβ +  and Y = Wγ + .

If C(X) = C(W), then PX does not depend on which of X or W is used; it depends only
on C(X) = C(W). As we will find out, the least-squares estimate of E(Y) is

Y
b = PX Y = Xβ
b = Wb
γ.

IMPLICATION : The β parameters in the model Y = Xβ + , where E() = 0, are

not really all that crucial. Because of this, it is standard to reparameterize linear models
(i.e., change the parameters) to exploit computational advantages, as we will soon see.
The essence of the model is that E(Y) ∈ C(X). As long as we do not change C(X), the
design matrix X and the corresponding model parameters can be altered in a manner
suitable to our liking.

EXAMPLE : Recall the simple linear regression model from Chapter 1 given by

Yi = β0 + β1 xi + i ,

for i = 1, 2, ..., n. Although not critical for this discussion, we will assume that 1 , 2 , ..., n
are uncorrelated random variables with mean 0 and common variance σ 2 > 0. Recall

PAGE 25
CHAPTER 2 STAT 714, J. TEBBS

that, in matrix notation,

     
Y1 1 x1 1
       
     
 Y2   1 x2  β0  2 
Yn×1 =  .  , Xn×2 = 
 
 .. ..
, β 2×1 =  , n×1 = ..
.
 .. 
  
 . .  β1  . 
     
Yn 1 xn n
As long as (x1 , x2 , ..., xn )0 is not a multiple of 1n and at least one xi 6= 0, then r(X) = 2
and (X0 X)−1 exists. Straightforward calculations show that
   
1 x2 x
P
n i xi n
+ (xi −x)2 − (xi −x)2
P P
X0 X =  P 0
P 2  , (X X) = 
−1 i i 
x 1
x
i i i ix − P
(xi −x)2
P
(xi −x)2
.
i i

and  P 
i Yi 
X0 Y =  P .
x Y
i i i

Thus, the (unique) least squares estimator is given by

   
β Y − β1 x
b = (X0 X)−1 X0 Y =  0  =  P
b b
β (xi −x)(Yi −Y )
.
β1
b iP
(xi −x)2 i

For the simple linear regression model, it can be shown (verify!) that the perpendicular
projection matrix PX is given by

PX = X(X0 X)−1 X0
 2

1 −x) 1 −x)(x2 −x) 1 −x)(xn −x)
1
+ P(x(x 2
1
+ (xP
2 ··· 1
+ (xP
2
 n i i −x) n i (xi −x) n i (xi −x) 
−x)2
 n + 1 −x)(x 2 −x) 2 −x)(xn −x)
 1 (xP 1 P(x2 1 (xP
+ ··· +

2 2 2
i (xi −x) n i (xi −x) n i (xi −x)

=  .. .. .. ..
.
.
 
 . . . 
 
2
1 −x)(xn −x) 2 −x)(xn −x) P(xn −x) 2
1
+ (xP (xi −x)2
1
+ (xP
2 ··· 1
+
n i n i (xi −x) n i (x i −x)

A reparameterization of the simple linear regression model Yi = β0 + β1 xi + i is

Yi = γ0 + γ1 (xi − x) + i

or Y = Wγ + , where
     
Y 1 x1 − x 1
 1       
 1 x2 − x
     
 Y2   γ0  2 
Yn×1 = 
 ..  , Wn×2 =  ..
 
..
,
 γ 2×1 =  , n×1 =
 ..
.

 .   . .  γ1  . 
     
Yn 1 xn − x n

PAGE 26
CHAPTER 2 STAT 714, J. TEBBS

To see why this is a reparameterized model, note that if we define

 
1 −x
U= ,
0 1

then W = XU and X = WU−1 (verify!) so that C(X) = C(W). Moreover, E(Y) =

Xβ = Wγ = XUγ. Taking P0 = (X0 X)−1 X0 leads to β = P0 Xβ = P0 XUγ = Uγ; i.e.,
   
β0 γ − γ1 x
β= = 0  = Uγ.
β1 γ1

To find the least-squares estimator for γ in the reparameterized model, observe that
   
1
n 0 0
W0 W =   and (W0 W)−1 =  n .
2 1
P
0 i (xi − x) 0 (xi −x)2
P
i

Note that (W0 W)−1 is diagonal; this is one of the benefits to working with this param-
eterization. The least squares estimator of γ is given by
   
γ
b0 Y
b = (W0 W)−1 W0 Y = 
γ = P ,
(xi −x)(Yi −Y )
γ
b1 iP
(xi −x)2 i

which is different than β.

b However, it can be shown directly (verify!) that the perpen-

dicular projection matrix onto C(W) is

PW = W(W0 W)−1 W0
 2

1 −x) 1 −x)(x2 −x) 1 −x)(xn −x)
1
+ P(x(x −x) 2
1
+ (xP
2 ··· 1
+ (xP
2
 n i i n i (xi −x) n i (xi −x) 
−x)2
 n + 1 −x)(x 2 −x) 2 −x)(xn −x)
 1 (xP 1 P(x2 1 (xP
+ ··· +

(xi −x)2 n i (xi −x)
2 n i (xi −x)
2 
= 
 i
.. .. ... ..
.

 . . . 
 
2
1 −x)(xn −x) 2 −x)(xn −x) P(xn −x) 2
1
+ (xP (xi −x)2
1
+ (xP
2 ··· 1
+
n i n i (xi −x) n i (x i −x)

which is the same as PX . Thus, the fitted values will be the same; i.e., Y
b = PX Y =

Xβ
b = Wb
γ = PW Y, and the analysis will be the same under both parameterizations.

Exercise: Show that the one way fixed effects ANOVA model Yij = µ + αi + ij , for
i = 1, 2, ..., a and j = 1, 2, ..., ni , and the cell means model Yij = µi + ij are reparameter-
izations of each other. Does one parameterization confer advantages over the other?

PAGE 27
CHAPTER 3 STAT 714, J. TEBBS

3 Estimability and Least Squares Estimators

Complementary reading from Monahan: Chapter 3 (except Section 3.9).

3.1 Introduction

REMARK : Estimability is one of the most important concepts in linear models. Consider
the general linear model
Y = Xβ + ,

where E() = 0. In our discussion that follows, the assumption cov() = σ 2 I is not
needed. Suppose that X is n × p with rank r ≤ p. If r = p (as in regression models), then
b = (X0 X)−1 X0 Y. If r < p,
estimability concerns vanish as β is estimated uniquely by β
(a common characteristic of ANOVA models), then β can not be estimated uniquely.
However, even if β is not estimable, certain functions of β may be estimable.

3.2 Estimability

DEFINITIONS :

1. An estimator t(Y) is said to be unbiased for λ0 β iff E{t(Y)} = λ0 β, for all β.

2. An estimator t(Y) is said to be a linear estimator in Y iff t(Y) = c + a0 Y, for

c ∈ R and a = (a1 , a2 , ..., an )0 , ai ∈ R.

3. A function λ0 β is said to be (linearly) estimable iff there exists a linear unbiased

estimator for it. Otherwise, λ0 β is nonestimable.

Result 3.1. Under the model assumptions Y = Xβ + , where E() = 0, a linear

function λ0 β is estimable iff there exists a vector a such that λ0 = a0 X; that is, λ0 ∈ R(X).
Proof. (⇐=) Suppose that there exists a vector a such that λ0 = a0 X. Then, E(a0 Y) =
a0 Xβ = λ0 β, for all β. Therefore, a0 Y is a linear unbiased estimator of λ0 β and hence

PAGE 28
CHAPTER 3 STAT 714, J. TEBBS

λ0 β is estimable. (=⇒) Suppose that λ0 β is estimable. Then, there exists an estimator

c+a0 Y that is unbiased for it; that is, E(c+a0 Y) = λ0 β, for all β. Note that E(c+a0 Y) =
c + a0 Xβ, so λ0 β = c + a0 Xβ, for all β. Taking β = 0 shows that c = 0. Successively
taking β to be the standard unit vectors convinces us that λ0 = a0 X; i.e., λ0 ∈ R(X). 

Example 3.1. Consider the one-way fixed effects ANOVA model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where E(ij ) = 0. Take a = 3 and ni = 2 so that

   
Y11 1 1 0 0
     
 Y12   1 1 0 0 µ
   

     
     
 Y21   1 0 1 0   α1 
Y= 
,
 X= 
 , and β = 
 
.

 Y22   1 0 1 0   α2 
     
   
 Y31   1 0 0 1  α3
   
Y32 1 0 0 1

Note that r(X) = 3, so X is not of full rank; i.e., β is not uniquely estimable. Consider
the following parametric functions λ0 β:

Parameter λ0 λ0 ∈ R(X)? Estimable?

λ01 β = µ λ01 = (1, 0, 0, 0) no no
λ02 β = α1 λ02 = (0, 1, 0, 0) no no
λ03 β = µ + α1 λ03 = (1, 1, 0, 0) yes yes
λ04 β = α1 − α2 λ04 = (0, 1, −1, 0) yes yes
λ05 β = α1 − (α2 + α3 )/2 λ05 = (0, 1, −1/2, −1/2) yes yes

Because λ03 β = µ + α1 , λ04 β = α1 − α2 , and λ05 β = α1 − (α2 + α3 )/2 are (linearly)

estimable, there must exist linear unbiased estimators for them. Note that
 
Y11 + Y12
E(Y 1+ ) = E
2
1 1
= (µ + α1 ) + (µ + α1 ) = µ + α1 = λ03 β
2 2

PAGE 29
CHAPTER 3 STAT 714, J. TEBBS

and that Y 1+ = c + a0 Y, where c = 0 and a0 = (1/2, 1/2, 0, 0, 0, 0). Also,

E(Y 1+ − Y 2+ ) = (µ + α1 ) − (µ + α2 )

= α1 − α2 = λ04 β

and that Y 1+ −Y 2+ = c+a0 Y, where c = 0 and a0 = (1/2, 1/2, −1/2, −1/2, 0, 0). Finally,
  
Y 2+ + Y 3+ 1
E Y 1+ − = (µ + α1 ) − {(µ + α2 ) + (µ + α3 )}
2 2
1
= α1 − (α2 + α3 ) = λ05 β.
2

Note that  
Y 2+ + Y 3+
Y 1+ − = c + a0 Y,
2
where c = 0 and a0 = (1/2, 1/2, −1/4, −1/4, −1/4, −1/4). 

REMARKS :

1. The elements of the vector Xβ are estimable.

2. If λ01 β, λ02 β, ..., λ0k β are estimable, then any linear combination of them; i.e.,
Pk 0
i=1 di λi β, where di ∈ R, is also estimable.

3. If X is n × p and r(X) = p, then R(X) = Rp and λ0 β is estimable for all λ.

DEFINITION : Linear functions λ01 β, λ02 β, ..., λ0k β are said to be linearly independent
if λ1 , λ2 , ..., λk comprise a set of linearly independent vectors; i.e., Λ = (λ1 λ2 · · · λk )
has rank k.

Result 3.2. Under the model assumptions Y = Xβ + , where E() = 0, we can

always find r = r(X) linearly independent estimable functions. Moreover, no collection
of estimable functions can contain more than r linearly independent functions.
Proof. Let ζ 0i denote the ith row of X, for i = 1, 2, ..., n. Clearly, ζ 01 β, ζ 02 β, ..., ζ 0n β are
estimable. Because r(X) = r, we can select r linearly independent rows of X; the corre-
sponding r functions ζ 0i β are linearly independent. Now, let Λ0 β = (λ01 β, λ02 β, ..., λ0k β)0
be any collection of estimable functions. Then, λ0i ∈ R(X), for i = 1, 2, ..., k, and hence

PAGE 30
CHAPTER 3 STAT 714, J. TEBBS

there exists a matrix A such that Λ0 = A0 X. Therefore, r(Λ0 ) = r(A0 X) ≤ r(X) = r.

Hence, there can be at most r linearly independent estimable functions. 

DEFINITION : A least squares estimator of an estimable function λ0 β is λ0 β,

b where
b = (X0 X)− X0 Y is any solution to the normal equations.
β

Result 3.3. Under the model assumptions Y = Xβ + , where E() = 0, if λ0 β is

estimable, then λ0 β
b = λ0 β
e for any two solutions β
b and β
e to the normal equations.

Proof. Suppose that λ0 β is estimable. Then λ0 = a0 X, for some a. From Result 2.5,

λ0 β
b = a0 Xβ
b = a0 PX Y

λ0 β
e = a0 Xβ
e = a0 PX Y.

This proves the result. 

Alternate proof. If β e both solve the normal equations, then X0 X(β
b and β b − β)
e = 0; that
e ∈ N (X0 X) = N (X). If λ0 β is estimable, then λ0 ∈ R(X) ⇐⇒ λ ∈ C(X0 ) ⇐⇒
b −β
is, β
λ⊥N (X). Thus, λ0 (β e = 0; i.e., λ0 β
b − β) b = λ0 β.
e 

IMPLICATION : Least squares estimators of (linearly) estimable functions are invariant

to the choice of generalized inverse used to solve the normal equations.

Example 3.2. In Example 3.1, we considered the one-way fixed effects ANOVA model
Yij = µ + αi + ij , for i = 1, 2, 3 and j = 1, 2. For this model, it is easy to show that
 
6 2 2 2
 
 
0
 2 2 0 0 
XX= 


 2 0 2 0 
 
2 0 0 2

and r(X0 X) = 3. Here are two generalized inverses of X0 X:

   
1
0 0 0 0 − 12 − 12 0
   2 
 0 12 0
 1 1
 −2 1
  
0  0 
(X0 X)−1 =   (X 0
X)−
2 =  2 .
 0 0 12
 1 1
 −2 2
  
0  1 0 
   
1
0 0 0 2
0 0 0 0

PAGE 31
CHAPTER 3 STAT 714, J. TEBBS

Note that
 
Y11
    
1 1 1 1 1 1 Y12 Y + Y12 + Y21 + Y12 + Y31 + Y32
 
  11
 
  
    
0
 1 1 0 0 0 0  Y21   Y11 + Y12 
XY =



=
 
.

 0 0 1 1 0 0  Y22   Y21 + Y22 
    
 
0 0 0 0 1 1  Y31  Y31 + Y32
 
Y32

Two least squares solutions (verify!) are thus

   
0 Y 3+
   
 Y − Y 3+
   
 Y
b = (X0 X)− X0 Y =  1+  e = (X0 X)− X0 Y =  1+
 
β 1 and β 2
.
 Y 2+ − Y 3+
   
 Y 2+  
   
Y 3+ 0

Recall our estimable functions from Example 3.1:

Parameter λ0 λ0 ∈ R(X)? Estimable?

λ03 β = µ + α1 λ03 = (1, 1, 0, 0) yes yes
λ04 β = α1 − α2 λ04 = (0, 1, −1, 0) yes yes
λ05 β = α1 − (α2 + α3 )/2 λ05 = (0, 1, −1/2, −1/2) yes yes

Note that

• for λ3 β = µ + α1 , the (unique) least squares estimator is

λ03 β
b = λ0 β
3
e = Y 1+ .

• for λ04 β = α1 − α2 , the (unique) least squares estimator is

λ04 β
b = λ0 β
4
e = Y 1+ − Y 2+ .

• for λ05 β = α1 − (α2 + α3 )/2, the (unique) least squares estimator is

e = Y 1+ − 1 (Y 2+ + Y 3+ ).
b = λ0 β
λ05 β 5
2

PAGE 32
CHAPTER 3 STAT 714, J. TEBBS

Finally, note that these three estimable functions are linearly independent since
 
1 0 0
 

  1 1 
 1 
Λ = λ3 λ4 λ5 =  
 0 −1 −1/2 
 
 
0 0 −1/2

has rank r(Λ) = 3. Of course, more estimable functions λ0i β can be found, but we can
find no more linearly independent estimable functions because r(X) = 3. 

Result 3.4. Under the model assumptions Y = Xβ + , where E() = 0, the least
squares estimator λ0 β
b of an estimable function λ0 β is a linear unbiased estimator of λ0 β.
b solves the normal equations. We know (by definition) that λ0 β
Proof. Suppose that β b is

the least squares estimator of λ0 β. Note that

λ0 β
b = λ0 {(X0 X)− X0 Y + [I − (X0 X)− X0 X]z}

= λ0 (X0 X)− X0 Y + λ0 [I − (X0 X)− X0 X]z.

Also, λ0 β is estimable by assumption, so λ0 ∈ R(X) ⇐⇒ λ ∈ C(X0 ) ⇐⇒ λ⊥N (X). Re-

sult MAR5.2 says that [I − (X0 X)− X0 X]z ∈ N (X0 X) = N (X), so λ0 [I − (X0 X)− X0 X]z =
0. Thus, λ0 β
b = λ0 (X0 X)− X0 Y, which is a linear estimator in Y. We now show that λ0 β
b

is unbiased. Because λ0 β is estimable, λ0 ∈ R(X) =⇒ λ0 = a0 X, for some a. Thus,

b = E{λ0 (X0 X)− X0 Y} = λ0 (X0 X)− X0 E(Y)

E(λ0 β)

= λ0 (X0 X)− X0 Xβ

= a0 X(X0 X)− X0 Xβ

= a0 PX Xβ = a0 Xβ = λ0 β. 

SUMMARY : Consider the linear model Y = Xβ + , where E() = 0. From the

definition, we know that λ0 β is estimable iff there exists a linear unbiased estimator for
it, so if we can find a linear estimator c+a0 Y whose expectation equals λ0 β, for all β, then
λ0 β is estimable. From Result 3.1, we know that λ0 β is estimable iff λ0 ∈ R(X). Thus,
if λ0 can be expressed as a linear combination of the rows of X, then λ0 β is estimable.

PAGE 33
CHAPTER 3 STAT 714, J. TEBBS

IMPORTANT : Here is a commonly-used method of finding necessary and sufficient

conditions for estimability in linear models with E() = 0. Suppose that X is n × p
with rank r < p. We know that λ0 β is estimable iff λ0 ∈ R(X).

• Typically, when we find the rank of X, we find r linearly independent columns of

X and express the remaining s = p − r columns as linear combinations of the r
linearly independent columns of X. Suppose that c1 , c2 , ..., cs satisfy Xci = 0, for
i = 1, 2, ..., s, that is, ci ∈ N (X), for i = 1, 2, ..., s. If {c1 , c2 , ..., cs } forms a basis
for N (X); i.e., c1 , c2 , ..., cs are linearly independent, then

λ0 c1 = 0

λ0 c2 = 0
..
.

λ0 cs = 0

are necessary and sufficient conditions for λ0 β to be estimable.

REMARK : There are two spaces of interest: C(X0 ) = R(X) and N (X). If X is n × p
with rank r < p, then dim{C(X0 )} = r and dim{N (X)} = s = p − r. Therefore, if
c1 , c2 , ..., cs are linearly independent, then {c1 , c2 , ..., cs } must be a basis for N (X). But,

λ0 β estimable ⇐⇒ λ0 ∈ R(X) ⇐⇒ λ ∈ C(X0 )

⇐⇒ λ is orthogonal to every vector in N (X)

⇐⇒ λ is orthogonal to c1 , c2 , ..., cs

⇐⇒ λ0 ci = 0, i = 1, 2, ..., s.

Therefore, λ0 β is estimable iff λ0 ci = 0, for i = 1, 2, ..., s, where c1 , c2 , ..., cs are s linearly

independent vectors satisfying Xci = 0.

TERMINOLOGY : A set of linear functions {λ01 β, λ02 β, ..., λ0k β} is said to be jointly
nonestimable if the only linear combination of λ01 β, λ02 β, ..., λ0k β that is estimable is
the trivial one; i.e., ≡ 0. These types of functions are useful in non-full-rank linear models
and are associated with side conditions.

PAGE 34
CHAPTER 3 STAT 714, J. TEBBS

3.2.1 One-way ANOVA

GENERAL CASE : Consider the one-way fixed effects ANOVA model Yij = µ + αi + ij ,
for i = 1, 2, ..., a and j = 1, 2, ..., ni , where E(ij ) = 0. In matrix form, X and β are
 
  µ
1n1 1n1 0n1 · · · 0n1  
 α1
   

 1n2 0n2 1n2 · · · 0n2 
   
Xn×p =  . and β p×1 =  α2 ,
 
. . .
 
.
 .. .. .. .. ..  
 ..
 

   . 
1na 0na 0na · · · 1na  
αa
P
where p = a+1 and n = i ni . Note that the last a columns of X are linearly independent
and the first column is the sum of the last a columns. Hence, r(X) = r = a and
s = p − r = 1. With c1 = (1, −10a )0 , note that Xc1 = 0 so {c1 } forms a basis for N (X).
Thus, the necessary and sufficient condition for λ0 β = λ0 µ + ai=1 λi αi to be estimable
P

is
a
X
0
λ c1 = 0 =⇒ λ0 = λi .
i=1

Here are some examples of estimable functions:

1. µ + αi

2. αi − αk
Pa Pa
3. any contrast in the α’s; i.e., i=1 λi αi , where i=1 λi = 0.

Here are some examples of nonestimable functions:

1. µ

2. αi
Pa
3. i=1 ni α i .

There is only s = 1 jointly nonestimable function. Later we will learn that jointly non-
estimable functions can be used to “force” particular solutions to the normal equations.

PAGE 35
CHAPTER 3 STAT 714, J. TEBBS

The following are examples of sets of linearly independent estimable functions (verify!):

1. {µ + α1 , µ + α2 , ..., µ + αa }

2. {µ + α1 , α1 − α2 , ..., α1 − αa }.

LEAST SQUARES ESTIMATES : We now wish to calculate the least squares estimates
of estimable functions. Note that X0 X and one generalized inverse of X0 X is given by
   
n n 1 n2 · · · na 0 0 0 ··· 0
   
 n1 n1 0 ··· 0   0 1/n1 0 ··· 0 
   
   
X0 X =  n 2 0 n 2 0
· · · 0  and (X X) =  0
−
1/n2 · · ·
   
0 0 
 .. .. .. ..   .. .. .. .. 
   
.. ..
 . . . . .   . . . . . 
   
na 0 0 · · · na 0 0 0 · · · 1/na

For this generalized inverse, the least squares estimate is

  P P   
0 0 0 ··· 0 i j Yij 0
  P   
 0 1/n1 0 ··· 0  j Y1j
  Y 1+
    

    
b = (X0 X)− X0 Y = 
β  0 0 1/n2 · · ·

0 
P
 =  Y 2+
 
.

j Y2j
 .. .. .. ..   ..   ..
    
..
.

 . . . .  .   . 
  P   
0 0 0 · · · 1/na j Yaj Y a+

REMARK : We know that this solution is not unique; had we used a different generalized
inverse above, we would have gotten a different least squares estimate of β. However, least
squares estimates of estimable functions λ0 β are invariant to the choice of generalized
inverse, so our choice of (X0 X)− above is as good as any other. From this solution, we
have the unique least squares estimates:

Estimable function, λ0 β Least squares estimate, λ0 β

b

µ + αi Y i+
αi − αk Y i+ − Y k+
Pa Pa Pa
i=1 λi αi , where i=1 λi = 0 i=1 λi Y i+

PAGE 36
CHAPTER 3 STAT 714, J. TEBBS

3.2.2 Two-way crossed ANOVA with no interaction

GENERAL CASE : Consider the two-way fixed effects (crossed) ANOVA model

Yijk = µ + αi + βj + ijk ,

for i = 1, 2, ..., a and j = 1, 2, ..., b, and k = 1, 2, ..., nij , where E(ij ) = 0. For ease of
presentation, we take nij = 1 so there is no need for a k subscript; that is, we can rewrite
the model as Yij = µ + αi + βj + ij . In matrix form, X and β are
 
µ
 
α1 
 

 
α2 
   

1b 1b 0b · · · 0b Ib 
.. 

. 
  

1b 0b 1b · · · 0b Ib
   
 
Xn×p = and β p×1 = αa  ,
 
.. .. .. . . . .

. .. ..
 
 . . .  



   β1 
1b 0b 0b · · · 1b Ib 



 β2 
.. 
 

 . 
 
βb
where p = a + b + 1 and n = ab. Note that the first column is the sum of the last b
columns. The 2nd column is the sum of the last b columns minus the sum of columns 3
through a + 1. The remaining columns are linearly independent. Thus, we have s = 2
linear dependencies so that r(X) = a + b − 1. The dimension of N (X) is s = 2. Taking
   
1 1
   
c1 =  −1a  and c2 =  0a
   

   
0b −1b
produces Xc1 = Xc2 = 0. Since c1 and c2 are linearly independent; i.e., neither is
a multiple of the other, {c1 , c2 } is a basis for N (X). Thus, necessary and sufficient
conditions for λ0 β to be estimable are
a
X
0
λ c1 = 0 =⇒ λ0 = λi
i=1
b
X
λ0 c2 = 0 =⇒ λ0 = λa+j .
j=1

PAGE 37
CHAPTER 3 STAT 714, J. TEBBS

Here are some examples of estimable functions:

1. µ + αi + βj

2. αi − αk

3. βj − βk
Pa Pa
4. any contrast in the α’s; i.e., i=1 λi αi , where i=1 λi = 0
Pb Pb
5. any contrast in the β’s; i.e., j=1 λa+j βj , where j=1 λa+j = 0.

Here are some examples of nonestimable functions:

1. µ

2. αi

3. βj
Pa
4. i=1 αi
Pb
5. j=1 βj .

We can find s = 2 jointly nonestimable functions. Examples of sets of jointly nones-

timable functions are

1. {αa , βb }
P P
2. { i αi , j βj }.

A set of linearly independent estimable functions (verify!) is

1. {µ + α1 + β1 , α1 − α2 , ..., α1 − αa , β1 − β2 , ..., β1 − βb }.

NOTE : When replication occurs; i.e., when nij > 1, for all i and j, our estimability
findings are unchanged. Replication does not change R(X). We obtain the following
least squares estimates:

PAGE 38
CHAPTER 3 STAT 714, J. TEBBS

Estimable function, λ0 β Least squares estimate, λ0 β

b

µ + αi + βj Y ij+
αi − αl Y i++ − Y l++
βj − βl Y +j+ − Y +l+
Pa Pa Pa
i=1 ci αi , with i=1 ci = 0 i=1 ci Y i++
Pb Pb Pb
j=1 di βj , with j=1 di = 0 j=1 di Y +j+

These formulae are still technically correct when nij = 1. When some nij = 0, i.e., there
are missing cells, estimability may be affected; see Monahan, pp 46-48.

3.2.3 Two-way crossed ANOVA with interaction

GENERAL CASE : Consider the two-way fixed effects (crossed) ANOVA model

Yijk = µ + αi + βj + γij + ijk ,

for i = 1, 2, ..., a and j = 1, 2, ..., b, and k = 1, 2, ..., nij , where E(ij ) = 0.

SPECIAL CASE : With a = 3, b = 2, and nij = 2, X and β are

   
1 1 0 0 1 0 1 0 0 0 0 0 µ
   
 1 1 0 0 1 0 1 0 0 0 0 0   α1
   

   
 1 1 0 0 0 1 0 1 0 0 0 0   α2
   

   
 1 1 0 0 0 1 0 1 0 0 0 0   α3
   

   
 1 0 1 0 1 0 0 0 1 0 0 0   β1
   

   
   
 1 0 1 0 1 0 0 0 1 0 0 0   β
 and β =  2

X=   
.

 1 0 1 0 0 1 0 0 0 1 0 0   γ11 
   
   
 1 0 1 0 0 1 0 0 0 1 0 0   γ12 
   
   
 1 0 0 1 1 0 0 0 0 0 1 0   γ21 
   
   
 1 0 0 1 1 0 0 0 0 0 1 0   γ22 
   
   
 1 0 0 1 0 1 0 0 0 0 0 1   γ31 
   
1 0 0 1 0 1 0 0 0 0 0 1 γ32

PAGE 39
CHAPTER 3 STAT 714, J. TEBBS

There are p = 12 parameters. The last six columns of X are linearly independent, and the
other columns can be written as linear combinations of the last six columns, so r(X) = 6
and s = p − r = 6. To determine which functions λ0 β are estimable, we need to find a
basis for N (X). One basis {c1 , c2 , ..., c6 } is
           


 −1 −1 0 0 0 −1 


            


 1   0   −1  0 0 1

            

      


            


 1   0   0  −1 0 1

            

      


           


1   0   0  0 0 0

            

       


            


 0   1   0  0 −1 1

            
      


            


           

 0   1   0   0   0   0 
 ,  ,  ,  ,  ,   .
−1
           


  0   0   1   0   1   

            

            



  0   0   1   0   0   0 


            


−1
            


  0   0   0   1   1   


            

            






   0
0     0 
  

 1 


 0 


 0 




            



  0   0   0   0   1   0 


           


 
 0 0 0 0 0 1 

Functions λ0 β must satisfy λ0 ci = 0, for each i = 1, 2, ..., 6, to be estimable. It should be

obvious that neither the main effect terms nor the interaction terms; i.e, αi , βj , γij , are
estimable on their own. The six αi + βj + γij “cell means” terms are, but these are not
that interesting. No longer are contrasts in the α’s or β’s estimable. Indeed, interaction
makes the analysis more difficult.

