Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling Nursel KOYUNCU Hacettepe University, Department of Statistics, Beytepe, Ankara, Turkey
[email protected]
Abstract In this paper new exponential type estimators are suggested for estimating the population mean in stratified random sampling. The biases and mean square errors of the suggested estimators are obtained. Also an empirical study is carried out to show the properties of the proposed estimators.
Keywords: Ratio estimator; Exponential estimator; Mean square error; Efficiency; Stratified random sampling.
1.
Introduction
In sampling theory the role of auxiliary information has a great importance. It is well known that using suitable auxiliary information such as population total, mean, skewness, correlation, attribute etc. can improve precision of estimates. Many authors have used this information in their ratio, product and exponential type estimators to get more efficient estimator under different sampling design. In stratified random sampling scheme the authors including Diana (1993), Singh and Vishwakarma (2008), Koyuncu and Kadilar (2009, 2010), Tailor et al. (2012), Koyuncu (2013), Yadav et al. (2014) etc. have suggested estimators using auxiliary information. In this paper we have tried to generalize Singh et al. (2008) and Yadav et al. (2014) estimators and suggested two exponential type estimators in stratified random sampling. Consider a finite population U u1 , u 2 ,..., u N of size N and let y and x, respectively, be the study and auxiliary variable associated with each unit u j j 1, 2, ...., N of the population. Assume that the population of size, N, is stratified into L strata with h-th stratum containing N h units, where h 1, 2, ..., L such that
L
N h 1
h
N . A simple random
sample of size nh is drawn without replacement from the h-th stratum such that L
n h 1
h
n . Let y hi , xhi denote the observed values of y and x on the i-th unit of the h-th nh
Nh L y hi y , y st Wh y h , and Yh hi , i 1 n h i 1 N h h 1 L N Y Yst WhYh be the sample and population means of y, respectively, where Wh h is N h 1 the stratum weight.
stratum, where i 1, 2, ..., N h . Moreover, let y h
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Nursel KOYUNCU
To obtain the bias and MSE, let us define e0 y st Y Y and e1 xst X X . Using these notations, E e0 E e1 0 , L
Vrs W h 1
r s h
E y h Yh xh X h . Y rXs r
s
(1)
Using (1), we can write L
Ee
2 0
Wh2 h S yh2 h 1
Y2
L
V20 , E e
2 1
Wh2 h S xh2 h 1
X2
L
E e0 e1
V02
W h 1
2 h
h S xyh
XY
V11
where
y Nh
S
2 yh
i 1
hi Yh
Nh 1 n and f h h . Nh
x Nh
2
,
S
2 xh
i 1
hi X h
Nh 1
x Nh
2
, S xyh
i 1
hi
X h y hi Yh Nh 1
, h
1 fh , nh
Singh et al. (2008) have suggested estimators of population mean using auxiliary information, given by X x1 (2) t1 y st exp 1 X 1 x1 X x2 t 2 y st exp 2 X 2 x2 L
x1 Wh x h C xh ,
where
h 1
(3) X 1 Wh X h C xh , L
h 1
L
x 2 Wh x h 2 h x
and
h 1
X 2 Wh X h 2 h x . L
h 1
To generalize t1 and t 2 estimators in a class we can define X x st k1 y st exp X x st 2 Ast
(4)
L
where Ast Wh Ah and Ah may be some population information of the auxiliary h 1
variable for h-th stratum such as S xh , coefficient of variation C xh , skewness 1h x , kurtosis 2 h x , correlation coefficient h xy . Note that t1 and t 2 estimators are member of k1 if we take Ah as C xh and 2 h x respectively as given in Table2. Expressing (4) in terms of e’s, we have
k
1
430
X 3X 2 X 2 Y Y e1 e e e e 1 0 0 1 2 2 X Ast 2 X Ast 8 X Ast
(5)
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
Squaring both sides of (5) and neglecting the terms of e’s having greater than two, we can write
k
1
2 X2 X Y Y 2 e02 e2 e0 e1 2 1 X Ast 4 X Ast
(6)
