Professional Documents
Culture Documents
Population I Population II
s y2 100 100
t1 223.14 228.70
t2 235.19 228.76
2007
Auxiliary Information and a priori
Values
in Construction of Improved
Estimators
Florentin Smarandache
Department of Mathematics,
University of New Mexico, Gallup, USA
1
2
In the front cover table the percent relative efficiency (PRE) of s y ,t1,t2,tr (in
2
optimum case) and tp (in optimum case) are computed with respect to s y .
Books on Demand
ProQuest Information & Learning
(University of Microfilm International)
300 N. Zeeb Road
P.O. Box 1346, Ann Arbor
MI 48106-1346, USA
Tel.: 1-800-521-0600 (Customer Service)
http://wwwlib.umi.com/bod/basic
Copyright 2007 by Renaissance High Press (Ann Arbor) and the Authors
Peer Reviewers:
Prof. Dr. Adel Helmy Phillips. Ain Shams University, 1 El-Sarayat st., Abbasia, 11517,
Cairo, Egypt.
(ISBN-10): 1-59973-046-4
(ISBN-13): 978-1-59973-046-2
(EAN): 978159973062
Printed in the United States of America
2
Contents
Preface ………………………………………………………………………………….. 4
4. Almost Unbiased Exponential Estimator for the Finite Population Mean ……... 41
3
Preface
4
It is common practice to use arithmetic mean while constructing estimators for
estimating population mean. Mohanty and Pattanaik (1984) used geometric mean and
harmonic mean, while constructing estimators for estimating population mean, using
multi-auxiliary variables. They have shown that in case of multi-auxiliary variables,
estimates based on geometric mean and harmonic means are less biased than Olkin’s
(1958) estimate based on arithmetic mean under certain conditions usually satisfied in
practice. For improving the precision in estimating the unknown mean Y of a finite
population by using the auxiliary variable x, which may be positively or negatively
correlated with y with known X ; the single supplementary variable is used by Bahl and
Tuteja (1991) for the exponential ratio and product type estimators. For estimating the
population mean Y of the study variable y, following Bahl and Tuteja (1991), a ratio-
product type exponential estimator has been proposed by using the known information of
population proportion possessing an attribute (highly correlated with y) in simple random
sampling. The proposed estimator has an improvement over mean per unit estimator,
ratio and product type exponential estimators as well as Naik and Gupta (1996)
estimators. The results have also been extended to the case of two-phase sampling.
estimators suggested by Bahl and Tuteja (1991) are biased. Biasedness of an estimator is
5
Hartley and Ross (1954) and Singh and Singh (1992) to construct either estimator with
this paper we have proposed an almost unbiased ratio and product type exponential
estimator for the finite population mean Y . It has been shown that Bahl and Tuteja
(1991) ratio and product type exponential estimators are particular members of the
proposed estimator.
The Authors
6
Ratio Estimators in Simple Random Sampling Using
(rsinghstat@yahoo.com)
Florentin Smarandache
USA
(smarand@unm.edu)
Abstract
Some ratio estimators for estimating the population mean of the variable under
study, which make use of information regarding the population proportion possessing
certain attribute, are proposed. Under simple random sampling without replacement
(SRSWOR) scheme, the expressions of bias and mean-squared error (MSE) up to the first
order of approximation are derived. The results obtained have been illustrated
7
1. Introduction
The use of auxiliary information can increase the precision of an estimator when
study variable y is highly correlated with auxiliary variable x. There exist situations when
information is available in the form of attribute φ , which is highly correlated with y. For
example
c) Amount of yield of wheat crop and a particular variety of wheat etc. (see
and φ i denote the observations on variable y and φ respectively for ith unit (i = 1,2,....N) .
Suppose there is a complete dichotomy in the population with respect to the presence or
absence of an attribute, say φ , and it is assumed that attribute φ takes only the two
= 0, otherwise.