3.3 Reparameterization

SETTING: Consider the general linear model

Model GL: Y = Xβ + , where E() = 0.

Assume that X is n × p with rank r ≤ p. Suppose that W is an n × t matrix such

that C(W) = C(X). Then, we know that there exist matrices Tp×t and Sp×t such that

PAGE 40
CHAPTER 3 STAT 714, J. TEBBS

W = XT and X = WS0 . Note that Xβ = WS0 β = Wγ, where γ = S0 β. The model

Model GL-R: Y = Wγ + , where E() = 0,

is called a reparameterization of Model GL.

REMARK : Since Xβ = WS0 β = Wγ = XTγ, we might suspect that the estimation

of an estimable function λ0 β under Model GL should be essentially the same as the
estimation of λ0 Tγ under Model GL-R (and that estimation of an estimable function
q0 γ under Model GL-R should be essentially the same as estimation of q0 S0 β under
Model GL). The upshot of the following results is that, in determining a least squares
estimate of an estimable function λ0 β, we can work with either Model GL or Model
GL-R. The actual nature of these conjectured relationships is now made precise.

Result 3.5. Consider Models GL and GL-R with C(W) = C(X).

1. PW = PX .

b is any solution to the normal equations W0 Wγ = W0 Y associated with Model

2. If γ
GL-R, then β γ is a solution to the normal equations X0 Xβ = X0 Y associated
b = Tb

with Model GL.

3. If λ0 β is estimable under Model GL and if γ

b is any solution to the normal equations
W0 Wγ = W0 Y associated with Model GL-R, then λ0 Tb
γ is the least squares
estimate of λ0 β.

4. If q0 γ is estimable under Model GL-R; i.e., if q0 ∈ R(W), then q0 S0 β is estimable

under Model GL and its least squares estimate is given by q0 γ
b , where γ
b is any
solution to the normal equations W0 Wγ = W0 Y.

Proof.
1. PW = PX since perpendicular projection matrices are unique.
2. Note that
X0 XTb
γ = X0 Wb
γ = X0 PW Y = X0 PX Y = X0 Y.

γ is a solution to the normal equations X0 Xβ = X0 Y.

Hence, Tb

PAGE 41
CHAPTER 3 STAT 714, J. TEBBS

3. This follows from (2), since the least squares estimate is invariant to the choice of the
solution to the normal equations.
4. If q0 ∈ R(W), then q0 = a0 W, for some a. Then, q0 S0 = a0 WS0 = a0 X ∈ R(X), so
that q0 S0 β is estimable under Model GL. From (3), we know the least squares estimate
of q0 S0 β is q0 S0 Tb
γ . But,

γ = a0 WS0 Tb
q0 S0 Tb γ = a0 XTb
γ = a0 Wb
γ = q0 γ
b. 

WARNING: The converse to (4) is not true; i.e., q0 S0 β being estimable under Model GL
doesn’t necessarily imply that q0 γ is estimable under Model GL-R. See Monahan, pp 52.

TERMINOLOGY : Because C(W) = C(X) and r(X) = r, Wn×t must have at least r
columns. If W has exactly r columns; i.e., if t = r, then the reparameterization of
Model GL is called a full rank reparameterization. If, in addition, W0 W is diagonal,
the reparameterization of Model GL is called an orthogonal reparameterization; see,
e.g., the centered linear regression model in Section 2 (notes).

NOTE : A full rank reparameterization always exists; just delete the columns of X that are
linearly dependent on the others. In a full rank reparameterization, (W0 W)−1 exists, so
the normal equations W0 Wγ = W0 Y have a unique solution; i.e., γ
b = (W0 W)−1 W0 Y.

DISCUSSION : There are two (opposing) points of view concerning the utility of full rank
reparameterizations.

• Some argue that, since making inferences about q0 γ under the full rank reparam-
eterized model (Model GL-R) is equivalent to making inferences about q0 S0 β in
the possibly-less-than-full rank original model (Model GL), the inclusion of the
possibility that the design matrix has less than full column rank causes a needless
complication in linear model theory.

• The opposing argument is that, since computations required to deal with the repa-
rameterized model are essentially the same as those required to handle the original
model, we might as well allow for less-than-full rank models in the first place.

PAGE 42
CHAPTER 3 STAT 714, J. TEBBS

• I tend to favor the latter point of view; to me, there is no reason not to include
less-than-full rank models as long as you know what you can and can not estimate.

Example 3.3. Consider the one-way fixed effects ANOVA model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where E(ij ) = 0. In matrix form, X and β are
 
  µ
1n1 1n1 0n1 · · · 0n1  
 α1
   

 1n2 0n2 1n2 · · · 0n2 
   
Xn×p =   and β p×1 =  ,

 .. .. .. .. ..   α2
 . . . . .   ..
 

   . 
1na 0na 0na · · · 1na  
αa
P
where p = a + 1 and n = i ni . This is not a full rank model since the first column is
the sum of the last a columns; i.e., r(X) = a.

Reparameterization 1: Deleting the first column of X, we have

     
1 0n1 · · · 0n1 µ + α1 µ1
 n1     
 0n2 1n2 · · · 0n2 
     
 µ + α2   µ2 
Wn×t =  . and γ = ≡ ,
.. ..  .. ..
  
. .. t×1
.
   
 . . .   .   . 
     
0na 0na · · · 1na µ + αa µa

where t = a and µi = E(Yij ) = µ + αi . This is called the cell-means model and is

written Yij = µi + ij . This is a full rank reparameterization with C(W) = C(X). The
least squares estimate of γ is
 
Y
 1+ 
 
 Y 2+
b = (W0 W)−1 W0 Y = 

γ  ..
.

 . 
 
Y a+

Exercise: What are the matrices T and S associated with this reparameterization?

PAGE 43
CHAPTER 3 STAT 714, J. TEBBS

Reparameterization 2: Deleting the last column of X, we have

   
1 1n1 0n1 · · · 0n1 µ + αa
 n1   
 1n2 0n2 1n2 · · · 0n2   α1 − αa
   

 . .. .. .. 
   
Wn×t =  .. ...
.  and γ t×1 =  α2 − αa ,
 
. .
..
   
· · · 1na−1 
   
 1na−1 0na−1 0na−1  . 
   
1na 0na 0na · · · 0na αa−1 − αa
where t = a. This is called the cell-reference model (what SAS uses by default). This
is a full rank reparameterization with C(W) = C(X). The least squares estimate of γ is
 
Y a+
 
 Y 1+ − Y a+ 
 
 
b = (W0 W)−1 W0 Y =  Y 2+ − Y a+  .
γ
 
..
 
 
 . 
 
Y (a−1)+ − Y a+

Reparameterization 3: Another reparameterization of the effects model uses

   
1 1n1 0n1 · · · 0n1 µ+α
 n1   
 1n2 0n2 1n2 · · · 0n2   α1 − α
   

 . .. .. .. 
   
Wn×t =  .. .. and γ =
.  α2 − α ,
 
. . .  t×1
..
   
 1na−1 0na−1 0na−1 · · · 1na−1 
   
 . 
   
1na −1na −1na · · · −1na αa−1 − α
where t = a and α = a−1 i αi . This is called the deviations from the mean
P
model.
This is a full rank reparameterization with C(W) = C(X).

Example 3.4. Two part multiple linear regression model. Consider the linear model
Y = Xβ + , where E() = 0. Suppose that X is full rank. Write X = (X1 X2 ) and
β = (β 01 , β 02 )0 so that the model can be written as

Model GL: Y = X1 β 1 + X2 β 2 + .

Now, set W1 = X1 and W2 = (I − PX1 )X2 , where PX1 = X1 (X01 X1 )−1 X01 is the
perpendicular projection matrix onto C(X1 ). A reparameterized version of Model GL is

Model GL-R: Y = W1 γ 1 + W2 γ 2 + ,

PAGE 44
CHAPTER 3 STAT 714, J. TEBBS

where E() = 0, W = (W1 W2 ) and

   
0 −1 0
γ1 β + (X1 X1 ) X1 X2 β 2
γ= = 1 .
γ2 β2

With this reparameterization, note that

   
0 0
X1 X 1 0 X1 Y
W0 W =   and W0 Y =  
0 0
0 X2 (I − PX1 )X2 X2 (I − PX1 )Y

so that
   
(X01 X1 )−1 X01 Y γ
b1
b = (W0 W)−1 W0 Y = 
γ ≡ .
{X02 (I − PX1 )X2 }−1 X02 (I − PX1 )Y γ
b2

In this reparameterization, W2 can be thought of as the “residual” from regressing (each

column of) X2 on X1 . A further calculation shows that
   
0 −1 0
β
b b − (X1 X1 ) X1 X2 β 2
γ b
b = (X0 X)−1 X0 Y =  1  =  1
β ,
β
b
2 γ
b 2

where note that (X01 X1 )−1 X01 X2 is the estimate obtained from “regressing” X2 on X1 .
Furthermore, the estimate γ
b 2 can be thought of as the estimate obtained from regressing
Y on W2 = (I − PX1 )X2 .

APPLICATION : Consider the two part full-rank regression model Y = X1 β 1 +X2 β 2 +,
where E() = 0. Suppose that X2 = x2 is n × 1 and that β 2 = β2 is a scalar. Consider
two different models:

Reduced model: Y = X1 β 1 + 

Full model: Y = X1 β 1 + x2 β2 + .

We use the term “reduced model” since C(X1 ) ⊂ C(X1 , x2 ). Consider the full model
Y = X1 β 1 + x2 β2 +  and premultiply by I − PX1 to obtain

(I − PX1 )Y = (I − PX1 )X1 β 1 + b2 (I − PX1 )x2 + (I − PX1 )

= b2 (I − PX1 )x2 + ∗ ,

PAGE 45
CHAPTER 3 STAT 714, J. TEBBS

where ∗ = (I − PX1 ). Now, note that

(I − PX1 )Y = Y − PX1 Y ≡ b
eY|X1 ,

say, are the residuals from regressing Y on X1 . Similarly, (I − PX1 )x2 ≡ b

ex2 |X1 are the
residuals from regressing x2 on X1 . We thus have the following induced linear model

ex2 |X1 + ∗ ,
eY|X1 = b2b
b

where E(∗ ) = 0. The plot of b

eY|X1 versus b
ex2 |X1 is called an added-variable plot (or
partial regression) plot. It displays the relationship between Y and x2 , after adjusting
for the effects of X1 being in the model.

• If a linear trend exists in this plot, this suggests that x2 enters into the (full) model
linearly. This plot can also be useful for detecting outliers and high leverage points.

• On the down side, added-variable plots only look at one predictor at a time so one
can not assess multicolinearity; that is, if the predictor x2 is “close” to C(X1 ), this
may not be detected in the plot.

• The slope of the least squares regression line for the added variable plot is

βb2 = [{(I − PX1 )x2 }0 (I − PX1 )x2 ]−1 {(I − PX1 )x2 }0 (I − PX1 )Y

= {x02 (I − PX1 )x2 }−1 x02 (I − PX1 )Y.

This is equal to the least squares estimate of β2 in the full model.

3.4 Forcing least squares solutions using linear constraints

REVIEW : Consider our general linear model Y = Xβ + , where E() = 0, and X is

an n × p matrix with rank r. The normal equations are X0 Xβ = X0 Y.

b = (X0 X)−1 X0 Y.
• If r = p, then a unique least squares solution exists; i.e., β

b = (X0 X)− X0 Y. This solution is not

• If r < p, then a least squares solution is β
unique; its value depends on which generalized inverse (X0 X)− is used.

PAGE 46
CHAPTER 3 STAT 714, J. TEBBS

Example 3.5. Consider the one-way fixed effects ANOVA model Yij = µ + αi + ij , for
i = 1, 2, ..., a and j = 1, 2, ..., ni , where E(ij ) = 0. The normal equations are
    P P 
n n1 n2 · · · na µ Y
    i j ij 
  P
 n1 n1 0 · · · 0   α1   Y1j 
  
    j 
X0 Xβ =  n2 0 n2 · · · 0   α2  =   = X0 Y,
    P 
j Y 2j
 .. .. .. . . . ..   ..   ..
    

 . . . .  .   . 
    P 
na 0 0 · · · na αa j ajY

or, written another way,

a
X
nµ + ni αi = Y++
i=1
ni µ + ni αi = Yi+ , i = 1, 2, ..., a,
P P P
where Yi+ = j Yij , for i = 1, 2, ..., a, and Y++ = i j Yij . This set of equations has no
unique solution. However, from our discussion on generalized inverses (and consideration
of this model), we know that

• if we set µ = 0, then we get the solution µ bi = Y i+ , for i = 1, 2, ..., a.

b = 0 and α
Pa
• if we set i=1 ni αi = 0, then we get the solution µ bi = Y i+ − Y ++ ,
b = Y ++ and α
for i = 1, 2, ..., a.

• if we set another nonestimable function equal to 0, we’ll get a different solution to

the normal equations.

Pa
REMARK : Equations like µ = 0 and i=1 ni αi = 0 are used to “force” a particular so-
lution to the normal equations and are called side conditions. Different side conditions
produce different least squares solutions. We know that in the one-way ANOVA model,
the parameters µ and αi , for i = 1, 2, ..., a, are not estimable (individually). Imposing
side conditions does not change this. My feeling is that when we attach side conditions to
force a unique solution, we are doing nothing more than solving a mathematical problem
that isn’t relevant. After all, estimable functions λ0 β have least squares estimates that
do not depend on which side condition was used, and these are the only functions we
should ever be concerned with.

PAGE 47
CHAPTER 3 STAT 714, J. TEBBS

REMARK : We have seen similar results for the two-way crossed ANOVA model. In
general, what and how many conditions should we use to “force” a particular solution to
the normal equations? Mathematically, we are interested in imposing additional linear
restrictions of the form Cβ = 0 where the matrix C does not depend on Y.

TERMINOLOGY : We say that the system of equations Ax = q is compatible if

c0 A = 0 =⇒ c0 q = 0; i.e., c ∈ N (A0 ) =⇒ c0 q = 0.

Result 3.6. The system Ax = q is consistent if and only if it is compatible.

Proof. If Ax = q is consistent, then Ax∗ = q, for some x∗ . Hence, for any c such that
c0 A = 0, we have c0 q = c0 Ax∗ = 0, so Ax = q is compatible. If Ax = q is compatible,
then for any c ∈ N (A0 ) = C(I − PA ), we have 0 = c0 q = q0 c = q0 (I − PA )z, for all
z. Successively taking z to be the standard unit vectors, we have q0 (I − PA ) = 0 =⇒
(I − PA )q = 0 =⇒ q = A(A0 A)− A0 q =⇒ q = Ax∗ , where x∗ = (A0 A)− A0 q. Thus,
Ax = q is consistent. 

AUGMENTED NORMAL EQUATIONS : We now consider adjoining the set of equations

Cβ = 0 to the normal equations; that is, we consider the new set of equations
   
X0 X X0 Y
 β =  .
C 0

These are called the augmented normal equations. When we add the constraint
Cβ = 0, we want these equations to be consistent for all Y. We now would like to find
a sufficient condition for consistency. Suppose that w ∈ R(X0 X) ∩ R(C). Note that

w ∈ R(X0 X) =⇒ w = X0 Xv1 , for some v1

w ∈ R(C) =⇒ w = −C0 v2 , for some v2 .

Thus, 0 = w − w = X0 Xv1 + C0 v2

=⇒ 0 = v10 X0 X + v20 C
   
0 0
X X X X
=⇒ 0 = (v10 v20 )   = v0  ,
C C

PAGE 48
CHAPTER 3 STAT 714, J. TEBBS

where v0 = (v10 v20 ). We want C chosen so that

   
0 0
XX XY
 β =  
C 0

is consistent, or equivalently from Result 3.6, is compatible. Compatibility occurs when

   
0 0
XX XY
v0   = 0 =⇒ v0   = 0.
C 0

Thus, we need v10 X0 Y = 0, for all Y. Successively taking Y to be standard unit vectors,
for i = 1, 2, ..., n, convinces us that v10 X0 = 0 ⇐⇒ Xv1 = 0 ⇐⇒ X0 Xv1 = 0 =⇒ w = 0.
Thus, the augmented normal equations are consistent when R(X0 X)∩R(C) = {0}. Since
R(X0 X) = R(X), a sufficient condition for consistency is R(X) ∩ R(C) = {0}. Now,
consider the parametric function λ0 Cβ, for some λ. We know that λ0 Cβ is estimable if
and only if λ0 C ∈ R(X). However, clearly λ0 C ∈ R(C). Thus, λ0 Cβ is estimable if and
only if λ0 Cβ = 0. In other words, writing
 
c01
 
c02
 
 
C=
 ..  ,

 . 
 
0
cs

the set of functions {c01 β, c02 β, ..., c0s β} is jointly nonestimable. Therefore, we can set
a collection of jointly nonestimable functions equal to zero and augment the normal
equations so that they remain consistent. We get a unique solution if
 
0
XX
r  = p.
C

Because R(X0 X) ∩ R(C) = {0},

 
X0 X
p = r  = r(X0 X) + r(C) = r + r(C),
C

showing that we need r(C) = s = p − r.

PAGE 49
CHAPTER 3 STAT 714, J. TEBBS

SUMMARY : To augment the normal equations, we can find a set of s jointly nonestimable
functions {c01 β, c02 β, ..., c0s β} with
 
c01
 
c02
 
 
r(C) = r 
 ..  = s.

 . 
 
c0s

Then,    
0 0
XX XY
 β =  
C 0
is consistent and has a unique solution. 

Example 3.5 (continued). Consider the one-way fixed effects ANOVA model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where E(ij ) = 0. The normal equations are
    P P 
n n1 n2 · · · na µ Y
    i j ij 
  P
 n1 n1 0 · · · 0   α 1   Y1j 
  
    j 
X0 Xβ =  n2 0 n2 · · · 0   α2  =   = X0 Y.
    P 
j Y2j
 .. .. .. . . ..   ..   ..
    
. .  .  

 . . . . 
    P 
na 0 0 · · · na αa j Yaj

We know that r(X) = r = a < p (this system can not be solved uniquely) and that
s = p − r = (a + 1) − a = 1. Thus, to augment the normal equations, we need to find
s = 1 (jointly) nonestimable function. Take c01 = (1, 0, 0, ..., 0), which produces
 
µ
 
 α1 
 
 
0
c1 β = (1 0 0 · · · 0)  α2  = µ.
 
 .. 
 
 . 
 
αa

PAGE 50
CHAPTER 3 STAT 714, J. TEBBS

For this choice of c1 , the augmented normal equations are

   P P 
n n1 n2 · · · na  
i j Yij
  µ  P 
 n1 n1 0 · · · 0   j Y1j
    
  
    α 1     
X0 X 0
P
 n2 0 n2 · · · 0  
    
j Y2j XY
 
β = 
α2  = 
 = .
 .. .. .. . . ..   ..
   
. . 

c1  . . . .
  . 0
  ..  
  
 
P
 na 0 0 · · · na  
    
j Yaj
 
  αa  
1 0 0 ··· 0 0

Solving this (now full rank) system produces the unique solution

µ
b = 0

α
bi = Y i+ i = 1, 2, ..., a.

You’ll note that this choice of c1 used to augment the normal equations corresponds to
specifying the side condition µ = 0. 
Pa
Exercise. Redo this example using (a) the side condition i=1 ni αi = 0, (b) the side
condition αa = 0 (what SAS does), and (c) using another side condition.

Example 3.6. Consider the two-way fixed effects (crossed) ANOVA model

Yij = µ + αi + βj + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., b, where E(ij ) = 0. For purposes of illustration, let’s
take a = b = 3, so that n = ab = 9 and p = a + b + 1 = 7. In matrix form, X and β are
 
1 1 0 0 1 0 0
   
 1 1 0 0 0 1 0  µ
 
   
 1 1 0 0 0 0 1   α1 
   
   
 1 0 1 0 1 0 0   α2 
   
   
X9×7 =  1 0 1 0 0 1 0  and β 7×1 =  α3  .
   
   
   
 1 0 1 0 0 0 1 β 
 1 
 
 
   
 1 0 0 1 1 0 0   β2 
   
 
 1 0 0 1 0 1 0  β3
 
1 0 0 1 0 0 1

PAGE 51
CHAPTER 3 STAT 714, J. TEBBS

We see that r(X) = r = 5 so that s = p − r = 7 − 5 = 2. The normal equations are

    P P 
9 3 3 3 3 3 3 µ i Y
j ij 
    P
 3 3 0 0 1 1 1   α1   Y
    
  Pj 1j 

 
 3 0 3 0 1 1 1   α2   Y2j 
    
    j 
X0 Xβ =  3 0 0 3 1 1 1   α3  =   = X0 Y.
    P 
j Y 3j
    
    P 
 3 1 1 1 3 0 0   β1   i Yi1

    
    P 
 3 1 1 1 0 3 0   β2   i Yi2 
    
P
3 1 1 1 0 0 3 β3 i Yi3

This system does not have a unique solution. To augment the normal equations, we will
need a set of s = 2 linearly independent jointly nonestimable functions. From Section
P P
3.2.2, one example of such a set is { i αi , j βj }. For this choice, our matrix C is
   
0
c1 0 1 1 1 0 0 0
C= = .
0
c2 0 0 0 0 1 1 1

Thus, the augmented normal equations become

   P P 
9 3 3 3 3 3 3 i j Yij
    P 
 3 3 0 0 1 1 1  µ Y
   
  Pj 1j
 
  
 3 0 3 0 1 1 1   α1 Y
    
  Pj 2j
  
  
 3 0 0 3 1 1 1   α2 j Y3j
        
  
0 0
XX    
  XY 

β =  3 1 1

1 3 0 0   α3 =
  P
= .
i Yi1

C 



 
  P

 0
 3 1 1 1 0 3 0   β1   i Yi2

    
    P 
 3 1 1 1 0 0 3   β2   i Yi3

    
   
 0 1 1 1 0 0 0  β3  0 
   
0 0 0 0 1 1 1 0

Solving this system produces the “estimates” of µ, αi and βj under the side conditions
P P
i αi = j βj = 0. These “estimates” are

µ
b = Y ++

bi = Y i+ − Y ++ , i = 1, 2, 3
α

βbj = Y +j − Y ++ , j = 1, 2, 3.

PAGE 52
CHAPTER 3 STAT 714, J. TEBBS

Exercise. Redo this example using (a) the side conditions αa = 0 and βb = 0 (what
SAS does) and (b) using another set of side conditions.

QUESTION : In general, can we give a mathematical form for the particular solution?
Note that we are now solving
   
0 0
XX XY
 β =  ,
C 0

which is equivalent to    
0 0
XX XY
 β =  
0
CC 0
since Cβ = 0 iff C0 Cβ = 0. Thus, any solution to this system must also satisfy

(X0 X + C0 C)β = X0 Y.

But,     
X X
r(X0 X + C0 C) = r (X0 C0 )   = r   = p,
C C
that is, X0 X + C0 C is nonsingular. Hence, the unique solution to the augmented normal
equations must be
b = (X0 X + C0 C)−1 X0 Y.
β

So, imposing s = p − r conditions Cβ = 0, where the elements of Cβ are jointly non-

estimable, yields a particular solution to the normal equations. Finally, note that by
Result 2.5 (notes),

b = X(X0 X + C0 C)−1 X0 Y
Xβ

= PX Y,

which shows that

PX = X(X0 X + C0 C)−1 X0

is the perpendicular projection matrix onto C(X). This shows that (X0 X + C0 C)−1 is a
(non-singular) generalized inverse of X0 X.

PAGE 53
CHAPTER 4 STAT 714, J. TEBBS

4 The Gauss-Markov Model

Complementary reading from Monahan: Chapter 4 (except Section 4.7).

4.1 Introduction

REVIEW : Consider the general linear model Y = Xβ + , where E() = 0.

• A linear estimator t(Y) = c + a0 Y is said to be unbiased for λ0 β if and only if

E{t(Y)} = λ0 β,

for all β. We have seen this implies that c = 0 and λ0 ∈ R(X); i.e., λ0 β is estimable.

• When λ0 β is estimable, it is possible to find several estimators that are unbiased for
λ0 β. For example, in the one-way (fixed effects) ANOVA model Yij = µ + αi + ij ,
with E(ij ) = 0, Y11 , (Y11 + Y12 )/2, and Y 1+ are each unbiased estimators of
λ0 β = µ + α1 (there are others too).

• If λ0 β is estimable, then the (ordinary) least squares estimator λ0 β,

b where β
b is any

solution to the normal equations X0 Xβ = X0 Y, is unbiased for λ0 β. To see this,

recall that
λ0 β
b = λ0 (X0 X)− X0 Y = a0 PX Y,

where λ0 = a0 X, for some a, that is, λ0 ∈ R(X), and PX is the perpendicular

projection matrix onto C(X). Thus,

E(λ0 β)
b = E(a0 PX Y) = a0 PX E(Y) = a0 PX Xβ = a0 Xβ = λ0 β.

GOAL: Among all linear unbiased estimators for λ0 β, we want to find the “best” linear
unbiased estimator in the sense that it has the smallest variance. We will show that
the least squares estimator λ0 β
b is the best linear unbiased estimator (BLUE) of λ0 β,

provided that λ0 β is estimable and cov() = σ 2 I.

PAGE 54
CHAPTER 4 STAT 714, J. TEBBS

4.2 The Gauss-Markov Theorem

Result 4.1. Consider the Gauss-Markov model Y = Xβ + , where E() = 0 and

cov() = σ 2 I. Suppose that λ0 β is estimable and let β
b denote any solution to the normal

equations X0 Xβ = X0 Y. The (ordinary) least squares estimator λ0 β

b is the best linear

unbiased estimator (BLUE) of λ0 β, that is, the variance of λ0 β

b is uniformly less than

that of any other linear unbiased estimator of λ0 β.

Proof. Suppose that θb = c + a0 Y is another linear unbiased estimator of λ0 β. From
Result 3.1, we know that c = 0 and that λ0 = a0 X. Thus, θb = a0 Y, where λ0 = a0 X.
b + (θb − λ0 β).
Now, write θb = λ0 β b Note that

b = var[λ0 β
var(θ) b + (θb − λ0 β)]
b

= var(λ0 β)
b + var(θb − λ0 β)
b +2cov(λ0 β,
b θb − λ0 β).
b
| {z }
≥0

We now show that cov(λ0 β,

b θb − λ0 β)
b = 0. Recalling that λ0 β
b = a0 PX Y, we have

b θb − λ0 β)
cov(λ0 β, b = cov(a0 PX Y, a0 Y − a0 PX Y)

= cov[a0 PX Y, a0 (I − PX )Y]

= a0 PX cov(Y, Y)[a0 (I − PX )]0

= σ 2 Ia0 PX (I − PX )a = 0,

since PX (I − PX ) = 0. Thus, var(θ) b showing that λ0 β

b ≥ var(λ0 β), b has variance no larger
b Equality results when var(θb − λ0 β)
than that of θ. b = 0. However, if var(θb − λ0 β)
b = 0,

then because E(θb − λ0 β)

b = 0 as well, θb − λ0 β
b is a degenerate random variable at 0; i.e.,

pr(θb = λ0 β)
b = 1. This establishes uniqueness. 

MULTIVARIATE CASE : Suppose that we wish to estimate simultaneously k estimable

linear functions  
λ01 β
 
 0 
0
 λ2 β 
Λβ =
 ..
,

 . 
 