Taking expectation of (6) we get the MSE of k1 , under the first order of approximation is given by L L Y2 Y L 2 2 2 2 MSE k1 Wh2 h S yh W S W S h h xyh 2 h h xh X Ast h1 4X Ast h1 h 1 L Y2 Y 2 2 Wh2 h S yh S S xh xyh 2 X A h 1 4 X A st st L 2 1 2 2 (7) Wh2h S yh Rk1S xh Rk1S xyh 4 h 1 Y where Rk1 . X Ast Secondly Singh et al. (2008) have suggested estimators using more auxiliary information X x3 (8) t 3 y st exp 3 X x 3 3 X x4 t 4 y st exp 4 X 4 x4
(9) X 3 Wh X h 2 h x C xh ,
L
L
where x3 Wh xh 2 h x C xh , h 1
h 1
x4 Wh xh C xh 2 h x , X 4 Wh X h C xh 2 h x . L
L
h 1
h 1
To generalize t 3 and t 4 estimators let us define following class
A* a * k 2 y st exp * st * st Ast a st 2 Bst
(10)
L
L
L
h 1
h 1
h 1
where Ast* Wh X h Ah , ast* Wh xh Ah , Bst Wh Bh , Ah and Bh may be some population information of the auxiliary variable for h-th stratum such as S xh , coefficient of variation C xh , skewness 1h x , kurtosis 2 h x , correlation coefficient h xy . To obtain the bias and MSE, let us define e1*
a
* st
W A x L
A
* st
* st
A
h 1
h
h
h
* st
A
Xh
r s t E yh Yh xh X h Whr s Ahs r *s t Y Ast X h 1 L
,
Vr*,s ,t
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
431
Nursel KOYUNCU
Using these notations we can write following notations L
* E e02 V200 V20
L
E e e * 0 1
W
2 h
h 1
E e1*2
Ah h S xyh
YAst*
W h 1
2 h
Ah2 h S xh2 *2 st
A
* V110
L
L
E e1e1*
Wh2 Ahh S x2 h 1
XAst* L
E e0 e1
W h 1
2 h
h S xyh
* V020
* , V011
E e12
W h 1
2 h
h S xh2
X2
* V101 V11 .
XY
Expressing (10) in terms of e’s, we have
2 Ast* Ast* 3 * k2 Y Y e 2 Ast* Bst 1 8 A* B st st
* V002 V02 ,
2
2
e1*2 e0
Ast* 1 * * 3 Ast e0 e1 2 8 Ast* Bst
2
e0 e1*2
(11)
Squaring both sides of (11) and neglecting the terms of e’s having greater than two, we can write
k
2 Ast* 1 2 2 *2 *2 * Y Y e A e e e 2 0 4 A* C 2 st 1 Ast* Cst 1 0 st st
(12)
Taking expectation both sides of (12) then we get corrected MSE of Singh et al. (2008) as follows: L L L 2 2 2 2 2 W S W A S Wh2 Ah h S xyh *2 * h h h xh h h yh A A st h 1 h 1 MSE k2 Y 2 h 1 2 * st 2 *2 * Y Ast YAst* Ast Cst 4 Ast Cst L Y 1 Y2 2 2 2 2 Wh h S yh * AS AS h 1 Ast Cst h xyh 4 Ast* Cst 2 h xh
1 2 Wh2h S yh Rk 2 S xyh Rk22 S xh2 4 h 1 YA where Rk 2 * h . Ast Bst L
432
(13)
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
Note that t 3 and t 4 estimators are member of k 2 for appropriate values of Ah and Bh as given in Table2. Also we have corrected the MSE of t3 and t4 estimators as given in Appendix with detail. Yadav et al. (2014) have suggested ratio and product exponential estimators are given by X x st a (14) t RC y st exp X a 1x st x st X b t PC y st exp X b 1x st
(15)
V2 L b a 2 MSEmin tPC MSEmin t RC Y 2V20 1 11 Wh2 h S yh 1 c2 V02V20 h 1
L
where c
W h 1
L
W h 1
2 h
2 h
(16)
h S xyh
h S xh2
.
L
W h 1
2 h
h S yh2
In sampling literature many authors have suggested estimators using auxiliary information. Koyuncu and Kadilar (2010) defined a general combined class in stratified random sampling to get a compact expression for the asymptotic variance and avoid tedious calculations, given by
t c g y st , u st
(17)
where u st x st X and g y st , u st is a function of y st and u st . To study the properties of t c we assume following regularity conditions:
y st , u st
assumes the value in a closed convex subset R2 of two
1.
The point
2.
The function g y st , u st is continuous and bounded in R2 ,
3.
g Y ,1 Y and g 0 Y ,1 1 , where g 0 Y ,1 denotes the first order partial derivative of g with respect to y st ,
4.
The first and second order partial derivatives of g y st , u st exist and are
dimensional real space containing the point Y ,1 ,
continuous and bounded in R2 . Note that this class contains many estimator. We can say that Yadav et al. (2014) ratio and product exponential estimators are member of this class. The bias and the MSE of t c are respectively given by Bt c V02 g 2 YV11 g 3 Y 2V20 g 4 ,
MSEt c Y 2V20 V02 g12 2YV11g1 .
