You are on page 1of 17

# Mathematical Theory and Modeling

## ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)

Vol.4, No.11, 2014

www.iiste.org

## Asymptotic Properties of Bayes Factor in One- Way Repeated

Measurements Model
Ameera Jaber Mohaisen and Khawla Abdul Razzaq Swadi
Mathematics Department College of Education for Pure Science AL-Basrah University-Iraq
E-mail: Khawla.swade@gmail.com

Abstract
In this paper, we consider the linear one- way repeated measurements model which has only one within units
factor and one between units factor incorporating univariate random effects as well as the experimental error
term. In this model, we investigate the consistency property of Bayes factor for testing the fixed effects linear
one- way repeated measurements model against the mixed one- way repeated measurements alternative model.
Under some conditions on the prior and design matrix, we identify the analytic form of the Bayes factor and
show that the Bayes factor is consistent.

Keywords: One- Way Repeated Measurements Model, ANOVA, Mixed model, Prior Distribution, Posterior
Distribution, covariance matrix, Bayes Factor, Consistent.

1. Introduction
Repeated measurements occur frequently in observational studies which are longitudinal in nature, and in
experimental studies incorporating repeated measures designs. For longitudinal studies, the underlying
metameter for the occasions at which measurements are taken is usually time. Here, interest often centers around
modeling the response as linear or nonlinear function of time, Repeated measures designs, on the other hand,
entail one or more response variables being measured repeatedly on each individual over arrange of conditions.
Here the metameter may be time or it may be a set of experimental conditions (e.g. ,dose levels of a drug).
Repeated measurements analysis is widely used in many fields, for example , in the health and life science,
epidemiology, biomedical, agricultural, industrial, psychological, educational researches and so on.,,
A balanced repeated measurements design means that the p occasions of measurements are the same for all of
the experimental units. A complete repeated measurements design means that measurements are available each
time point for each experimental unit. A typical repeated measurements design consist of experimental units or
subjects randomized to a treatment (a between-units factor) and remaining on the assigned treatment throughout
the course of the experiment. ,,
A one way repeated measurements analysis of variance refers to the situation with only one within-units
factor and a multi-way repeated measurements analysis of variance to the situation with more than one within
units factor. This is because the number of within factors, and not the number or between factor, dictates the
complexity of repeated measurements analysis of variance.,,,,
In this paper, we consider the linear one- way repeated measurements model which has only one within units
factor and one between units factor incorporating univariate random effects as well as the experimental error
term, we can represent repeated measurements model as a mixed model.
The asymptotic properties of the Bayes factor have been studied mainly in nonparametric density estimation
problems related to goodness of fit testing ,,,,,,. Choi, Taeryon and et al in (2009) 
studied the semiparametric additive regression models as the encompassing model with algebraic smoothing and
obtained the closed form of the Bayes factor and studied the asymptotic behavior of the Bayes factor based on
the closed form. Mohaisen, A. J. and Abdulhussain , A. M. in (2013) investigated large sample properties of
the Bayes factor for testing the pure polynomial component of spline null model whose mean function consists
of only the polynomial component against the fully spline semiparametric alternative model whose mean
function comprises both the pure polynomial and the component spline basis functions. In this paper, we
investigate the consistency property of Bayes factor for testing the fixed effects linear one- way repeated
measurements model against the mixed one- way repeated measurements alternative model. Under some
conditions on the prior and design matrix, we identify the analytic form of the Bayes factor and show that the
Bayes factor is consistent.

143

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

## 2. One- Way Repeated Measurements Model

Consider the model
yijk = + j + i(j) + k + ()jk + eijk

(1)

Where
i=1,.,n is an index for experimental unit within group j ,
j=1,,q is an index for levels of the between-units factor (Group) ,
k=1,,p is an index for levels of the within-units factor (Time) ,
yijk

## is the added effect for treatment group j ,

i(j)

is the random effect for due to experimental unit i within treatment group j ,

e ijk

## For the parameterization to be of full rank, we imposed the

following set of conditions
qj=1 j=0

pk=1 k=0

pk=1()jk=0

qj=1()jk=0

## for each j=1,,q .

And we assumed that the eijk and i(j) are indepndent with
eijk

~ i.i.d N (0,2
e)

ij

(2)

~ i.i.d N (0,2
)

## The model (1) is rewritten as follows

= + +

(3)

Where
y111
y112
Y=[ ]
ynqp

nqp1

1(1)

b = = 1(2)

[ n(q) ]

11
q1
, = [ p1 ]
()qp1

,
nq1

(qp+q+p+1)1

e111
e112
=[ ]
enqp

1P1
, Z=(
0P1

0P1
)
1P1 nqpnq

and = (1 , 2 , 3 , 4 ) is an
nqp1

144

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

## We assume < 1 + + + + , + = 1 + + + + , ( is integer number and greater

than 1 ), and we assume that the design matrix as the following way:

(4)

where:
= [ ] and be the [ (1 + + + )] matrix, and let = [ ].
We would like to choose between a Bayesian mixed one- way repeated measurements model and its one- way
repeated measurements model (the model without random effects) by the criterion of the Bayes factor for two
hypotheses,
0 : = +
i . e 0 : yijk = + j + k + ()jk + eijk
versus
1 : = + +

i.e

## 1 : yijk = + j + i(j) + k + ()jk + eijk .

(5)

We assume and are independent and the prior distribution for the parameters are
~ (0, 0 ) , where
2
01
01
0
1

0 = 2 1+++ =
[
and ~ (0, 1 )

01

01
01

0
0

, where
211

1 = 2 = 2 =

[ 0

0
212

0
0

(6)

2
()
]

(1+++)(1+++)

2 ]

(7)

2.Posterior distribution
From the model (1), we have the following:
| ~ ( , 2 ) , ~ (0 , 0 + 1 )
where
= + j + i(j) + k + ()jk + eijk ,for, = 1,2, , , = 1, , = 1, , .

(8)

## Then, the posterior distribution 1 ( | ) is the multivariate normal with

( | , 2 ) = 2 (2 + 2 )1 , where N=1+q+p+qp+nq and
2
1 1
( | , 2 ) = (1
= 2 (2 + 2 )1 2
2 + ( ) )
where
2
01
01
0
1

2 = 0 + 1 =
[
211
0

0
212

01

01
01

0
0

2
0

0
0

2
()
]

2 ]

0
[ 0
Then, ~ (0, 2 + 2 ).
(9)
2

## Hence, the distribution of under 0 and 1 are respectively (0, + 0 ) and

(0, 2 + 0 + 1 ).
(10)
We investigate the consistency of the Bayes factor, under the one- way repeated measurements model with
design matrix of the random effects () to the first terms. We get
yijk = + j + K + k + ()jk + eijk , K = 1, . , ( (1 + + + )).The covariance matrix of the
distribution of under 1 as:

145

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

2 + 0 + 1, = 2 +

211

## where 1,= 2 (1+++) =

212

0
[

(11)

2(1+++) ]

(1+++)(1+++)

01
= 2 (1+++) and = [ 0
] , where 01 , 01 are the matrices for zeros element with size
01 1,
(1 + + + ) ( (1 + + + ))
and
( (1 + + + )) (1 + + + )
,
respectively.
Then, we can rewrite (11) by using as:
[

+ ]
.

(12)

Then, the covariance matrix of the distribution of under 0 , can be represented as:

2 + 0 + 0 = 2 +
, where:

01
= [ 0
]
01 0
and 0 is the matrix for zeros element with size ( (1 + + + )) ( (1 + + + )), then
we can rewrite the covariance matrix of the distribution of under 0 , the following:
[

+ ]

(13)

and the covariance matrix of the distribution of under 1 , can be represented as:
2 + 0 + 1 = 2 + [ ] [ ] , where:

01
= [ 0
]
01 1
then we can rewrite the covariance matrix of the distribution of under 1 , the following:
2

[ ] [

+ ] [ ]

(14)

Lemma(1):Let the covariance matrices of the distribution of under and 1 be defined as in (12) and
(13) respectively, and suppose satisfy (4). Then, the following hold.

1
=
=
2 det(2 + 0 ) = 1+++ (

+ 2 )(

+ 2 ) (

+ 2 ) (

2
+ ()
)

(2 )(1+++)
3 det(2 + 0 + 1, ) =
() (

+ 2 ) (

+ 2 ) (

4 (2 + 0 )1 =

1
2

+ 2 ) (

2
+ ()
)
1

5 (2 + 0 + 1, )1 =

+ ]

1
2

+ 2 )(1+++)

+ ]

Proof:
1 By multiplying and
to equation (4) from right and left side, we have:

Multiplying (
)1 to the above equation from left side, we get:

(
)1 =
(
)1

=
2

2 2 + 0 = [

+ ]
, and by using (4), we get

146

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014
(

+ 2 )

01

01
= [ , ]

+ 2 )

01

01

01

01

01

+ 2 )
(

2
+ ()
)

+ 2 )

01

01

[ , ]

0
2

(1+++) ]

01

01

01

01
2

01

01

+ 2 )

01
=

www.iiste.org

+ 2 )

2
+ ()
)

0
2 (1+++) ]

Then
det(2 + 0 ) = 1+++ (

(2 )(1+++)

3 2 + 0 + 1, = [
(

2
+ 2 )

01

= [ ]

01

01

01
=

+ 2 )
(

01

01

2
+ ()
)

01

01

[ ]

(1+++) + 2 (1+++) )

01

01

01

2
+ 2 )

