Janusz L. Wywiał
Uniwersytet Ekonomiczny w Katowicach
SAMPLING DESIGNS PROPORTIONATE TO NON-NEGATIVE FUNCTIONS OF TWO QUANTILES OF AUXILIARY VARIABLE
1. Sampling design
Let U be a fixed population of size N. The observation of a variable un- der study and an auxiliary variable are denoted by yi and xi, i=1,...,N, re- spectively. Moreover, let xi ≤xi+1, i=1,..., N−1. Our problem is estimation of the population average
= ∑
∈U k
y
ky N 1
.
Let us consider the sample space S of the samples s of the fixed effective size 1<n <N. The sampling design is denoted by
P (s )
. We assume that0 ) ( ≥ s
P
for all s∈S and∑
∈=
s S
P ( s ) 1
.Let
( X
( j))
be the sequence of the order statistics of observations of auxilia- ry variable in the sample s. It is well known that the sample quantile of order aα
∈(0;1) is defined as follows:Q
s,α= X
(r) wherer = [ n α ] + 1
, the func- tion[ n α ]
means the integer part of the value[ n α ]
, r =1,2,...,n. Let us note thatX
(r)= Q
s,α forn r n
r − 1 ≤ α <
. In this paper it will be more conveniently to consider the order statistic than the quantile.Let
G ( r , u , i , j ) = { s : X
(r)= x
i, X
(u)= x
j}
, r=1,...,n−1; u=2,...,n, ur < be the set of all samples whose r-th and u-th order statistics of the auxi- liary variable are equal to xi and
x
j, respectively where r≤i< j≤N−n+u. Moreover,U U
-n rN
S
+
= +
−
− +
=
=
r i
u n N
r u i j
j i u r
G ( , , , )
(1)The size of the set G(r,u,i,j) is denoted by ))
, , , ( ( )
, , ,
(r u i j Card G r u i j
g = and
⎟⎟ ⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
= −
u n
j N r
u i j r
j i i u r
g 1
1 1
) 1 , , ,
(
(2)Let
f ( x
j, x
j, c )
be non-negative function of values of the order statisticsX
(r) andX
(r) and⎟⎟ ⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
= −
u n
j N r
u i j r
j i i u r
g 1
1 1
) 1 , , ,
(
(3)The straightfoward generalization of the Wywiał’s (2009) sampling design is as follows.
Definition 1.1. The conditional sampling design proportional to the non- negative functions of the order statistics
X
(u), X
(r) is as follows:) , , (
) , , ) (
|
(
( ) ( ),
z r u c
x X X c f
s
P
ru=
u r (4)where i < j and r≤i≤N−n+rand r<u≤ j≤N−n+u. Particularly, let
f ( x
j, x
i, c ) = x
j− x
i and⎩⎨
⎧
<
−
≥
−
= −
. 0
) , , ,
( for x x c
c x x for x c x
x x f
i j
i j i j i
j (5)
We say that the above sampling design is (conditional) unconditional when
)
0
( > c
c=0. In general this concept is agree with definition of the conditional sampling design considered by Tillé (1999; 2006).As it is well known the inclusion probability of the first order is determined
by the equations:
= ∈ = ∑
∈} :
{ ,
,
( : ) ( | )
) , ,
(
ru sk s ruk
r u c P s k s P s c
π
, k=1,...,N. We assume that if x≤0 then
δ ( x ) = 0
otherwiseδ ( x ) = 1
. Let us note that)
1 ( ) 1 ( )
( x δ x − = δ x −
δ
.Theorem 1.1. Under the sampling design
P
r,u( s )
the inclusion probabili- ties of the first degree are as follows. If k <r,) , , 1 (
1 2
2 )
, , (
) ( ) 1 ) (
, ,
( f x x c
u n
j N r
u i j r i c
u r z
k r c r
u
r j i
r n N
r i
u n N
r u i j
k ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
= −
∑ ∑
− += +
−
− +
=
δ
π δ (6)
If r≤k≤N−n+u,
− ⋅
= −
) , , (
) ( ) 1 ) (
, ,
( z r u c
k r c r
u
k r
δ π δ
⎜ +
⎜
⎝
⎛ ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
−
⋅ −
∑ ∑
+ −=
−
− +
=
) , , 1 ( 1 1
1 1
) 1 ( ) (
1 1
c x x u f
n j N r
u i j r u i
n u
k j i
r u k
r i
k
r u i j
δ δ
+ ⎟⎟⎠ +
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
−
∑
− +=
) , , 1 ( 1 1
) 1 1
( f x x c
r u
i k r i u
n k u N
k j i
r u k
r i
δ
∑ ∑
−= +
− +
=
⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
−
− +
−
−
+ 1
1
) , , 1 (
1 1
) 1 1 (
) (
) (
k r i
i j u
n N
k j
c x x u f n
j N r u
i j r r i
u k u n N r
k δ δ
δ (7)
⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
− +
−
+
∑
− +− +
=
) , , 1 (
1 1
) 1 1
( f x x c
u n
j N r
u k j r
k k r n
N j k
u n N
r u k j
δ
⎟⎟
⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
− +
−
+
∑ ∑
− ++
= +
−
− +
= r n N
k i
i j u
n N
r u i j
c x x u f n
j N r
u i j r r i
k r n N
1
) , , 1 (
1 2
) 2 1 ( )
( δ
δ
If k>N−n+u,
) , , 1 (
1 1
1 1
1 )
, , (
) ) (
, ,
( f x x c
u n
j N r u
i j r i c
u r z
u n N c k
u
r j i
r n N
r i
u n N
r u i j
k ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− − +
= −
∑ ∑
− += +
−
− +
=
π δ (8)
The inclusion probabilities of the second order are defined by
∑
∈ ∈=
∈
∈
=
=
} , : {
, ,
,
,
( , , ) ( , , ) ( : , ) ( | )
s t s k s
u r u
r k
t t
k
r u c π r u P s k s t s P s c
π
where k<t=1,...,N.
Theorem 1.2. The inclusion probabilities of the second degree of the sam- pling design
P
r,u( s | c )
are as follows.+
∈
∈ +
∈
= +
∈
= ( , ) ( , ) ( , )
) , ,
(
3 ( ) 3 2 3,t
r u c P k t s P X
ux
kt s P k s t s
π
k+
∈
∈ +
∈
∈ +
∈
=
+ P ( X
(r)x
k, t s
3) P ( k s
1, t s
3) P ( k s
2, X
(u)x
t)
(9)+
∈ +
=
∈ +
=
=
+ P ( X
(r)x
k, X
(u)x
t) P ( k s
1, X
(u)x
t) P ( k , t s
2)
) , ( ) ,
( ) , ( ) ,
( X
( )x t s
2P k s
1t s
2P k s
1X
( )x P k t s
1P
r=
k∈ + ∈ ∈ + ∈
r∈
t+ ∈
+
where
( − + − ⋅
−
= −
∈ ( k N n u )
) c , u , r ( z
) u n ) ( s t , k (
P δ 1 δ
3
. ⎟⎟⎠ + − + − + − ⋅
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
∑ ∑
−+ −= +
−
− +
=
) ( _ 1 (
) , , 2 ( 2 1
1 1
1 f x x c N n u k k u
u n
j N r u
i j r i