N n
Let A = ∑ φi and a = ∑ φ i denote the total number of units in the population and
i =1 i =1
A a
sample respectively possessing attribute φ . Let P = and p = denote the proportion
N n
Taking into consideration the point biserial correlation between a variable and an
attribute, Naik and Gupta (1996) defined ratio estimator of population mean when the
8
prior information of population proportion of units, possessing the same attribute is
available, as follows:
⎛P⎞
t NG = y⎜⎜ ⎟⎟ (1.1)
⎝p⎠
here y is the sample mean of variable of interest. The MSE of t NG up to the first order of
approximation is
⎛1− f ⎞ 2
MSE ( t NG ) = ⎜ [
⎟ S y + R 1 Sφ − 2R 1S yφ
2 2
] (1.2)
⎝ n ⎠
where
n
f= ,
Y
R1 = , S =
2 1 N
∑ (y i − Y )2 , S =
2
φ
1 N
∑ (φi − P )2 ,
N − 1 i =1 N − 1 i=1
y
N P
∑ (φi − P )(y i − Y ) .
1 N
S yφ =
N − 1 i=1
In the present paper, some ratio estimators for estimating the population mean of
the variable under study, which make use of information regarding the population
proportion possessing certain attribute, are proposed. The expressions of bias and MSE
have been obtained. The numerical illustrations have also been done by taking some
y + b φ ( P − p)
t1 = P = R *P (2.1)
p
s yφ y + b φ ( P − p) ⎛ 1 ⎞n
where b φ = , R* = , s φ2 = ⎜ ⎟∑ (φ i − p ) and
2
s φ2 p ⎝ n − 1 ⎠ i =1
⎛ 1 ⎞n
s yφ = ⎜ ⎟∑ (φ i − p )(y i − Y ) .
⎝ n − 1 ⎠ i =1
9
Remark 1: When we put b φ = 0 in (2.1), the proposed estimator turns to the Naik and
MSE of this estimator can be found by using Taylor series expansion given by
∂f (c, d) ∂f (c, d)
f (p, y) ≅ f (P, y) + (p − P) + ( y − Y) (2.2)
∂c P ,Y
∂c P ,Y
Expression (2.2) can be applied to the proposed estimator in order to obtain MSE
equation as follows:
⎛ y bφP ⎞ 1
≅ −⎜⎜ 2 + 2 ⎟⎟ (p − P) + ( y − Y)
⎝p p ⎠ P ,Y p P,Y
E(R − R1 )
* 2
≅
(Y + B P ) φ
2
V ( p) −
2( Y + Bφ P) 1
Cov(p, y) + 2 V( y)
P4 P 3
P
1 ⎧ ( Y + Bφ P ) 2( Y + Bφ P) ⎫
2
≅ 2⎨ 2
V ( p) − Cov(p, y) + V( y)⎬ (2.3)
P ⎩ P P ⎭
Sφy ρ pbS y
where B φ = = .
Sφ2 Sφ
S yφ
ρ pb = , is the point biserial correlation coefficient.
S ySφ
Now,
MSE ( t1 ) = P 2 E(R1 − R φ ) 2
≅
(Y + B P ) V(p ) − 2(Y + B P ) Cov(p, y ) + V(y)
φ
2
φ
(2.4)
2
P P
10
Simplifying (2.4), we get MSE of t1 as
⎛1− f
MSE (t1 ) ≅ ⎜ [
⎞ 2 2
(
⎟ R 1 Sφ + Sy 1 − ρpb
2 2
)] (2.5)
⎝ n ⎠
Several authors have used prior value of certain population parameters (s) to find
more precise estimates. Searls (1964) used Coefficient of Variation (CV) of study
(1964) work, Sen (1978), Sisodiya and Dwivedi (1981), and Upadhyaya and Singh
(1984) used the known CV of the auxiliary character for estimating population mean of a
study character in ratio method of estimation. The use of prior value of Coefficient of
Kurtosis in estimating the population variance of study character y was first made by
Singh et al. (1973). Later, used by and Searls and Intarapanich (1990), Upadhyaya and
Singh (1999), Singh (2003) and Singh et al. (2004) in the estimation of population mean
of study character. Recently Singh and Tailor (2003) proposed a modified ratio estimator
In next section, we propose some ratio estimators for estimating the population
mean of the variable under study using known parameters of the attribute φ such as
coefficient of variation Cp, Kurtosis (β2 (φ) ) and point biserial correlation coefficient ρpb .