+ 2 ) (

2
2
+ ()
)

+ ]

+ 2 )

01

01

2
(
+ 2 )

+ 2 ) (

+ 2 )

01

01

01
2

+ 2 )(

2
+ ()
)

0
(

(1+++) + 2 (1+++) )

Then

4 (2 + 0 )1 = { [

=
=
=

()2

+ ]

1

1
()2

+ 2 ) (

+ ]
}1

1
1

+ ]
[

+ 2 )(

2 )(1+++)

det(2 + 0 + 1, ) = () (

+ ]

1
1

+ ]

(by lemma 1)

5 (2 + 0 + 1, )1 = { [

+ ]
}1

147

+ 2 ) (

2
+ ()
) (

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

=
=
=

+ ]

()2

1
()2

1
1

www.iiste.org

+ ]

+ ]

+ ]

## 3- The Bayes factor

The Bayes factor for testing problem (5) is given by:
m(yijk | )
01 (yijk ) =
m(yijk |1 )
1

2
2
2
2
(1+++)
21+++ ( +2
)( +2) ( +2
) ( +2
) (2
)

()

1
2
1
)

[
+ ]
}
2

1
2

(
exp{

1
1
1
exp{
(
[

2
2
2
2
2
2
2
2() ( +2 )( +2 ) ( +2 ) ( +2 ) ( +2 )

()

2 )( + 2 ) ( + 2 ) ( + 2 )
() ( +

()

][
+ ]

+ ]

[ ] [

[ ] )}

( +2 )

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( 2 )(1+++)
1+++ ( +

()

( [

+ ]

exp{

22

[ ] ) }.

(15)

Now let the approximation of 01 with design matrix of the random effects () to the first terms as
following
2

01 =

2 )( + 2 ) ( + 2 ) ( + 2 ) ( + 2 )(1+++)
() ( +

()

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( 2 )(1+++)
1+++ ( +

()

( [

+ ]

+ ]

exp{

22

) }.

(16)

Then,
2

log01 =

1
2

( [

log

+ ]

(1+++)

() ( +2 )

+ ]

+ ]

( [

+ ]

)
2

() ( +2 )(1+++)

1+++ (2 )(1+++)

2 { {[

()

] }
}
2

22

(1+++)

1+++ (2 )(1+++)

2 log01 = log

= log

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( 2 )(1+++)
1+++ ( +

()

2 log01 = log

2 )( + 2 ) ( + 2 ) ( + 2 ) ( + 2 )
() ( +

()

()(1+++) ( +2 )

(1+++)

(2 )(1+++)

()2 {

148

+ ]

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014
0

01

01
01

0
0

01

[01

01

= log(

01

www.iiste.org

01

0
0

0
0

0
0

2 +2 (1+++)
2

()2 2
2
( +2 )2

)(1+++)

{
}

where

1
()2

01

01

01
01

0
0

01

[01

01

01

0
0

0
0

0
0

()2 2

(2 +2 )2

)(1+++)

The Bayes factor for testing the model in (5) satisfy is consistent if 
01 =
,
0 under 0 ,
01 = 0
, 1 under 1 ,

(17)
(18)

## Lemma(2):let = (111 , , ) where are independent normal random variable

with mean + j + k + ()jk and variance 2 = 1, then there exist a positive constant 1 such that
1

(1+++)

,1

[=1

log(1 + 2 )
] > 1 , with probability tending to 1, i.e. log01
nqp2

nqp

## , in probability, where Snqp,1 = i=1

(1+nqp2
)

Proof:
Let Xnqp = ( X111 , X121 , , Xnqp ) are independent standard normal random variables, Note that Ynqp =d Xnqp +
and
T
Ynqp
Q nqp Ynqp = (Xnqp + )T Q nqp (X nqp + )
T
T
= Xnqp
Q nqp Xnqp + ()T Q nqp () + Xnqp
Q nqp () + ()T Q nqp Xnqp .
By the property of the design matrix (4) get
01

01

01

01

01
01

0
0

01

0
0

0
0

0
0

[01

()T

Q nqp =

()T

1
(nqp)2

Mnqp

T
Mnqp
=

1
()2

0
01

01
0

## [, , , ()] (1+++ , 0(1+++ ) 01

01

01
[

0
0

()2 2
2
( +2 )2

01
0
0
0

01
0
0
0

() 2
2

nqp2

(1+nqp2
)

= Snqp,1

Note that

149

( + )2

T
T
Then Ynqp
Q nqp Ynqp = Xnqp
Q nqp Xnqp .
Now by using the expectation formula of the quadratic form,,,, get
nqp

01
0
0
0

## (1+++ , 0(1+++ ) [, , , ()] = 0.

T
0 < (Xnqp
Q nqp Xnqp ) = tr(Q nqp ) = i=1

)(1+++)

