Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
Introduction to Sampling Theory
Lecture 37Two Stage Sampling (Subsampling)
ShalabhDepartment of Mathematics and Statistics
Indian Institute of Technology Kanpur
1
Slides can be downloaded from http://home.iitk.ac.in/~shalab/sp
2
Two Stage Sampling With Unequal First Stage Units:
Consider two stage sampling when the first stage units are of
unequal size and SRSWOR is employed at each stage.
Let value of jth second stage unit of the ith first stage unit.
number of second stage units in ith first stage unit.
total number of second stage units in the population.
number of second stage units to be selected from ith
first stage units, if it is in the sample.
total number of second stage units in the sample.
:ijy
:iM
01
:N
ii
M M
:im
01
:n
ii
m m
3
( )1
1
1
1 1 1
1
1
1
1
1
1
1
1
i
i
i
N
i
m
i m ijji
M
i ijji
N
ii
MN N
ij i i Ni j i
i iNi
ii
ii
N
ii
y ym
Y yM
Y y YN
y M YY u Y
MN NM
MuM
M MN
Two Stage Sampling With Unequal First Stage Units:
4
Population
Cluster2
M2 Units
Cluster 1
M1 Units
Cluster N
MN Units
Cluster 2
M2 Units
Cluster 1
M1 Units
Cluster n
Mn Units
…
…
Population N Clusters
First stage sample n clusters
Cluster 2
m2 Units
Cluster 1
m1 Units
Cluster n
mn Units…
Second stage sample n clusters
Two Stage Sampling With Equal First Stage Units:
5
Two Stage Sampling With Equal First Stage Units:
Now we consider different estimators for the estimation of
population mean.
1. Estimator Based on the First Stage Unit Means in the Sample: Bias
2 ( )1
1ˆi
n
S i mi
Y y yn
6
1. Estimator Based on the First Stage Unit Means in the Sample: Bias
2 ( )1
1 2 ( )1
11
1
1( )
1 ( )
1
[ ]
1
.
i
i
N
n
S i mi
n
i mi
n
ii
i i
N
ii
E y E yn
E E yn
E Yn
m M
YN
Y
Y
Since a sample of size is selected out of units by SRSWOR
7
1. Estimator Based on the First Stage Unit Means inthe Sample: BiasSo is a biased estimator of and its bias is given by
This bias can be estimated by
2Sy Y
2 2
1 1
1 1 1
1
( ) ( )
1 1
1 1
1 ( )( ).
S S
N N
i i ii i
N N N
i i i ii i i
N
Ni ii
Bias y E y Y
Y M YN NM
M Y Y MNM N
M M Y YNM
2 ( ) 2
1
1( ) ( )( )( 1)
n
S i i mi Si
NBias y M m y yNM n
8
1. Estimator Based on the First Stage Unit Means in theSample: Biaswhich can be seen as follows:
where
2 1 2 ( ) 21
1
1
1 1( ) ( )( ) |1
1 1 ( )( )1
1 ( )( )
n
S i i mi Si
n
i i ni
N
Ni ii
N
NE Bias y E E M m y y nNM n
N E M m Y yNM n
M M Y YNM
Y Y
1
1 .n
n ii
y Yn
9
1. Estimator Based on the First Stage Unit Means in theSample: Bias
An unbiased estimator of the population mean is thus obtained
as
Note that the bias arises due to the inequality of sizes of the first
stage units and probability of selection of second stage units varies
from one first stage to another.
2 ( ) 21
1 1 ( )( ).1
n
S i i mi Si
Ny M m y yNM N
Y
10
1. Estimator Based on the First Stage Unit Means in theSample: Variance
TheMSE can be obtained as
2 2 2
( )21 1
2 22
1
2 2
1
( ) ( | ) ( | )
1 1 ( | )
1 1 1 1 1
1 1 1 1 1
S S S
n n
i i mii i
n
b ii i i
N
b ii i i
Var y E Var y n Var E y n
Var y E Var y in n
S E Sn N n m M
S Sn N Nn m M
22
2 2
1 1
1 1, .1 1
iMN
Nb i i ij ii ji
S Y Y S y YN M
where
22 2 2( ) ( ) ( ) .S S SMSE y Var y Bias y
11
1. Estimator Based on the First Stage Unit Means in the Sample: Estimation of VarianceConsider mean square between cluster means in the sample
It can be shown that
22( ) 2
1
1 .1
n
b i mi Si
s y yn
2 2 2
1
2 2( )
1
2 2 2
1
2 2
1 1
2 2 2
1 1 1( )
1 ( )1
1( ) ( )1
1 1 1 1 1 1 .