3. Suggested Estimators
y + b φ ( P − p)
t= (m1P + m 2 ) (3.1)
(m1p + m 2 )
where m1 (≠ 0) , m2 are either real number or the functions of the known parameters of
11
The following scheme presents some of the important estimators of the population
Estimator Values of
m1 m2
y + b φ ( P − p) 1 1
t1 = P
p
y + b φ ( P − p) 1 β2 (φ)
t2 = [P + β2 (φ)]
(p + β2 (φ))
y + b φ ( P − p)
t3 =
(p + Cp ) P + Cp [ ] 1 Cp
y + b φ ( P − p) ρpb
t4 =
(p + ρpb )) P + ρpb [ ] 1
y + b φ ( P − p) β2 (φ)
t5 = [
(pβ2 (φ) + Cp ) Pβ2 (φ) + Cp ] Cp
y + b φ ( P − p) β2 (φ)
t6 = [
(pCp + β2 (φ)) PCp + β2 (φ) ] Cp
y + b φ ( P − p) ρpb
t7 = [
(pCp + ρpb ) PCp + ρpb ] Cp
y + b φ ( P − p) ρpb
t8 = [
(pρpb + Cp ) Pρpb + Cp ] Cp
y + b φ ( P − p) β2 (φ) ρpb
t9 = [
(pβ2 (φ) + ρpb ) Pβ2 (φ) + ρpb ]
y + b φ ( P − p) ρpb β2 (φ)
t10 = [
(pρpb + β2 (φ)) Pρpb + β2 (φ) ]
12
Following the approach of section 2, we obtain the MSE expression for these
proposed estimators as –
⎛1− f
MSE ( t i ) ≅ ⎜
⎞
[ ]
⎟ R iSφ + Sy (1 − ρpb ) ,
2 2 2
( i = 1,2,3,....,10 ) (3.2)
⎝ n ⎠
Y Y Y Y
where R 1 = , R2 = , R3 = , R4 = ,
P P + β2 (φ) P + Cp P + ρpb
4. Efficiency comparisons
⎛1− f ⎞ 2
V ( y) = ⎜ ⎟Sy
⎝ n ⎠
(4.1)
V ( y) − MSE ( t i ) ≥ 0 , i = 1,2,.....,10
Sφ2 2
⇒ ρ > 2 Ri
2
pb
Sy
(4.2)
When this condition is satisfied, proposed estimators are more efficient than
13
Now, we compare the MSE of the proposed estimators with the MSE of
Naik and Gupta [2] estimator t NG . From (3.2) and (1.1) we have
Sφ2 2
2
pb
Sy
[
⇒ ρ ≥ 2 R i − R φ2 + 2R φ K yp ] (4.3)
Cy
where K yp = ρ yp .
Cp
5. Empirical Study
The data for the empirical study is taken from natural population data set
In the below table 5.1 percent relative efficiencies (PRE) of various estimators
14
Table 5.1: PRE of different estimators of Y with respect to y .
y 100
t NG 11.61
t1 7.36
t2 236.55
t3 227.69
t4 208.09
t5 185.42
t6 230.72
t7 185.27
t8 230.77
t9 152.37
t10 237.81
which uses some known values of population proportion performs better than the usual
Conclusion:
15
We have suggested some ratio estimators for estimating Y which uses
some known value of population proportion. For practical purposes the choice of the
References
Jhajj, H. S., Sharma, M. K. and Grover, L. K., A family of estimators of population mean
using information on auxiliary attribute. Pakistan Journal of Statistics, 22
(1), 43-50 (2006).
Naik, V. D. and Gupta, P. C., A note on estimation of mean with known population
proportion of an auxiliary character. Journal of the Indian Society of
Agricultural Statistics, 48 (2), 151-158 (1996).
Ray, S. K. and Singh, R. K., Difference-cum-ratio type estimators. Journal of the Indian
Statistical Association, 19, 147-151 (1981).
Searls, D. T., The utilization of known coefficient of variation in the estimation procedure.
Journal of the American Statistical Association, 59, 1125-1126 (1964).
Searls, D. T. and Intarapanich, P., A note on an estimator for the variance that utilizes the
kurtosis. The American Statistician, 44, 295-296 (1990).
Sen, A. R., Estimation of the population mean when the coefficient of variation is known.
Communications in Statistics Theory and Methods A, 7, 657-672 (1978).
Singh, G. N., On the improvement of product method of estimation in sample surveys.
Journal of the Indian Society of Agricultural Statistics, 56 (3), 267-275
(2003).