)(1+++)
]

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

Q2nqp

(nqp)4

Mnqp

01

01
01

0
0

01

[01
0
01

01
0

01

01

01
[

(nqp)3

Mnqp

01

01
0
0
0

01
0
0
0

01

01

01
01

0
0

01

0
0

01

01

0
0

0
0

0
0

01
0
0
0

[01

www.iiste.org

()2 2
2
( +2 )2

)(1+++)

MTnqp

() 2
2

T
Mnqp
Mnqp

( + )2

)(1+++)
]

01

01

0
0

0
0

0
0

T
Mnqp

()2 2
2
( +2 )2

)2 (1+++)

and
1

tr(Q2nqp ) =

(nqp)2

nqp
i=1 (

(nqp)2 2
2
)
1+nqp2

nqp2

nqp
= i=1 (

1+nqp2

)2 = nqp3 (

1+nqp2

)2 = Snqp,2 .

## Using the variance formula of the quadratic form,,,, get

nqp

T
var(Xnqp
Q nqp Xnqp ) = 2tr(Q2nqp ) = 2Snqp,2 = 2 i=1 (
Snqp,3 Snqp,1

Let cnqp =
Pr {

2Snqp,1

)2 .

## , where Snqp,3 = =1 log(1 + 2 ) = log(1 + 2 ) . Then,

nqp(1+q+p+qp)

[i=1

Snqp,1

nqp2

1+nqp2

T
log(1 + nqp2 )2 Ynqp
Q nqp Ynqp ] cnqp } = Pr [

{YT
nqp Qnqp Ynqp }
Snqp,1

Snqp,3
Snqp,1

cnqp ]
= Pr [
= Pr [

{YT
nqp Qnqp Ynqp }

Snqp,1
1

E(YT
nqp Qnqp Ynqp )
Snqp,1

var(YT
nqp Qnqp Ynqp )

Snqp,3
Snqp,1

cnqp

T
T
{(Ynqp
Q nqp Ynqp ) E(Ynqp
Q nqp Ynqp )}

Snqp,1

Snqp,2 Snqp,1
(
)2
2Snqp,1

S2
nqp,1

nqp

8Snqp,2

(Snqp,3 Snqp,1 )2

E(YT
nqp Qnqp Ynqp )
Snqp,1

Snqp,3 E(YT
nqp Qnqp Ynqp )
Snqp,1

cnqp ]

## 0, from the Chebyshev inequality .

Since Snqp,3 Snqp,1 c1 Snqp,1 for some c1 > 0 and let cnqp c1 /2.
Then, there exists a positive constant C1 = c1 /2 such as
1

nqp(1+q+p+qp)

Snqp,1

[i=1

T
log(1 + nqp2 )2 Ynqp
Q nqp Ynqp ] > C1 , with probability tending to 1.

Then,
nqp
T
01 = nqp(1+q+p+qp)
2logB
log(1 + nqp2 )2 Ynqp
Q nqp Ynqp
in probability under H0 i.e.
i=1
nqp

01
logB

Lemma 3: Suppose we have the one- way repeated measurements model, then
1

,1

+ ]

[ ] [

+ ]

[ ] )

Proof:
2

Since[ ] [

+ ] [ ] [

+ ]
is the covariance matrix of

1
2

Thus, (

+ ]

[ ] [

2
1

]
)

1
2

150

+ ]

[ ] [

+ ]

[ ] ), also

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

2 { ((

+ ]

2
(1+=+)
(
=1

2 )

+ ]

## by the Cauchy-Schwarz inequality, where

1

) ( ) (

= { ((

+ ]

+ 2 ) + =1 (

+ ]
1

=[

www.iiste.org

+ 2 ) + =1 (

+ ]
)2 )}2

+ 2 ) + =1 (

2
+ ()
))

1
2

}] .

Since
2

2
((2+2)+(2+2)+(2 +2 )+(2+2() ))

1
((2 )+(2 )+(2 )+(2() ))

(1+++)
(
=1

2 [

(1+++)
2

( +2 )

2
((2)+(2 )+(2)+(2() ))

Then,

2
2
2
2
2
+2 )+( +2 )+( +2 )+( +2 ))
((

()

= (1), and

+ 2 )

2
+ 2 ) + =1 ( + 2 ) + =1 ( + 2 ) + =1 ( + ()
))
((

+ ]

( +2 )

()3
2
(2 +2 )

()3
2

(2 )

= ().

= {()2 }.