(18) (19)
where Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
433
Nursel KOYUNCU
g g1 u st
, g2 y st Y ,u st 1
1 2g 2 u st2
, g3 y st Y ,u st 1
By using the optimal value of g1*
1 2g 2 y st u st
, g4 y st Y ,u st 1
1 2g 2 y st2
. y st Y ,u st 1
Y V11 , the minimum MSE of the estimators in the V02
class t c is found as
V112 MSE t c min Y V20 1 (20) V02V20 which is also the MSE of combined regression type estimators and minimum MSE of Yadav et. al [8]. 2
2.
Proposed Exponential Type Estimators
Motivated by Koyuncu (2012) we propose the following estimator Ast X x st x st y N w1 y st w2 exp X Ast X x st 2 Bst
(21)
Some new estimators, which are generated from (21) for different combinations of Ah , Bh and are given in Table 2. To obtain the MSE, we are applying the same procedure as follows: XAst X 2 Ast2 3 2 (22) y N w1Y 1 e0 w2 1 e1 1 e1 e ... 2 1 2 X A B 8 X A B st st st st 1 1 3 y N Y Y w1 1 w1 Y e0 Y e1 Y e1e0 2Y e12 2 2 8 (23) 1 2 1 1 3 2 2 2 w2 1 e1 e1 e1 e1 e1 2 2 2 8
where
XAst . XAst Bst
Squaring both sides of (23) and taking expectation we have
MSE y N E Y 2 Y 2 w12 1 e02 2 e12 2e0 e1 w22 1 2 2 1 e12 3 3 Y 2 w1 e0 e1 2 2 e12 Y w2 2 1e12 2 4 4 2 2 Y w1 w2 2 2 e0 e1 2 1 2 e1
The MSE of the proposed estimator is given by
MSE y N Y 2 w12 H w22 B Y 2 w1 D Yw2 G Y 2 Yw1 w2 F
434
(24)
(25)
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
where H 1 V20 2V02 2V11 , B 1 2 2 1 2 V02 ,
3 3 D V11 2 2V02 , G 2 1V02 2 4 4 2 F 2 2 V11 2 1 2 V02 The optimum values of w1 and w2 , obtained by minimizing (25) respectively, are given by GF 2 DB DF 2GH (26) w1 opt w2 opt Y 2 4BH F 4HB F 2 Substituting these values in (25) we get the minimum MSE of yN as follows:
BD 2 DFG HG 2 MSE y N Y 1 4 BH F 2 2
(27)
Note that the optimum choice of the constants w1 and w2 involve unknown parameters. These quantities can be guessed quite accurately through pilot sample survey or sample data or experience gathered in due course of time (Koyuncu and Kadilar (2009, 2010). Secondly we consider following estimator * Ast* ast* * xst (28) yM w1 yst w2 exp * * X Ast ast 2 Bst Some new estimators, which are generated from (28) for different combinations of Ah , Bh and are given in Table 2. Expressing (28) in terms of e’s, we have
yM w y w 1 e1 * 1 st
y
* 2
2 * 3 Ast* A * *2 st 1 e e ... 1 2 1 * * 8 Ast Bst 2 Ast Bst
(29)
1 2 1 * * Y w1*Y Y w2* w2* e1 w1 Ye0 e1* 2 2 1 3 3 w2* * e1e1* w1* *2Ye1*2 w2* *2 e1*2 2 8 8
M
(30)
Ast* where * . Ast Bst *
Squaring both sides of (30) and taking expectation we have 3 MSE y M E Y 2 w1Y 2 2 * e0 e1* *2 e1*2 w12Y 2 1 e02 *2 e1*2 2 * e0 e1* 4 3 (31) Y w2 2 *e1e1* 1e12 *2 e1*2 w22 1 2 1 e12 *2 e1*2 2e1e1* 4 * * 2 * w1 w2Y 2 2 e1e1 1e1 2 e0 e1* 2 *2 e1*2 2e0 e1
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
435
Nursel KOYUNCU
The MSE of the proposed estimator is given by MSE yM Y 2 w1*2Y 2 H * w2*2 B* w1*Y 2 D* Yw2*G* w1*w2*YF * where
* * * H * 1 V200 *2V020 2 *V110
* * * B* 1 2 1 V002 *2V020 2 *V011
(32)
3 * * D* 2 *V110 *2V020 4 3 * * * G* 2 * V011 1V002 *2V020 4 * * * * * * * * F 2 2 V011 1V002 2 V110 2 *2V020 2 V101
The optimum values of w1 and w2 are given
w1* opt
G * F * 2 D* B * 4B* A* F *2
w2* opt Y
D* F * 2G* H * 4B* A* F *2
(33)
The minimum MSE of yM at optimum values of w1* and w2* , is given by
B * D *2 D * F *G * H *G *2 MSE y M Y 2 1 4 B * H * F *2
3.