Singh H. P. and Tailor, R., Use of known correlation coefficient in estimating the finite
population mean. Statistics in Transition, 6, 555-560 (2003).
Singh H. P., Tailor, R., Tailor, R. and Kakran, M. S., An improved estimator of
population mean using power transformation. Journal of the Indian
Society of Agricultural Statistics, 58 (2), 223-230 (2004).
16
Singh, J., Pandey, B. N. and Hirano, K., On the utilization of a known coefficient of
kurtosis in the estimation procedure of variance. Annals of the Institute of
Statistical Mathematics, 25, 51-55 (1973).
17
Ratio-Product Type Exponential Estimator for Estimating Finite
Population Mean Using Information on Auxiliary Attribute
Florentin Smarandache
Chair of Department of Mathematics, University of New Mexico, Gallup, USA
(smarand@unm.edu)
Abstract
attribute is easily available, see Jhajj et.al. (2006). For estimating the population mean Y
of the study variable y, following Bahl and Tuteja (1991), a ratio-product type
exponential estimator has been proposed by using the known information of population
The expressions for the bias and the mean-squared error (MSE) of the estimator and its
minimum value have been obtained. The proposed estimator has an improvement over
mean per unit estimator, ratio and product type exponential estimators as well as Naik
and Gupta (1996) estimators. The results have also been extended to the case of two
phase sampling. The results obtained have been illustrated numerically by taking some
18
1. Introduction
In survey sampling, the use of auxiliary information can increase the precision of
an estimator when study variable y is highly correlated with the auxiliary variable x. but
in several practical situations, instead of existence of auxiliary variables there exists some
auxiliary attributes, which are highly correlated with study variable y, such as
(i) Amount of milk produced and a particular breed of cow. (ii) Yield of wheat crop and
In such situations, taking the advantage of point biserial correlation between the
study variable and the auxiliary attribute, the estimators of parameters of interest can be
observations on variable y and φ respectively for the ith unit ( i = 1,2,..., N ). We note that
N n
A = ∑ φ i and a = ∑ φ i denote the total number of units in the population and sample
i =1 i =1
A a
respectively possessing attribute φ . Let P = and p = denote the proportion of units
N n
assuming the knowledge of the population proportion P, Naik and Gupta (1996) defined
ratio and product estimators of population when the prior information of population
proportion of units, possessing the same attribute is available. Naik and Gupta (1996)
19
⎛P⎞
t 1 = y⎜⎜ ⎟⎟ (1.1)
⎝p⎠
⎛p⎞
t 2 = y⎜ ⎟ (1.2)
⎝P⎠
[
MSE (t 1 ) = f 1 Y 2 C 2y + C 2p (1 − 2K p ) ] (1.3)
[
MSE (t 2 ) = f 1 Y 2 C 2y + C 2p (1 + 2K p ) ] (1.4)
S 2y S φ2 1 1 Cy 2 1 N
where C 2y =
Y2
, C 2p =
P2
, f1 = − , K p = ρ pb
n N Cp
, Sy = ∑
N − 1 i =1
( y i − Y) 2 ,
1 N 1 ⎛ N ⎞
S φ2 = ∑
N − 1 i =1
( φ i − P ) 2
, S yφ = ⎜ ∑ y i φ i − NPY ⎟ and
N − 1 ⎝ i =1 ⎠
S yφ
ρ pb = is the point biserial correlation coefficient.