Since = 1 from the distances between the elements of design matrix , then 2 .
1

Thus ,
1
,1

,1

+ ]

1
,1
1

,1

2
,1

[ ] [

[ ] [

+ ]

+ ] [ ]

[ ] ) =
1

[ ] )

+ ]

2
+
=(1+++) (

[ ] ) )

]
} [ ] [

+ ]

1
2

2 [

+
=(1+++) 1

+ ]

+ ]

2
,1

1
2

[ ] [

=(1+++) 2

2
()2 2

(1+2
)

(1+2 )
()2

Thus
1

,1

+ ]

1
2

[ ] [

+ ]

[ ] )

(1+2 )
()2

Theorem(1): Suppose that the true model is the one- way repeated measurements model with fixed effect

only (the model under 0 in 5), and let 0 denote the true distribution of the data, with probability density
function 0 ( yijk |, j , k , ()jk ) = {( j k ()jk )/ } , where () is the standard normal
density, then, the Bayes factor is consistent under the null hypothesis 0 i.e

lim 01 = in probability 0 .

Proof:
1

det(2 +0 +1 )

det(2 +0 )

log 01 = log
1

det(2 +0 +1 )

det(2 +0 + 1, )

= log

( (2 + 0 )1 (2 + 0 + 1 )1 )
2

1
2

log

det(2 +0 + 1, )
det(2 +0 )

151

(2 + 0 ) 1
2

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

(2 + 0 + 1, )1 )
((2 + 0 + 1, )1 (2 + 0 +
2

1 )1 )
1
= log01 + log
2

det(2 +0 +1 )
det(2 +0 + 1, )

((2 + 0 + 1, )1 (2 + 0 +
2

1 )1 )
Since (2 + 0 + 1 ) (2 + 0 + 1, ) is a positive definite matrix,
(2 + 0 + 1 ) (2 + 0 + 1, ), thus
1

log

det(2 +0 +1 )

det(2 +0 + 1, )

(
((2 + 0 + 1, )1 (2 + 0 + 1 )1 ) ) = 2 ((2 +

0 + 1, )1 (2 + 0 + 1 )1 ) + () ((2 + 0 + 1, )1
(2 + 0 + 1 )1 )() ,
since (2 + 0 + 1, )1 (2 + 0 + 1 )1 , (2 + 0 + 1, )1
and (2 + 0 + 1 )1 are all nonnegative definite matrices, it follows that
0 () ((2 + 0 + 1, )1 (2 + (0 + 1 )1 )() ()
(2 + 0 + 1, )1 ().
By the property of design matrix (4), we get

=
= (1+++ , 0(1+++ ).
Now as the proof of lemma (3), we get:
() (2 + 0 + 1, )1 () =

1
()2

() [

+ ]

()

1
[, , , ()] (1+++ , 0(1+++ )
()2

+ 2 )

01

01

01

01

01

01

+ 2 )
0

01
[

01

+ 2 )

2
+ ()
)

0
(

+ 2 )(1+++) ]

## (1+++ , 0(1+++ ) [, , , ()]

= 2 [

+ 2 ]

+ =1 2 [

+ 2 ]

+ =1 2 [

+ 2 ]

+ =1()2 [

2
+ ()
]

= (1).
Since {
1
,1

1
((2
,1

+ 0 + 1, )1 (2 + 0 + 1 )1 ) > }

(( 2

1
2

1
(
+0 + 1, ) ( +0 +1 ) ) )

((2 +0 + 1, )1 (2 +0 +1 )1

,1

() (2 +0 + 1, )1 ()

1
,1

## Since (2 + 0 + 1, )1 (2 + 0 + 1 )1 is nonnegative definite

((2 + 0 + 1, )1 (2 + 0 + 1 )1 ) is nonnegative .

Therefore,
1

((2 + 0 + 1, )1 (2 + 0 + 1 )1 ) = (,1 )
2
1

((2 + 0 + 1, )1 (2 + 0 + 1 )1 )
2

,1
= (1)
Then, as in lemma (2) there exists a constant 1 such that

152

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014
1
,1

log 01

,1

www.iiste.org

## log 01 + (1) > 1 + (1).

That is,

01 exp( ,1 { 1 + (1)})

in probability 0

Lemma(4):
Assuming that the true model is the mixed one-way repeated measurements model (the model under 1 in 5). Let
= (1 , , ) where are independent normal random variables with mean + j + i(j) + k +

## ()jk and variance 2 = 1 and let = ( ) ( ).

Then, lim

= 1, in probability 1 .

Proof:
From the moment formula of the quadratic form of normal random variables, the expectation and variance of

quadratic form
are given by:

(
) = ( ) + ( ) ( )
(1+++)

= =1

(1+2 )

+ ( ) ( )

= ,1 + , and

(
) = 2 ( 2 ) + 4( ) 2 ( ).
Note

= ( ) ( )

= (, , , () , (1+++)
)

0
01

01
0

01

01

01
[

01
0
0
0

01
0
0
0

01
0
0
0

2
() 2
2
2
( + )2

0
01

(, , , () , (1+++)
) 1

(, , , ()

]
01

0
0

[01

01

()2 2
2
( +2 )2

01

, (1+++)
)

= (1+++)
(

)(1+++)

01

01

(, , , () , (1+++) )

0
0
0

()2 2

(2 +2 )2

) (1+++)

(1+++)
) (1+++) (1+++) = =1

()2 2 2
1+2

## since < , we have

2

1+2
2
consider
>0,

2 =1

= ( ,1 ), then = (2 ).