λ0k β

PAGE 55
CHAPTER 4 STAT 714, J. TEBBS

where λ0i = a0i X, for some ai ; i.e., λ0i ∈ R(X), for i = 1, 2, ..., k. We say that Λ0 β is
estimable if and only if λ0i β, i = 1, 2, ..., k, are each estimable. Put another way, Λ0 β is
estimable if and only if Λ0 = A0 X, for some matrix A.

Result 4.2. Consider the Gauss-Markov model Y = Xβ + , where E() = 0 and

cov() = σ 2 I. Suppose that Λ0 β is any k-dimensional estimable vector and that c+A0 Y is
any vector of linear unbiased estimators of the elements of Λ0 β. Let β
b denote any solution

to the normal equations. Then, the matrix cov(c + A0 Y) − cov(Λ0 β)

b is nonnegative

definite.
Proof. It suffices to show that x0 [cov(c + A0 Y) − cov(Λ0 β)]x
b ≥ 0, for all x. Note that

x0 [cov(c + A0 Y) − cov(Λ0 β)]x

b = x0 cov(c + A0 Y)x − x0 cov(Λ0 β)x
b

= var(x0 c + x0 A0 Y) − var(x0 Λ0 β).

b

But x0 Λ0 β = x0 A0 Xβ (a scalar) is estimable since x0 A0 X ∈ R(X). Also, x0 c + x0 A0 Y =

x0 (c+A0 Y) is a linear unbiased estimator of x0 Λ0 β. The least squares estimator of x0 Λ0 β
b Thus, by Result 4.1, var(x0 c + x0 A0 Y) − var(x0 Λ0 β)
is x0 Λ0 β. b ≥ 0. 

OBSERVATION : Consider the Gauss-Markov linear model Y = Xβ +, where E() = 0

and cov() = σ 2 I. If X is full rank, then X0 X is nonsingular and every linear combination
of λ0 β is estimable. The (ordinary) least squares estimator of β is β
b = (X0 X)−1 X0 Y. It

is unbiased and

b = cov[(X0 X)−1 X0 Y] = (X0 X)−1 X0 cov(Y)[(X0 X)−1 X0 ]0

cov(β)

= (X0 X)−1 X0 σ 2 IX(X0 X)−1 = σ 2 (X0 X)−1 .

Note that this is not correct if X is less than full rank.

Example 4.1. Recall the simple linear regression model

Yi = β0 + β1 xi + i ,

for i = 1, 2, ..., n, where 1 , 2 , ..., n are uncorrelated random variables with mean 0 and
common variance σ 2 > 0 (these are the Gauss Markov assumptions). Recall that, in

PAGE 56
CHAPTER 4 STAT 714, J. TEBBS

matrix notation,
     
Y1 1 x1 1
       
     
 Y2   1 x2  β0  2 
Y=
 ..
,
 X=
 .. ..
,
 β= , =
 ..
.

 .   . .  β1  . 
     
Yn 1 xn n

The least squares estimator of β is

   
βb0 Y − βb1 x
b = (X0 X)−1 X0 Y = 
β = P .
(xi −x)(Yi −Y )
βb1 iP
2
i (xi −x)

The covariance matrix of β

b is
 
1 2
+ P x 2 −P x
2
b = σ 2 (X0 X)−1 = σ 2  n i (xi −x) i (xi −x)
cov(β) .
− P (xxi −x)2 P 1 2
i i (xi −x)

4.3 Estimation of σ 2 in the GM model

REVIEW : Consider the Gauss-Markov model Y = Xβ+, where E() = 0 and cov() =
σ 2 I. The best linear unbiased estimator (BLUE) for any estimable function λ0 β is λ0 β,
b

where β
b is any solution to the normal equations. Clearly, E(Y) = Xβ is estimable and

the BLUE of E(Y) is

b = X(X0 X)−1 X0 Y = PX Y = Y,
Xβ b

the perpendicular projection of Y onto C(X); that is, the fitted values from the least
squares fit. The residuals are given by

e=Y−Y
b b = Y − PX Y = (I − PX )Y,

the perpendicular projection of Y onto N (X0 ). Recall that the residual sum of squares
is
b 0 (Y − Xβ)
b = (Y − Xβ)
Q(β) e0 b
b =b e = Y0 (I − PX )Y.

We now turn our attention to estimating σ 2 .

PAGE 57
CHAPTER 4 STAT 714, J. TEBBS

Result 4.3. Suppose that Z is a random vector with mean E(Z) = µ and covariance
matrix cov(Z) = Σ. Let A be nonrandom. Then

E(Z0 AZ) = µ0 Aµ + tr(AΣ).

Proof. Note that Z0 AZ is a scalar random variable; hence, Z0 AZ = tr(Z0 AZ). Also, recall
that expectation E(·) and tr(·) are linear operators. Finally, recall that tr(AB) = tr(BA)
for conformable A and B. Now,

E(Z0 AZ) = E[tr(Z0 AZ)] = E[tr(AZZ0 )]

= tr[AE(ZZ0 )]

= tr[A(Σ + µµ0 )]

= tr(AΣ) + tr(Aµµ0 )

= tr(AΣ) + tr(µ0 Aµ) = tr(AΣ) + µ0 Aµ. 

REMARK : Finding var(Z0 AZ) is more difficult; see Section 4.9 in Monahan. Consider-
able simplification results when Z follows a multivariate normal distribution.

APPLICATION : We now find an unbiased estimator of σ 2 under the GM model. Suppose

that Y is n × 1 and X is n × p with rank r ≤ p. Note that E(Y) = Xβ. Applying Result
4.3 directly with A = I − PX , we have

E[Y0 (I − PX )Y] = (Xβ)0 (I − PX )Xβ +tr[(I − PX )σ 2 I]

| {z }
= 0
2
= σ [tr(I) − tr(PX )]

= σ 2 [n − r(PX )] = σ 2 (n − r).

Thus,
b2 = (n − r)−1 Y0 (I − PX )Y
σ

is an unbiased estimator of σ 2 in the GM model. In non-matrix notation,

n
X
b2 = (n − r)−1
σ (Yi − Ybi )2 ,
i=1

where Ybi is the least squares fitted value of Yi .

PAGE 58
CHAPTER 4 STAT 714, J. TEBBS

ANOVA: Consider the Gauss-Markov model Y = Xβ + , where E() = 0 and cov() =

σ 2 I. Suppose that Y is n × 1 and X is n × p with rank r ≤ p. Recall from Chapter 2
the basic form of an ANOVA table (with corrected sums of squares):

Source df SS MS F
Model (Corrected) r−1 SSR = Y0 (PX − P1 )Y MSR = SSR
r−1
F = MSR
MSE

Residual n−r SSE = Y0 (I − PX )Y MSE = SSE

n−r

Total (Corrected) n−1 SST = Y0 (I − P1 )Y

NOTES :

• The degrees of freedom associated with each SS is the rank of its appropriate
perpendicular projection matrix; that is, r(PX −P1 ) = r −1 and r(I−PX ) = n−r.

• Note that

cov(Y,
b be) = cov[PX Y, (I − PX )Y] = PX σ 2 I(I − PX ) = 0.

That is, the least squares fitted values are uncorrelated with the residuals.

• We have just shown that E(MSE) = σ 2 . If Xβ ∈

/ C(1), that is, the independent
variables in X add to the model (beyond an intercept, for example), then

E(SSR) = E[Y0 (PX − P1 )Y] = (Xβ)0 (PX − P1 )Xβ + tr[(PX − P1 )σ 2 I]

= (Xβ)0 Xβ + σ 2 r(PX − P1 )

= (Xβ)0 Xβ + (r − 1)σ 2 .

Thus,
E(MSR) = (r − 1)−1 E(SSR) = σ 2 + (r − 1)−1 (Xβ)0 Xβ.

• If Xβ ∈ C(1), that is, the independent variables in X add nothing to the model,
then (Xβ)0 (PX − P1 )Xβ = 0 and MSR and MSE are both unbiased estimators
of σ 2 . If this is true, F should be close to 1. Large values of F occur when
(Xβ)0 (PX − P1 )Xβ is large, that is, when Xβ is “far away” from C(1), that is,
when the independent variables in X are more relevant in explaining E(Y).

PAGE 59
CHAPTER 4 STAT 714, J. TEBBS

4.4 Implications of model selection

REMARK : Consider the linear model Y = Xβ + , where Y is n × 1 and X is n × p with

rank r ≤ p. We now investigate two issues: underfitting (model misspecification) and
overfitting. We assume that E() = 0 and cov() = σ 2 I, that is, our usual Gauss-Markov
assumptions.

4.4.1 Underfitting (Misspecification)

UNDERFITTING: Suppose that, in truth, the “correct” model for Y is

Y = Xβ + Wδ + ,

where E() = 0 and cov() = σ 2 I. The vector η = Wδ includes the variables and
coefficients missing from Xβ. If the analyst uses Y = Xβ +  to describe the data,
s/he is missing important variables that are in W, that is, the analyst is misspecifying
the true model by underfitting. We now examine the effect of underfitting on (a) least
squares estimates of estimable functions and (b) the estimate of the error variance σ 2 .

CONSEQUENCES : Suppose that λ0 β is estimable under Y = Xβ + , where E() = 0

and cov() = σ 2 I; i.e., λ0 = a0 X, for some vector a. The least squares estimator of λ0 β
b = λ0 (X0 X)− X0 Y. If Wδ = 0, then Y = Xβ +  is the correct model
is given by λ0 β
and E(λ0 β)
b = λ0 β. If Wδ 6= 0, then, under the correct model,

E(λ0 β)
b = E[λ0 (X0 X)− X0 Y] = λ0 (X0 X)− X0 E(Y)

= λ0 (X0 X)− X0 (Xβ + Wδ)

= a0 X(X0 X)− X0 Xβ + a0 X(X0 X)− X0 Wδ

= a0 PX Xβ + a0 PX Wδ

= λ0 β + a0 PX Wδ,

showing that λ0 β
b is no longer unbiased, in general. The amount of the bias depends on

where Wδ is. If η = Wδ is orthogonal to C(X), then PX Wδ = 0 and the estimation of

λ0 β with λ0 β
b is unaffected. Otherwise, PX Wδ 6= 0 and the estimate of λ0 β is biased.

PAGE 60
CHAPTER 4 STAT 714, J. TEBBS

CONSEQUENCES : Now, let’s turn to the estimation of σ 2 . Under the correct model,

E[Y0 (I − PX )Y] = (Xβ + Wδ)0 (I − PX )(Xβ + Wδ) + tr[(I − PX )σ 2 I]

= (Wδ)0 (I − PX )Wδ + σ 2 (n − r),

where r = r(X). Thus,

E(MSE) = σ 2 + (n − r)−1 (Wδ)0 (I − PX )Wδ,

b2 = MSE is unbiased if and only if Wδ ∈ C(X).

that is, σ

4.4.2 Overfitting

OVERFITTING: Suppose that, in truth, the correct model for Y is

Y = X1 β 1 + ,

where E() = 0 and cov() = σ 2 I, but, instead, we fit

Y = X1 β 1 + X2 β 2 + ,

that is, the extra variables in X2 are not needed; i.e., β 2 = 0. Set X = [X1 X2 ] and
suppose that X and X1 have full column rank (i.e., a regression setting). The least
e 1 = (X0 X1 )−1 X0 Y. We know that
squares estimator of β 1 under the true model is β 1 1

E(β
e 1) = β1

e ) = σ 2 (X0 X1 )−1 .
cov(β 1 1

On the other hand, the normal equations associated with the larger (unnecessarily large)
model are
    
X01 X1 X01 X2 β1 X01 Y
X0 Xβ = X0 Y ⇐⇒   = 
X02 X1 X02 X2 β2 X02 Y

and the least squares estimator of β is

   −1  
0 0 0
β X X X X X Y
b= 1 = 1 1 1 2
b
β   1 .
β
b2 X02 X1 X02 X2 X02 Y

PAGE 61
CHAPTER 4 STAT 714, J. TEBBS

The least squares estimator β

b is still unbiased in the unnecessarily large model, that is,

E(β
b ) = β
1 1

E(β
b ) = 0.
2

Thus, we assess the impact of overfitting by looking at cov(β

b ). Under the unnecessarily
1

large model,

b ) = σ 2 (X0 X1 )−1 + σ 2 (X0 X1 )−1 X0 X2 [X0 (I − PX )X2 ]−1 X0 X1 (X0 X1 )−1 ;

cov(β 1 1 1 1 2 1 2 1

see Exercise A.72 (pp 268) in Monahan. Thus,

e 1 ) = σ 2 (X0 X1 )−1 X0 X2 [X0 (I − PX )X2 ]−1 X0 X1 (X0 X1 )−1 .

b 1 ) − cov(β
cov(β 1 1 2 1 2 1

• If the columns of X2 are each orthogonal to C(X1 ), then X01 X2 = 0 and

   
0 0 0
X1 X1 X1 X2 XX 0
X0 X =  = 1 1 ;
0 0 0
X2 X1 X2 X2 0 X 2 X2

b = (X0 X1 )−1 X0 Y = β
i.e., X0 X is block diagonal, and β 1 1 1
e . This would mean that
1

using the unnecessarily large model has no effect on our estimate of β 1 . However,
the precision with which we can estimate σ 2 is affected since r(I−PX ) < r(I−PX1 );
that is, we have fewer residual degrees of freedom.

• If the columns of X2 are not all orthogonal to C(X1 ), then

b ) 6= σ 2 (X0 X1 )−1 .
cov(β 1 1

Furthermore, as X2 gets “closer” to C(X1 ), then X02 (I − PX1 )X2 gets “smaller.”
This makes [X02 (I − PX1 )X2 ]−1 “larger.” This makes cov(β
b ) “larger.”
1

• Multicollinearity occurs when X2 is “close” to C(X1 ). Severe multicollinearity

can greatly inflate the variances of the least squares estimates. In turn, this can
have a deleterious effect on inference (e.g., confidence intervals too wide, hypothesis
tests with no power, predicted values with little precision, etc.). Various diagnostic
measures exist to assess multicollinearity (e.g., VIFs, condition numbers, etc.); see
the discussion in Monahan, pp 80-82.

PAGE 62
CHAPTER 4 STAT 714, J. TEBBS

4.5 The Aitken model and generalized least squares

TERMINOLOGY : The general linear model

Y = Xβ + ,

where E() = 0 and cov() = σ 2 V, V known, is called the Aitken model. It is

more flexible than the Guass-Markov (GM) model, because the analyst can incorporate
different correlation structures with the observed responses. The GM model is a special
case of the Aitken model with V = I; that is, where responses are uncorrelated.

REMARK : In practice, V is rarely known; rather, it must be estimated. We will discuss

this later. We assume that V is positive definite (pd), and, hence nonsingular, for reasons
that will soon be obvious. Generalizations are possible; see Christensen (Chapter 10).

RECALL: Because V is symmetric, we can write V in its Spectral Decomposition; i.e.,

V = QDQ0 ,

where Q is orthogonal and D is the diagonal matrix consisting of λ1 , λ2 , ..., λn , the eigen-
values of V. Because V is pd, we know that λi > 0, for each i = 1, 2, ..., n. The symmetric
square root of V is
V1/2 = QD1/2 Q0 ,
√ √ √
where D1/2 = diag( λ1 , λ2 , ..., λn ). Note that V1/2 V1/2 = V and that V−1 =
V−1/2 V−1/2 , where
V−1/2 = QD−1/2 Q0
√ √ √
and D−1/2 = diag(1/ λ1 , 1/ λ2 , ..., 1/ λn ).

TRANSFORMATION : Consider the Aitken model Y = Xβ + , where E() = 0 and

cov() = σ 2 V, where V is known. Premultiplying by V−1/2 , we get

V−1/2 Y = V−1/2 Xβ + V−1/2 

⇐⇒ Y∗ = Uβ + ∗ ,

PAGE 63
CHAPTER 4 STAT 714, J. TEBBS

where Y∗ = V−1/2 Y, U = V−1/2 X, and ∗ = V−1/2 . It is easy to show that Y∗ =

Uβ + ∗ is now a GM model. To see this, note that

E(∗ ) = V−1/2 E() = V−1/2 0 = 0

and
cov(∗ ) = V−1/2 cov()V−1/2 = V−1/2 σ 2 VV−1/2 = σ 2 I.

Note also that R(X) = R(U), because V−1/2 is nonsingular. This means that λ0 β is
estimable in the Aitken model if and only if λ0 β is estimable in the transformed GM
model. The covariance structure on  does not affect estimability.

AITKEN EQUATIONS : In the transformed model, the normal equations are

U0 Uβ = U0 Y∗ .

However, note that

U0 Uβ = U0 Y∗ ⇐⇒ (V−1/2 X)0 V−1/2 Xβ = (V−1/2 X)0 V−1/2 Y

⇐⇒ X0 V−1 Xβ = X0 V−1 Y

in the Aitken model. The equations

X0 V−1 Xβ = X0 V−1 Y

are called the Aitken equations. These should be compared with the normal equations

X0 Xβ = X0 Y

in the GM model. In general, we will denote by β

b GLS and β
b OLS the solutions to the Aitken

and normal equations, respectively. “GLS” stands for generalized least squares. “OLS”
stands for ordinary least squares.

GENERALIZED LEAST SQUARES : Any solution β

b GLS to the Aitken equations is called

a generalized least squares (GLS) estimator of β. It is not necessarily unique (unless

X is full rank). The solution β
b
GLS minimizes

Q∗ (β) = (Y∗ − Uβ)0 (Y∗ − Uβ) = (Y − Xβ)0 V−1 (Y − Xβ).

PAGE 64
CHAPTER 4 STAT 714, J. TEBBS

When X is full rank, the unique GLS estimator is

0 −1
β
b
GLS = (X V X)−1 X0 V−1 Y.

When X is not full rank, a GLS estimator is

b GLS = (X0 V−1 X)− X0 V−1 Y.

β

NOTE : If V is diagonal; i.e., V = diag(v1 , v2 , ..., vn ), then

n
X
∗ 0 −1
Q (β) = (Y − Xβ) V (Y − Xβ) = wi (Yi − x0i β)2 ,
i=1

where wi = 1/vi and x0i is the ith row of X. In this situation, β

b
GLS is called the weighted

least squares estimator.

Result 4.4. Consider the Aitken model Y = Xβ+, where E() = 0 and cov() = σ 2 V,
where V is known. If λ0 β is estimable, then λ0 β
b 0
GLS is the BLUE for λ β.

Proof. Applying the Gauss-Markov Theorem to the transformed model Y∗ = Uβ + ∗ ,

b GLS is the BLUE of λ0 β among all linear unbiased estimators
the GLS estimator λ0 β
involving Y∗ = V−1/2 Y. However, any linear estimator in Y can be obtained from Y∗
because V−1/2 is invertible. Thus, λ0 β
b
GLS is the BLUE. 

REMARK : If X is full rank, then estimability concerns vanish (as in the GM model) and
b GLS = (X0 V−1 X)−1 X0 V−1 Y is unique. In this case, straightforward calculations show
β
that E(β
b ) = β and
GLS

b ) = σ 2 (X0 V−1 X)−1 .

cov(β GLS

Example 4.2. Heteroscedastic regression through the origin. Consider the regression
model Yi = βxi + i , for i = 1, 2, ..., n, where E(i ) = 0, var(i ) = σ 2 g 2 (xi ), for some real
function g(·), and cov(i , j ) = 0, for i 6= j. For this model,
     
Y x g 2 (x1 ) 0 ··· 0
 1   1   
g 2 (x2 ) ···
     
 Y2  x   0 0
 , X =  2  , and V = 
 
Y=  ..  ..  .. .. .. ..
.
.
  
 .   .   . . . 
     
Yn xn 0 0 ··· g 2 (xn )

PAGE 65
CHAPTER 4 STAT 714, J. TEBBS

The OLS estimator of β = β is given by

Pn
0 −1 0 xi Y i
βOLS = (X X) X Y = Pi=1
b n 2
.
i=1 xi

The GLS estimator of β = β is given by

Pn 2
0 −1 i=1 xi Yi /g (xi )
−1 0 −1
βbGLS = (X V X) X V Y = P n 2 2
.
i=1 xi /g (xi )

Which one is better? Both of these estimators are unbiased, so we turn to the variances.
Straightforward calculations show that
Pn
σ2 2 2
i=1 xi g (xi )
var(βbOLS ) = 2
( ni=1 x2i )
P
2
σ
var(βbGLS ) = Pn 2 2
.
i=1 xi /g (xi )

We are thus left to compare

Pn 2 2
i=1 xi g (xi ) 1
2 with Pn 2
.
( ni=1 x2i ) 2
i=1 xi /g (xi )
P

Write x2i = ui vi , where ui = xi g(xi ) and vi = xi /g(xi ). Applying Cauchy-Schwartz’s

inequality, we get
n
!2 n
!2 n n n n
X X X X X X
x2i = ui vi ≤ u2i vi2 = x2i g 2 (xi ) x2i /g 2 (xi ).
i=1 i=1 i=1 i=1 i=1 i=1

Thus, Pn 2 2
1 i=1 xi g (xi )
Pn 2 2 ≤ 2 =⇒ var(βbGLS ) ≤ var(βbOLS ).
i=1 xi /g (xi ) ( ni=1 x2i )
P

This result should not be surprising; after all, we know that βbGLS is BLUE. 

Result 4.5. An estimate β

b is a generalized least squares estimate if and only if Xβ
b=

AY, where A = X(X0 V−1 X)− X0 V−1 .

Proof. The GLS estimate; i.e., the OLS estimate in the transformed model Y∗ = Uβ+∗ ,
where Y∗ = V−1/2 Y, U = V−1/2 X, and ∗ = V−1/2 , satisfies

V−1/2 X[(V−1/2 X)0 V−1/2 X]− (V−1/2 X)0 V−1/2 Y = V−1/2 Xβ,
b

by Result 2.5. Multiplying through by V1/2 and simplifying gives the result. 

PAGE 66
CHAPTER 4 STAT 714, J. TEBBS

Result 4.6. A = X(X0 V−1 X)− X0 V−1 is a projection matrix onto C(X).
Proof. We need to show that

(a) A is idempotent

(b) Aw ∈ C(X), for any w

(c) Az = z, for all z ∈ C(X).

The perpendicular projection matrix onto C(V−1/2 X) is

V−1/2 X[(V−1/2 X)0 V−1/2 X]− (V−1/2 X)0 ,

which implies that

V−1/2 X[(V−1/2 X)0 V−1/2 X]− (V−1/2 X)0 V−1/2 X = V−1/2 X.

This can also be written as

V−1/2 AX = V−1/2 X.

Premultiplying by V1/2 gives AX = X. Thus,

AA = AX(X0 V−1 X)− X0 V−1 = X(X0 V−1 X)− X0 V−1 = A,

showing that A is idempotent. To show (b), note Aw = X(X0 V−1 X)− X0 V−1 w ∈ C(X).
To show (c), it suffices to show C(A) = C(X). But, A = X(X0 V−1 X)− X0 V−1 implies
that C(A) ⊂ C(X) and AX = X implies that C(X) ⊂ C(A). 

Result 4.7. In the Aitken model, if C(VX) ⊂ C(X), then the GLS and OLS estimates
will be equal; i.e., OLS estimates will be BLUE in the Aitken model.
Proof. The proof proceeds by showing that A = X(X0 V−1 X)− X0 V−1 is the perpendicular
projection matrix onto C(X) when C(VX) ⊂ C(X). We already know that A is a
projection matrix onto C(X). Thus, all we have to show is that if w⊥C(X), then Aw = 0.
If V is nonsingular, then r(VX) = r(X). The only way this and C(VX) ⊂ C(X) holds
is if C(VX) = C(X), in which case VXB1 = X and VX = XB2 , for some matrices
B1 and B2 . Multiplying through by V−1 gives XB1 = V−1 X and X = V−1 XB2 .
Thus, C(V−1 X) = C(X) and C(V−1 X)⊥ = C(X)⊥ . If w⊥C(X), then w⊥C(V−1 X); i.e.,
w ∈ N (X0 V−1 ). Since Aw = X(X0 V−1 X)−1 X0 V−1 w = 0, we are done. 

PAGE 67
CHAPTER 5 STAT 714, J. TEBBS

5 Distributional Theory

Complementary reading from Monahan: Chapter 5.

5.1 Introduction

PREVIEW : Consider the Gauss-Markov linear model Y = Xβ + , where Y is n × 1, X

is an n × p with rank r ≤ p, β is p × 1, and  is n × 1 with E() = 0 and cov() = σ 2 I.
In addition to the first two moment assumptions, it is common to assume that  follows
a multivariate normal distribution. This additional assumption allows us to formally
pursue various questions dealing with inference. In addition to the multivariate normal
distribution, we will also examine noncentral distributions and quadratic forms.

RECALL: If Z ∼ N (0, 1), then the probability density function (pdf) of Z is given by

1 2
fZ (z) = √ e−z /2 I(z ∈ R).
2π
The N (µ, σ 2 ) family is a location-scale family generated by the standard density fZ (z).

TERMINOLOGY : The collection of pdfs

   
1 x−µ
LS(f ) = fX (·|µ, σ) : fX (x|µ, σ) = fZ ; µ ∈ R, σ > 0
σ σ

is a location-scale family generated by fZ (z); see Casella and Berger, Chapter 3. That
is, if Z ∼ fZ (z), then
 
1 x−µ
X = σZ + µ ∼ fX (x|µ, σ) = fZ .
σ σ

APPLICATION : With the standard normal density fZ (z), it is easy to see that
 
1 x−µ 1 1 2
fX (x|µ, σ) = fZ =√ e− 2σ2 (x−µ) I(x ∈ R).
σ σ 2πσ
That is, any normal random variable X ∼ N (µ, σ 2 ) may be obtained by transforming
Z ∼ N (0, 1) via X = σZ + µ.

PAGE 68
CHAPTER 5 STAT 714, J. TEBBS

5.2 Multivariate normal distribution

5.2.1 Probability density function

STARTING POINT : Suppose that Z1 , Z2 , ..., Zp are iid standard normal random vari-
ables. The joint pdf of Z = (Z1 , Z2 , ..., Zp )0 is given by
p
Y
fZ (z) = fZ (zi )
i=1
 p p
1 −
Pp 2
i=1 zi /2
Y
= √ e I(zi ∈ R)
2π i=1

= (2π)−p/2 exp(−z0 z/2)I(z ∈ Rp ).

If Z has pdf fZ (z), we say that Z has a standard multivariate normal distribution;
i.e., a multivariate normal distribution with mean 0p×1 and covariance matrix Ip . We
write Z ∼ Np (0, I).

MULTIVARIATE NORMAL DISTRIBUTION : Suppose that Z ∼ Np (0, I). Suppose

that V is symmetric and positive definite (and, hence, nonsingular) and let V1/2 be the
symmetric square root of V. Define the transformation

Y = V1/2 Z + µ,

where Y and µ are both p × 1. Note that

E(Y) = E(V1/2 Z + µ) = µ,

since E(Z) = 0, and

cov(Y) = cov(V1/2 Z + µ) = V1/2 cov(Z)V1/2 = V,

since cov(Z) = I. The transformation y = g(z) = V1/2 z+µ is linear in z (and hence, one-
to-one) and the pdf of Y can be found using a transformation. The inverse transformation
is z = g −1 (y) = V−1/2 (y − µ). The Jacobian of the inverse transformation is
 −1 
 ∂g (y)  −1/2
 ∂y  = |V |,
 

PAGE 69
CHAPTER 5 STAT 714, J. TEBBS

where |A| denotes the determinant of A. The matrix V−1/2 is pd; thus, its determinant
is always positive. Thus, for y ∈ Rp ,

fY (y) = fZ {g −1 (y)}|V−1/2 |

= |V|−1/2 fZ {V−1/2 (y − µ)}

= (2π)−p/2 |V|−1/2 exp[−{V−1/2 (y − µ)}0 V−1/2 (y − µ)/2]

= (2π)−p/2 |V|−1/2 exp{−(y − µ)0 V−1 (y − µ)/2}.

If Y ∼ fY (y), we say that Y has a multivariate normal distribution with mean µ

and covariance matrix V. We write Y ∼ Np (µ, V).

IMPORTANT : In the preceding derivation, we assumed V to be pd (hence, nonsingular).