S y Sφ
Following Bahl and Tuteja (1991), we propose the following ratio and product
exponential estimators
⎛P−p⎞
t 3 = y exp⎜⎜ ⎟⎟ (1.5)
⎝P+ p⎠
⎛p−P⎞
t 4 = y exp⎜⎜ ⎟⎟ (1.6)
⎝p+P⎠
To obtain the bias and MSE of t3 to the first degree of approximation, we define
ey =
(y − Y ) , e =
(p − P ) , therefore E (e ) = 0. i = ( y, φ) ,
φ i
Y P
20
Expressing (1.5) in terms of e’s, we have
⎡ P − P(1 + e φ ) ⎤
t 3 = Y (1 + e y ) exp ⎢ ⎥
⎢⎣ P + P(1 + e φ ) ⎥⎦
⎡ − eφ ⎤
= Y (1 + e y ) exp ⎢ ⎥ (2.1)
⎢⎣ (2 + e φ ) ⎥⎦
Expanding the right hand side of (2.1) and retaining terms up to second powers of e’s, we
have
⎡ e φ e φ2 e y e φ ⎤
t 3 = Y ⎢1 + e y − + − ⎥ (2.2)
⎣⎢ 2 8 2 ⎥⎦
Taking expectations of both sides of (2.2) and then subtracting Y from both sides, we get
C 2p ⎛ 1 ⎞
B(t 3 ) = f 1 Y ⎜ − Kp ⎟ (2.3)
2 ⎝4 ⎠
⎡ eφ ⎤
(t 3 − Y ) ≅ Y ⎢e y − ⎥ (2.4)
⎣ 2⎦
Squaring both sides of (2.4) and then taking expectations we get MSE of the estimator t3,
⎡ 1 ⎤
MSE ( t 3 ) = f 1 Y 2 ⎢C 2y + C 2p ( − K p )⎥ (2.5)
⎣ 4 ⎦
To obtain the bias and MSE of t4 to the first degree of approximation, we express (1.6) in
terms of e’s
21
⎡ P(1 + e φ ) − P ⎤
t 4 = Y (1 + e y ) exp ⎢ ⎥ (2.6)
⎢⎣ P(1 + e φ ) + P ⎥⎦
and following the above procedure, we get the bias and MSE of t4 as follows
C 2p ⎛ 1 ⎞
B(t 4 ) = f 1 Y ⎜ + Kp ⎟ (2.7)
2 ⎝4 ⎠
⎡ 1 ⎤
MSE ( t 4 ) = f 1 Y 2 ⎢C 2y + C 2p ( + K p )⎥ (2.8)
⎣ 4 ⎦
estimator is more efficient than the ratio (product) estimator except when the regression
line of y on x passes through the neighborhood of the origin, in which case the
efficiencies of these estimators are almost equal. Also in many practical situations the
regression line does not pass through the neighborhood of the origin. In these situations,
the ratio estimator does not perform as good as the linear regression estimator. The ratio
estimator does not perform well as the linear regression estimator does.
⎡ ⎛P−p⎞ ⎛ p − P ⎞⎤
t 5 = y ⎢α exp⎜⎜ ⎟⎟ + (1 − α ) exp⎜⎜ ⎟⎟⎥ (3.1)
⎣ ⎝P+p⎠ ⎝ p + P ⎠⎦
⎛P−p⎞
For α =1, t5 reduces to the estimator t 3 = y exp⎜⎜ ⎟⎟ and for α = 0, it reduces to
⎝P+ p⎠
⎛p−P⎞
t 4 = y exp⎜⎜ ⎟⎟ .
⎝p+P⎠
22
Bias and MSE of t5:
⎡ ⎧− eφ ⎫ ⎧ e φ ⎫⎤
= Y (1 + e y )⎢α exp⎨ ⎬ + (1 − α) exp⎨ ⎬⎥ (3.2)
⎣ ⎩ 2 ⎭ ⎩ 2 ⎭⎦
Expanding the right hand side of (3.2) and retaining terms up to second powers of e’s, we
have
⎡ eφ e φ2 ⎤
t 5 = Y ⎢1 + e y + − αe φ + + e y e φ − αe y e φ ⎥ (3.3)
⎢⎣ 2 8 ⎥⎦
Taking expectations of both sides of (3.3) and then subtracting Y from both sides, we get
⎡ C 2p ⎛1 ⎞⎤
B( t 5 ) = f 1 Y ⎢ + ρ pb C y C p ⎜ − α ⎟⎥ (3.4)
⎢⎣ 8 ⎝2 ⎠⎥⎦
⎡ ⎛1 ⎞⎤
(t 5 − Y ) ≅ Y ⎢e y + e φ ⎜ − α ⎟ ⎥ (3.5)
⎣ ⎝2 ⎠⎦
Squaring both sides of (3.5) and then taking expectations we get MSE of the estimator t5,
⎡ ⎛1 ⎞ ⎛1 ⎞⎤
MSE ( t 5 ) = f1 Y 2 ⎢C 2y + C 2p ⎜ + α 2 − α ⎟ + 2ρ pb C y C p ⎜ − α ⎟⎥ (3.6)
⎣ ⎝4 ⎠ ⎝2 ⎠⎦
2K p + 1
α= = α 0 (Say) (3.7)
2
23
Substitution of (3.7) in (3.1) yields the optimum estimator for t5 as (t5)opt (say)with
minimum MSE as
4. Efficiency comparisons
In this section, the conditions for which the proposed estimator t5 is better than y , t1,
var( y) = f 1 Y 2 C 2y (4.1)
To compare the efficiency of the proposed estimator t5 with the existing estimator, from
MSE ( t 1 ) − M( t 5 ) 0 = (C p − ρ pb C y ) ≥ 0 .