Now we can

1+2

1
2

(1+++)
= =1

1+2

2 2 2

1+2
2

2 2

1+2

## Similarly to the previous calculation, we have

( )
( ) = (, , , () , (1+++)
)

153

()3

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014
0
01

01
0

01

01

01
[

01
0
0
0

01
0
0
0

=
=

www.iiste.org

01
0
0
0

) (1+++)

( + )2

()2 2

(
((1+++) (2 +2 )2
1

(1+++)
=1

2
Since ( )
( )

() 2
2

(, , , () , (1+++) )

()4 4 2

(1+2 )2

3 4 2
()
2
=1
(1+2 )

)2 (1+++) (1+++)

.
2 4
()

2 =1

(1+2 )

= ( ,2 ) (as in case

of ( ) ( )).
Now for previous and we can find

(1+++)
2
( )
( ) = =1

()3 4 2

2
(1+2 )

2

(

2)

1+

(
) ( ,1 ).

## Furthermore, note that ( ) ( ) ( ) 2 ( ).

By the Chebshev inequality get
{|

Since lim
lim

(
)

(
)

2 2

(
)

|> }
,1

= lim (

= 1 , in probability

= (2 )

0, where > 0 .

)=1

Lemma(5):
Suppose we have the mixed one- way repeated measurements model, then
lim

log

det(2 +0 +1 )
det(2 +0 + 1, )

=0

Proof:

2
2
det(2 + 0 + 1 ) =1 (2 + 2 2 + =1 2 2 + =1 2 2 +
=1 () +

Since

2 2

= () =1 (

2
2
2 2
2 2 +
2 2
2 2
(

## =1 +=1 +=1 () +=1

).

And we have
det(2 + 0 + 1, ) = () (

det(2 +0 +1 )
log

+ 2 )(

+ 2 ) (

+ 2 ) (

2
+ ()
) (

1).

## = log det(2 + 0 + 1 ) log det(2 + 0 +

det(2 +0 +1, )

1, )

{()log() + =1 log (

log (

)log (

+ 2 ) + log (
2

2
2
2 2
2 2 +
2 2
2 2
(

+ 2 ) + log (

+ 2 ) + log (

2
+ ()
) + (1 + + +

Then
1

log

det(2 +0 +1 )
det(2 +0 + 1, )

Theorem(2):
154

)} {log() +

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

Suppose that the true model is the mixed one- way repeated measurements model (the model under 1 in 5),

and let 1
denote the true distribution of the data, with probability density function
1 (yijk |j , i(j) , k , ()jk ) = {(yijk ( + j + i(j) + k + ()jk ))/ }, where () is the standard normal
density. Then the Bayes factor is consistent under the alternative hypothesis 1 i.e

lim 01 = 0 in probability 1 .

Proof:
1

det(2 +0 +1 )

det(2 +0 )

log 01 = log

((2 + 0 ) 1 (2 + 0 + 1 ) )
2

=
1

log

det(2 +0 +1 )

0 +

1
1, ) )

det(2 +0 + 1, )

log

det(2 +0 + 1, )
det(2 +0 )

((2
2

+ 0 +

((2 + 0 ) 1 (2 +

2
1
1, )

(2 + 0 + 1 ) )
1
1
1
= log01
((2 + 0 ) 1 (2 + 0 + 1, ) )
2
2
1
1
((2 + 0 + 1, ) (2 + 0 + 1 ) ) , where
log01 =
1
2

log

det(2 +0 + 1, )
det(2 +0 )

((2 + 0 ) 1 (2 + 0 + 1, ) ) .
2

## From lemma (1), we have

det(2 +0 +1, )

2
2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( + 2 )(1+++)
() ( +

()

2
2
2
2
2

2
2
2
1+++

(
+ )(
+ ) (
+ ) (
+() ) (2 )(1+++)

(1+++)
2
+ 2 )
(1+++)
(1+++) (

2 +2
(1+++)
2
2

det(2 +0 )

=(

=(1 +

2
2

log (1 +

(1+++)

2
2

= ()

(1+++)

= ().

## And from lemma (4) = (2 ). Then

1

log

det(2 +0 + 1, )
det(2 +0 )

= (1 )

0.