If V is singular, then the distribution of Y is concentrated in a subspace of Rp , with
dimension r(V). In this situation, the density function of Y does not exist.

5.2.2 Moment generating functions

REVIEW : Suppose that X is a random variable with cumulative distribution function

FX (x) = P (X ≤ x). If E(etX ) < ∞ for all |t| < δ, ∃δ > 0, then
Z
tX
MX (t) = E(e ) = etx dFX (x)
R

is defined for all t in an open neighborhood about zero. The function MX (t) is called the
moment generating function (mgf ) of X.

Result 5.1.

1. If MX (t) exists, then E(|X|j ) < ∞, for all j ≥ 1, that is, the moment generating
function characterizes an infinite set of moments.

2. MX (0) = 1.

3. The jth moment of X is given by


j
∂ MX (t) 
E(X j ) = .

∂t j


t=0

PAGE 70
CHAPTER 5 STAT 714, J. TEBBS

4. Uniqueness. If X1 ∼ MX1 (t), X2 ∼ MX2 (t), and MX1 (t) = MX2 (t) for all t in an
open neighborhood about zero, then FX1 (x) = FX2 (x) for all x.

5. If X1 , X2 , ..., Xn are independent random variables with mgfs MXi (t), i = 1, 2, ..., n,
and Y = a0 + ni=1 ai Xi , then
P

n
Y
MY (t) = ea0 t MXi (ai t).
i=1

Result 5.2.

1. If X ∼ N (µ, σ 2 ), then MX (t) = exp(µt + t2 σ 2 /2), for all t ∈ R.

2. If X ∼ N (µ, σ 2 ), then Y = a + bX ∼ N (a + bµ, b2 σ 2 ).

3. If X ∼ N (µ, σ 2 ), then Z = σ1 (X − µ) ∼ N (0, 1).

TERMINOLOGY : Define the random vector X = (X1 , X2 , ..., Xp )0 and let t = (t1 , t2 , ..., tp )0 .
The moment generating function for X is given by
Z
0
MX (t) = E{exp(t X)} = exp(t0 x)dFX (x),
Rp

provided that E{exp(t0 X)} < ∞, for all ||t|| < δ, ∃δ > 0.

Result 5.3.

1. If MX (t) exists, then MXi (ti ) = MX (t∗i ), where t∗i = (0, ..., 0, ti , 0, ..., 0)0 . This
implies that E(|Xi |j ) < ∞, for all j ≥ 1.

2. The expected value of X is


∂MX (t) 
E(X) = .
∂t 

t=0

3. The p × p second moment matrix


2
0 ∂ MX (t) 
E(XX ) = .

0
∂t∂t 

t=0

Thus, 
∂ 2 MX (t) 
E(Xr Xs ) = .
∂tr ts 

tr =ts =0

PAGE 71
CHAPTER 5 STAT 714, J. TEBBS

4. Uniqueness. If X1 and X2 are random vectors with MX1 (t) = MX2 (t) for all t in
an open neighborhood about zero, then FX1 (x) = FX2 (x) for all x.

5. If X1 , X1 , ..., Xn are independent random vectors, and

n
X
Y = a0 + Ai X i ,
i=1

for conformable a0 and Ai ; i = 1, 2, ..., n, then

n
Y
MY (t) = exp(a00 t) MXi (A0i t).
i=1

6. Let X = (X01 , X02 , ..., X0m )0 and suppose that MX (t) exists. Let MXi (ti ) denote the
mgf of Xi . Then, X1 , X2 , ..., Xm are independent if and only if
n
Y
MX (t) = MXi (ti )
i=1

for all t = (t01 , t02 , ..., t0m )0 in an open neighborhood about zero.

Result 5.4. If Y ∼ Np (µ, V), then MY (t) = exp(t0 µ + t0 Vt/2).

Proof. Exercise.

5.2.3 Properties

Result 5.5. Let Y ∼ Np (µ, V). Let a be p × 1, b be k × 1, and A be k × p. Then

1. X = a0 Y ∼ N (a0 µ, a0 Va).

2. X = AY + b ∼ Nk (Aµ + b, AVA0 ).

Result 5.6. If Y ∼ Np (µ, V), then any r × 1 subvector of Y has an r-variate normal
distribution with the same means, variances, and covariances as the original distribution.
Proof. Partition Y = (Y10 , Y20 )0 , where Y1 is r × 1. Partition µ = (µ01 , µ02 )0 and
 
V11 V12
V= 
V21 V22

PAGE 72
CHAPTER 5 STAT 714, J. TEBBS

accordingly. Define A = (Ir 0), where 0 is r ×(p−r). Since Y1 = AY is a linear function

of Y, it is normally distributed. Since Aµ = µ1 and AVA0 = V11 , we are done. 

COROLLARY : If Y ∼ Np (µ, V), then Yi ∼ N (µi , σi2 ), for i = 1, 2, ..., p.

WARNING: Joint normality implies marginal normality. That is, if Y1 and Y2 are
jointly normal, then they are marginally normal. However, if Y1 and Y2 are marginally
normal, this does not necessarily mean that they are jointly normal.

APPLICATION : Consider the general linear model

Y = Xβ + ,

where  ∼ Nn (0, σ 2 I). Note that E(Y) = Xβ and that V = cov(Y) = σ 2 I. Fur-
thermore, because Y is a linear combination of , it is also normally distributed;
i.e., Y ∼ Nn (Xβ, σ 2 I). With PX = X(X0 X)− X0 , we know that Y
b = PX Y and

e = (I − PX )Y. Now,
b

E(Y)
b = E(PX Y) = PX E(Y) = PX Xβ = Xβ

and
b = cov(PX Y) = PX cov(Y)P0 = σ 2 PX IPX = σ 2 PX ,
cov(Y) X

since PX is symmetric and idempotent. Also, Y

b = PX Y is a linear combination of Y,

so it also has normal distribution. Putting everything together, we have

b ∼ Nn (Xβ, σ 2 PX ).
Y

e ∼ Nn {0, σ 2 (I − PX )}.
Exercise: Show that b

5.2.4 Less-than-full-rank normal distributions

TERMINOLOGY : The random vector Yp×1 ∼ Np (µ, V) is said to have a p-variate

normal distribution with rank k if Y has the same distribution as µp×1 + Γ0p×k Zk×1 ,
where Γ0 Γ = V, r(V) = k < p, and Z ∼ Nk (0, I).

PAGE 73
CHAPTER 5 STAT 714, J. TEBBS

Example 5.1. Suppose that k = 1, Z1 ∼ N (0, 1), and Y = (Y1 , Y2 )0 , where

     
0 γ1 γZ
Y = +  Z1 =  1 1  ,
0 γ2 γ2 Z1

where Γ0 = (γ1 , γ2 )0 and r(Γ) = 1. Since r(Γ) = 1, this means that at least one of γ1 and
γ2 is not equal to zero. Without loss, take γ1 6= 0, in which case
γ2
Y2 = Y1 .
γ1
Note that E(Y) = 0 = (0, 0)0 and
   
2 2 2 2
 γ Z
1 1 γ γ Z
1 2 1
 γ γ γ
1 2
cov(Y) = E(YY0 ) = E   = 1  = Γ0 Γ = V.
 γ γ Z 2 γ 2Z 2  γ1 γ2 γ22
1 2 1 2 1

Note that |V| = 0. Thus, Y2×1 is a random vector with all of its probability mass located
in the linear subspace {(y1 , y2 ) : y2 = γ2 y1 /γ1 }. Since r(V) = 1 < 2, Y does not have a
density function. 

5.2.5 Independence results

Result 5.7. Suppose that Y ∼ N (µ, V), where

     
Y1 µ1 V V12 · · · V1m
     11 
 V21 V22 · · · V2m
     
 Y2   µ2  
Y =  .  , µ =  .  , and V = 
   
 .. .. .. ..
.
 ..   ..  .

 . . . 
     
Ym µm Vm1 Vm2 · · · Vmm
Then, Y1 , Y2 , ..., Ym are jointly independent if and only if Vij = 0, for all i 6= j.
Proof. Sufficiency (=⇒): Suppose Y1 , Y2 , ..., Ym are jointly independent. For all i 6= j,

Vij = E{(Yi − µi )(Yj − µj )0 }

= E(Yi − µi )E{(Yj − µj )0 } = 0.

Necessity (⇐=): Suppose that Vij = 0 for all i 6= j, and let t = (t01 , t02 , ..., t0m )0 . Note
that
m
X m
X
0
t Vt = t0i Vii ti and t µ = 0
t0i µi .
i=1 i=1

PAGE 74
CHAPTER 5 STAT 714, J. TEBBS

Thus,

MY (t) = exp(t0 µ + t0 Vt/2)

m m
!
X X
= exp t0i µi + t0i Vii ti /2
i=1 i=1
m
Y m
Y
= exp(t0i µi + t0i Vii ti /2) = MYi (ti ). 
i=1 i=1

Result 5.8. Suppose that X ∼ N (µ, Σ), and let Y1 = a1 + B1 X and Y2 = a2 + B2 X,

for nonrandom conformable ai and Bi ; i = 1, 2. Then, Y1 and Y2 are independent if
and only if B1 ΣB02 = 0.
Proof. Write Y = (Y1 , Y2 )0 as
     
Y a B
 1  =  1  +  1  X = a + BX.
Y2 a2 B2

Thus, Y is a linear combination of X; hence, Y follows a multivariate normal distribution

(i.e., Y1 and Y2 are jointly normal). Also, cov(Y1 , Y2 ) = cov(B1 X, B2 X) = B1 ΣB02 .
Now simply apply Result 5.7. 

REMARK : If X1 ∼ N (µ1 , Σ1 ), X2 ∼ N (µ2 , Σ2 ), and cov(X1 , X2 ) = 0, this does not

necessarily mean that X1 and X2 are independent! We need X = (X01 , X02 )0 to be jointly
normal.

APPLICATION : Consider the general linear model

Y = Xβ + ,

where  ∼ Nn (0, σ 2 I). We have already seen that Y ∼ Nn (Xβ, σ 2 I). Also, note that
with PX = X(X0 X)− X0 ,    
Y
b PX
 =  Y,
e
b I − PX

a linear combination of Y. Thus, Y

b and b
e are jointly normal. By the last result, we
know that Y
b and b
e are independent since

cov(Y,
b be) = PX σ 2 I(I − PX )0 = 0.

PAGE 75
CHAPTER 5 STAT 714, J. TEBBS

That is, the fitted values and residuals from the least-squares fit are independent. This
explains why residual plots that display nonrandom patterns are consistent with a vio-
lation of our model assumptions.

5.2.6 Conditional distributions

RECALL: Suppose that (X, Y )0 ∼ N2 (µ, Σ), where

   
2
µX σX ρσX σY
µ=  and Σ =  
µY ρσX σY σY2

and ρ = corr(X, Y ). The conditional distribution of Y , given X, is also normally dis-

tributed, more precisely,

Y |{X = x} ∼ N µY + ρ(σY /σX )(x − µX ), σY2 (1 − ρ2 ) .



It is important to see that the conditional mean E(Y |X = x) is a linear function of x.

Note also that the conditional variance var(Y |X = x) is free of x.

EXTENSION : We wish to extend the previous result to random vectors. In particular,

suppose that X and Y are jointly multivariate normal with ΣXY 6= 0. That is, suppose
     
X  µ ΣX ΣY X 
  ∼ N  X ,  ,
Y  µ
Y Σ XY Σ Y


and assume that ΣX is nonsingular. The conditional distribution of Y given X is

Y|{X = x} ∼ N (µY |X , ΣY |X ),

where
µY |X = µY + ΣY X Σ−1
X (x − µX )

and
ΣY |X = ΣY − ΣY X Σ−1
X ΣXY .

Again, the conditional mean µY |X is a linear function of x and the conditional covariance
matrix ΣY |X is free of x.

PAGE 76
CHAPTER 5 STAT 714, J. TEBBS

5.3 Noncentral χ2 distribution

RECALL: Suppose that U ∼ χ2n ; that is, U has a (central) χ2 distribution with n > 0
degrees of freedom. The pdf of U is given by
1 n
−1
fU (u|n) = u 2 e−u/2 I(u > 0).
Γ( n2 )2n/2
The χ2n family of distributions is a gamma(α, β) subfamily with shape parameter α = n/2
and scale parameter β = 2. Note that E(U ) = n, var(U ) = 2n, and MU (t) = (1−2t)−n/2 ,
for t < 1/2.

RECALL: If Z1 , Z2 , ..., Zn are iid N (0, 1), then U1 = Z12 ∼ χ21 and
n
X
0
ZZ= Zi2 ∼ χ2n .
i=1

Proof. Exercise.

TERMINOLOGY : A univariate random variable V is said to have a noncentral χ2

distribution with degrees of freedom n > 0 and noncentrality parameter λ > 0 if it has
the pdf
∞  −λ j 
X e λ 1 n+2j
−1
fV (v|n, λ) = v 2 e−v/2 I(v > 0) .
j=0
j! Γ( n+2j
2
)2(n+2j)/2
| {z }
fU (v|n+2j)

We write V ∼ χ2n (λ). When λ = 0, the χ2n (λ) distribution reduces to the central χ2n
distribution. In the χ2n (λ) pdf, notice that e−λ λj /j! is the jth term of a Poisson pmf with
parameter λ > 0.

Result 5.9. If V |W ∼ χ2n+2W and W ∼ Poisson(λ), then V ∼ χ2n (λ).

Proof. Note that
∞
X
fV (v) = fV,W (v, j)
j=0
∞
X
= fW (j)fV |W (v|j)
j=0
∞  −λ j 
X e λ 1 n+2j
−1
= v 2 e−v/2 I(v > 0). 
j=0
j! Γ( n+2j
2
)2 (n+2j)/2

PAGE 77
CHAPTER 5 STAT 714, J. TEBBS

Result 5.10. If V ∼ χ2n (λ), then

 
−n/2 2tλ
MV (t) = (1 − 2t) exp ,
1 − 2t

for t < 1/2.

Proof. The mgf of V , by definition, is MV (t) = E(etV ). Using an iterated expectation,
we can write, for t < 1/2,

MV (t) = E(etV ) = E{E(etV |W )},

where W ∼ Poisson(λ). Note that E(etV |W ) is the conditional mgf of V , given W . We

know that V |W ∼ χ2n+2W ; thus, E(etV |W ) = (1 − 2t)−(n+2W )/2 and
∞
e−λ λj
X  
tV −(n+2j)/2
E{E(e |W )} = (1 − 2t)
j=0
j!
∞
−λ −n/2
X λj
= e (1 − 2t) (1 − 2t)−j
j=0
j!
∞ λ

j
X 1−2t
−λ −n/2
= e (1 − 2t)
j=0
j!
 
−λ −n/2 λ
= e (1 − 2t) exp
1 − 2t
 
−n/2 2tλ
= (1 − 2t) exp .
1 − 2t

MEAN AND VARIANCE : If V ∼ χ2n (λ), then

E(V ) = n + 2λ and var(V ) = 2n + 8λ.

Result 5.11. If Y ∼ N (µ, 1), then U = Y 2 ∼ χ21 (λ), where λ = µ2 /2.

Outline of the proof. The proof proceeds by finding the mgf of U and showing that it
equals  
−n/2 2tλ
MU (t) = (1 − 2t) exp ,
1 − 2t
with n = 1 and λ = µ2 /2. Note that
Z
tY 2 2 1 1 2
tU
MU (t) = E(e ) = E(e )= ety √ e− 2 (y−µ) dy
R 2π

PAGE 78
CHAPTER 5 STAT 714, J. TEBBS

Now, combine the exponents in the integrand, square out the (y − µ)2 term, combine like
terms, complete the square, and collapse the expression to (1 − 2t)−1/2 exp{µ2 t/(1 − 2t)}
times some normal density that is integrated over R. 

Result 5.12. If U1 , U2 , ..., Um are independent random variables, where Ui ∼ χ2ni (λi );
i = 1, 2, ..., m, then U = i Ui ∼ χ2n (λ), where n = i ni and λ = i λi .
P P P

Result 5.13. Suppose that V ∼ χ2n (λ). For fixed n and c > 0, the quantity Pλ (V > c)
is a strictly increasing function of λ.
Proof. See Monahan, pp 106-108. 

IMPLICATION : If V1 ∼ χ2n (λ1 ) and V2 ∼ χ2n (λ2 ), where λ2 > λ1 , then pr(V2 > c) >
pr(V1 > c). That is, V2 is (strictly) stochastically greater than V1 , written V2 >st V1 .
Note that
V2 >st V1 ⇐⇒ FV2 (v) < FV1 (v) ⇐⇒ SV2 (v) > SV1 (v),

for all v, where FVi (·) denotes the cdf of Vi and SVi (·) = 1 − FVi (·) denotes the survivor
function of Vi .

5.4 Noncentral F distribution

RECALL: A univariate random variable W is said to have a (central) F distribution with

degrees of freedom n1 > 0 and n2 > 0 if it has the pdf
 n1 /2
Γ( n1 +n
2
2
) n1
n2
w(n1 −2)/2
fW (w|n1 , n2 ) =  (n1 +n2 )/2 I(w > 0).
n1 n2 n1 w
Γ( 2 )Γ( 2 ) 1 + n2

We write W ∼ Fn1 ,n2 . The moment generating function for the F distribution does not
exist in closed form.

RECALL: If U1 and U2 are independent central χ2 random variables with degrees of

freedom n1 and n2 , respectively, then
U1 /n1
W = ∼ Fn1 ,n2 .
U2 /n2

PAGE 79
CHAPTER 5 STAT 714, J. TEBBS

TERMINOLOGY : A univariate random variable W is said to have a noncentral F

distribution with degrees of freedom n1 > 0 and n2 > 0 and noncentrality parameter
λ > 0 if it has the pdf
 (n1 +2j)/2
∞  −λ j 
X e λ Γ( n1 +2j+n
2
2
) n1 +2j
n2
w(n1 +2j−2)/2
fW (w|n1 , n2 , λ) = (n1 +2j+n2 )/2 I(w > 0).
j!

j=0 Γ( n1 +2j
2
)Γ( n22 ) 1+ n1 w
n2

We write W ∼ Fn1 ,n2 (λ). When λ = 0, the noncentral F distribution reduces to the
central F distribution.

MEAN AND VARIANCE : If W ∼ Fn1 ,n2 (λ), then

 
n2 2λ
E(W ) = 1+
n2 − 2 n1

and
2n22 (n1 + 2λ)2
 
n1 + 4λ
var(W ) = 2 + .
n1 (n2 − 2) (n2 − 2)(n2 − 4) n2 − 4
E(W ) exists only when n2 > 2 and var(W ) exists only when n2 > 4. The moment
generating function for the noncentral F distribution does not exist in closed form.

Result 5.14. If U1 and U2 are independent random variables with U1 ∼ χ2n1 (λ) and
U2 ∼ χ2n2 , then
U1 /n1
W = ∼ Fn1 ,n2 (λ).
U2 /n2
Proof. See Searle, pp 51-52.

Result 5.15. Suppose that W ∼ Fn1 ,n2 (λ). For fixed n1 , n2 , and c > 0, the quantity
Pλ (W > c) is a strictly increasing function of λ. That is, if W1 ∼ Fn1 ,n2 (λ1 ) and
W2 ∼ Fn1 ,n2 (λ2 ), where λ2 > λ1 , then pr(W2 > c) > pr(W1 > c); i.e., W2 >st W1 .

REMARK : The fact that the noncentral F distribution tends to be larger than the
central F distribution is the basis for many of the tests used in linear models. Typically,
test statistics are used that have a central F distribution if the null hypothesis is true
and a noncentral F distribution if the null hypothesis is not true. Since the noncentral
F distribution tends to be larger, large values of the test statistic are consistent with

PAGE 80
CHAPTER 5 STAT 714, J. TEBBS

the alternative hypothesis. Thus, the form of an appropriate rejection region is to reject
H0 for large values of the test statistic. The power is simply the probability of rejection
region (defined under H0 ) when the probability distribution is noncentral F . Noncentral
F distributions are available in most software packages.

5.5 Distributions of quadratic forms

GOAL: We would like to find the distribution of Y0 AY, where Y ∼ Np (µ, V). We will
obtain this distribution by taking steps. Result 5.16 is a very small step. Result 5.17 is
a large step, and Result 5.18 is the finish line. There is no harm in assuming that A is
symmetric.

Result 5.16. Suppose that Y ∼ Np (µ, I) and define

p
X
0 0
W = Y Y = Y IY = Yi2 .
i=1

Result 5.11 says Yi2 ∼ χ21 (µ2i /2), for i = 1, 2, ..., p. Thus, from Result 5.12,
p
X
0
YY= Yi2 ∼ χ2p (µ0 µ/2).
i=1

LEMMA: The p × p symmetric matrix A is idempotent of rank s if and only if there

exists a p × s matrix P1 such that (a) A = P1 P01 and (b) P01 P1 = Is .
Proof. (⇐=) Suppose that A = P1 P01 and P01 P1 = Is . Clearly, A is symmetric. Also,

A2 = P1 P01 P1 P01 = P1 P01 = A.

Note also that r(A) = tr(A) = tr(P1 P01 ) = tr(P01 P1 ) = tr(Is ) = s. Now, to go the other
way (=⇒), suppose that A is a symmetric, idempotent matrix of rank s. The spectral
decomposition of A is given by A = QDQ0 , where D = diag(λ1 , λ2 , ..., λp ) and Q is
orthogonal. Since A is idempotent, we know that s of the eigenvalues λ1 , λ2 , ..., λp are
equal to 1 and other p − s eigenvalues are equal to 0. Thus, we can write
  
 Is 0 0
 P
A = QDQ0 = P1 P2    1  = P1 P01 .
0 0 P02

PAGE 81
CHAPTER 5 STAT 714, J. TEBBS

Thus, we have shown that (a) holds. To show that (b) holds, note that because Q is
orthogonal,
   
P01   P01 P1 P01 P2
Ip = Q0 Q =   P1 P2 = .
P02 P02 P1 P02 P2
It is easy to convince yourself that P01 P1 is an identity matrix. Its dimension is s × s
because tr(P01 P1 ) = tr(P1 P01 ) = tr(A) = r(A), which equals s by assumption. 

Result 5.17. Suppose that Y ∼ Np (µ, I). If A is idempotent of rank s, then Y0 AY ∼

χ2s (λ), where λ = 21 µ0 Aµ.
Proof. Suppose that Y ∼ Np (µ, I) and that A is idempotent of rank s. By the last
lemma, we know that A = P1 P01 , where P1 is p × s, and P01 P1 = Is . Thus,

Y0 AY = Y0 P1 P01 Y = X0 X,

where X = P01 Y. Since Y ∼ Np (µ, I), and since X = P01 Y is a linear combination of Y,
we know that
X ∼ Ns (P01 µ, P01 IP1 ) ∼ Ns (P01 µ, Is ).

Result 5.16 says that Y0 AY = X0 X ∼ χ2s {(P01 µ)0 (P01 µ)/2}. But,
1 1 1
λ ≡ (P01 µ)0 P01 µ = µ0 P1 P01 µ = µ0 Aµ. 
2 2 2

Result 5.18. Suppose that Y ∼ Np (µ, V), where r(V) = p. If AV is idempotent of

rank s, then Y0 AY ∼ χ2s (λ), where λ = 21 µ0 Aµ.
Proof. Since Y ∼ Np (µ, V), and since X = V−1/2 Y is a linear combination of Y, we
know that
X ∼ Np (V−1/2 µ, V−1/2 VV−1/2 ) ∼ Np (V−1/2 µ, Ip ).

Now,
Y0 AY = X0 V1/2 AV1/2 X = X0 BX,

where B = V1/2 AV1/2 . Recall that V1/2 is the symmetric square root of V. From Result
5.17, we know that Y0 AY = X0 BX ∼ χ2s (λ) if B is idempotent of rank s. However, note
that
r(B) = r(V1/2 AV1/2 ) = r(A) = r(AV) = s,

PAGE 82
CHAPTER 5 STAT 714, J. TEBBS

since AV has rank s (by assumption) and V and V1/2 are both nonsingular. Also, AV
is idempotent by assumption so that

AV = AVAV ⇒ A = AVA

⇒ V1/2 AV1/2 = V1/2 AV1/2 V1/2 AV1/2

⇒ B = BB.

Thus, B is idempotent of rank s. This implies that Y0 AY = X0 BX ∼ χ2s (λ). Noting

that
1 1 1
λ = (V−1/2 µ)0 B(V−1/2 µ) = µ0 V−1/2 V1/2 AV1/2 V−1/2 µ = µ0 Aµ
2 2 2
completes the argument. 

Example 5.2. Suppose that Y = (Y1 , Y2 , ..., Yn )0 ∼ Nn (µ1, σ 2 I), so that µ = µ1 and
V = σ 2 I, where 1 is n × 1 and I is n × n. The statistic
n
X
2
(n − 1)S = (Yi − Y )2 = Y0 (I − n−1 J)Y = Y0 AY,
i=1

where A = I − n−1 J. Thus, consider the quantity

(n − 1)S 2
2
= Y0 BY,
σ

where B = σ −2 A = σ −2 (I − n−1 J). Note that BV = σ −2 (I − n−1 J)σ 2 I = I − n−1 J = A,

which is idempotent with rank

r(BV) = tr(BV) = tr(A) = tr(I − n−1 J) = n − n−1 n = n − 1.

Result 5.18 says that (n − 1)S 2 /σ 2 = Y0 BY ∼ χ2n−1 (λ), where λ = 21 µ0 Bµ. However,

1 1
λ = µ0 Bµ = (µ1)0 σ −2 (I − n−1 J)µ1 = 0,
2 2

since µ1 ∈ C(1) and I − n−1 J is the ppm onto C(1)⊥ . Thus,

(n − 1)S 2
2
= Y0 BY ∼ χ2n−1 ,
σ

a central χ2 distribution with n − 1 degrees of freedom. 

PAGE 83
CHAPTER 5 STAT 714, J. TEBBS

Example 5.3. Consider the general linear model

Y = Xβ + ,

where X is n × p with rank r ≤ p and  ∼ Nn (0, σ 2 I). Let PX = X(X0 X)− X0 denote the
perpendicular projection matrix onto C(X). We know that Y ∼ Nn (Xβ, σ 2 I). Consider
the (uncorrected) partitioning of the sums of squares given by

Y0 Y = Y0 PX Y + Y0 (I − PX )Y.

• We first consider the residual sum of squares Y0 (I − PX )Y. Dividing this quantity
by σ 2 , we get

Y0 (I − PX )Y/σ 2 = Y0 {σ −2 (I − PX )}Y = Y0 AY,

where A = σ −2 (I − PX ). With V = σ 2 I, note that

AV = σ −2 (I − PX )σ 2 I = I − PX ,

an idempotent matrix with rank

r(I − PX ) = tr(I − PX ) = tr(I) − tr(PX ) = tr(I) − r(PX ) = n − r,

since r(PX ) = r(X) = r, by assumption. Result 5.18 says that

Y0 (I − PX )Y/σ 2 = Y0 AY ∼ χ2n−r (λ),

where λ = 21 µ0 Aµ. However,

1 1
λ = µ0 Aµ = (Xβ)0 σ −2 (I − PX )Xβ = 0,
2 2
because Xβ ∈ C(X) and I − PX projects onto the orthogonal complement N (X0 ).
Thus, we have shown that Y0 (I − PX )Y/σ 2 ∼ χ2n−r , a central χ2 distribution with
n − r degrees of freedom.

• Now, we turn our attention to the (uncorrected) model sum of squares Y0 PX Y.

Dividing this quantity by σ 2 , we get

Y0 PX Y/σ 2 = Y0 (σ −2 PX )Y = Y0 BY,

PAGE 84
CHAPTER 5 STAT 714, J. TEBBS

where B = σ −2 PX . With V = σ 2 I, note that

BV = σ −2 PX σ 2 I = PX ,

an idempotent matrix with rank r(PX ) = r(X) = r. Result 5.18 says that

Y0 PX Y/σ 2 = Y0 BY ∼ χ2r (λ),

where λ = 21 µ0 Bµ. Note that

1 1
λ = µ0 Bµ = (Xβ)0 σ −2 PX Xβ = (Xβ)0 Xβ/2σ 2 .
2 2

That is, Y0 PX Y/σ 2 has a noncentral χ2 distribution with r degrees of freedom and
noncentrality parameter λ = (Xβ)0 Xβ/2σ 2 .