2
(4.3)
MSE ( t 2 ) − M( t 5 ) 0 = (C p + ρ pb C y ) ≥ 0 .
2
(4.4)
2
⎛ C 2p ⎞
MSE ( t 3 ) − M ( t 5 ) 0 = ⎜ − ρ pb C y ⎟ ≥ 0 . (4.5)
⎜ 2 ⎟
⎝ ⎠
2
⎛ C 2p ⎞
MSE ( t 4 ) − M ( t 5 ) 0 = ⎜ + ρ pb C y ⎟ ≥ 0 . (4.6)
⎜ 2 ⎟
⎝ ⎠
Using (4.2)-(4.6), we conclude that the proposed estimator t5 outperforms y , t1, t2, t3, and
t4 .
5. Empirical study
We now compare the performance of various estimators considered here using the
24
Population 1. [Source: Sukhatme and Sukhatme (1970), p. 256]
The percent relative efficiency (PRE’s) of the estimators y , t1-t4 and (t5)opt with respect to
unusual unbiased estimator y have been computed and compiled in table 5.1.
Population
I II
y 100 100
t1 11.63 1.59
t2 5.07 1.94
t3 66.24 5.57
t4 14.15 8.24
25
Table 5.1 shows that the proposed estimator t5 under optimum condition performs
better than the usual sample mean y , Naik and Gupta (1996) estimators (t1 and t2) and
the ratio and product type exponential estimators (t3 and t4).
6. Double sampling
In some practical situations when P is not known a priori, the technique of two-phase
sampling is used. Let p’ denote the proportion of units possessing attribute φ in the first
phase sample of size n’; p denote the proportion of units possessing attribute φ in the
second phase sample of size n < n’ and y denote the mean of the study variable y in the
When P is not known, two-phase ratio and product type exponential estimator are
given by
⎛ p'− p ⎞
t 6 = y exp⎜⎜ ⎟⎟ (6.1)
⎝ p'+ p ⎠
⎛ p − p' ⎞
t 7 = y exp⎜⎜ ⎟⎟ (6.2)
⎝ p + p' ⎠
such that
and
( ) ( )
E e 2y = f 1C 2y , E e φ2 = f 1C 2p , E (e' φ ) = f 2 C 2p , E (e φ e' φ ) = f 2 ρ pb C y C p .
2
26
1 1
where f 2 = − .
n' N
⎡ e' φ − e φ ⎤
= Y (1 + e y ) exp ⎢ ⎥ (6.3)
⎣ 2 ⎦
Expanding the right hand side of (6.3) and retaining terms up to second powers of e’s, we
have
Taking expectations of both sides of (6.4) and then subtracting Y from both sides, we get
C 2p
B(t 6 ) = f 3 Y (1 − 2K )
p (6.5)
4
⎡ ( e' φ − e φ ) ⎤
(t 6 − Y ) ≅ Y ⎢e y + ⎥ (6.6)
⎣ 2 ⎦
Squaring both sides of (6.6) and then taking expectations we get MSE of the estimator t6,
⎡ C 2p ⎤
MSE ( t 6 ) = Y 2 ⎢f 1C 2y + f 3 (1 − 4K p )⎥ (6.7)
⎢⎣ 4 ⎥⎦
