Dependent Observations

(I) Quasi-likelihood estimating equations

Suppose now that positive-definite

V (Y ) = 2V ( ) , where V ( )

is a symmetric

nn

matrix of known functions

Vij ( ) ,

no

longer diagonal. The quasi-score function is

( ) =
and

D tV

(y

),

where

D = [Dir ]n p

Dir =
1

i r

. The score function has the

following properties,

E [U ( )] = 0, Cov[U ( )] =

U ( ) t 1 = = D V D i E . 2
of the estimating

Under suitable limiting conditions, the root equation

t V U = D

( )

(y

)=

0 ,

is approximately unbiased for distributed with limiting variance

and asymptotically normally

2 D tV 1 D Cov
Note:

()

1 = i .

Block-diagonal covariance matrices arise most commonly in longitudinal studies, in which repeat measurements made on the same subject are usually positively correlated. A well-known application is the generalized estimating equation proposed by Liang and Zeger (1986).

(II) Quasi-likelihood function

The quasi-score function for dependent observations is different from the one for independent observations in that U r ( ) U s ( ) (for dependent observatio ns ) s r

2Q ( ) U r ( ) U s ( ) 2Q ( ) (for indpendent observatio ns ) = = = r s s r s r

The line integral

Q ( , y , t (s )) =

t 1 ( ) ( ) y t [ V t ] n n dt n 1 (s ) , 1 n t ( s )= y

t ( s 1 )=
0

along a smooth path t (s

in

from t (s 0

)=

to

t (s 0

)=

is path independent if and only if

2Q( , y, t (s )) 2Q( , y, t (s )) = r s s r
where y =

[y 1

y2

yn

]t

Note:
Let

P ( x , y ) F = , t (s ) = ( ) Q x y ,

x (s ) y (s ) . Then,

( )
t a

t (b )

Fdt (s ) = P [x (s ), y (s )]x ' (s ) + Q [x (s ), y (s )]y ' (s ) ds .

b a

For example,

2 xy , t (s ) = F = 2 2 + x y

x (s ) s y (s ) = s 2 , 1 s 3 .
2

Then,
t (3 ) t 1

( ) F dt (s ) = [2 s
3 1

1 + s + s 2 s ds =
2 4

) ]

(4 s
3 1

+ 2 s 5 ds

968 3

If V

( ) =

''

( ),

= b ' ( ), = b

( ) ( ) , b( )
' 1

is some function, then can construct

Q ( , y , t (s )) is path-independent. Thus, we
k

( ) =

j =1

A tj W

(A

)A j

which will ensure the path-independent of construct the

Q ( , y , t (s )) . After we
further construct the

V 1 ( )

we

can

quasi-likelihood function. Consider the straight-line path

t (s
so that

)=

y +

y )s , 0 s 1 ,

t (0 ) = y and t (1) = . Provided the straight-line path

1 ( ( ) ) [ ] s V t s ds 0 (y )
1

exists, the quasi-likelihood function is given by

t 1 Q ( , y ) = ( y ) 2

If

V 1 ( )

is approximately linear in t over the straight-line path,

the above integral may be approximated by

Q ( , y )

1 t ( ) y V 1 ( )( y ) 2 3 1 t ( ) y V 1 ( y )( y ) 2 6