In the last calculation, note that λ = (Xβ)0 Xβ/2σ 2 = 0 iff Xβ = 0. In this case, both
quadratic forms Y0 (I − PX )Y/σ 2 and Y0 PX Y/σ 2 have central χ2 distributions. 

5.6 Independence of quadratic forms

GOALS : In this subsection, we consider two problems. With Y ∼ N (µ, V), we would
like to establish sufficient conditions for

(a) Y0 AY and BY to be independent, and

(b) Y0 AY and Y0 BY to be independent.

Result 5.19. Suppose that Y ∼ Np (µ, V). If BVA = 0, then Y0 AY and BY are
independent.
Proof. We may assume that A is symmetric. Write A = QDQ0 , where D =
diag(λ1 , λ2 , ..., λp ) and Q is orthogonal. We know that s ≤ p of the eigenvalues
λ1 , λ2 , ..., λp are nonzero where s = r(A). We can thus write
  
 D1 0 0
 P
A = QDQ0 = P1 P2    1  = P1 D1 P01 ,
0 0 P02

PAGE 85
CHAPTER 5 STAT 714, J. TEBBS

where D1 = diag(λ1 , λ2 , ..., λs ). Thus,

Y0 AY = Y0 P1 D1 P01 Y = X0 D1 X,

where X = P01 Y. Notice that

       
0
BY B  Bµ BVB BVP1 
 = Y ∼ N  ,  .
X P01  P0 µ
1 P01 VB0 P01 VP1 

Suppose that BVA = 0. Then,

0 = BVA = BVP1 D1 P01 = BVP1 D1 P01 P1 .

But, because Q is orthogonal,

   
P01   P01 P1 P01 P2
Ip = Q0 Q =   P1 P2 = .
P02 P02 P1 P02 P2

This implies that P01 P1 = Is , and, thus,

0 = BVP1 D1 P01 P1 = BVP1 D1 = BVP1 D1 D−1

1 = BVP1 .

Therefore, cov(BY, X) = 0, that is, X and BY are independent. But, Y0 AY = X0 D1 X,

a function of X. Thus, Y0 AY and BY are independent as well. 

Example 5.4. Suppose that Y = (Y1 , Y2 , ..., Yn )0 ∼ Nn (µ1, σ 2 I), where 1 is n × 1 and I
is n × n, so that µ = µ1 and V = σ 2 I. Recall that

(n − 1)S 2 = Y0 (I − n−1 J)Y = YAY,

where A = I − n−1 J. Also,

Y = n−1 10 Y = BY,

where B = n−1 10 . These two statistics are independent because

BVA = n−1 10 σ 2 I(I − n−1 J) = σ 2 n−1 10 (I − n−1 J) = 0,

because I − n−1 J is the ppm onto C(1)⊥ . Since functions of independent statistics are
also independent, Y and S 2 are also independent. 

PAGE 86
CHAPTER 5 STAT 714, J. TEBBS

Result 5.20. Suppose that Y ∼ Np (µ, V). If BVA = 0, then Y0 AY and Y0 BY are
independent.
Proof. Write A and B in their spectral decompositions; that is, write
  
 D1 0 0
 P
A = PDP0 = P1 P2    1  = P1 D1 P01 ,
0 0 P02

where D1 = diag(λ1 , λ2 , ..., λs ) and s = r(A). Similarly, write

  
 R1 0 0
 Q
B = QRQ0 = Q1 Q2    1  = Q1 R1 Q01 ,
0 0 Q02

where R1 = diag(γ1 , γ2 , ..., γt ) and t = r(B). Since P and Q are orthogonal, this implies
that P01 P1 = Is and Q01 Q1 = It . Suppose that BVA = 0. Then,

0 = BVA = Q1 R1 Q01 VP1 D1 P01

= Q01 Q1 R1 Q01 VP1 D1 P01 P1

= R1 Q01 VP1 D1

= R−1 0 −1
1 R1 Q1 VP1 D1 D1

= Q01 VP1

= cov(Q01 Y, P01 Y).

Now,      
P01  P0 µ 0
P VP1 0 
 Y ∼ N  1 , 1  .
Q01  Q0 µ
1 0 Q01 VQ1 
That is, P01 Y and Q01 Y are jointly normal and uncorrelated; thus, they are independent.
So are Y0 P1 D1 P01 Y and Y0 Q1 R1 Q01 Y. But A = P1 D1 P01 and B = Q1 R1 Q01 , so we are
done. 

Example 5.5. Consider the general linear model

Y = Xβ + ,

where X is n × p with rank r ≤ p and  ∼ Nn (0, σ 2 I). Let PX = X(X0 X)− X0 denote
the perpendicular projection matrix onto C(X). We know that Y ∼ Nn (Xβ, σ 2 I). In

PAGE 87
CHAPTER 5 STAT 714, J. TEBBS

Example 5.3, we showed that

Y0 (I − PX )Y/σ 2 ∼ χ2n−r ,

and that
Y0 PX Y/σ 2 ∼ χ2r (λ),

where λ = (Xβ)0 Xβ/2σ 2 . Note that

Y0 (I − PX )Y/σ 2 = Y0 {σ −2 (I − PX )}Y = Y0 AY,

where A = σ −2 (I − PX ). Also,

Y0 PX Y/σ 2 = Y0 (σ −2 PX )Y = Y0 BY,

where B = σ −2 PX . Applying Result 5.20, we have

BVA = σ −2 PX σ 2 Iσ −2 (I − PX ) = 0,

that is, Y0 (I − PX )Y/σ 2 and Y0 PX Y/σ 2 are independent quadratic forms. Thus, the
statistic

Y0 PX Y/r
F =
Y0 (I − PX )Y/(n − r)
σ −2 Y0 PX Y/r  0 2

= ∼ Fr,n−r (Xβ) Xβ/2σ ,
σ −2 Y0 (I − PX )Y/(n − r)

a noncentral F distribution with degrees of freedom r (numerator) and n − r (denomi-

nator) and noncentrality parameter λ = (Xβ)0 Xβ/2σ 2 .

OBSERVATIONS :

• Note that if Xβ = 0, then F ∼ Fr,n−r since the noncentrality parameter λ = 0.

• On the other hand, as the length of Xβ gets larger, so does λ. This shifts the
noncentral Fr,n−r {(Xβ)0 Xβ/2σ 2 } distribution to the right, because the noncentral
F distribution is stochastically increasing in its noncentrality parameter.

• Therefore, large values of F are consistent with large values of ||Xβ||.

PAGE 88
CHAPTER 5 STAT 714, J. TEBBS

5.7 Cochran’s Theorem

REMARK : An important general notion in linear models is that sums of squares like
Y0 PX Y and Y0 Y can be “broken down” into sums of squares of smaller pieces. We now
discuss Cochran’s Theorem (Result 5.21), which serves to explain why this is possible.

Result 5.21. Suppose that Y ∼ Nn (µ, σ 2 I). Suppose that A1 , A2 , ..., Ak are n × n
symmetric and idempotent matrices, where r(Ai ) = si , for i = 1, 2, ..., k. If A1 + A2 +
· · · + Ak = In , then Y0 A1 Y/σ 2 , Y0 A2 Y/σ 2 , ..., Y0 Ak Y/σ 2 follow independent χ2si (λi )
distributions, where λi = µ0 Ai µ/2σ 2 , for i = 1, 2, ..., k, and ki=1 si = n.
P

Outline of proof. If A1 + A2 + · · · + Ak = In , then Ai idempotent implies that both (a)

Ai Aj = 0, for i 6= j, and (b) ki=1 si = n hold. That Y0 Ai Y/σ 2 ∼ χ2si (λi ) follows from
P

Result 5.18 with V = σ 2 I. Because Ai Aj = 0, Result 5.20 with V = σ 2 I guarantees

that Y0 Ai Y/σ 2 and Y0 Aj Y/σ 2 are independent. 

IMPORTANCE : We now show how Cochran’s Threorem can be used to deduce the joint
distribution of the sums of squares in an analysis of variance. Suppose that we partition
the design matrix X and the parameter vector β in Y = Xβ +  into k + 1 parts, so that
 
β
 0 
 
 β1 
Y = (X0 X1 · · · Xk ) 
 ..  ,

 . 
 
βk

where the dimensions of Xi and β i are n × pi and pi × 1, respectively, and ki=0 pi = p.

P

We can now write this as a k + 1 part model (the full model):

Y = X0 β 0 + X1 β 1 + · · · + Xk β k + .

Now consider fitting each of the k submodels

Y = X0 β 0 + 

Y = X0 β 0 + X1 β 1 + 
..
.

Y = X0 β 0 + X1 β 1 + · · · + Xk−1 β k−1 + ,

PAGE 89
CHAPTER 5 STAT 714, J. TEBBS

and let R(β 0 , β 1 , ..., β i ) denote the regression (model) sum of squares from fitting the
ith submodel, for i = 0, 1, ..., k; that is,

−
R(β 0 , β 1 , ..., β i ) = Y0 (X0 X1 · · · Xi ) [(X0 X1 · · · Xi )0 (X0 X1 · · · Xi )] (X0 X1 · · · Xi )0 Y

= Y0 PX∗i Y,

where
−
PX∗i = (X0 X1 · · · Xi ) [(X0 X1 · · · Xi )0 (X0 X1 · · · Xi )] (X0 X1 · · · Xi )0

is the perpendicular projection matrix onto C(X∗i ), where X∗i = (X0 X1 · · · Xi ) and i =
0, 1, ..., k. Clearly,
C(X∗0 ) ⊂ C(X∗1 ) ⊂ · · · ⊂ C(X∗k−1 ) ⊂ C(X).

We can now partition the total sums of squares as

Y0 Y = R(β 0 ) + [R(β 0 , β 1 ) − R(β 0 )] + [R(β 0 , β 1 , β 2 ) − R(β 0 , β 1 )] + · · ·

+ [R(β 0 , β 1 , ..., β k ) − R(β 0 , β 1 , ..., β k−1 )] + [Y0 Y − R(β 0 , β 1 , ..., β k )]

= Y0 A0 Y + Y0 A1 Y + Y0 A2 Y + · · · + Y0 Ak Y + Y0 Ak+1 Y,

where
A0 = PX∗0

Ai = PX∗i − PX∗i−1 i = 1, 2, ..., k − 1,

Ak = PX − PX∗k−1

Ak+1 = I − PX .

Note that A0 + A1 + A2 + · · · + Ak+1 = I. Note also that the Ai matrices are symmetric,
for i = 0, 1, ..., k + 1, and that

A2i = (PX∗i − PX∗i−1 )(PX∗i − PX∗i−1 )

= P2X∗i − PX∗i−1 − PX∗i−1 + P2X∗i−1

= PX∗i − PX∗i−1 = Ai ,

for i = 1, 2, ..., k, since PX∗i PX∗i−1 = PX∗i−1 and PX∗i−1 PX∗i = PX∗i−1 . Thus, Ai is idempo-
tent for i = 1, 2, ..., k. However, clearly A0 = PX∗0 = X0 (X00 X0 )− X00 and Ak+1 = I − PX

PAGE 90
CHAPTER 5 STAT 714, J. TEBBS

are also idempotent. Let

s0 = r(A0 ) = r(X0 )

si = r(Ai ) = tr(Ai ) = tr(PX∗i ) − tr(PX∗i−1 ) = r(PX∗i ) − r(PX∗i−1 ), i = 1, 2, ..., k,

sk+1 = r(Ak+1 ) = n − r(X).

Pk+1
It is easy to see that i=0 si = n.

APPLICATION : Consider the Gauss-Markov model Y = Xβ + , where  ∼ Nn (0, σ 2 I)

so that Y ∼ Nn (Xβ, σ 2 I). Cochran’s Theorem applies and we have

1 0
Y A0 Y ∼ χ2s0 [λ0 = (Xβ)0 A0 Xβ/2σ 2 ]
σ2
1 0
Y Ai Y ∼ χ2si [λi = (Xβ)0 Ai Xβ/2σ 2 ], i = 1, 2, ..., k,
σ2
1 0 1
2
Y Ak+1 Y = 2 Y0 (I − PX )Y ∼ χ2sk+1 ,
σ σ

where sk+1 = n − r(X). Note that the last quadratic follows a central χ2 distribution
because λk+1 = (Xβ)0 (I − PX )Xβ/2σ 2 = 0. Cochran’s Theorem also guarantees that the
quadratic forms Y0 A0 Y/σ 2 , Y0 A1 Y/σ 2 , ..., Y0 Ak Y/σ 2 , Y0 Ak+1 Y/σ 2 are independent.

ANOVA TABLE : The quadratic forms Y0 Ai Y, for i = 0, 1, ..., k + 1, and the degrees of
freedom si = r(Ai ) are often presented in the following ANOVA table:

Source df SS Noncentrality
β0 s0 R(β 0 ) λ0 = (Xβ)0 A0 Xβ/2σ 2
β 1 (after β 0 ) s1 R(β 0 , β 1 ) − R(β 0 ) λ1 = (Xβ)0 A1 Xβ/2σ 2
β 2 (after β 0 , β 1 ) s2 R(β 0 , β 1 , β 2 ) − R(β 0 , β 1 ) λ2 = (Xβ)0 A2 Xβ/2σ 2
.. .. .. ..
. . . .
β k (after β 0 , ..., β k−1 ) sk R(β 0 , ..., β k ) − R(β 0 , ..., β k−1 ) λk = (Xβ)0 Ak Xβ/2σ 2
Residual sk+1 Y0 Y − R(β 0 , ..., β k ) λk+1 = 0
Total n Y0 Y (Xβ)0 Xβ/2σ 2

2
Note that if X0 = 1, then β 0 = µ, R(β 0 ) = Y0 P1 Y = nY . The R(·) notation
will come in handy when we talk about hypothesis testing later. The sums of squares

PAGE 91
CHAPTER 5 STAT 714, J. TEBBS

R(β 0 ), R(β 0 , β 1 ) − R(β 0 ), ..., R(β 0 , ..., β k ) − R(β 0 , ..., β k−1 ) are called the sequential
sums of squares. These correspond to the Type I sums of squares printed out by SAS
in the ANOVA and GLM procedures. We will also use the notation

R(β i |β 0 , β 1 , ..., β i−1 ) = R(β 0 , β 1 , ..., β i ) − R(β 0 , β 1 , ..., β i−1 ).

Example 5.6. Consider the one-way (fixed effects) analysis of variance model

Yij = µ + αi + ij ,

for i = 1, 2, ..., a and j = 1, 2, ..., ni , where ij ∼ iid N (0, σ 2 ) random variables. In matrix
form, Y, X, and β are
 
    µ
Y 1n1 1n1 0n1 · · · 0n1  
 11    
 α1


· · · 0n2 
   
 Y12   1n2 0n2 1n2  
Yn×1 = , Xn×p = ..  , and β p×1 = ,
 
 ..   .. .. .. ...
 α2
 .   . . . .  
 ..


     . 
Yana 1na 0na 0na · · · 1na  
αa
P
where p = a + 1 and n = i ni . Note that we can write

X = (X0 X1 ) ,

where X0 = 1,  
1n1 0n1 · · · 0n1
 
· · · 0n2 
 
 0n2 1n2
X1 = 
 .. .. ... ..  ,

 . . . 
 
0na 0na · · · 1na
and β = (β 0 , β 01 )0 , where β 0 = µ and β 1 = (α1 , α2 , ..., αa )0 . That is, we can express this
model in the form
Y = X0 β 0 + X1 β 1 + .

The submodel is
Y = X0 β 0 + ,

PAGE 92
CHAPTER 5 STAT 714, J. TEBBS

where X0 = 1 and β 0 = µ. We have

A 0 = P1

A 1 = PX − P1

A 2 = I − PX .

These matrices are clearly symmetric and idempotent. Also, note that

s0 + s1 + s2 = r(P1 ) + r(PX − P1 ) + r(I − PX ) = 1 + (a − 1) + (n − a) = n

and that A0 + A1 + A2 = I. Therefore, Cochran’s Theorem applies and we have

1 0
Y P1 Y ∼ χ21 [λ0 = (Xβ)0 P1 Xβ/2σ 2 ]
σ2
1 0
Y (PX − P1 )Y ∼ χ2a−1 [λ1 = (Xβ)0 (PX − P1 )Xβ/2σ 2 ]
σ2
1 0
Y (I − PX )Y ∼ χ2n−a .
σ2

Cochran’s Theorem also guarantees the quadratic forms Y0 P1 Y/σ 2 , Y0 (PX − P1 )Y/σ 2 ,
and Y0 (I − PX )Y/σ 2 are independent. The sums of squares, using our new notation, are

Y0 P1 Y = R(µ)

Y0 (PX − P1 )Y = R(µ, α1 , ..., αa ) − R(µ) ≡ R(α1 , ..., αa |µ)

Y0 (I − PX )Y = Y0 Y − R(µ, α1 , ..., αa ).

The ANOVA table for this one-way analysis of variance model is

Source df SS Noncentrality
µ 1 R(µ) λ0 = (Xβ)0 P1 Xβ/2σ 2
α1 , ..., αa (after µ) a−1 R(µ, α1 , ..., αa ) − R(µ) λ1 = (Xβ)0 (PX − P1 )Xβ/2σ 2
Residual n−a Y0 Y − R(µ, α1 , ..., αa ) 0
Total n Y0 Y (Xβ)0 Xβ/2σ 2

F STATISTIC : Because
1 0
Y (PX − P1 )Y ∼ χ2a−1 (λ1 )
σ2

PAGE 93
CHAPTER 5 STAT 714, J. TEBBS

and
1 0
Y (I − PX )Y ∼ χ2n−a ,
σ2
and because these two quadratic forms are independent, it follows that
Y0 (PX − P1 )Y/(a − 1)
F = ∼ Fa−1,n−a (λ1 ).
Y0 (I − PX )Y/(n − a)
This is the usual F statistic to test H0 : α1 = α2 = · · · = αa = 0.

• If H0 is true, then Xβ ∈ C(1) and λ1 = (Xβ)0 (PX −P1 )Xβ/2σ 2 = 0 since PX −P1
is the ppm onto C(1)⊥
C(X) . In this case, F ∼ Fa−1,n−a , a central F distribution. A

level α rejection region is therefore RR = {F : F > Fa−1,n−a,α }.

/ C(1) and λ1 = (Xβ)0 (PX − P1 )Xβ/2σ 2 > 0. In

• If H0 is not true, then Xβ ∈
this case, F ∼ Fa−1,n−a (λ1 ), which is stochastically larger than the central Fa−1,n−a
distribution. Therefore, if H0 is not true, we would expect F to be large.

EXPECTED MEAN SQUARES : We already know that

MSE = (n − a)−1 Y0 (I − PX )Y

is an unbiased estimator of σ 2 , i.e., E(MSE) = σ 2 . Recall that if V ∼ χ2n (λ), then

E(V ) = n + 2λ. Therefore,
 
1 0
E 2 Y (PX − P1 )Y = (a − 1) + (Xβ)0 (PX − P1 )Xβ/σ 2
σ
and

E(MSR) = E (a − 1)−1 Y0 (PX − P1 )Y = σ 2 + (Xβ)0 (PX − P1 )Xβ/(a − 1)

 

Xa
= σ2 + ni (αi − α+ )2 /(a − 1),
i=1

where α+ = a−1 ai=1 αi . Again, note that if H0 : α1 = α2 = · · · = αa = 0 is true, then

P

MSR is also an unbiased estimator of σ 2 . Therefore, values of

Y0 (PX − P1 )Y/(a − 1)
F =
Y0 (I − PX )Y/(n − a)
should be close to 1 when H0 is true and larger than 1 otherwise.

PAGE 94
CHAPTER 6 STAT 714, J. TEBBS

6 Statistical Inference

Complementary reading from Monahan: Chapter 6 (and revisit Sections 3.9 and 4.7).

6.1 Estimation

PREVIEW : Consider the general linear model Y = Xβ + , where X is n × p with

rank r ≤ p and  ∼ Nn (0, σ 2 I). Note that this is the usual Gauss-Markov model with
the additional assumption of normality. With this additional assumption, we can now
rigorously pursue questions that deal with statistical inference. We start by examining
minimum variance unbiased estimation and maximum likelihood estimation.

SUFFICIENCY : Under the assumptions stated above, we know that Y ∼ Nn (Xβ, σ 2 I).
Set θ = (β 0 , σ 2 )0 . The pdf of Y, for all y ∈ Rn , is given by

fY (y|θ) = (2π)−n/2 (σ 2 )−n/2 exp{−(y − Xβ)0 (y − Xβ)/2σ 2 }

= (2π)−n/2 (σ 2 )−n/2 exp{−y0 y/2σ 2 + y0 Xβ/σ 2 − (Xβ)0 Xβ/2σ 2 }

= (2π)−n/2 (σ 2 )−n/2 exp{−(Xβ)0 Xβ/2σ 2 } exp{−y0 y/2σ 2 + β 0 X0 y/σ 2 }

= h(y)c(θ) exp{w1 (θ)t1 (y) + w2 (θ)t2 (y)},

where h(y) = (2π)−n/2 I(y ∈ Rn ), c(θ) = (σ 2 )−n/2 exp{−(Xβ)0 Xβ/2σ 2 }, and

w1 (θ) = −1/2σ 2 t1 (y) = y0 y

w2 (θ) = β/σ 2 t2 (y) = X0 y,

that is, Y has pdf in the exponential family (see Casella and Berger, Chapter 3). The
family is full rank (i.e., it is not curved), so we know that T(Y) = (Y0 Y, X0 Y) is
a complete sufficient statistic for θ. We also know that minimum variance unbiased
estimators (MVUEs) of functions of θ are unbiased functions of T(Y).

Result 6.1. Consider the general linear model Y = Xβ + , where X is n × p with rank
r ≤ p and  ∼ Nn (0, σ 2 I). The MVUE for an estimable function Λ0 β is given by Λ0 β,
b

PAGE 95
CHAPTER 6 STAT 714, J. TEBBS

b = (X0 X)− X0 Y is any solution to the normal equations. The MVUE for σ 2 is
where β

MSE = (n − r)−1 Y0 (I − PX )Y
0
= (n − r)−1 (Y0 Y − β
b X0 Y),

where PX is the perpendicular projection matrix onto C(X).

Proof. Both Λ0 β
b and MSE are unbiased estimators of Λ0 β and σ 2 , respectively. These

estimators are also functions of T(Y) = (Y0 Y, X0 Y), the complete sufficient statistic.
Thus, each estimator is the MVUE for its expected value. 

MAXIMUM LIKELIHOOD: Consider the general linear model Y = Xβ + , where X is

n × p with rank r ≤ p and  ∼ Nn (0, σ 2 I). The likelihood function for θ = (β 0 , σ 2 )0 is

L(θ|y) = L(β, σ 2 |y) = (2π)−n/2 (σ 2 )−n/2 exp{−(y − Xβ)0 (y − Xβ)/2σ 2 }.

Maximum likelihood estimators for β and σ 2 are found by maximizing

n n
log L(β, σ 2 |y) = − log(2π) − log σ 2 − (y − Xβ)0 (y − Xβ)/2σ 2
2 2

with respect to β and σ 2 . For every value of σ 2 , maximizing the loglikelihood is the same
as minimizing Q(β) = (y − Xβ)0 (y − Xβ), that is, the least squares estimator

b = (X0 X)− X0 Y,
β

b 0 (y − Xβ)
is also an MLE. Now substitute (y − Xβ) b = y0 (I − PX )y in for Q(β) and

differentiate with respect to σ 2 . The MLE of σ 2 is

σ 2
bMLE = n−1 Y0 (I − PX )Y.

Note that the MLE for σ 2 is biased. The MLE is rarely used in practice; MSE is the
conventional estimator for σ 2 .

INVARIANCE : Under the normal GM model, the MLE for an estimable function Λ0 β
is Λ0 β,
b where β
b is any solution to the normal equations. This is true because of the

invariance property of maximum likelihood estimators (see, e.g., Casella and Berger,
Chapter 7). If Λ0 β is estimable, recall that Λ0 β
b is unique even if β
b is not.

PAGE 96
CHAPTER 6 STAT 714, J. TEBBS

6.2 Testing models

PREVIEW : We now provide a general discussion on testing reduced versus full models
within a Gauss Markov linear model framework. Assuming normality will allow us to
derive the sampling distribution of the resulting test statistic.

PROBLEM : Consider the linear model

Y = Xβ + ,

where r(X) = r ≤ p, E() = 0, and cov() = σ 2 I. Note that these are our usual GM
model assumptions. For the purposes of this discussion, we assume that this model (the
full model) is a “correct” model for the data. Consider also the linear model

Y = Wγ + ,

where E() = 0, cov() = σ 2 I and C(W) ⊂ C(X). We call this a reduced model because
the estimation space is smaller than in the full model. Our goal is to test whether or not
the reduced model is also correct.

• If the reduced model is also correct, there is no reason not to use it. Smaller
models are easier to interpret and fewer degrees of freedom are spent in estimating
σ 2 . Thus, there are practical and statistical advantages to using the reduced model
if it is also correct.

• Hypothesis testing in linear models essentially reduces to putting a constraint on

the estimation space C(X) in the full model. If C(W) = C(X), then the Wγ model
is a reparameterization of the Xβ model and there is nothing to test.

RECALL: Let PW and PX denote the perpendicular projection matrices onto C(W) and
C(X), respectively. Because C(W) ⊂ C(X), we know that PX − PW is the ppm onto
C(PX − PW ) = C(W)⊥
C(X) .

GEOMETRY : In a general reduced-versus-full model testing framework, we start by as-

suming the full model Y = Xβ +  is essentially “correct” so that E(Y) = Xβ ∈ C(X).

PAGE 97
CHAPTER 6 STAT 714, J. TEBBS

If the reduced model is also correct, then E(Y) = Wγ ∈ C(W) ⊂ C(X). Geometri-
cally, performing a reduced-versus-full model test therefore requires the analyst to decide
whether E(Y) is more likely to be in C(W) or C(X) − C(W). Under the full model,
our estimate for E(Y) = Xβ is PX Y. Under the reduced model, our estimate for
E(Y) = Wγ is PW Y.

• If the reduced model is correct, then PX Y and PW Y are estimates of the same
thing, and PX Y − PW Y = (PX − PW )Y should be small.

• If the reduced model is not correct, then PX Y and PW Y are estimating different
things, and PX Y − PW Y = (PX − PW )Y should be large.

• The decision about reduced model adequacy therefore hinges on assessing whether
(PX − PW )Y is large or small. Note that (PX − PW )Y is the perpendicular
projection of Y onto C(W)⊥
C(X) .

MOTIVATION : An obvious measure of the size of (PX − PW )Y is its squared length,

that is,
{(PX − PW )Y}0 (PX − PW )Y = Y0 (PX − PW )Y.

However, the length of (PX − PW )Y is also related to the sizes of C(X) and C(W). We
therefore adjust for these sizes by using

Y0 (PX − PW )Y/r(PX − PW ).

We now compute the expectation of this quantity when the reduced model is/is not
correct. For notational simplicity, set r∗ = r(PX − PW ). When the reduced model is
correct, then
1 
E{Y0 (PX − PW )Y/r∗ } = (Wγ)0 (PX − PW )Wγ + tr{(PX − PW )σ 2 I}

r ∗
1
= ∗ {σ 2 tr(PX − PW )}
r
1
= ∗ (r∗ σ 2 ) = σ 2 .
r
This is correct because (PX − PW )Wγ = 0 and tr(PX − PW ) = r(PX − PW ) = r∗ .
Thus, if the reduced model is correct, Y0 (PX − PW )Y/r∗ is an unbiased estimator of σ 2 .

PAGE 98
CHAPTER 6 STAT 714, J. TEBBS

When the reduced model is not correct, then

1 
E{Y0 (PX − PW )Y/r∗ } = (Xβ)0 (PX − PW )Xβ + tr{(PX − PW )σ 2 I}

r ∗
1
= ∗ {(Xβ)0 (PX − PW )Xβ + r∗ σ 2 }
r
= σ 2 + (Xβ)0 (PX − PW )Xβ/r∗ .