i j r
n N
r i
u n N
r u i j
δ δ
. ⎟⎟
⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
∑ ∑
− + −= +
−
−
=
) , , 2 ( 2 1
1 1
1
1
c x x u f
n j N r u
i j r i
i j r
u N
r i
u n N
k j
, (10)
+ ⋅
− +
−
−
= −
∈
= ( , , )
) 1 (
) ( ) ) (
,
( ( ) 3
c u r z
k u n N u n k s N
t x X
P u k δ δ δ
. ( , , )
1 1 1
1 1
1 f x x c
r u
i k r
i u
n k N
i j r
n k
r i
∑
− += ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
− , (11)
+ ⋅ +
−
−
−
−
= −
∈
∈ ( , , )
) 1 (
) ( ) 1 ) (
,
(
2 3c u r z
r u k t u n r
s u t s k
P δ δ δ
⎜ +
⎜
⎝
⎛ ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
− +
−
⋅
∑ ∑
−=
−
− +
=
) , , 1 ( 1 2
2 1
1 1 (
1 1
c x x u f
n j N r
u i j r t i
u n
N j i
k r i
t r u i j
δ (12)
⎟⎟⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− − +
−
+
∑ ∑
−= +
−
− +
=
) , , 1 ( 1 2
2 1
) 1 (
1
c x x u f
n j N r
u i j r u i
n N
t j i
k r i
u n N
r u i j
δ ,
+ ⋅
−
−
= −
∈
∈ ( , , )
) 1 (
) ( ) ) (
,
( ( ) 3
c u r z
r k k N u s n
t x X
P r k
δ δ δ
⎜ +
⎜
⎝
⎛ ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
− +
− +
−
−
⋅
∑
−− +
=
) , , 1 (
1 1
1 1
) 1 1 (
) (
1
c x x u f n
j N r u
k j r
t k u n N r u k
t j i
t r u k j
δ
δ (13)
)
) , , 1 ( 1 1
1 1
) 1 (
) 1
( f x x c
u n
j N r u
k j r
u k n N t r k n
N j k
u n N
r u k
j ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− − +
− + +
−
−
+
∑
−+− +
=
δ
δ ,
(
− − + − ⋅−
= −
∈
∈ ( ) ( )
) , , (
) ( ) 1 ) (
,
( 1 3 r k t N n u
c u r z
u n s r
t s X
P δ δ δ δ
⋅
− +
− +
−
−
⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅
∑ ∑
− + −= +
−
− +
=
) ( ) 1 (
) ( ) , , 1 ( 1 1
1 2
2 f x x c r k N n u t t u
u n
j N r u
i j r i
i j r
n N
r i
u n N
r u i j
δ δ
δ
⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅ −
∑ ∑
+ − −=
−
− +
=
) , , 1 (
1 1
1 2
2
1 1
c x x u f
n j N r u
i j r
i
i j r
u t
r i
t r u i j
(14)
⋅
−
− +
−
− +
− +
−
+δ(k r 1)δ(t N n u)δ(N n r k)δ(t u)
⎟⎟
⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅ −
∑ ∑
+ − −+
=
−
− +
=
) , , 1 (
1 1
1 2
2
1
1 1
c x x u f
n j N r
u i j r i
i j r
u t
k i
t
r u i j
,
) , , 1 ( 1 1
1 )
, , (
) ( ) ) (
, (
1 )
(
2 f x x c
r u
i t r i u
n t N c u r z
r u u x t
X s X
P j i
r u t
r i t
u −
∑
+−= ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
−
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
− −
= −
=
∈ δ δ .
⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅
−∑ ∑
+ −−
=
−
− +
=
) , , 1 (
1 1
1 2
2
1 1
c x x u f
n j N r
u i j r
i
i j r
u t
r i
t r u i j
(15)
⋅
−
− +
−
− +
− +
−
+ δ ( k r 1 ) δ ( t N n u ) δ ( N n r k ) δ ( t u )
⎟ ⎟
⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅
−∑ ∑
+ −−
+
=
−
− +
=
) , , 1 (
1 1
1 2
2
1 1