1 1 1( )
Also
So
Thus
i
i
N
b b ii i i
m
i ij i miji
M
i i ij iji
n N
i ii ii i i i
b b ii i i
E s S SN m M
s y ym
E s S y YM
E s Sn m M N m M
E s S E sn m M
1
n
12
1. Estimator Based on the First Stage Unit Means in the Sample: Estimation of Varianceand an unbiased estimator of is
So an estimator of the variance can be obtained by replacing
by their unbiased estimators as
2bS
2 2 2
1
1 1 1ˆ .n
b b ii i i
S s sn m M
2 2b iS Sand
2 22
1
1 1 1 1 1ˆ ˆ( ) .N
S b ii i i
Var y S Sn N Nn m M
13
2. Estimation Based on First Stage Unit Totals
where
Bias:
Thus is an unbiased estimator of .
( )*2 ( )
1 1
1 1ˆ n ni i mi
S i i mii i
M yY y u y
n M n
.ii
MuM
*2 ( )
1
2 ( )1
1 1
1( )
1 ( | )
1 1 .
n
S i i mii
n
i i mii
n N
i i i ii i
E y E u yn
E u E y in
E u Y u Y Yn N
*2Sy Y
14
2. Estimation Based on First Stage Unit Totals : Variance
* * *2 2 2
2( )2
1 1
*2 2 2
1
2 2
1
*2 2
1
( ) ( | ) ( | )
1 1 ( ) |
1 1 1 1 1
1 ( )1
1 ( ) .1
wherei
S S S
n n
i i i i mii i
N
b i ii i i
M
i ij iji
N
b i ii
Var y Var E y n E Var y n
Var u Y E u Var y in n
S u Sn N nN m M
S y YM
S u Y YN
15
3. Estimator Based on Ratio Estimator
where
This estimator can be seen as if arising by the ratio method of
estimation as follows:
*( ) ( )** 1 1 2
2
1 1
ˆ
n n
i i mi i i mii i S
S n nn
i ii i
M y u yyY yuM u
1
1, .n
ii n i
i
Mu u uM n
16
3. Estimator Based on Ratio Estimator:
be the values of study variable and auxiliary variable in
reference to the ratio method of estimation. Then
The corresponding ratio estimator of is
* * *2
1
* *
1
*
1
1
1
1* 1.
n
i Si
n
i ni
N
ii
y y yn
x x un
X XN
***2
2*ˆ * 1 .*
SR S
n
yyY X yx u
Y
* *( ) , 1, 2,...,i
i i i mi iMy u y x i NM
Let and
17
3. Estimator Based on Ratio Estimator:
So the bias and mean squared error of can be obtained
directly from the results of the bias and MSE of the ratio
estimator.
Recall that in ratio method of estimation, the bias of ratio
estimator up to second order of approximation is
where
**2Sy
2
2
2
ˆ( ) ( 2 )
( ) ( , )
ˆ( ) ( ) ( ) 2 ( , )
R x x y
R
N nBias y Y C C CNn
Var x Cov x yYX XY
MSE Y Var y R Var x RCov x y
.YRX
18
3. Estimator Based on Ratio Estimator Bias:
The bias of up to second order of approximation is
where is the mean of auxiliary variable similar to as
**2Sy
* * *** 2 2 2
2 2
( ) ( , )( ) S S SS
Var x Cov x yBias y YX XY
*
2Sy
*2 ( )
1
1 .n
S i mii
x xn
*2Sx
19
3. Estimator Based on Ratio Estimator:Bias
Now we find
where
* *2 2( , ).S SCov x y
* *2 2 ( ) ( ) ( ) ( )
1 1 1 1
2( ) ( ) ( ) ( )2
1 1 1
1 1 1 1( , ) , ,
1 1 1( ), ( ) ( , ) |
n n n n
S S i i mi i i mi i i mi i i mii i i i
n n n
i i mi i i mi i i mi i mii i i