Thus, if the reduced model is not correct, Y0 (PX − PW )Y/r∗ is estimating something
larger than σ 2 . Of course, σ 2 is unknown, so it must be estimated. Because the full
model is assumed to be correct,

MSE = (n − r)−1 Y0 (I − PX )Y,

the MSE from the full model, is an unbiased estimator of σ 2 .

TEST STATISTIC : To test the reduced model versus the full model, we use
Y0 (PX − PW )Y/r∗
F = .
MSE
Using only our GM model assumptions (i.e., not necessarily assuming normality), we can
surmise the following:

• When the reduced model is correct, the numerator and denominator of F are both
unbiased estimators of σ 2 , so F should be close to 1.

• When the reduced model is not correct, the numerator in F is estimating something
larger than σ 2 , so F should be larger than 1. Thus, values of F much larger than
1 are not consistent with the reduced model being correct.

• Values of F much smaller than 1 may mean something drastically different; see
Christensen (2003).

OBSERVATIONS : In the numerator of F , note that

Y0 (PX − PW )Y = Y0 PX Y − Y0 PW Y = Y0 (PX − P1 )Y − Y0 (PW − P1 )Y,

which is the difference in the regression (model) sum of squares, corrected or uncorrected,
from fitting the two models. Also, the term

r∗ = r(PX − PW ) = tr(PX − PW ) = tr(PX ) − tr(PW ) = r(PX ) − r(PW ) = r − r0 ,

PAGE 99
CHAPTER 6 STAT 714, J. TEBBS

say, where r0 = r(PW ) = r(W). Thus, r∗ = r − r0 is the difference in the ranks of the
X and W matrices. This also equals the difference in the model degrees of freedom from
the two ANOVA tables.

REMARK : You will note that we have formulated a perfectly sensible strategy for testing
reduced versus full models while avoiding the question, “What is the distribution of F ?”
Our entire argument is based on first and second moment assumptions, that is, E() = 0
and cov() = σ 2 I, the GM assumptions. We now address the distributional question.

DISTRIBUTION OF F: To derive the sampling distribution of

Y0 (PX − PW )Y/r∗
F = ,
MSE

we require that  ∼ Nn (0, σ 2 I), from which it follows that Y ∼ Nn (Xβ, σ 2 I). First,
we handle the denominator MSE = Y0 (I − PX )Y/(n − r). In Example 5.3 (notes), we
showed that
Y0 (I − PX )Y/σ 2 ∼ χ2n−r .

This distributional result holds regardless of whether or not the reduced model is correct.
Now, we turn our attention to the numerator. Take A = σ −2 (PX − PW ) and consider
the quadratic form
Y0 AY = Y0 (PX − PW )Y/σ 2 .

With V = σ 2 I, the matrix

AV = σ −2 (PX − PW )σ 2 I = PX − PW

is idempotent with rank r(PX − PW ) = r∗ . Therefore, we know that Y0 AY ∼ χ2r∗ (λ),

where
1 1
λ = µ0 Aµ = 2 (Xβ)0 (PX − PW )Xβ.
2 2σ
Now, we make the following observations:

• If the reduced model is correct and Xβ ∈ C(W), then (PX − PW )Xβ = 0 because
PX − PW projects onto C(W)⊥
C(X) . This means that the noncentrality parameter

λ = 0 and Y0 (PX − PW )Y/σ 2 ∼ χ2r∗ , a central χ2 distribution.

PAGE 100
CHAPTER 6 STAT 714, J. TEBBS

• If the reduced model is not correct and Xβ ∈

/ C(W), then (PX − PW )Xβ 6= 0 and
λ > 0. In this event, Y0 (PX − PW )Y/σ 2 ∼ χ2r∗ (λ), a noncentral χ2 distribution
with noncentrality parameter λ.

• Regardless of whether or not the reduced model is correct, the quadratic forms
Y0 (PX − PW )Y and Y0 (I − PX )Y are independent since

(PX − PW )σ 2 I(I − PX ) = σ 2 (PX − PW )(I − PX ) = 0.

CONCLUSION : Putting this all together, we have that

Y0 (PX − PW )Y/r∗ σ −2 Y0 (PX − PW )Y/r∗
F = = −2 0 ∼ Fr∗ ,n−r (λ),
MSE σ Y (I − PX )Y/(n − r)
where r∗ = r − r0 and
1
λ= (Xβ)0 (PX − PW )Xβ.
2σ 2
If the reduced model is correct, that is, if Xβ ∈ C(W), then λ = 0 and F ∼ Fr∗ ,n−r , a
central F distribution. Note also that if the reduced model is correct,
n−r
E(F ) = ≈ 1.
n−r−2
This reaffirms our (model free) assertion that values of F close to 1 are consistent with the
reduced model being correct. Because the noncentral F family is stochastically increasing
in λ, larger values of F are consistent with the reduced model not being correct.

SUMMARY : Consider the linear model Y = Xβ + , where X is n × p with rank r ≤ p

and  ∼ Nn (0, σ 2 I), Suppose that we would like to test

H0 : Y = Wγ + 
versus
H1 : Y = Xβ + ,

where C(W) ⊂ C(X). An α level rejection region is

RR = {F : F > Fr∗ ,n−r,α },

where r∗ = r − r0 , r0 = r(W), and Fr∗ ,n−r,α is the upper α quantile of the Fr∗ ,n−r
distribution.

PAGE 101
CHAPTER 6 STAT 714, J. TEBBS

Example 6.1. Consider the simple linear regression model

Yi = β0 + β1 (xi − x) + i ,

for i = 1, 2, ..., n, where i ∼ iid N (0, σ 2 ). In matrix notation,

     
Y1 1 x1 − x 1
       
 1 x2 − x 
     
 Y2  β 2
, β =  0 ,
 
Y=  ..  , X =  ..
 
..  =
 ..
,

 .   . .  β1  . 
     
Yn 1 xn − x n

where  ∼ Nn (0, σ 2 I). Suppose that we would like to test whether the reduced model

Yi = β0 + i ,

for i = 1, 2, ..., n, also holds. In matrix notation, the reduced model can be expressed as
     
Y 1 
 1     1 
     
 Y2   1   2 
Y=  .. 
, W = 
 ..  = 1, γ = β0 ,  =  ..  ,
  
 .   .   . 
     
Yn 1 n

where  ∼ Nn (0, σ 2 I) and 1 is an n × 1 vector of ones. Note that C(W) ⊂ C(X) with
r0 = 1, r = 2, and r∗ = r − r0 = 1. When the reduced model is correct,

Y0 (PX − PW )Y/r∗
F = ∼ F1,n−2 ,
MSE

where MSE is the mean-squared error from the full model. When the reduced model is
not correct, F ∼ F1,n−2 (λ), where

1
λ = (Xβ)0 (PX − PW )Xβ
2σ 2
X n
2
= β1 (xi − x)2 /2σ 2 .
i=1

Exercises: (a) Verify that this expression for the noncentrality parameter λ is correct.
(b) Suppose that n is even and the values of xi can be selected anywhere in the interval
(d1 , d2 ). How should we choose the xi values to maximize the power of a level α test?

PAGE 102
CHAPTER 6 STAT 714, J. TEBBS

6.3 Testing linear parametric functions

PROBLEM : Consider our usual Gauss-Markov linear model with normal errors; i.e.,
Y = Xβ + , where X is n × p with rank r ≤ p and  ∼ Nn (0, σ 2 I). We now consider
the problem of testing
H0 : K0 β = m
versus
H1 : K0 β 6= m,

where K is a p × s matrix with r(K) = s and m is s × 1.

Example 6.2. Consider the regression model Yi = β0 + β1 xi1 + β2 xi2 + β3 xi3 + β4 xi4 + i ,
for i = 1, 2, ..., n. Express each hypothesis in the form H0 : K0 β = m:

1. H0 : β1 = 0

2. H0 : β3 = β4 = 0

3. H0 : β1 + β3 = 1, β2 − β4 = −1

4. H0 : β2 = β3 = β4 .

Example 6.3. Consider the analysis of variance model Yij = µ + αi + ij , for i = 1, 2, 3, 4
and j = 1, 2, ..., ni . Express each hypothesis in the form H0 : K0 β = m:

1. H0 : µ + α1 = 5, α3 − α4 = 1

2. H0 : α1 − α2 = α3 − α4

3. H0 : α1 − 2 = 13 (α2 + α3 + α4 ).

TERMINOLOGY : The general linear hypothesis H0 : K0 β = m is said to be testable

iff K has full column rank and each component of K0 β is estimable. In other words, K0 β
contains s linearly independent estimable functions. Otherwise, H0 : K0 β = m is said to
be nontestable.

PAGE 103
CHAPTER 6 STAT 714, J. TEBBS

GOAL: Our goal is to develop a test for H0 : K0 β = m, K0 β estimable, in the Gauss-

Markov model with normal errors. We start by noting that the BLUE of K0 β is K0 β,
b

where β
b is a least squares estimator of β. Also,

K0 β
b = K0 (X0 X)− X0 Y,

a linear function of Y, so K0 β
b follows an s-variate normal distribution with mean

E(K0 β)
b = K0 β and covariance matrix

cov(K0 β)
b = K0 cov(β)K
b = σ 2 K0 (X0 X)− X0 X(X0 X)− K = σ 2 K0 (X0 X)− K = σ 2 H,

where H = K0 (X0 X)− K. That is, we have shown K0 β

b ∼ Ns (K0 β, σ 2 H).

NOTE : In the calculation above, note that K0 (X0 X)− X0 X(X0 X)− K = K0 (X0 X)− K only
because K0 β is estimable; i.e., K0 = A0 X for some A. It is also true that H is nonsingular.

LEMMA: If K0 β is estimable, then H is nonsingular.

Proof. First note that H is an s × s matrix. We can write

H = K0 (X0 X)− K = A0 PX A = A0 PX PX A = A0 P0X PX A,

since K0 = A0 X for some A (this follows since K0 β is estimable). Therefore,

r(H) = r(A0 P0X PX A) = r(PX A)

and

s = r(K) = r(X0 A) = r(X0 PX A) ≤ r(PX A) = r[X(X0 X)− X0 A] ≤ r(X0 A) = s.

Therefore, r(H) = r(PX A) = s, showing that H is nonsingular. 

IMPLICATION : The lemma above is important, because it convinces us that the distri-
bution of K0 β
b is full rank. Subtracting m, we have

K0 β
b − m ∼ Ns (K0 β − m, σ 2 H).

F STATISTIC : Now, consider the quadratic form

(K0 β
b − m)0 (σ 2 H)−1 (K0 β
b − m).

PAGE 104
CHAPTER 6 STAT 714, J. TEBBS

Note that (σ 2 H)−1 σ 2 H = Is , an idempotent matrix with rank s. Therefore,

b − m)0 (σ 2 H)−1 (K0 β

(K0 β b − m) ∼ χ2 (λ),
s

where the noncentrality parameter

1
λ= (K0 β − m)0 H−1 (K0 β − m).
2σ 2

We have already shown that Y0 (I − PX )Y/σ 2 ∼ χ2n−r . Also,

(K0 β b − m) and Y0 (I − PX )Y/σ 2

b − m)0 (σ 2 H)−1 (K0 β

are independent (verify!). Thus, the ratio

b − m)0 H−1 (K0 β

(K0 β b − m)/s
F =
MSE
(K β − m) H (K0 β
0b 0 −1 b − m)/s
=
Y0 (I − PX )Y/(n − r)
(K0 β
b − m)0 (σ 2 H)−1 (K0 β
b − m)/s
= ∼ Fs,n−r (λ),
σ −2 Y0 (I − PX )Y/(n − r)

where λ = (K0 β − m)0 H−1 (K0 β − m)/2σ 2 . Note that if H0 : K0 β = m is true, the
noncentrality parameter λ = 0 and F ∼ Fs,n−r . Therefore, an α level rejection region for
the test of H0 : K0 β = m versus H1 : K0 β 6= m is

RR = {F : F > Fs,n−r,α }.

SCALAR CASE : We now consider the special case of testing H0 : K0 β = m when

r(K) = 1, that is, K0 β is a scalar estimable function. This function (and hypothesis) is
perhaps more appropriately written as H0 : k0 β = m to emphasize that k is a p×1 vector
and m is a scalar. Often, k is chosen in a way so that m = 0 (e.g., testing a contrast in
an ANOVA model, etc.). The hypotheses in Example 6.2 (#1) and Example 6.3 (#3)
are of this form. Testing a scalar hypothesis is a mere special case of the general test we
have just derived. However, additional flexibility results in the scalar case; in particular,
we can test for one sided alternatives like H1 : k0 β > m or H1 : k0 β < m. We first discuss
one more noncentral distribution.

PAGE 105
CHAPTER 6 STAT 714, J. TEBBS

TERNINOLOGY : Suppose Z ∼ N (µ, 1) and V ∼ χ2k . If Z and V are independent, then

Z
T =p
V /k
follows a noncentral t distribution with k degrees of freedom and noncentrality pa-
rameter µ. We write T ∼ tk (µ). If µ = 0, the tk (µ) distribution reduces to a central t
distribution with k degrees of freedom. Note that T ∼ tk (µ) implies T 2 ∼ F1,k (µ2 /2).

TEST PROCEDURE : Consider the Gauss-Markov linear model with normal errors; i.e.,
Y = Xβ + , where X is n × p with rank r ≤ p and  ∼ Nn (0, σ 2 I). Suppose that k0 β
is estimable; i.e., k0 ∈ R(X), and that our goal is to test

H0 : k0 β = m
versus
H1 : k0 β 6= m

or versus a suitable one-sided alternative H1 . The BLUE for k0 β is k0 β,

b where β
b is a

least squares estimator. Straightforward calculations show that

0b 0
b ∼ N {k0 β, σ 2 k0 (X0 X)− k} =⇒ Z = p k β − k β
k0 β ∼ N (0, 1).
σ 2 k0 (X0 X)− k
We know that V = Y0 (I − PX )Y/σ 2 ∼ χ2n−r and that Z and V are independent (verify!).
Thus, p
b − k0 β)/ σ 2 k0 (X0 X)− k
(k0 β b − k0 β
k0 β
T = p = p ∼ tn−r .
σ −2 Y0 (I − PX )Y/(n − r) MSE k0 (X0 X)− k
When H0 : k0 β = m is true, the statistic
k0 β
b −m
T =p ∼ tn−r .
MSE k0 (X0 X)− k
Therefore, an α level rejection region, when H1 is two sided, is

RR = {T : T ≥ tn−r,α/2 }.

One sided tests use rejection regions that are suitably adjusted. When H0 is not true,
T ∼ tn−r (µ), where
k0 β − m
µ= p .
σ 2 k0 (X0 X)− k
This distribution is of interest for power and sample size calculations.

PAGE 106
CHAPTER 6 STAT 714, J. TEBBS

6.4 Testing models versus testing linear parametric functions

SUMMARY : Under our Gauss Markov linear model Y = Xβ + , where X is n × p

with rank r ≤ p and  ∼ Nn (0, σ 2 I), we have presented F statistics to test (a) a reduced
model versus a full model in Section 6.2 and (b) a hypothesis defined by H0 : K0 β = m,
for K0 β estimable, in Section 6.3. In fact, testing models and testing linear parametric
functions essentially is the same thing, as we now demonstrate. For simplicity, we take
m = 0, although the following argument can be generalized.

DISCUSSION : Consider the general linear model Y = Xβ + , where X is n × p with

rank r ≤ p and β ∈ Rp . Consider the testable hypothesis H0 : K0 β = 0. We can write
this hypothesis in the following way:

H0 : Y = Xβ +  and K0 β = 0.

We now find a reduced model that corresponds to this hypothesis. Note that K0 β = 0
holds if and only if β⊥C(K). To identify the reduced model, pick a matrix U such that
C(U) = C(K)⊥ . We then have

K0 β = 0 ⇐⇒ β⊥C(K) ⇐⇒ β ∈ C(U) ⇐⇒ β = Uγ,

for some vector γ. Substituting β = Uγ into the linear model Y = Xβ +  gives the
reduced model Y = XUγ + , or letting W = XU, our hypothesis above can be written

H0 : Y = Wγ + ,

where C(W) ⊂ C(X).

OBSERVATION : When K0 β is estimable, that is, when K0 = D0 X for some n × s matrix

D, we can find the perpendicular projection matrix for testing H0 : K0 β = 0 in terms
of D and PX . From Section 6.2, recall that the numerator sum of squares to test the
reduced model Y = Wγ +  versus the full model Y = Xβ +  is Y0 (PX − PW )Y, where
PX − PW is the ppm onto C(PX − PW ) = C(W)⊥
C(X) . For testing the estimable function

K0 β = 0, we now show that the ppm onto C(PX D) is also the ppm onto C(W)⊥
C(X) ; i.e.,

that C(PX D) = C(W)⊥

C(X) .

PAGE 107
CHAPTER 6 STAT 714, J. TEBBS

PROPOSITION : C(PX − PW ) = C(W)⊥ ⊥

C(X) = C(XU)C(X) = C(PX D).

Proof. We showed C(PX − PW ) = C(W)⊥

C(X) in Chapter 2 and W = XU, so the second

equality is obvious. Suppose that v ∈ C(XU)⊥ 0 0

C(X) . Then, v XU = 0, so that X v⊥C(U).

Because C(U) = C(K)⊥ , we know that X0 v ∈ C(K) = C(X0 D), since K0 = D0 X. Thus,

v = PX v = X(X0 X)− X0 v ∈ C[X(X0 X)− X0 D] = C(PX D).

Suppose that v ∈ C(PX D). Clearly, v ∈ C(X). Also, v = PX Dd, for some d and

v0 XU = d0 D0 PX XU = d0 D0 XU = d0 K0 U = 0,

because C(U) = C(K)⊥ . Thus, v ∈ C(XU)⊥

C(X) .

IMPLICATION : It follows immediately that the numerator sum of squares for testing
the reduced model Y = Wγ +  versus the full model Y = Xβ +  is Y0 MPX D Y, where

MPX D = PX D[(PX D)0 (PX D)]− (PX D)0 = PX D(D0 PX D)− D0 PX

is the ppm onto C(PX D). If  ∼ Nn (0, σ 2 I), the resulting test statistic

Y0 MPX D Y/r(MPX D )
F = 0 ∼ Fr(MPX D ),r(I−PX ) (λ),
Y (I − PX )Y/r(I − PX )

where the noncentrality parameter

1
λ= (Xβ)0 MPX D Xβ.
2σ 2

GOAL: Our goal now is to show that the F statistic above is the same F statistic we
derived in Section 6.3 with m = 0, that is,

(K0 β)
b 0 H−1 K0 β/s
b
F = .
Y0 (I − PX )Y/(n − r)

Recall that this statistic was derived for the testable hypothesis H0 : K0 β = 0. First,
we show that r(MPX D ) = s, where, recall, s = r(K). To do this, it suffices to show
that r(K) = r(PX D). Because K0 β is estimable, we know that K0 = D0 X, for some D.
Writing K = X0 D, we see that for any vector a,

X0 Da = 0 ⇐⇒ Da⊥C(X),

PAGE 108
CHAPTER 6 STAT 714, J. TEBBS

which occurs iff PX Da = 0. Note that the X0 Da = 0 ⇐⇒ PX Da = 0 equivalence

implies that

N (X0 D) = N (PX D) ⇐⇒ C(D0 X)⊥ = C(D0 PX )⊥ ⇐⇒ C(D0 X) = C(D0 PX )

so that r(D0 X) = r(D0 PX ). But K0 = D0 X, so r(K) = r(D0 X) = r(D0 PX ) = r(PX D),

which is what we set out to prove. Now, consider the quadratic form Y0 MPX D Y, and
let β
b denote a least squares estimator of β. Result 2.5 says that Xβ
b = PX Y, so that
b = D0 X β
K0 β b = D0 PX Y. Substitution gives

Y0 MPX D Y = Y0 PX D(D0 PX D)− D0 PX Y

= (K0 β)
b 0 [D0 X(X0 X)− X0 D]− K0 β
b

= (K0 β)
b 0 [K0 (X0 X)− K]− K0 β.
b

Recalling that H = K0 (X0 X)− K and that H is nonsingular (when K0 β is estimable)

should convince you that the numerator sum of squares in

Y0 MPX D Y/r(MPX D )
F =
Y0 (I − PX )Y/r(I − PX )

and
b 0 H−1 K0 β/s
(K0 β) b
F =
Y0 (I − PX )Y/(n − r)
are equal. We already showed that r(MPX D ) = s, and because r(I − PX ) = n − r, we
are done. 

6.5 Likelihood ratio tests

6.5.1 Constrained estimation

REMARK : In the linear model Y = Xβ + , where E() = 0 (note the minimal assump-
tions), we have, up until now, allowed the p × 1 parameter vector β to take on any value
in Rp , that is, we have made no restrictions on the parameters in β. We now consider
the case where β is restricted to the subspace of Rp consisting of values of β that satisfy

PAGE 109
CHAPTER 6 STAT 714, J. TEBBS

P0 β = δ, where P is a p × q matrix and δ is q × 1. To avoid technical difficulties, we

will assume that the system P0 β = δ is consistent; i.e., δ ∈ C(P0 ). Otherwise, the set
{β ∈ Rp : P0 β = δ} could be empty.

PROBLEM : In the linear model Y = Xβ +, where E() = 0, we would like to minimize

Q(β) = (Y − Xβ)0 (Y − Xβ)

subject to the constraint that P0 β = δ. Essentially, this requires us to find the minimum
value of Q(β) over the linear subspace {β ∈ Rp : P0 β = δ}. This is a restricted mini-
mization problem and standard Lagrangian methods apply; see Appendix B in Monahan.
The Lagrangian a(β, θ) is a function of β and the Lagrange multipliers in θ and can be
written as
a(β, θ) = (Y − Xβ)0 (Y − Xβ) + 2θ 0 (P0 β − δ).

Taking partial derivatives, we have

∂a(β, θ)
= −2X0 Y + 2X0 Xβ + 2Pθ
∂β
∂a(β, θ)
= 2(P0 β − δ).
∂θ

Setting these equal to zero leads to the restricted normal equations (RNEs), that is,
    
0 0
XX P β XY
  = .
0
P 0 θ δ

Denote by β
b and θ
H
bH the solutions to the RNEs, respectively. The solution β
b is called
H

a restricted least squares estimator.

DISCUSSION : We now present some facts regarding this restricted linear model and its
(restricted) least squares estimator. We have proven all of these facts for the unrestricted
model; restricted versions of the proofs are all in Monahan.

1. The restricted normal equations are consistent; see Result 3.8, Monahan (pp 62-63).

b minimizes Q(β) over the set T ≡ {β ∈ Rp : P0 β = δ}; see Result

2. A solution β H

3.9, Monahan (pp 63).

PAGE 110
CHAPTER 6 STAT 714, J. TEBBS

3. The function λ0 β is estimable in the restricted model

Y = Xβ + , E() = 0, P0 β = δ,

if and only if λ0 = a0 X + d0 P0 , for some a and d, that is,

 
X
λ0 ∈ R  .
0
P

See Result 3.7, Monahan (pp 60).

4. If λ0 β is estimable in the unrestricted model; i.e., the model without the linear
restriction, then λ0 β is estimable in the restricted model. The converse is not true.

5. Under the GM model assumptions, if λ0 β is estimable in the restricted model, then

λ0 β
b H is the BLUE of λ0 β in the restricted model. See Result 4.5, Monahan (pp

89-90).

6.5.2 Testing procedure

SETTING: Consider the Gauss Markov linear model Y = Xβ + , where X is n × p with

rank r ≤ p and  ∼ Nn (0, σ 2 I). We are interested in deriving the likelihood ratio test
(LRT) for
H0 : K0 β = m
versus
H1 : K0 β 6= m,

where K is a p × s matrix with r(K) = s and m is s × 1. We assume that H0 : K0 β = m

is testable, that is, K0 β is estimable.

RECALL: A likelihood ratio testing procedure is intuitive. One simply compares the
maximized likelihood over the restricted parameter space (that is, the space under H0 )
to the maximized likelihood over the entire parameter space. If the former is small when
compared to the latter, then there a large amount of evidence against H0 .

PAGE 111
CHAPTER 6 STAT 714, J. TEBBS

DERIVATION : Under our model assumptions, we know that Y ∼ Nn (Xβ, σ 2 I). The
likelihood function for θ = (β 0 , σ 2 )0 is

L(θ|y) = L(β, σ 2 |y) = (2πσ 2 )−n/2 exp{−Q(β)/2σ 2 },

where Q(β) = (y − Xβ)0 (y − Xβ). The unrestricted parameter space is

Θ = {θ : β ∈ Rp , σ 2 ∈ R+ }.

The restricted parameter space, that is, the parameter space under H0 : K0 β = m, is

Θ0 = {θ : β ∈ Rp , K0 β = m, σ 2 ∈ R+ }.

The likelihood ratio statistic is

supΘ0 L(θ|Y)
λ ≡ λ(Y) = .
supΘ L(θ|Y)

We reject the null hypothesis H0 for small values of λ = λ(Y). Thus, to perform a level
α test, reject H0 when λ < c, where c ∈ (0, 1) is chosen to satisfy PH0 {λ(Y) ≤ c} = α.
We have seen (Section 6.1) that the unrestricted MLEs of β and σ 2 are

b = (X0 X)− X0 Y Q(β)

b
β b2 =
and σ .
n

Similarly, maximizing L(θ|y) over Θ0 produces the solutions β e2 = Q(β

b H and σ b H )/n,

where β
b H is any solution to
    
0 0
XX K β XY
  = ,
K0 0 θ m

the restricted normal equations. Algebra shows that

n/2 ( )n/2
2
b2

L(β H e |Y)
b ,σ σ Q(β)
b
λ= = = .
L(β,
b σb2 |Y) e2
σ Q(β
b )
H

More algebra shows that

( )n/2
Q(β)
b {Q(β
b H ) − Q(β)}/s
b
< c ⇐⇒ > c∗ ,
Q(β
b )
H Q(β)/(n
b − r)

PAGE 112
CHAPTER 6 STAT 714, J. TEBBS

where s = r(K) and c∗ = s−1 (n − r)(c−2/n − 1). Furthermore, Monahan’s Theorem 6.1
(pp 139-140) shows that when K0 β is estimable,

b H ) − Q(β)
Q(β b − m)0 H−1 (K0 β
b = (K0 β b − m),

where H = K0 (X0 X)− K. Applying this result, and noting that Q(β)/(n
b − r) = MSE,
we see that

{Q(β
b H ) − Q(β)}/s
b (K0 β
b − m)0 H−1 (K0 β
b − m)/s
> c∗ ⇐⇒ F = > c∗ .
Q(β)/(n − r)
b MSE

That is, the LRT specifies that we reject H0 when F is large. Choosing c∗ = Fs,n−r,α
provides a level α test. Therefore, under the Gauss Markov model with normal errors,
the LRT for H0 : K0 β = m is the same test as that in Section 6.3.

6.6 Confidence intervals

6.6.1 Single intervals

PROBLEM : Consider the Gauss Markov linear model Y = Xβ + , where X is n × p

with rank r ≤ p and  ∼ Nn (0, σ 2 I). Suppose that λ0 β estimable, that is, λ0 = a0 X, for
some vector a. Our goal is to write a 100(1 − α) percent confidence interval for λ0 β.

DERIVATION : We start with the obvious point estimator λ0 β,

b the least squares esti-

mator (and MLE) of λ0 β. Under our model assumptions, we know that

λ0 β
b ∼ N {λ0 β, σ 2 λ0 (X0 X)− λ}

and, hence,
λ0 β
b − λ0 β
Z=p ∼ N (0, 1).
σ 2 λ0 (X0 X)− λ
If σ 2 was known, our work would be done as Z is a pivot. More likely, this is not the
case, so we must estimate it. An obvious point estimator for σ 2 is MSE, where

MSE = (n − r)−1 Y0 (I − PX )Y.