Then,
1

log01 =

log

det(1, )

det(2 +0 )

1
((2

+ 0 ) 1 (2 + 0 +

1, ) ) ).
1

Since (2 + 0 ) 1 (2 + 0 + 1, )
((2

1
(4), lim

+ 0 ) 1 (2 + 0 + 1, ) ) =

log01

log

1, ) ) )
1

1
.

from lemma

= 1 , then

= , then
1

log01

det(2 +0 + 1, )
det(2 +0 )

(0 1) =

, in probability

1
((2

+ 0 ) 1 (2 + 0 +

1
2

1 .

## Now from lemma (5),

1

log

det(2 +0 +1 )
det(2 +0 + 1, )

0, and
((2 + 0 + 1, ) (2 +
2

(2

+ 0 + 1 )

155

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

Therefore, by combining the above results, there exist 2 > 0 such that
1

lim sup

Then , 01

log 01

## , with probability tending to 1.

0, in probability 1

4.Conclusions
1- The posterior distribution 1 ( | ) is the multivariate normal with
( | , 2 ) = 2 (2 + 2 )1 , where N=1+q+p+qp+nq and
2
1 1
( | , 2 ) = (1
= 2 (2 + 2 )1 2 , where
2 + ( ) )
= + j + i(j) + k + ()jk + eijk ,for, = 1,2, , , = 1, , = 1, , , and

2 = 0 + 1 =
[
211

01

01

01

01
01

0
0

01
0
0

2
()
]

212

2 ]
0
[ 0
2-The Bayes factor for testing problem 0 : yijk = + j + k + ()jk + eijk versus
1 : yijk = + j + i(j) + k + ()jk + eijk is given by:

01 =

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 )
() ( +

()

( +2 )

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( 2 )(1+++)
1+++ ( +

()

( [

+ ]

+ ]

exp{

22

) }.

3-The approximation of 01 with design matrix of the random effects () to the first terms as following
2

01 =

2 )( + 2 ) ( + 2 ) ( + 2 ) ( + 2 )(1+++)
() ( +

()

2
2
2
2
2 )( + 2 ) ( + 2 ) ( + 2 ) ( 2 )(1+++)
1+++ ( +

()

( [

+ ]

exp{

22

+ ]

) }.

## 4- If we have the one- way repeated measurements model, then

1
,1

+ ]

[ ] [

+ ]

[ ] )

5-If the true model is the one-way repeated measurement model with fixed effects only then the Bayes factor is

consistent under the null hypothesis 0 . This implies that, lim 01 = in probability 0 .
6- If we have the mixed one- way repeated measurements model, then
lim

log

det(2 +0 +1 )
det(2 +0 + 1, )

= 0.

7- If the true model is the mixed one- way repeated measurements model, then the Bayes factor is consistent

## under the alternative hypothesis 1 .This implies that, lim 01 = 0 in probability 1 .

5.References
 Aerts, M. , Claeskens, G. and Hart, J.D., (2004). Bayesian motivated tests of function fit and their
asymptotic frequentist properties, Ann. Statist. 32 (6), 2580-2615.
 Al-Mouel, A.H.S.,(2004) . Multivariate Repeated Measures Models and Comparison of Estimators, h. D.
Thesis, East China Normal University, China.
 Al-Mouel , A.H.S., and Wang, J.L,(2004). One Way Multivariate Repeated Measurements
analysis of
Variance Model, Applied Mathematics a Journal of Chinese Universities, 19,4,pp435448.

156

## Mathematical Theory and Modeling

ISSN 2224-5804 (Paper) ISSN 2225-0522 (Online)
Vol.4, No.11, 2014

www.iiste.org

 Arnold, Steven F., (1981). The theory of linear models and multivariate analysis. John Wiley and
Sons, New York. Chichester. Brisbane. Toronto .
 Berger, J. O., (1985). Statistical Decision Theory and Bayesian Analysis. Springer, New York.
 Choi, T., Lee, J. and Roy, A., (2008). A note on the Bayes factor in a semiparametric regression
model, Journal of Multivariate Analysis. 1316-1327.
 Dass, S.C. and Lee, J., (2004). A note on the consistency of Bayes factors for testing point null
versus
non-parametric alternatives, J. Statist. plan. Inference 119 (1) 143-152.
 Ghosh, J.K. , Delampady, M. and Samanta, T., (2006). Introduction to Bayesian Analysis: Theory and
Methods, Springer, New York.
 Gosal, G., Lember, J. and Van der Vaart, A., (2008). Nonparametric Bayesian model selection and
averaging, Electron. J. Stat. 2 63-89.
 Graybill, F.A., (1976). Theory and Application of the linear Model, Duxbury press, North Scituate, Mass,
xiv+704.
 Mohaisen, A. J. and Abdulhussain , A. M., (2013). Large Sample Property of The Bayes Factor in a Spline
Semiparametric Regression Model , Mathematical Theory and Modeling www.iiste.org ISSN
(Paper)2224-5804 ISSN (Online)2225-0522 Vol 3, No. 11, 2013.
 Neter, J. and Wasserman, W. , (1974). Applied linear statistical models, regression analysis of
variance
and experimental designs, Richard. D. IRWIN, INC .
 Papoulis,A.,(1997). Markoff Sequences in Probability, Random Variables, and Stochastic
Processes,
2nd ed. New York . McGraw-Hill, PP. 528-535, 1984.
 Tong, Y.L., (1989). The Multivariate Normal Distribution, Springer series in statistics, Springer.
 Verdinelli, I. and Wasserman, L., (1998). Bayesian goodness of-fit testing using infinitedimensional
exponential families, Ann. Stat.26 (4) 1215-1241.
 Vinish, R. Mc , Rousseau, J. and Mengerson, K., (2008). Bayesian goodness of-fit testing with
mixtures of triangular distributions, Scand. J, statist. in press (doi:10.1111/j.14679469.2008.00620.x).
 Vonesh, E.F. and Chinchilli, V.M., (1997). Linear and Nonlinear Models for the Analysis of
Repeated
Measurements, Marcel Dakker, Inc., New York.