To obtain the bias and MSE of t7 to the first degree of approximation, we express (6.2) in
terms of e’s as
27
⎡ P(1 + e φ ) − P(1 + e' φ ) ⎤
t 7 = Y (1 + e y ) exp ⎢ ⎥
⎢⎣ P(1 + e φ ) + P(1 + e' φ ) ⎥⎦
⎡ e φ − e' φ ⎤
= Y (1 + e y ) exp ⎢ ⎥ (6.8)
⎣ 2 ⎦
Expanding the right hand side of (6.8) and retaining terms up to second powers of e’s, we
have
Taking expectations of both sides of (6.9) and then subtracting Y from both sides, we get
C 2p
B(t 7 ) = f 3 Y (1 + 2K )
p (6.10)
4
⎡ ( e φ − e' φ ) ⎤
(t 7 − Y ) ≅ Y ⎢e y + ⎥ (6.11)
⎣ 2 ⎦
Squaring both sides of (6.11) and then taking expectations we get MSE of the estimator
⎡ C 2p ⎤
MSE ( t 7 ) = Y 2 ⎢f1C 2y + f 3 (1 + 4K p )⎥ (6.12)
⎢⎣ 4 ⎥⎦
⎡ ⎛ p'−p ⎞ ⎛ p − p' ⎞ ⎤
t 8 = y ⎢α 1 exp⎜⎜ ⎟⎟ + (1 − α 1 ) exp⎜⎜ ⎟⎟⎥ (7.1)
⎣ ⎝ p '+ p ⎠ ⎝ p + p ' ⎠⎦
28
⎛ p'− p ⎞
For α 1 =1, t8 reduces to the estimator t 6 = y exp⎜⎜ ⎟⎟ and for α 1 = 0, it reduces to
⎝ p'+ p ⎠
⎛ p − p' ⎞
t 7 = y exp⎜⎜ ⎟⎟ .
⎝ p + p' ⎠
Expanding the right hand side of (7.2) and retaining terms up to second powers of e’s, we
have
2
eφ e' φ e φ2 e' φ e yeφ e y e' φ
t 8 = Y[1 + e y + − − α 1 e φ + α 1 e' φ + + + −
2 2 8 8 2 2
e φ e' φ
− + α 1 e y e' φ − α 1 e y e φ ] (7.3)
4
Taking expectations of both sides of (7.3) and then subtracting Y from both sides, we get
C 2p ⎡ ⎛ 1 ⎞⎤
B( t 8 ) = f 3 Y ⎢1 − 8K p ⎜ α 1 − ⎟ ⎥ (7.4)
8 ⎣ ⎝ 2 ⎠⎦
⎡ ⎛ 1⎞ ⎛ 1⎞ ⎤
( t 8 − Y ) ≅ Y ⎢e y − ⎜ α 1 − ⎟e φ + ⎜ α 1 − ⎟e ' φ ⎥ (7.5)
⎣ ⎝ 2⎠ ⎝ 2⎠ ⎦
29
Squaring both sides of (7.5) and then taking expectations we get MSE of the estimator t8,
⎡ ⎛ 1 ⎞⎧⎛ 1⎞ ⎫⎤
MSE ( t 8 ) = Y 2 ⎢f 1C 2y + f 3 C 2p ⎜ α 1 − ⎟⎨⎜ α 1 − ⎟ − 2K p ⎬⎥ (7.6)
⎣ ⎝ 2 ⎠⎩⎝ 2⎠ ⎭⎦
2K p + 1
α1 = = α 10 (Say) (7.7)
2
Substitution of (7.7) in (7.1) yields the optimum estimator for t8 as (t8)opt (say) with
minimum MSE as
8. Efficiency comparisons
[
MSE (t 9 ) = Y 2 f 1C 2y + f 3 C 2p (1 − 2K p ) ] (8.1)
[
MSE (t 10 ) = Y 2 f 1C 2y + f 3 C 2p (1 + 2K p ) ] (8.2)
2
⎛ Cp ⎞
MSE ( t 6 ) − M( t 8 ) 0 = f 3 ⎜⎜ − ρ pb C y ⎟⎟ ≥ 0 . (8.4)
⎝ 2 ⎠
2
⎛ Cp ⎞
MSE ( t 7 ) − M( t 8 ) 0 = f 3 ⎜⎜ + ρ pb C y ⎟⎟ ≥ 0 . (8.5)
⎝ 2 ⎠
MSE ( t 9 ) − M ( t 8 ) 0 = f 3 (C p − ρ pb C y ) ≥ 0 .
2
(8.6)
30
MSE ( t 10 ) − M( t 8 ) 0 = f 3 (C p + ρ pb C y ) ≥ 0 .