(34)
Numerical Example
To analyze the performance of proposed estimator we use the data concerning the number of teachers as the study variable and the number of students as the auxiliary variable in both primary and secondary schools for 923 districts at 6 regions (as 1:Marmara 2:Agean 3:Mediterranean 4:Central Anatolia 5:Black Sea 6:East and Southeast Anatolia) in Turkey in 2007 (Source: Koyuncu and Kadilar (2009)). The summary statistics about the population is given in Table 1. We have generated some new estimators using suitable values for Ah , Bh and . The MSE values of the k1 , k 2 , y N and y M estimators have been obtained. These values are given in Table 2. When we examine Table 2, we observe that the proposed y M 4 estimator has the smallest. From Table 2 we can conclude that using suitable auxiliary information is very important otherwise we can get worst estimates in each class. 4.
Conclusion
In this paper we have tried to generalize Yadav et al. (2014) estimators and corrected mean square error formula of Singh et al. (2008) and suggested two exponential type estimators in stratified random sampling. To see the performance of estimators we used a real data. We investigate the usage of suitable information of auxiliary variable. We have found that suggested exponential class of estimators is highly efficient than Yadav et al. (2014) and combined regression estimator. 436
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
References 1.
Diana G. A class of estimators of the population mean in stratified random sampling, Statistica, V.1, 1993, pp.59-66.
2.
Singh H.P., Vishwakarma G.K. A family of estimators of population mean using auxiliary information in stratified sampling, Communications in Statistics-Theory and Methods, V.37, 2008, pp.1038-1050.
3.
Koyuncu N., Kadilar C. Family of estimators of population mean using two auxiliary variables in stratified random sampling, Communications in Statistics: Theory and Methods, V.38, N.14, 2009, pp.2398-2417.
4.
Koyuncu N., Kadilar C. On the family of estimators of population mean in stratified sampling, Pakistan Journal of Statistics, V.26, N.2, 2010, pp.427-443.
5.
Koyuncu N., Kadilar C. On improvement in estimating population mean in stratified random sampling, Journal of Applied Statistics, V.37, N.6, 2010, pp.999-1013.
6.
Tailor R., Chouhan S., Tailor R., Garg N. A ratio-cum-product estimator of population mean in stratified random sampling using two auxiliary variables, Statistica, V.LXXII. N.3, 2012, pp. 287-297.
7.
Koyuncu N. Families of estimators for population mean using information on auxiliary attribute in stratified random sampling, Gazi University Journal of Science, V.26, N.2, 2013, pp.181-193.
8.
Yadav R., Upadhyaya L.N., Singh H.P., Chatterjee S. Improved ratio and product exponential type estimators for finite population mean in stratified random sampling, Communications in Statistics: Theory and Methods, V.43, N.16, 2014, pp.3269-3285.
9.
Singh R., Kumar M., Singh R. D., Chaudhary M.K. Exponential ratio type estimators in stratified random sampling, presented in International Symposium on Optimization and Statistics (I.S.O.S.) at AMU, Aligarh, India, 29–31 Dec., 2008.
10.
Koyuncu N. Efficient estimators of population mean using auxiliary attributes, Applied Mathematics and Computation, V.218, 2012, pp.10900-10905.