1
c x x u f
n j N r
u i j r
i
i j r
u t
k i
t r u i j
,
=
∈
= , )
( X
(r)x
kX
(u)x
tP
(
( , , )1 1 1
1 )
, , (
) 1 (
) 1
( f x x c
u n
t N r
u k t r k c
u r z
t u n N r
k
i
⎟⎟ j
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
− +
− +
=
δ
−δ
, (16)
− ⋅
= −
=
∈ ( , , )
) ( ) 1 ) (
,
(
1 ( )c u r z
r u x r
X s k
P
u tδ δ
⎜⎜⎝
⎛ ⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
−
−
⋅
∑
− +=
) , 1 ( 1 2
) 2 1 (
)
( j i
r u t
r i
x x r f
u i t r
i u
n t u N
t k
r
δ
δ
(17), ) , 1 ( 1 2
) 2 (
) 1 (
1 ⎟⎟⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− − +
− +
−
⋅
∑
− ++
= j i
r u t
k i
x x r f
u i t r
i u
n t k N r u t r
k
δ
δ
− ⋅
−
− +
−
= −
∈ ( , , )
) 2 (
) 1 ) (
,
(
2c u r z
r u k
t r s u
t k
P δ δ
(18)
)
, , 3 (
3 1
1
1 1
c x x u f n
j N r
u i j r
i
i j k
r i
u n N
t
j
⎟⎟ ⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅ ∑ ∑
−−
= +
− +
=
,
− ⋅
− +
+
−
= −
∈
∈ ( , , )
) 1 (
) 1 ) (
,
(
( ) 2c u r z
r u r
k n s N
t x X
P
r kδ δ
) , , 2 (
2 1
1 f x x c
u n
j N r
u k j r
k
i j u
n N
r u k
j ⎟⎟
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅ −
∑
− +− +
=
,
(19)
− ⋅ +
−
−
−
= −
∈
∈ ( , , )
) )
1 (
) 1 ) (
,
(
1 2c u r z
t u n N r
u s r
t s k
P δ δ δ
⎜ ⎜
⎝
⎛ ⎟⎟ +
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
⋅ ∑ ∑
−= +
− +
=
) , , 2 (
2 2
) 2 (
1 1
c x x u f n
j N r
u i j r
k i
r
j it r i
u n N
t j
δ
(20), ) , , 2 (
2 2
) 2 1 (
1
1 1
⎟ ⎟
⎠
⎟⎟ ⎞
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
− + −
−
+ ∑ ∑
−+
= +
− +
=
c x x u f n
j N r
u i j r
r i
k
j it k i
u n N
t j
δ
− ⋅ +
−
−
= −
=
∈ ( , , )
) (
) 1 ( ) 1 ) (
,
(
1 ()c u r z
t u n N k
x r X s k
P
r tδ δ δ
(21) )
, , 1 (
1 2
2
1
c x x u f n
j N r
u k j r
k
k j u
n N
t
j ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
⋅ −
∑
− ++
=
,
⎜ +
⎜
⎝
⎛ ⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
= −
∈
∑ ∑
− += +
−
− +
=
) , 1 (
1 3
) 3 ) (
, , (
) 2 ) (
,
( 1
r n N
r i
u n N
r u i j
i jx c x u f n
j N r u
i j r t i
c r u r z s r t k
P δ δ
).
, 1 (
1 3
) 3 1 (
∑ ∑
− +1 +=
+
−
− +
=
⎟⎟
⎠
⎜⎜ ⎞
⎝
⎛
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
⎟⎟ −
⎠
⎜⎜ ⎞
⎝
⎛
−
− −
−
+
N n rt i
u n N
r u i j
i j
x c x u f n
j N r
u i j r
r i δ t
(22)
2. Sampling scheme
The sampling scheme implementing the sampling design
P
r,u( s | c )
P is as fol- lows. Firstly, population elements are ordered according to increasing values of the auxiliary variable. Let s=s1∪{i}∪s3 ∪{j}∪s3, s1 ={k:k∈U,xk <xi},} ,
:
2
{ k k U x
jx
kx
is = ∈ > >
ands
3= { k : k ∈ U , x
k> x
j}
. Moreover, let ), 1 ( } { ) 1 , 1 ( } { ) 1 , 1
( i i U i j j U j N
U
U= − ∪ ∪ + − ∪ ∪ + where U(1,i−1)=
) 1 , ...