PAGE 113
CHAPTER 6 STAT 714, J. TEBBS

We consider the quantity

b − λ0 β
λ0 β
T =p
MSE λ0 (X0 X)− λ
and subsequently show that T ∼ tn−r . Note that
b − λ0 β
λ0 β
T = p
MSE λ0 (X0 X)− λ
(λ0 β
b − λ0 β)/ σ 2 λ0 (X0 X)− λ
p
“N (0, 1)”
= p ∼p 2 .
−1 0
(n − r) Y (I − PX )Y/σ 2 “χn−r ”/(n − r)

To verify that T ∼ tn−r , it remains only to show that Z and Y0 (I − PX )Y/σ 2 are
independent, or equivalently, that λ0 β
b and Y0 (I − PX )Y are, since λ0 β
b is a function of

Z and since σ 2 is not random. Note that

λ0 β
b = a0 X(X0 X)− X0 Y = a0 PX Y,

a linear function of Y. Using Result 5.19, λ0 β

b and Y0 (I − PX )Y are independent since

a0 PX σ 2 I(I − PX ) = 0. Thus, T ∼ tn−r , i.e., t is a pivot, so that

!
b − λ0 β
λ0 β
pr −tn−r,α/2 < p < tn−r,α/2 = 1 − α.
MSE λ0 (X0 X)− λ

Algebra shows that this probability statement is the same as

 q q 
0b 0 0 − 0 0b 0 0 −
pr λ β − tn−r,α/2 MSE λ (X X) λ < λ β < λ β + tn−r,α/2 MSE λ (X X) λ = 1−α,

showing that
q
0b
λ β ± tn−r,α/2 MSE λ0 (X0 X)− λ

is a 100(1 − α) percent confidence interval for λ0 β.

Example 6.4. Recall the simple linear regression model

Yi = β0 + β1 xi + i ,

for i = 1, 2, ..., n, where 1 , 2 , ..., n are iid N (0, σ 2 ). Recall also that the least squares
estimator of β = (β0 , β1 )0 is
   
βb0 Y − βb1 x
b = (X0 X)−1 X0 Y = 
β = P ,
(xi −x)(Yi −Y )
βb1 iP
2
i (xi −x)

PAGE 114
CHAPTER 6 STAT 714, J. TEBBS

and that the covariance matrix of β

b is
 
1 2
+ P x − P (xxi −x)2
b = σ 2 (X0 X)−1 = σ 2  n (x i −x)2
cov(β) i i .
x 1
−P 2
P 2
i (xi −x) i (xi −x)

We now consider the problem of writing a 100(1 − α) percent confidence interval for

E(Y |x = x0 ) = β0 + β1 x0 ,

the mean response of Y when x = x0 . Note that E(Y |x = x0 ) = β0 + β1 x0 = λ0 β, where

λ0 = (1 x0 ). Also, λ0 β is estimable because this is a regression model so our previous
work applies. The least squares estimator (and MLE) of E(Y |x = x0 ) is

λ0 β
b = βb0 + βb1 x0 .

Straightforward algebra (verify!) shows that

  
1 x2 x
  + (xi −x)2 − (xi −x)2
P P 1
λ0 (X0 X)−1 λ = 1 x0 
n i i  
− P (xxi −x)2 P 1
(xi −x)2
x 0
i i

1 (x0 − x)2
= +P 2
.
n i (xi − x)

Thus, a 100(1 − α) percent confidence interval for λ0 β = E(Y |x = x0 ) is

s  
1 (x − x)2
0
(βb0 + βb1 x0 ) ± tn−2,α/2 MSE +P 2
. 
n i (xi − x)

6.6.2 Multiple intervals

PROBLEM : Consider the Gauss Markov linear model Y = Xβ +, where X is n×p with
rank r ≤ p and  ∼ Nn (0, σ 2 I). We now consider the problem of writing simultaneous
confidence intervals for the k estimable functions λ01 β, λ02 β, ..., λ0k β. Let the p × k matrix
Λ = (λ1 λ2 · · · λk ) so that  
λ01 β
 
 0 
0
 λ2 β 
τ =Λβ =
 ..
.

 . 
 
λ0k β

PAGE 115
CHAPTER 6 STAT 714, J. TEBBS

Because τ = Λ0 β is estimable, it follows that

τb = Λ0 β
b ∼ Nk (Λ0 β, σ 2 H),

where H = Λ0 (X0 X)− Λ. Furthermore, because λ01 β, 2

b ..., λ0 β
b λ0 β,
k
b are jointly normal, we

have that
b ∼ N (λ0 β, σ 2 hjj ),
λ0j β j

where hjj is the jth diagonal element of H. Using our previous results, we know that

λ0j β
p
b ± tn−r,α/2 σb2 hjj ,

b2 = MSE, is a 100(1 − α) percent confidence interval for λ0j β, that is,

where σ
 
pr λ0j β b2 hjj < λ0j β < λ0j β
p p
b − tn−r,α/2 σ b + tn−r,α/2 σb2 hjj = 1 − α.

This statement is true for a single interval.

SIMULTANEOUS COVERAGE : To investigate the simultaneous coverage probability

of the set of intervals

{λ0j β
p
b ± tn−r,α/2 σb2 hjj , j = 1, 2, ..., k},

let Ej denote the event that interval j contains λ0j β, that is, pr(Ej ) = 1 − α, for j =
1, 2, ..., k. The probability that each of the k intervals includes their target λ0j β is

k
! k
!
\ [
pr Ej = 1 − pr Ej ,
j=1 j=1

by DeMorgan’s Law. In turn, Boole’s Inequality says that

k
! k
[ X
pr Ej ≤ pr(E j ) = kα.
j=1 j=1

Thus, the probability that each interval contains its intended target is
k
!
\
pr Ej ≥ 1 − kα.
j=1

PAGE 116
CHAPTER 6 STAT 714, J. TEBBS

Obviously, this lower bound 1 − kα can be quite a bit lower than 1 − α, that is, the
simultaneous coverage probability of the set of

{λ0j β
p
b ± tn−r,α/2 b2 hjj , j = 1, 2, ..., k}
σ

can be much lower than the single interval coverage probability.

GOAL: We would like the set of intervals {λ0j β

p
b±d σb2 hjj , j = 1, 2, ..., k} to have a
simultaneous coverage probability of at least 1 − α. Here, d represents a probability
point that guarantees the desired simultaneous coverage. Because taking d = tn−r,α/2
does not guarantee this minimum, we need to take d to be larger.

BONFERRONI : From the argument on the last page, it is clear that if one takes d =
tn−r,α/2k , then
k
!
\
pr Ej ≥ 1 − k(α/k) = 1 − α.
j=1

Thus, 100(1 − α) percent simultaneous confidence intervals for λ01 β, λ02 β, ..., λ0k β are

λ0j β
p
b ± tn−r,α/2k σb2 hjj

for j = 1, 2, ..., k.

SCHEFFÉ : The idea behind Scheffé’s approach is to consider an arbitrary linear combi-
nation of τ = Λ0 β, say, u0 τ = u0 Λ0 β and construct a confidence interval
√ √
C(u, d) = (u0 τb − d σb2 u0 Hu, u0 τb + d σb2 u0 Hu),

where d is chosen so that

pr{u0 τ ∈ C(u, d), for all u} = 1 − α.

Since d is chosen in this way, one guarantees the necessary simultaneous coverage proba-
bility for all possible linear combinations of τ = Λ0 β (an infinite number of combinations).
Clearly, the desired simultaneous coverage is then conferred for the k functions of interest
τj = λ0j β, j = 1, 2, ..., k; these functions result from taking u to be the standard unit
vectors. The argument in Monahan (pp 144) shows that d = (kFk,n−r,α )1/2 .

PAGE 117
CHAPTER 7 STAT 714, J. TEBBS

7 Appendix

7.1 Matrix algebra: Basic ideas

TERMINOLOGY : A matrix A is a rectangular array of elements; e.g.,

 
3 5 4
A= .
1 2 8

The (i, j)th element of A is denoted by aij . The dimensions of A are m (the number of
rows) by n (the number of columns). If m = n, A is square. If we want to emphasize
the dimension of A, we can write Am×n .

TERMINOLOGY : A vector is a matrix consisting of one column or one row. A column

vector is denoted by an×1 . A row vector is denoted by a1×n . By convention, we assume
a vector is a column vector, unless otherwise noted; that is,
 
a
 1 
 
 a2  0
 
a=  .. 
 a = a1 a2 · · · an .
 . 
 
an

TERMINOLOGY : If A = (aij ) is an m × n matrix, the transpose of A, denoted by A0

or AT , is the n × m matrix (aji ). If A0 = A, we say A is symmetric.

Result MAR1.1.

(a) (A0 )0 = A

(b) For any matrix A, A0 A and AA0 are symmetric

(c) A = 0 iff A0 A = 0

(d) (AB)0 = B0 A0

(e) (A + B)0 = A0 + B0

PAGE 118
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : The n × n identity matrix I is given by

 
1 0 ··· 0
 
 0 1 ···
 
0 
I = In = 
 .. .. . . .. 
 ;
 . . . . 
 
0 0 ··· 1
n×n

that is, aij = 1 for i = j, and aij = 0 when i 6= j. The n × n matrix of ones J is
 
1 1 ··· 1
 
1 ··· 1 
 
 1
J = Jn =  .. .. . . .. 
 ;
 . . . . 
 
1 1 ··· 1
n×n

that is, aij = 1 for all i and j. Note that J = 110 , where 1 = 1n is an n × 1 (column)
vector of ones. The n × n matrix where aij = 0, for all i and j, is called the null matrix,
or the zero matrix, and is denoted by 0.

TERMINOLOGY : If A is an n × n matrix, and there exists a matrix C such that

AC = CA = I,

then A is nonsingular and C is called the inverse of A; henceforth denoted by A−1 .

If A is nonsingular, A−1 is unique. If A is a square matrix and is not nonsingular, A is
singular.

SPECIAL CASE : The inverse of the 2 × 2 matrix

   
a b 1 d −b
A=  is given by A−1 =  .
c d ad − bc −c a

SPECIAL CASE : The inverse of the n × n diagonal matrix

   
a 0 ··· 0 a−1 0 · · · 0
 11   11 
 0 a22 · · · 0   0 a−1 ··· 0
   
−1 22 
A= .

. . .
 is given by A =  . . .. ..
.
 .. .. .. ..   .. .. .
  
. 
   
0 0 · · · ann 0 0 · · · a−1
nn

PAGE 119
CHAPTER 7 STAT 714, J. TEBBS

Result MAR1.2.

(a) A is nonsingular iff |A| =

6 0.

(b) If A and B are nonsingular matrices, (AB)−1 = B−1 A−1 .

(c) If A is nonsingular, then (A0 )−1 = (A−1 )0 .

7.2 Linear independence and rank

TERMINOLOGY : The m × 1 vectors a1 , a2 , ..., an are said to be linearly dependent

if and only if there exist scalars c1 , c2 , ..., cn such that
n
X
ci ai = 0
i=1

and at least one of the ci ’s is not zero; that is, it is possible to express at least one vector
as a nontrivial linear combination of the others. If
n
X
ci ai = 0 =⇒ c1 = c2 = · · · = cn = 0,
i=1

then a1 , a2 , ..., an are linearly independent. If m < n, then a1 , a2 , ..., an must be

linearly dependent.

NOTE : If a1 , a2 , ..., an denote the columns of an m × n matrix A; i.e.,

 
A= a1 a2 · · · an ,

then the columns of A are linearly independent if and only if Ac = 0 ⇒ c = 0, where

c = (c1 , c2 , ..., cn )0 . Thus, if you can find at least one nonzero c such that Ac = 0, the
columns of A are linearly dependent.

TERMINOLOGY : The rank of a matrix A is defined as

r(A) = number of linearly independent columns of A

= number of linearly independent rows of A.

PAGE 120
CHAPTER 7 STAT 714, J. TEBBS

The number of linearly independent rows of any matrix is always equal to the number of
linearly independent columns. Alternate notation for r(A) is rank(A).

TERMINOLOGY : If A is n × p, then r(A) ≤ min{n, p}.

• If r(A) = min{n, p}, then A is said to be of full rank.

• If r(A) = n, we say that A is of full row rank.

• If r(A) = p, we say that A is of full column rank.

• If r(A) < min{n, p}, we say that A is less than full rank or rank deficient.

• Since the maximum possible rank of an n × p matrix is the minimum of n and p,

for any rectangular (i.e., non-square) matrix, either the rows or columns (or both)
must be linearly dependent.

Result MAR2.1.

(a) For any matrix A, r(A0 ) = r(A).

(b) For any matrix A, r(A0 A) = r(A).

(c) For conformable matrices, r(AB) ≤ r(A) and r(AB) ≤ r(B).

(d) If B is nonsingular, then, r(AB) = r(A).

(e) For any n × n matrix A, r(A) = n ⇐⇒ A−1 exists ⇐⇒ |A| =

6 0.

(f) For any matrix An×n and vector bn×1 , r(A, b) ≥ r(A); i.e., the inclusion of a
column vector cannot decrease the rank of a matrix.

ALTERNATE DEFINITION : An m × n matrix A has rank r if the dimension of the

largest possible nonsingular submatrix of A is r × r.

APPLICATION TO LINEAR MODELS : Consider our general linear model

Y = Xβ + ,

PAGE 121
CHAPTER 7 STAT 714, J. TEBBS

where Y is an n×1 vector of observed responses, X is an n×p matrix of fixed constants, β

is a p×1 vector of fixed but unknown parameters, and  is an n×1 vector of (unobserved)
random errors with zero mean. If X is n × p, then X0 X is p × p. Furthermore, if r(X) = p
(i.e., it is full column rank), then r(X0 X) = p. Thus, we know that (X0 X)−1 exists. On
the other hand, if X has rank r < p, then r(X0 X) < p and (X0 X)−1 does not exist.
Consider the normal equations (which will be motivated later):

X0 Xβ = X0 Y.

We see that left multiplication by (X0 X)−1 produces the solution

b = (X0 X)−1 X0 Y.
β

This is the unique solution to the normal equations (since inverses are unique). Note
that if r(X) = r < p, then a unique solution to the normal equations does not exist.

TERMINOLOGY : We say that two vectors a and b are orthogonal, and write a⊥b, if
their inner product is zero; i.e.,
a0 b = 0.

Vectors a1 , a2 , ..., an are mutually orthogonal if and only if a0i aj = 0 for all i 6= j. If
a1 , a2 , ..., an are mutually orthogonal, then they are also linearly independent (verify!).
The converse is not necessarily true.

TERMINOLOGY : Suppose that a1 , a2 , ..., an are orthogonal. If a0i ai = 1, for all

i = 1, 2, ..., n, we say that a1 , a2 , ..., an are orthonormal.

TERMINOLOGY : Suppose that a1 , a2 , ..., an are orthogonal. Then

ci = ai /||ai ||,

where ||ai || = (a0i ai )1/2 , i = 1, 2, ..., n, are orthonormal. The quantity ||ai || is the length
of ai . If a1 , a2 , ..., an are the columns of A, then A0 A is diagonal; similarly, C0 C = I.

TERMINOLOGY : Let A be an n × n (square) matrix. We say that A is orthogonal

if A0 A = I = AA0 , or equivalently, if A0 = A−1 . Note that if A is orthogonal, then

PAGE 122
CHAPTER 7 STAT 714, J. TEBBS

||Ax|| = ||x||. Geometrically, this means that multiplication of x by A only rotates the
vector x (since the length remains unchanged).

7.3 Vector spaces

TERMINOLOGY : Let V ⊆ Rn be a set of n × 1 vectors. We call V a vector space if

(i) x1 ∈ V, x2 ∈ V ⇒ x1 + x2 ∈ V, and

(ii) x ∈ V ⇒ cx ∈ V for c ∈ R.

That is, V is closed under addition and scalar multiplication.

TERMINOLOGY : A set of n × 1 vectors S ⊆ Rn is a subspace of V if S is a vector

space and S ⊆ V; i.e., if x ∈ S ⇒ x ∈ V.

TERMINOLOGY : We say that subspaces S1 and S2 are orthogonal, and write S1 ⊥S2 ,
if x01 x2 = 0, for all x1 ∈ S1 and for all x2 ∈ S2 .

Example. Suppose that V = R3 . Then, V is a vector space.

Proof. Suppose x1 ∈ V and x2 ∈ V. Then, x1 + x2 ∈ V and cx1 ∈ V for all c ∈ R. 

Example. Suppose that V = R3 . The subspace consisting of the z-axis is

  


 0 


  
S1 =  0  : for z ∈ R .
 

   


 z 


The subspace consisting of the x-y plane is

  


 x 


  
S2 =  y  : for x, y ∈ R .
 

   


 0 


PAGE 123
CHAPTER 7 STAT 714, J. TEBBS

It is easy to see that S1 and S2 are orthogonal. That S1 is a subspace is argued as follows.
Clearly, S1 ⊆ V. Now, suppose that x1 ∈ S1 and x2 ∈ S1 ; i.e.,
   
0 0
   
x1 =  0  and x2 =  0  ,
   
   
z1 z2

for z1 , z2 ∈ R. Then,  
0
 
x 1 + x2 =   ∈ S1
 
0
 
z1 + z2
and  
0
 
cx1 =  0  ∈ S1 ,
 
 
cz1
for all c ∈ R. Thus, S1 is a subspace. That S2 is a subspace follows similarly. 

TERMINOLOGY : Suppose that V is a vector space and that x1 , x2 , ..., xn ∈ V. The set
of all linear combinations of x1 , x2 , ..., xn ; i.e.,
( n
)
X
S= x∈V:x= ci x i
i=1

is a subspace of V. We say that S is generated by x1 , x2 , ..., xn . In other words, S is

the space spanned by x1 , x2 , ..., xn , written S = span{x1 , x2 , ..., xn }.

Example. Suppose that V = R3 and let

   
1 1
   
x1 =  1  and x2 =  0 .
   
   
1 0

For c1 , c2 ∈ R, the linear combination c1 x1 + c2 x2 = (c1 + c2 , c1 , c1 )0 . Thus, the space

spanned by x1 and x2 is the subspace S which consists of all the vectors in R3 of the
form (a, b, b)0 , for a, b ∈ R. 

PAGE 124
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : Suppose that S is a subspace of V. If {x1 , x2 , ..., xn } is a linearly

independent spanning set for S, we call {x1 , x2 , ..., xn } a basis for S. In general, a basis
is not unique. However, the number of vectors in the basis, called the dimension of S,
written dim(S), is unique.

Result MAR3.1. Suppose that S and T are vector spaces. If S ⊆ T , and dim(S) =
dim(T ), then S = T .
Proof. See pp 244-5 in Monahan. 

TERMINOLOGY : The subspaces S1 and S2 are orthogonal complements in Rm if

and only if S1 ⊆ Rm , S2 ⊆ Rm , S1 and S2 are orthogonal, S1 ∩ S2 = {0}, dim(S1 ) = r,
and dim(S2 ) = m − r.

Result MAR3.2. Let S1 and S2 be orthogonal complements in Rm . Then, any vector

y ∈ Rm can be uniquely decomposed as y = y1 + y2 , where y1 ∈ S1 and y2 ∈ S2 .
Proof. Suppose that the decomposition is not possible; that is, suppose that y is linearly
independent of basis vectors in both S1 and S2 . However, this would give m + 1 linearly
independent vectors in Rm , which is not possible. Thus, the decomposition must be
possible. To establish uniqueness, suppose that y = y1 + y2 and y = y1∗ + y2∗ , where
y1 , y1∗ ∈ S1 and y2 , y2∗ ∈ S2 . Then, y1 −y1∗ = y2∗ −y2 . But, y1 −y1∗ ∈ S1 and y2∗ −y2 ∈ S2 .
Thus, both y1 − y1∗ and y2∗ − y2 must be the 0 vector. 

NOTE : In the last result, note that we can write

||y||2 = y0 y = (y1 + y2 )0 (y1 + y2 ) = y10 y1 + 2y10 y2 + y20 y2 = ||y1 ||2 + ||y2 ||2 .

This is simply Pythagorean’s Theorem. The cross product term is zero since y1 and
y2 are orthogonal.

TERMINOLOGY : For the matrix

 
Am×n = a1 a2 · · · an ,

PAGE 125
CHAPTER 7 STAT 714, J. TEBBS

where aj is m × 1, the column space of A,

( n
)
X
C(A) = x ∈ Rm : x = cj aj ; cj ∈ R
j=1
= {x ∈ Rm : x = Ac; c ∈ Rn },

is the set of all m × 1 vectors spanned by the columns of A; that is, C(A) is the set of all
vectors that can be written as a linear combination of the columns of A. The dimension
of C(A) is the column rank of A.

TERMINOLOGY : Let  
b01
 
b02
 
 
Am×n =
 ..
,

 . 
 
b0m
where bi is n × 1. Denote
m
X
n
R(A) = {x ∈ R : x = di bi ; di ∈ R}
i=1
= {x ∈ Rn : x0 = d0 A; d ∈ Rm }.

We call R(A) the row space of A. It is the set of all n × 1 vectors spanned by the rows
of A; that is, the set of all vectors that can be written as a linear combination of the
rows of A. The dimension of R(A) is the row rank of A.

TERMINOLOGY : The set N (A) = {x : Ax = 0} is called the null space of A, denoted

N (A). The dimension of N (A) is called the nullity of A.

Result MAR3.3.

(a) C(B) ⊆ C(A) iff B = AC for some matrix C.

(b) R(B) ⊆ R(A) iff B = DA for some matrix D.

(c) C(A), R(A), and N (A) are all vector spaces.

(d) R(A0 ) = C(A) and C(A0 ) = R(A).

PAGE 126
CHAPTER 7 STAT 714, J. TEBBS

(e) C(A0 A) = C(A0 ) and R(A0 A) = R(A).

(f) For any A and B, C(AB) ⊆ C(A). If B is nonsingular, then C(AB) = C(A).

Result MAR3.4. If A has full column rank, then N (A) = {0}.

Proof. Suppose that A has full column rank. Then, the columns of A are linearly
independent and the only solution to Ax = 0 is x = 0. 

Example. Define    
1 1 2 3
   
A= 1 0 3 and c =  −1  .
   

   
1 0 3 −1
The column space of A is the set of all linear combinations of the columns of A; i.e., the
set of vectors of the form
 
c1 + c2 + 2c3
 
c1 a1 + c2 a2 + c3 a3 =  ,
 
c1 + 3c3
 
c1 + 3c3

where c1 , c2 , c3 ∈ R. Thus, the column space C(A) is the set of all 3 × 1 vectors of
the form (a, b, b)0 , where a, b ∈ R. Any two vectors of {a1 , a2 , a3 } span this space. In
addition, any two of {a1 , a2 , a3 } are linearly independent, and hence form a basis for
C(A). The set {a1 , a2 , a3 } is not linearly independent since Ac = 0. The dimension of
C(A); i.e., the rank of A, is r = 2. The dimension of N (A) is 1, and c forms a basis for
this space. 

Result MAR3.5. For an m × n matrix A with rank r ≤ n, the dimension of N (A) is

n − r. That is, dim{C(A)} + dim{N (A)} = n.
Proof. See pp 241-2 in Monahan. 

Result MAR3.6. For an m×n matrix A, N (A0 ) and C(A) are orthogonal complements
in Rm .
Proof. Both N (A0 ) and C(A) are vector spaces with vectors in Rm . From the last
result, we know that dim{C(A)} = rank(A) = r, say, and dim{N (A0 )} = m − r, since

PAGE 127
CHAPTER 7 STAT 714, J. TEBBS

r = rank(A) = rank(A0 ). Now we need to show N (A0 ) ∩ C(A) = {0}. Suppose x is in

both spaces. If x ∈ C(A), then x = Ac for some c. If x ∈ N (A0 ), then A0 x = 0. Thus,

A0 x = A0 Ac = 0 =⇒ c0 A0 Ac = 0 =⇒ (Ac)0 Ac = 0 =⇒ Ac = x = 0.

To finish the proof, we need to show that N (A0 ) and C(A) are orthogonal spaces. Suppose
that x1 ∈ C(A) and x2 ∈ N (A0 ). It suffices to show that x01 x2 = 0. But, note that
x1 ∈ C(A) =⇒ x1 = Ac, for some c. Also, x2 ∈ N (A0 ) =⇒ A0 x2 = 0. Since x01 x2 =
(Ac)0 x2 = c0 A0 x2 = c0 0 = 0, the result follows. 

Result MAR3.7. Suppose that S1 and T1 are orthogonal complements. Suppose that
S2 and T2 are orthogonal complements. If S1 ⊆ S2 , then T2 ⊆ T1 .
Proof. See pp 244 in Monahan. 

7.4 Systems of equations

REVIEW : Consider the system of equations Ax = c. If A is square and nonsingular,

then there is a unique solution to the system and it is x = A−1 c. If A is not nonsingular,
then the system can have no solution, finitely many solutions, or infinitely many solutions.

TERMINOLOGY : The linear system Ax = c is consistent if there exists an x∗ such

that Ax∗ = c; that is, if c ∈ C(A).

REMARK : We will show that

• for every m × n matrix A, there exists a n × m matrix G such that AGA = A.

• for a consistent system Ax = c, if AGA = A, then x∗ = Gc is a solution.

Result MAR4.1. Suppose that Ax = c is consistent. If G is a matrix such that

AGA = A, then x∗ = Gc is a solution to Ax = c.
Proof. Because Ax = c is consistent, there exists an x∗ such that Ax∗ = c. Note that
AGc = AGAx∗ = Ax∗ = c. Thus, x∗ = Gc is a solution. 

PAGE 128
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : A matrix G that satisfies AGA = A is called a generalized inverse

of A and is denoted by A− . That is,

AGA = A =⇒ AA− A = A.

If A is square and nonsingular, then the generalized inverse of A is A−1 since AA− A =
AA−1 A = A.

NOTES :

• Every matrix A, regardless of its dimension, has a generalized inverse.

• Generalized inverses are not unique unless A is nonsingular.

• If A is m × n, then A− is n × m.

• A generalized inverse of A, A symmetric, is not necessarily symmetric. However, a

symmetric generalized inverse can always be found. We will thus assume that the
generalized inverse of a symmetric matrix is symmetric.

• If G is a generalized inverse of A, then G0 is a generalized inverse of A0 .

• Monahan uses Ag to denote generalized inverse, but I will use A− .

Example. Consider the matrices

   
4 1 2 1/3 −1/3 0
   
A= 1 1 5  and G =  −1/3 0 .
   
4/3
   
3 1 3 0 0 0

Note that r(A) = 2 because −a1 + 6a2 − a3 = 0. Thus A−1 does not exist. However, it
is easy to show that AGA = A; thus, G is a generalized inverse of A. 

Result MAR4.2. Let A be an m × n matrix with r(A) = r. If A can be partitioned

as follows  
C D
A= ,
E F

PAGE 129
CHAPTER 7 STAT 714, J. TEBBS

where r(A) = r(C) = r, and Cr×r is nonsingular, then

 
−1
C 0
G= 
0 0

is a generalized inverse of A. This result essentially shows that every matrix has a
generalized inverse (see Results A.10 and A.11, Monahan). Also, it gives a method to
compute it.

COMPUTATION : This is an algorithm for finding a generalized inverse A− for A, any

m × n matrix of rank r.

1. Find any r × r nonsingular submatrix C. It is not necessary that the elements of

C occupy adjacent rows and columns in A.

2. Find C−1 and (C−1 )0 .

3. Replace the elements of C by the elements of (C−1 )0 .

4. Replace all other elements of A by zeros.

5. Transpose the resulting matrix.

Result MAR4.3. Let Am×n , xn×1 , cm×1 , and In×n be matrices, and suppose that
Ax = c is consistent. Then, x∗ is a solution to Ax = c if and only if

x∗ = A− c + (I − A− A)z,

for some z ∈ Rn . Thus, we can generate all solutions by just knowing one of them; i.e.,
by knowing A− c.
Proof. (⇐=) We know that x∗ = A− c is a solution (Result MAR4.1). Suppose that
x∗ = A− c + (I − A− A)z, for some z ∈ Rn . Thus,

Ax∗ = AA− c + (A − AA− A)z

= AA− c = Ax∗ = c;

PAGE 130
CHAPTER 7 STAT 714, J. TEBBS

that is, x∗ = A− c + (I − A− A)z solves Ax = c. Conversely, (=⇒) suppose that x∗ is a

solution to Ax = c. Now,

x∗ = A − c + x∗ − A − c

= A− c + x∗ − A− Ax∗ = A− c + (I − A− A)x∗ .