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
437
Nursel KOYUNCU
Table 1:
438
Data Statistics
N 1 =127
N 2 =117
N 3 =103
N 4 =170
N 5 =205
N 6 =201
n1 =31
n2 =21
n3 =29
n4 =38
n5 =22
n6 =39
S y1 =883.835
S y 2 =644.922
S y 3 =1033.467
S y 4 =810.585
S y 5 =403.654
S y 6 =711.723
Y1 =703.74
Y2 =413
Y3 =573.17
Y4 =424.66
Y5 =267.03
Y6 =393.84
S x1 =30486.751
S x 2 =15180.769
S x 3 =27549.697
S x 4 =18218.931
S x 5 =8497.776
S x 6 =23094.141
X 1 =20804.59
X 2 =9211.79
X 3 =14309.30
X 4 =9478.85
X 5 =5569.95
X 6 =12997.59
S xy1 =25237153.52
S xy 2 =9747942.85
S xy 3 =28294397.04
S xy 4 =14523885.53
S xy 5 =3393591.75
S xy 6 =15864573.97
xy1 =0.936
xy 2 =0.996
xy 3 =0.994
xy 4 =0.983
xy 5 =0.989
xy 6 =0.965
2 x1 =4.593
2 x2 =18.543
2 x3 =15.446
2 x4 =10.162
2 x5 =21.947
2 x6 =23.114
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
Table 2:
Some members of k1 , k 2 , y N and y M and MSE values
Ah
Bh
MSE
t1
C xh
602.5943
t2
2 h x
603.8937
t3
2 h x
C xh
688.4160
t4
C xh
2 h x
589.5579
y N1
2 h x
C xh
0
53.1396
yN 2
2 h x
C xh
1
556.4491
yN3
2 h x
C xh
2
342.4030
yN 4
C xh
2 h x
0
52.1180
yN5
C xh
2 h x
1
557.3786
yN6
C xh
2 h x
2
342.5917
yN7
h xy
C xh
0
52.9641
yN8
h xy
C xh
1
556.6087
yN9
h xy
C xh
2
342.4354
yM1
C xh
2 h x
0
51.1939
yM 2
C xh
2 h x
1
551.1354
yM 3
C xh
2 h x
2
325.882
yM 4
h xy
C xh
0
48.8562
yM 5
h xy
C xh
1
505.5486
yM 6
h xy
C xh
2
331.6634
b a MSEmin t PC MSEmin t RC
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
194.2832
439
Nursel KOYUNCU
Appendix
A* a * k 2 y st exp * st * st Ast a st 2 Bst
(1)
If we take Ah as 2h x and Bh as Cxh in Ast* , ast* and Bst respectively we can get L
L
L
h 1
h 1
h 1
Ast* Wh X h 2 h x , ast* Wh xh 2 h x , Bst WhCxh . Rewriting these values in (1)
we get Singh et al (2008) estimator. X x3 t 3 y st exp 3 X x 3 3
(2) X 3 Wh X h 2 h x C xh
L
L
where x3 Wh xh 2 h x C xh h 1
h 1
W A x
Xh
L
To find the MSE formula let us define e0
y
L
such that E e0 E e1* 0 , E e02 L
E e e * 0 1
W h 1
2 h
W h 1
st
Y
S yh2
2 h h
Y
Y
2
a * A* , e st * st Ast * 1
L
, E e1*2
W h 1
2 h
h 1
h
h
h
Ast*
Ah2 h S xh2
Ast*2
,
Ah h S xyh
YAst*
Expressing k 2 with e term, we have
Ast* Ast* 1 e1* k2 Y 1 e0 exp * Ast Ast* 1 e1* 2 Bst 2 * * 2 A A e e*2 1 3 1 3 st st * * *2 * * k2 Y 1 A e e e A e e 0 st 0 1 0 1 2 Ast* Bst st 1 8 A* B 2 1 2 8 A* B 2 st st st st
* 2 * 2 * A A e e*2 (3) A 3 1 3 st st * *2 * * st k2 Y Y e e e A e e 0 st 0 1 0 1 2 Ast* Bst 1 8 A* B 2 1 2 8 A* B 2 st st st st
Squaring both sides of (3) and neglecting the terms of e’s having power greater than two, we have
2 Ast*2 Ast* 2 *2 2 * k Y Y e e e e 2 0 1 0 * 4 A* B 2 1 Ast Bst st st 440
(4)
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
Improved Exponential Type Estimators for Finite Population Mean in Stratified Random Sampling
Taking expectation of both sides of (4) we get MSE of estimator k 2
1 Ast*2 MSE k2 Y 2 2 4 Ast* Bst
L
Wh2 Ah2h S xh2 h 1
*2 st
A
L
Wh2h S yh2 h 1
Y
2
L
* st
A A Bst * st
W h 1
Ah h S xyh Ast* Y
2 h
YAh 1 Y 2 Ah2 2 2 W S yh * S S h 1 Ast Bst xyh 4 Ast* Bst 2 xh L
2 h h
1 2 Wh2h S yh Rk 2 S xyh Rk22 S xh2 4 h 1 L
where Rk 2
YAh . A Bst * st
If we take Ah as 2h x and Bh as Cxh in MSE k2 we get corrected MSE t3 as 2 2 L Y 2h x Y 2h x 1 2 2 2 MSE t3 Wh h S yh L S xyh S xh 2 L L 4 L h 1 Wh X h 2 h x WhCxh Wh X h 2 h x WhCxh h 1 h 1 h 1 h 1 L 1 2 Wh2h S yh Rt 3 S xyh Rt23 S xh2 4 h 1
where Rt 3
Y 2h x Wh X h 2 h x Cxh h 1 L
.
Pak.j.stat.oper.res. Vol.XII No.3 2016 pp429-441
441