Cov x y Cov E u x u y E Cov u x u yn n n n
Cov u E x u E y E u Cov x y in n n
Cov
22
1 1 1
* 2
1
1 1 1 1 1,
1 1 1 1 1
n n n
i i i i i ixyi i i i i
N
bxy i ixyi i i
u X u Y E u Sn n n m M
S u Sn N nN m M
*
1
1
1 ( )( )1
1 ( )( ).1
i
N
bxy i i i ii
M
ixy ij i ij iji
S u X X uY YN
S x X y YM
20
3. Estimator Based on Ratio Estimator: Bias
Similarly, can be obtained by replacing x in place of y
in as
Substituting and in we obtain
the approximate bias as
* *2 2( , )S SCov x y
*2( )SVar x
* *2 2 22
1
*2 2
1
*2 2
1
1 1 1 1 1( )
1 ( )1
1 ( ) .1
i
N
S bx i ixi i i
N
bx i iiM
ix ij iii
Var x S u Sn N nN m M
S u X XN
S x XM
where
**2 2** 2
2 2 21
1 1 1 1 1( ) .N
bxy ixybx ixS i
i i i
S SS SBias y Y un N X XY nN m M X XY
**2( ),SBias y*
2( )SVar x* *2 2( , )S SCov x y
21
3. Estimator Based on Ratio Estimator: MSE** * * * * *2 *
2 2 2 2 2
** *2 2 22
1
** *2 2 22
1
( ) ( ) 2 ( , ) ( )
1 1 1 1 1( )
1 1 1 1 1( )
S S S S S
N
S by i iyi i i
N
S bx i ixi i i
MSE y Var y R Cov x y R Var x
Var y S u Sn N nN m M
Var x S u Sn N nN m M
* ** * 22 2
1
*2 2
1
*2 2
1
*
1 1 1 1 1( , )
1 ( )1
1 ( )1
.
i
N
S S bxy i ixyi i i
N
by i ii
M
iy ij iji
Cov x y S u Sn N nN m M
S u Y YN
S y YM
YR YX
where
22
3. Estimator Based on Ratio Estimator: MSE
Thus
Also
** *2 * * *2 *2 2 2 * *2 22
1
1 1 1 1 1( ) 2 2 .N
S by bxy bx i iy ixy ixi i i
MSE y S R S R S u S R S R Sn N nN m M
2** 2 * 2 2 * *2 22
1 1
1 1 1 1 1 1( ) 2 .1
N N
S i i i i iy ixy ixi i i i
MSE y u Y R X u S R S R Sn N N nN m M
23
3. Estimator Based on Ratio Estimator: Estimate ofVarianceConsider
It can be shown that
So
* * *( ) 2 ( ) 2
1
( ) ( )1
11
1 .1
n
bxy i i mi S i i mi Si
n
ixy ij i mi ij i miji
s u y y u x xn
s x x y ym
* * 2
1
2 2
1 1
1 1 1
( ) .
1 1 1 1 1 1 .
N
bxy bxy i ixyi i i
ixy ixy
n N
i ixy i ixyi ii i i i
E s S u SN m M
E s S
E u s u Sn m M N m M
24
3. Estimator Based on Ratio Estimator: Estimate ofVarianceThus
Also
* * 2
1
*2 *2 2 2
1
*2 *2 2 2
1
1 1 1ˆ
1 1 1ˆ
1 1 1ˆ .
n
bxy bxy i ixyi i i
n
bx bx i ixi i i
n
by by i iyi i i
S s u sn m M
S s u sn m M
S s u sn m M
2 2 2 2
1 1
2 2 2 2
1 1
1 1 1 1 1 1
1 1 1 1 1 1 .
n N
i ix i ixi ii i i i
n N
i iy i iyi ii i i i
E u s u Sn m M N m M
E u s u Sn m M N m M
25
3. Estimator Based on Ratio Estimator: Estimate ofVarianceA consistent estimator of MSE of can be obtained by
substituting the unbiased estimators of respective statistics
in as
where
**2Sy
**2( )SMSE y
** *2 * * *2 *2 2 2 * *2 22
1
2 2 2 * *2 2( ) ( )
1 1
1 1 1 1 1( ) 2 2
1 1 1 1 1 1* 21
n
S by bxy bx i iy ixy ixi i i
n n
i mi i mi i iy ixy ixi i i i
MSE y s r s r s u s r s r sn N nN m M
y r x u s r s r sn N n nN m M
** 2
*2
.S
S
yrx