Thus, x∗ = A− c + (I − A− A)z, where z = x∗ . Note that if A is nonsingular, A− = A−1

and x∗ = A−1 c + (I − A−1 A)z = A−1 c; i.e., there is just one solution. 

NOTE : Consider the general form of the solution to Ax = c (which is assumed to be

consistent); i.e., x∗ = A− c+(I−A− A)z. We call A− c a particular solution. The term
(I − A− A)z is the general solution to the homogeneous equations Ax = 0, producing
vectors in N (A).

COMPARE : Suppose that X1 , X2 , ..., Xn is an iid sample from fX (x; θ) and let X =
(X1 , X2 , ..., Xn )0 . Suppose also that θb1 = θb1 (X) is an unbiased estimator of θ; that is,
Eθ [θb1 (X)] = θ for all θ ∈ Θ, say. The general form of an unbiased estimator for θ is

θe = θb1 + T,

where Eθ (T ) = 0 for all θ ∈ Θ.

APPLICATION : Consider the general linear model

Y = Xβ + ,

where Y is an n × 1 vector of observed responses, X is an n × p matrix of rank r < p, β is

a p × 1 vector of fixed but unknown parameters, and  is an n × 1 vector of (unobserved)
random errors. The normal equations are given by

X0 Xβ = X0 Y.

The normal equations are consistent (see below). Thus, the general form of the least
squares estimator is given by

b = (X0 X)− X0 Y + [I − (X0 X)− X0 X]z,

β

PAGE 131
CHAPTER 7 STAT 714, J. TEBBS

where z ∈ Rp . Of course, if r(X) = p, then (X0 X)−1 exists, and the unique solution
becomes
b = (X0 X)−1 X0 Y.
β

PROPOSITION : The normal equations X0 Xβ = X0 Y are consistent.

Proof. First, we will state and prove the following lemma.
LEMMA: For any matrix X, and for any matrices A and B,

X0 XA = X0 XB ⇐⇒ XA = XB.

Proof. The necessity part (⇐=) is obvious. For the sufficiency part (=⇒), note that

X0 XA = X0 XB =⇒ X0 XA − X0 XB = 0

=⇒ (A − B)0 (X0 XA − X0 XB) = 0

=⇒ (A − B)0 X0 (XA − XB) = 0

=⇒ (A0 X0 − B0 X0 )(XA − XB) = 0

=⇒ (XA − XB)0 (XA − XB) = 0.

This can only be true if XA − XB = 0. Thus, the lemma is proven. Now, let (X0 X)−
denote a generalized inverse of X0 X so that X0 X(X0 X)− X0 X = X0 X. Taking A0 =
X0 X(X0 X)− and B0 = I in the lemma, we have

X0 X(X0 X)− X0 X = X0 X =⇒ X0 X(X0 X)− X0 = X0

=⇒ X0 X(X0 X)− X0 Y = X0 Y.

b = (X0 X)− X0 Y is a solution to the normal equations. Hence, the

This implies that β
normal equations are consistent. 

Example. Consider the one-way fixed effects ANOVA model

Yij = µ + αi + ij ,

for i = 1, 2 and j = 1, 2, ..., ni , where n1 = 2 and n2 = 3. It is easy to show that

 
5 2 3
 
X0 X =  2 2 0 .
 
 
3 0 3

PAGE 132
CHAPTER 7 STAT 714, J. TEBBS

One generalized inverse of X0 X is

 
0 0 0
 
(X0 X)−
1 =  0 1/2 ,
 
0
 
0 0 1/3
and a solution to the normal equations (based on this generalized inverse) is
  
0 0 0 Y + Y12 + Y21 + Y22 + Y23
   11 
b = (X0 X)− X0 Y =  0 1/2 0  
β
  
Y11 + Y12

1 1 
  
0 0 1/3 Y21 + Y22 + Y23
   
0 0
   
=  1 = 1+  .
   
2
(Y 11 + Y12 )   Y
   
1
3
(Y21 + Y22 + Y23 ) Y 2+
Another generalized inverse of X0 X is
 
1/3 −1/3 0
 
(X0 X)− =  −1/3 0 ,
 
2 5/6
 
0 0 0
and a solution to the normal equations (based on this generalized inverse) is
  
1/3 −1/3 0 Y + Y12 + Y21 + Y22 + Y23
   11 
b = (X0 X)− X0 Y = 
β −1/3 5/6 0

Y + Y

2 2    11 12 
  
0 0 0 Y21 + Y22 + Y23
   
1
3
(Y 21 + Y 22 + Y 23 ) Y 2+
   
=  12 (Y11 + Y12 ) − 13 (Y21 + Y22 + Y23 )  =  Y 1+ − Y 2+ .
   
   
0 0
The general solution is given by

b = (X0 X)− X0 Y + [I − (X0 X)− X0 X]z

β 1 1
      
0 1 0 0 z z1
    1   
=  Y 1+  +  −1 0 0   z2  =  Y 1+ − z1 ,
      
      
Y 2+ −1 0 0 z3 Y 2+ − z1
where z = (z1 , z2 , z3 )0 ∈ R3 . Furthermore, we see that the first particular solution
corresponds to z1 = 0 while the second corresponds to z1 = Y 2+ . 

PAGE 133
CHAPTER 7 STAT 714, J. TEBBS

7.5 Perpendicular projection matrices

TERMINOLOGY : A square matrix P is idempotent if P2 = P.

TERMINOLOGY : A square matrix P is a projection matrix onto the vector space S

if and only if

1. P is idempotent

2. Px ∈ S, for any x

3. z ∈ S =⇒ Pz = z (projection).

Result MAR5.1. The matrix P = AA− projects onto C(A).

Proof. Clearly, AA− is a square matrix. Note that AA− AA− = AA− . Note that
AA− x = A(A− x) ∈ C(A). Finally, if z ∈ C(A), then z = Ax, for some x. Thus,
AA− z = AA− Ax = Ax = z. 

NOTE : In general, projection matrices are not unique. However, if we add the require-
ment that Pz = 0, for any z⊥S, then P is called a perpendicular projection matrix,
which is unique. These matrices are important in linear models.

Result MAR5.2. The matrix I − A− A projects onto N (A).

Proof. Clearly, I − A− A is a square matrix. Note that

(I − A− A)(I − A− A) = I − 2A− A + A− A = I − A− A.

For any x, note that (I − A− A)x ∈ N (A) because A(I − A− A)x = 0. Finally, if
z ∈ N (A), then Az = 0. Thus, (I − A− A)z = z − A− Az = z. 

Example. Consider the (linear) subspace of R2 defined by

   
 2a 
S= z:z=   , for a ∈ R
 a 

and take  
0.8 0.4
P= .
0.4 0.2

PAGE 134
CHAPTER 7 STAT 714, J. TEBBS

Exercise. Show that P is a projection matrix onto S. Show that I − P is a projection

matrix onto S ⊥ , the orthogonal complement of S.

Result MAR5.3. The matrix M is a perpendicular projection matrix onto C(M) if and
only if M is symmetric and idempotent.
Proof. (=⇒) Suppose that M is a perpendicular projection matrix onto C(M) and write
v = v1 + v2 , where v1 ∈ C(M) and v2 ⊥C(M). Also, let w = w1 + w2 , where w1 ∈ C(M)
and w2 ⊥C(M). Since (I − M)v = (I − M)v2 and Mw = Mw1 = w1 , we get

w0 M0 (I − M)v = w10 M0 (I − M)v2 = w10 v2 = 0.

This is true for any v and w, so it must be true that M0 (I − M) = 0 =⇒ M0 = M0 M.

Since M0 M is symmetric, so is M0 , and this, in turn, implies that M = M2 . (⇐=) Now,
suppose that M is symmetric and idempotent. If M = M2 and v ∈ C(M), then since
v = Mb, for some b, we have that Mv = MMb = Mb = v (this establishes that M is
a projection matrix). To establish perpendicularity, note that if M0 = M and w⊥C(M),
then Mw = M0 w = 0, because the columns of M are in C(M). 

Result MAR5.4. If M is a perpendicular projection matrix onto C(X), then C(M) =

C(X).
Proof. We need to show that C(M) ⊆ C(X) and C(X) ⊆ C(M). Suppose that v ∈ C(M).
Then, v = Mb, for some b. Now, write b = b1 + b2 , where b1 ∈ C(X) and b2 ⊥C(X).
Thus, v = Mb = M(b1 + b2 ) = Mb1 + Mb2 = b1 ∈ C(X). Thus, C(M) ⊆ C(X). Now
suppose that v ∈ C(X). Since M is a perpendicular projection matrix onto C(X), we
know that v = Mv = M(v1 + v2 ), where v1 ∈ C(X) and v2 ⊥C(X). But, M(v1 + v2 ) =
Mv1 , showing that v ∈ C(M). Thus, C(X) ⊆ C(M) and the result follows. 

Result MAR5.5. Perpendicular projection matrices are unique.

Proof. Suppose that M1 and M2 are both perpendicular projection matrices onto any
arbitrary subspace S ⊆ Rn . Let v ∈ Rn and write v = v1 + v2 , where v1 ∈ S and v2 ⊥S.
Since v is arbitrary and M1 v = v1 = M2 v, we have M1 = M2 . 

Result MAR5.6. If M is the perpendicular projection matrix onto C(X), then I − M

PAGE 135
CHAPTER 7 STAT 714, J. TEBBS

is the perpendicular projection matrix onto N (X0 ).

Sketch of Proof. I − M is symmetric and idempotent so I − M is the perpendicular
projection matrix onto C(I − M). Show that C(I − M) = N (X0 ); use Result MAR5.5. 

7.6 Trace, determinant, and eigenproblems

TERMINOLOGY : The sum of the diagonal elements of a square matrix A is called the
trace of A, written tr(A), that is, for An×n = (aij ),
n
X
tr(A) = aii .
i=1

Result MAR6.1.

1. tr(A ± B) = tr(A) ± tr(B)

2. tr(cA) = ctr(A)

3. tr(A0 ) = tr(A)

4. tr(AB) = tr(BA)
Pn Pn
5. tr(A0 A) = i=1 j=1 a2ij .

TERMINOLOGY : The determinant of a square matrix A is a real number denoted by

|A| or det(A).

Result MAR6.2.

1. |A0 | = |A|

2. |AB| = |BA|

3. |A−1 | = |A|−1

4. |A| = 0 iff A is singular

Qn
5. For any n × n upper (lower) triangular matrix, |A| = i=1 aii .

PAGE 136
CHAPTER 7 STAT 714, J. TEBBS

REVIEW : The table below summarizes equivalent conditions for the existence of an
inverse matrix A−1 (where A has dimension n × n).

A−1 exists A−1 does not exist

A is nonsingular A is singular
|A| =
6 0 |A| = 0
A has full rank A has less than full rank
r(A) = n r(A) < n
A has LIN rows (columns) A does not have LIN rows (columns)
Ax = 0 has one solution, x = 0 Ax = 0 has many solutions

EIGENVALUES : Suppose that A is a square matrix and consider the equations Au =

λu. Note that
Au = λu ⇐⇒ Au − λu = (A − λI)u = 0.

If u 6= 0, then A − λI must be singular (see last table). Thus, the values of λ which
satisfy Au = λu are those values where

|A − λI| = 0.

This is called the characteristic equation of A. If A is n × n, then the characteristic

equation is a polynomial (in λ) of degree n. The roots of this polynomial, say, λ1 , λ2 , ..., λn
are the eigenvalues of A (some of these may be zero or even imaginary). If A is a
symmetric matrix, then λ1 , λ2 , ..., λn must be real.

EIGENVECTORS : If λ1 , λ2 , ..., λn are eigenvalues for A, then vectors ui satisfying

Aui = λi ui ,

for i = 1, 2, ..., n, are called eigenvectors. Note that

Aui = λi ui =⇒ Aui − λi ui = (A − λi I)ui = 0.

From our discussion on systems of equations and consistency, we know a general solution
for ui is given by ui = [I − (A − λi I)− (A − λi I)]z, for z ∈ Rn .

PAGE 137
CHAPTER 7 STAT 714, J. TEBBS

Result MAR6.3. If λi and λj are eigenvalues of a symmetric matrix A, and if λi 6= λj ,

then the corresponding eigenvectors, ui and uj , are orthogonal.
Proof. We know that Aui = λi ui and Auj = λj uj . The key is to recognize that

λi u0i uj = u0i Auj = λj u0i uj ,

which can only happen if λi = λj or if u0i uj = 0. But λi 6= λj by assumption. 

PUNCHLINE : For a symmetric matrix A, eigenvectors associated with distinct eigen-

values are orthogonal (we’ve just proven this) and, hence, are linearly independent. If
the symmetric matrix A has an eigenvalue λk , of multiplicity mk , then we can find mk
orthogonal eigenvectors of A which correspond to λk (Searle, pp 291). This leads to the
following result (c.f., Christensen, pp 402):

Result MAR6.4. If A is a symmetric matrix, then there exists a basis for C(A) con-
sisting of eigenvectors of nonzero eigenvalues. If λ is a nonzero eigenvalue of multiplicity
m, then the basis will contain m eigenvectors for λ. Furthermore, N (A) consists of the
eigenvectors associated with λ = 0 (along with 0).

SPECTRAL DECOMPOSITION : Suppose that An×n is symmetric with eigenvalues

λ1 , λ2 , ..., λn . The spectral decomposition of A is given by A = QDQ0 , where

• Q is orthogonal; i.e., QQ0 = Q0 Q = I,

• D = diag(λ1 , λ2 , ..., λn ), a diagonal matrix consisting of the eigenvalues of A; note

that r(D) = r(A), because Q is orthogonal, and

• the columns of Q are orthonormal eigenvectors of A.

Result MAR6.5. If A is an n × n symmetric matrix with eigenvalues λ1 , λ2 , ..., λn , then

Qn
1. |A| = i=1 λi
Pn
2. tr(A) = i=1 λi .

PAGE 138
CHAPTER 7 STAT 714, J. TEBBS

NOTE : These facts are also true for a general n × n matrix A.

Proof (in the symmetric case). Write A in its Spectral Decomposition A = QDQ0 .
Note that |A| = |QDQ0 | = |DQ0 Q| = |D| = ni=1 λi . Also, tr(A) = tr(QDQ0 ) =
Q

tr(DQ0 Q) = tr(D) = ni=1 λi . 

P

Result MAR6.6. Suppose that A is symmetric. The rank of A equals the number of
nonzero eigenvalues of A.
Proof. Write A in its spectral decomposition A = QDQ0 . Because r(D) = r(A) and
because the only nonzero elements in D are the nonzero eigenvalues, the rank of D must
be the number of nonzero eigenvalues of A. 

Result MAR6.7. The eignenvalues of an idemptotent matrix A are equal to 0 or 1.

Proof. If λ is an eigenvalue of A, then Au = λu. Note that A2 u = AAu = Aλu =
λAu = λ2 u. This shows that λ2 is an eigenvalue of A2 = A. Thus, we have Au = λu
and Au = λ2 u, which implies that λ = 0 or λ = 1. 

Result MAR6.8. If the n × n matrix A is idempotent, then r(A) = tr(A).

Proof. From the last result, we know that the eignenvalues of A are equal to 0 or 1. Let
v1 , v2 , ..., vr be a basis for C(A). Denote by S the subspace of all eigenvectors associated
with λ = 1. Suppose v ∈ S. Then, because Av = v ∈ C(A), v can be written
as a linear combination of v1 , v2 , ..., vr . This means that any basis for C(A) is also a
basis for S. Furthermore, N (A) consists of eigenvectors associated with λ = 0 (because
Av = 0v = 0). Thus,

n = dim(Rn ) = dim[C(A)] + dim[N (A)]

= r + dim[N (A)],

showing that dim[N (A)] = n − r. Since A has n eigenvalues, all are accounted for λ = 1
P
(with multiplicity r) and for λ = 0 (with multiplicity n − r). Now tr(A) = i λi = r,
the multiplicity of λ = 1. But r(A) = dim[C(A)] = r as well. 

PAGE 139
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : Suppose that x is an n × 1 vector. A quadratic form is a function

f : Rn → R of the form
n X
X n
f (x) = aij xi xj = x0 Ax.
i=1 j=1

The matrix A is called the matrix of the quadratic form.

Result MAR6.9. If x0 Ax is any quadratic form, there exists a symmetric matrix B

such that x0 Ax = x0 Bx.
Proof. Note that x0 A0 x = (x0 Ax)0 = x0 Ax, since a quadratic form is a scalar. Thus,

1 0 1
x0 Ax = x Ax + x0 A0 x
2  2 
1 1
= x0 A + A0 x = x0 Bx,
2 2

where B = 12 A + 12 A0 . It is easy to show that B is symmetric. 

UPSHOT : In working with quadratic forms, we can, without loss of generality, assume
that the matrix of the quadratic form is symmetric.

TERMINOLOGY : The quadratic form x0 Ax is said to be

• nonnegative definite (nnd) if x0 Ax ≥ 0, for all x ∈ Rn .

• positive definite (pd) if x0 Ax > 0, for all x 6= 0.

• positive semidefinite (psd) if x0 Ax is nnd but not pd.

TERMINOLOGY : A symmetric n × n matrix A is said to be nnd, pd, or psd if the

quadratic form x0 Ax is nnd, pd, or psd, respectively.

Result MAR6.10. Let A be a symmetric matrix. Then

1. A pd =⇒ |A| > 0

2. A nnd =⇒ |A| ≥ 0.

Result MAR6.11. Let A be a symmetric matrix. Then

PAGE 140
CHAPTER 7 STAT 714, J. TEBBS

1. A pd ⇐⇒ all eigenvalues of A are positive

2. A nnd ⇐⇒ all eigenvalues of A are nonnegative.

Result MAR6.12. A pd matrix is nonsingular. A psd matrix is singular. The converses

are not true.

CONVENTION : If A1 and A2 are n × n matrices, we write A1 ≥nnd A2 if A1 − A2 is

nonnegative definite (nnd) and A1 ≥pd A2 if A1 − A2 is positive definite (pd).

Result MAR6.13. Let A be an m × n matrix of rank r. Then A0 A is nnd with rank

r. Furthermore, A0 A is pd if r = n and is psd if r < n.
Proof. Let x be an n × 1 vector. Then x0 (A0 A)x = (Ax)0 Ax ≥ 0, showing that A0 A is
nnd. Also, r(A0 A) = r(A) = r. If r = n, then the columns of A are linearly independent
and the only solution to Ax = 0 is x = 0. This shows that A0 A is pd. If r < n, then
the columns of A are linearly dependent; i.e., there exists an x 6= 0 such that Ax = 0.
Thus, A0 A is nnd but not pd, so it must be psd. 

RESULT : A square matrix A is pd iff there exists a nonsingular lower triangular matrix
L such that A = LL0 . This is called the Choleski Factorization of A. Monahan
proves this result (see pp 258), provides an algorithm on how to find L, and includes an
example.

RESULT : Suppose that A is symmetric and pd. Writing A in its Spectral Decompo-
sition, we have A = QDQ0 . Because A is pd, λ1 , λ2 , ..., λn , the eigenvalues of A, are
√ √ √
positive. If we define A1/2 = QD1/2 Q0 , where D1/2 = diag( λ1 , λ2 , ..., λn ), then A1/2
is symmetric and

A1/2 A1/2 = QD1/2 Q0 QD1/2 Q0 = QD1/2 ID1/2 Q0 = QDQ0 = A.

The matrix A1/2 is called the symmetric square root of A. See Monahan (pp 259-60)
for an example.

PAGE 141
CHAPTER 7 STAT 714, J. TEBBS

7.7 Random vectors

TERMINOLOGY : Suppose that Y1 , Y2 , ..., Yn are random variables. We call

 
Y
 1 
 
 Y2 
Y=  .. 

 . 
 
Yn
a random vector. The joint pdf of Y is denoted by fY (y).

DEFINTION : Suppose that E(Yi ) = µi , var(Yi ) = σi2 , for i = 1, 2, ..., n, and cov(Yi , Yj ) =
σij , for i 6= j. The mean of Y is
   
E(Y1 ) µ1
   
   
 E(Y2 )   µ2 
µ = E(Y) = 
 ..
=
  ..
.

 .   . 
   
E(Yn ) µn
The variance-covariance matrix of Y is
 
σ 2 σ12 · · · σ1n
 1 
 σ21 σ22 · · · σ2n
 

Σ = cov(Y) =  .. .. ... ..
.

 . . . 
 
σn1 σn2 · · · σn2
n


NOTE : Note that Σ contains the variances σ12 , σ22 , ..., σn2 on the diagonal and the 2

covariance terms cov(Yi , Yj ), for i < j, as the elements strictly above the diagonal. Since
cov(Yi , Yj ) = cov(Yj , Yi ), it follows that Σ is symmetric.

EXAMPLE : Suppose that Y1 , Y2 , ..., Yn is an iid sample with mean E(Yi ) = µ and
variance var(Yi ) = σ 2 and let Y = (Y1 , Y2 , ..., Yn )0 . Then µ = E(Y) = µ1n and
Σ = cov(Y) = σ 2 In .

EXAMPLE : Consider the GM linear model Y = Xβ+. In this model, the random errors
1 , 2 , ..., n are uncorrelated random variables with zero mean and constant variance σ 2 .
We have E() = 0n×1 and cov() = σ 2 In .

PAGE 142
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : Suppose that Z11 , Z12 , ..., Znp are random variables. We call
 
Z Z12 · · · Z1p
 11 
 Z21 Z22 · · ·
 
Z2p 
Zn×p =  .. .. .. .. 

 . . . . 
 
Zn1 Zn2 · · · Znp

a random matrix. The mean of Z is

 
E(Z11 ) E(Z12 ) · · · E(Z1p )
 
 E(Z21 ) E(Z22 ) · · · E(Z2p )
 

E(Z) =  .. .. .. ..
 .
.
 
 . . . 
 
E(Zn1 ) E(Zn2 ) · · · E(Znp )
n×p

Result RV1. Suppose that Y is a random vector with mean µ. Then

Σ = cov(Y) = E[(Y − µ)(Y − µ)0 ] = E(YY0 ) − µµ0 .

Proof. That cov(Y) = E[(Y − µ)(Y − µ)0 ] follows straightforwardly from the definition
of variance and covariance in the scalar case. Showing this equals E(YY0 )−µµ0 is simple
algebra. 

DEFINITION : Suppose that Yp×1 and Xq×1 are random vectors with means µY and
µX , respectively. The covariance between Y and X is the p × q matrix defined by

cov(Y, X) = E{(Y − µY )(X − µX )0 } = (σij )p×q ,

where
σij = E[{Yi − E(Yi )}{Xj − E(Xj )}] = cov(Yi , Xj ).

DEFINITION : Random vectors Yp×1 and Xq×1 are uncorrelated if cov(Y, X) = 0p×q .

Result RV2. If cov(Y, X) = 0, then cov(Y, a + BX) = 0, for all nonrandom con-
formable a and B. That is, Y is uncorrelated with any linear function of X.

PAGE 143
CHAPTER 7 STAT 714, J. TEBBS

TERMINOLOGY : Suppose that var(Yi ) = σi2 , for i = 1, 2, ..., n, and cov(Yi , Yj ) = σij ,
for i 6= j. The correlation matrix of Y is the n × n matrix
 
1 ρ12 · · · ρ1n
 
 ρ21 1 · · · ρ2n 
 
R = (ρij ) = 
 .. .. ... ..  ,

 . . . 
 
ρn1 ρn2 · · · 1

where, recall, the correlation ρij is given by

σij
ρij = ,
σi σj

for i, j = 1, 2, ..., n.

TERMINOLOGY : Suppose that Y1 , Y2 , ..., Yn are random variables and that a1 , a2 , ..., an
are constants. Define a = (a1 , a2 , ..., an )0 and Y = (Y1 , Y2 , ..., Yn )0 . The random variable
n
X
X = a0 Y = ai Y i
i=1

is called a linear combination of Y1 , Y2 , ..., Yn .

Result RV3. If a = (a1 , a2 , ..., an )0 is a vector of constants and Y = (Y1 , Y2 , ..., Yn )0 is a

random vector with mean µ = E(Y), then

E(a0 Y) = a0 µ.

Proof. The quantity a0 Y is a scalar so E(a0 Y) is also a scalar. Note that

n
! n n
X X X
E(a0 Y) = E ai Y i = ai E(Yi ) = ai µi = a0 µ. 
i=1 i=1 i=1

Result RV4. Suppose that Y = (Y1 , Y2 , ..., Yn )0 is a random vector with mean µ =
E(Y), let Z be a random matrix, and let A and B (a and b) be nonrandom conformable
matrices (vectors). Then

1. E(AY) = Aµ

2. E(a0 Zb) = a0 E(Z)b.

PAGE 144
CHAPTER 7 STAT 714, J. TEBBS

3. E(AZB) = AE(Z)B.

Result RV5. If a = (a1 , a2 , ..., an )0 is a vector of constants and Y = (Y1 , Y2 , ..., Yn )0 is a

random vector with mean µ = E(Y) and covariance matrix Σ = cov(Y), then

var(a0 Y) = a0 Σa.

Proof. The quantity a0 Y is a scalar random variable, and its variance is given by

var(a0 y) = E{(a0 Y − a0 µ)2 } = E[{a0 (Y − µ)}2 ] = E{a0 (Y − µ)a0 (Y − µ)}.

But, note that a0 (Y − µ) is a scalar, and hence equals (Y − µ)0 a. Using this fact, we
can rewrite the last expectation to get

E{a0 (Y − µ)(Y − µ)0 a} = a0 E{(Y − µ)(Y − µ)0 }a = a0 Σa. 

Result RV6. Suppose that Y = (Y1 , Y2 , ..., Yn )0 is a random vector with covariance
matrix Σ = cov(Y), and let a and b be conformable vectors of constants. Then

cov(a0 Y, b0 Y) = a0 Σb.

Result RV7. Suppose that Y = (Y1 , Y2 , ..., Yn )0 is a random vector with mean µ = E(Y)
and covariance matrix Σ = cov(Y). Let b, A, and B denote nonrandom conformable
vectors/matrices. Then

1. E(AY + b) = Aµ + b

2. cov(AY + b) = AΣA0

3. cov(AY, BY) = AΣB0 .

Result RV8. A variance-covariance matrix Σ = cov(Y) is nonnegative definite.

Proof. Suppose that Yn×1 has variance-covariance matrix Σ. We need to show that
a0 Σa ≥ 0, for all a ∈ Rn . Consider X = a0 Y, where a is a conformable vector of
constants. Then, X is scalar and var(X) ≥ 0. But, var(X) = var(a0 Y) = a0 Σa. Since a
is arbitrary, the result follows. 

PAGE 145
CHAPTER 7 STAT 714, J. TEBBS

Result RV9. If Y = (Y1 , Y2 , ..., Yn )0 is a random vector with mean µ = E(Y) and
covariance matrix Σ, then P {(Y − µ) ∈ C(Σ)} = 1.
Proof. Without loss, take µ = 0, and let MΣ be the perpendicular projection matrix
onto C(Σ). We know that Y = MΣ Y + (I − MΣ )Y and that

E{(I − MΣ )Y} = (I − MΣ )E(Y) = 0,

since µ = E(Y) = 0. Also,

cov{(I − MΣ )Y} = (I − MΣ )Σ(I − MΣ )0 = (Σ − MΣ Σ)(I − MΣ )0 = 0,

since MΣ Σ = Σ. Thus, we have shown that P {(I − MΣ )Y = 0} = 1, which implies

that P (Y = MΣ Y) = 1. Since MΣ Y ∈ C(Σ), we are done. 

IMPLICATION : Result RV9 says that there exists a subset C(Σ) ⊆ Rn that contains Y
with probability one (i.e., almost surely). If Σ is positive semidefinite (psd), then Σ is
singular and C(Σ) is concentrated in a subspace of Rn , where the subspace has dimension
r = r(Σ), r < n. In this situation, the pdf of Y may not exist.

Result RV10. Suppose that X, Y, and Z are n × 1 vectors and that X = Y + Z. Then

1. E(X) = E(Y) + E(Z)

2. cov(X) = cov(Y) + cov(Z) + 2cov(Y, Z)

3. if Y and Z are uncorrelated, then cov(X) = cov(Y) + cov(Z).

PAGE 146