157

## Business, Economics, Finance and Management Journals

European Journal of Business and Management
Research Journal of Finance and Accounting
Journal of Economics and Sustainable Development
Information and Knowledge Management
Journal of Developing Country Studies
Industrial Engineering Letters

EJBM@iiste.org
RJFA@iiste.org
JESD@iiste.org
IKM@iiste.org
DCS@iiste.org
IEL@iiste.org

## Physical Sciences, Mathematics and Chemistry Journals

Journal of Natural Sciences Research
Journal of Chemistry and Materials Research
Journal of Mathematical Theory and Modeling
Advances in Physics Theories and Applications
Chemical and Process Engineering Research

JNSR@iiste.org
CMR@iiste.org
MTM@iiste.org
APTA@iiste.org
CPER@iiste.org

## Engineering, Technology and Systems Journals

Computer Engineering and Intelligent Systems
Innovative Systems Design and Engineering
Journal of Energy Technologies and Policy
Information and Knowledge Management
Journal of Control Theory and Informatics
Journal of Information Engineering and Applications
Industrial Engineering Letters
Journal of Network and Complex Systems

CEIS@iiste.org
ISDE@iiste.org
JETP@iiste.org
IKM@iiste.org
CTI@iiste.org
JIEA@iiste.org
IEL@iiste.org
NCS@iiste.org

## Environment, Civil, Materials Sciences Journals

Journal of Environment and Earth Science
Journal of Civil and Environmental Research
Journal of Natural Sciences Research

JEES@iiste.org
CER@iiste.org
JNSR@iiste.org

## Life Science, Food and Medical Sciences

Advances in Life Science and Technology
Journal of Natural Sciences Research
Journal of Biology, Agriculture and Healthcare
Journal of Food Science and Quality Management
Journal of Chemistry and Materials Research

ALST@iiste.org
JNSR@iiste.org
JBAH@iiste.org
FSQM@iiste.org
CMR@iiste.org

## Education, and other Social Sciences

Journal of Education and Practice
Journal of Law, Policy and Globalization
Journal of New Media and Mass Communication
Journal of Energy Technologies and Policy
Historical Research Letter
Public Policy and Administration Research
International Affairs and Global Strategy
Research on Humanities and Social Sciences
Journal of Developing Country Studies
Journal of Arts and Design Studies

## PAPER SUBMISSION EMAIL

JEP@iiste.org
JLPG@iiste.org
NMMC@iiste.org
JETP@iiste.org
HRL@iiste.org
PPAR@iiste.org
IAGS@iiste.org
RHSS@iiste.org
DCS@iiste.org
ADS@iiste.org

The IISTE is a pioneer in the Open-Access hosting service and academic event management.
The aim of the firm is Accelerating Global Knowledge Sharing.
More information about the firm can be found on the homepage:
http://www.iiste.org
CALL FOR JOURNAL PAPERS
There are more than 30 peer-reviewed academic journals hosted under the hosting platform.
Prospective authors of journals can find the submission instruction on the following
page: http://www.iiste.org/journals/ All the journals articles are available online to the
readers all over the world without financial, legal, or technical barriers other than those
inseparable from gaining access to the internet itself. Paper version of the journals is also
available upon request of readers and authors.

MORE RESOURCES
Book publication information: http://www.iiste.org/book/

## IISTE Knowledge Sharing Partners

EBSCO, Index Copernicus, Ulrich's Periodicals Directory, JournalTOCS, PKP Open
Archives Harvester, Bielefeld Academic Search Engine, Elektronische Zeitschriftenbibliothek
EZB, Open J-Gate, OCLC WorldCat, Universe Digtial Library , NewJour, Google Scholar