, 1
( −
= i , U(i+1, j−1)=(i+1,..., j−1), U(j+1,N)=(j+1,...,N). Let
)
);
1 , 1 (
( U i s
S −
be sample space of the samples
1 of size ));
1 , 1 ( ( ,
1SU i j s
r− + − be sample space of the sample
s
2 of size ));
, 1 ( ( ,
1SU j N s
r
u− − + be sample space of the sample
s
2 of size n −u. Similarly,S = S ( U , s )
.The sampling scheme is given by the following of probabilities:
) ( )
| ( ' ) ( )
| ( ) ( )
|
(
1 1 , 2 2 , 3 3,
s c P s p i c P s p j c P s
P
ru=
ru ru (23)where
1 3
3 1 2
2 1 1
1 , ( )
1 ) 1
( 1 ,
) 1 (
−
−
−
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
= −
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
−
−
= −
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛
−
= −
u n
j s N
r P u
i s j
r P s i
P (24)
) , (
) , ,
) ( ,
| (
)
| (
) ( ,
) ( )
( , )
( )
(
, P X x c
c x X x X c P
x X x X P c i P
j u u r
j u i r u r j
u i r u
r =
=
= =
=
=
= (25)
∑
− +=
=
=
= N n r
r i
i j j
u u r u
r f x x c g r u i j
u r c z x X P c j
p ( , ) ( , , , )
) , , ( ) 1 , (
)
| (
' , , ( ) (26)
) , , (
) , , , ( ) , ) (
| ( )
,
| (
) , , , (
, )
( )
(
,
z r u c
j i u r g c x x c f s P c
x X x X
P
j ij i u r G s
u r j
u i r u
r
∑
∈
=
=
=
(27)In order to select the sample s, firstly the j-th element of the population should be selected, according to the probability function
p '
r,u= ( j | c )
. Next, the i-the element of the population should be drawn according to the probability functionp '
r,u= ( i | c )
. Finally, the sampless
1,s
2 and s3 should be selected, according to the sampling designsP
1( s
1)
,P
2( s
2)
and P3(s3), respectively.
3. Some sampling strategies
The well known Horvitz-Thompson (1952) estimator is as follows:
∑
∈=
s
k k
k s
HT
y y N
π 1
, (28)
The statistic is unbiased estimator of the population mean value if
π
k >0 for k=1,...,N. The variance and its estimator are determined by the expressions (32) and (34), respectively.The particular case of the above estimator is the well known sampling desi- gn of the simple sample drawn without replacement is as follows:
1 0( )
−
⎟⎟⎠
⎜⎜ ⎞
⎝
=⎛ n s N
P . The variance of the mean from the simple sample
∑
∈=
s k
s
y
ky n 1
drawn without replacement is s
v
ynN
n s N
P y
D
2( ,
0( )) = −
where∑
=− −
=
Nk k
y
y y
v N
1
)
21 (
1
.The results of the previous chapter lead to construction of the regression sampling strategy for the population mean =
∑
ks
s yi
y n1 . We assume that
i i
i a bx e
y = + + for all i ∈U,
∑
∈
=
U i
e
i0
and the residuals of that linear re- gression function are not correlated with the auxiliary variable. The linear corre- lation coefficient between the variables y and x will be denoted byρ
. Let) ,
( X
(r)Y
r be two dimensional random variable whereX
(r) is the r-th order statistic of an auxiliary variable and Yr is the variable under study. Wywiał (2009) considered the following estimator:)
(
,, , , ,
,us HTs rus HTs
r
y b x x
y = + −
(29)where
) ( ) ( , ,
r u
r u s
u
r X X
Y b Y
−
= − (30)
Wywiał (2009) showed that under the sampling design stated in the defini- tion 1.1 the parameters of the following strategies
y
r,u,s, P
r,u( s | c ))
, are approximately as follows.( y P s c ) y
E
r,u,s,
r,u( | ) ≈
+
−
≈ ( , ( | )) 2 ( , , ( | ))
))
| ( ,
( , , , , ,
2 , ,
,
2 y P s c D y P s c bCov y x P s c
D rus ru HTs ru HTs HTs ru (31)
))
| ( ,
( , ,
2
2D x P s c
b HTs ru
+ where
⎟⎟⎠
⎜⎜ ⎞
⎝
⎛ Δ
=
∑∑
∈ ∈U
k l
l k k U
l l u k
s r s HT HT
x y c N
s P x y
Cov , , , 2 ,
π π
)) 1
| ( , ,
( (32)
l k l k l
k,
= π
,− π , π
Δ
,))
| ( , , ( ))
| ( ,
( , , , , ,
2 x P s c Cov x x P s c
D HTs ru = HTs HTs ru ,
))
| ( , , ( ))
| (
,
, , , ,,
2