2
(8.7)
From (8.3)-(8.7), we conclude that our proposed estimator t8 is better than y , t6, t7, t9, and
t10.
9. Empirical study
The various results obtained in the previous section are now examined with the
Cp = 2.7005.
Cp =1.34701.
Population
I II
y 100 100
t6 40.59 25.42
t7 21.90 40.89
t9 11.16 8.89
31
Table 9.1 shows that the proposed estimator t8 under optimum condition performs better
than the usual sample mean y , t6, t7, t9, and t10.
References
Bahl, S. and Tuteja, R.K. (1991): Ratio and Product type exponential estimator,
Information and Optimization sciences, Vol.XII, I, 159-163.
Jhajj, H. S., Sharma, M. K. and Grover, L. K. (2006): A family of estimators of
population mean using information on auxiliary attribute. Pak. J. Statist.,
22 (1), 43-50.
Mukhopadhyaya, P.(2000): Theory and methods of survey sampling. Prentice Hall of
India, New Delhi, India.
Naik,V.D. and Gupta, P.C. (1996): A note on estimation of mean with known population
proportion of an auxiliary character. Jour. Ind. Soc. Agr. Stat., 48(2),151-
158.
Shabbir,J. and Gupta, S.(2007) : On estimating the finite population mean with known
population proportion of an auxiliary variable. Pak. J. Statist.,23(1),1-9.
Singh, H.P. and Espejo, M.R. (2003): On linear regression and ratio-product estimation
of a finite population mean. The statistician, 52, 1, 59-67.
Sukhatme, P.V. and Sukhatme, B.V. (1970): Sampling theory of surveys with
applications. Iowa State University Press, Ames, U.S.A.
32
Improvement in Estimating the Population Mean Using Exponential
Estimator in Simple Random Sampling
Florentin Smarandache
Department of Mathematics, University of New Mexico, Gallup, USA
(fsmarandache@yahoo.com)
Abstract
This study proposes some exponential ratio-type estimators for estimating the
population mean of the variable under study, using known values of certain population
parameter(s). Under simple random sampling without replacement (SRSWOR) scheme,
mean square error (MSE) equations of all proposed estimators are obtained and compared
with each other. The theoretical results are supported by a numerical illustration.
1. Introduction
Consider a finite population U = U1 , U 2 ,....., U N of N unites. Let y and x stand for the
variable under study and auxiliary variable respectively. Let ( yi , x i ) , i = 1,2,......, n denote
the values of the units included in a sample s n of size n drawn by simple random
sampling without replacement (SRSWOR). The auxiliary information has been used in
improving the precision of the estimate of a parameter (see Cochran (1977), Sukhatme
33
and Sukhatme (1970) and the reference cited there in). Out of many methods, ratio and
product methods of estimation are good illustrations in this context.
In order to have a survey estimate of the population mean Y of the study
character y, assuming the knowledge of the population mean X of the auxiliary character
x, the well-known ratio estimator is –
⎛X⎞
t r = y⎜⎜ ⎟⎟ (1.1)
⎝x⎠
Bahl and Tuteja (1991) suggested an exponential ratio type estimator as –
⎡X − x⎤
t1 = y exp ⎢ ⎥ (1.2)
⎣X + x⎦
Several authors have used prior value of certain population parameter(s) to find more
precise estimates. Sisodiya and Dwivedi (1981), Sen (1978) and Upadhyaya and Singh
(1984) used the known coefficient of variation (CV) of the auxiliary character for
estimating population mean of a study character in ratio method of estimation. The use of
prior value of coefficient of kurtosis in estimating the population variance of study
character y was first made by Singh et.al.(1973). Later used by Singh and Kakaran (1993)
in the estimation of population mean of study character. Singh and Tailor (2003)
proposed a modified ratio estimator by using the known value of correlation coefficient.
Kadilar and Cingi (2006(a)) and Khoshnevisan et.al.(2007) have suggested modified ratio
estimators by using different pairs of known value of population parameter(s).
In this paper, under SRSWOR, we have suggested improved exponential ratio-
type estimators for estimating population mean using some known value of population
parameter(s).
2. The suggested estimator
Following Kadilar and Cingi (2006(a)) and Khoshnevisan (2007), we define modified
exponential estimator for estimating Y as –
⎡ (aX + b) − (ax + b) ⎤
t = y exp ⎢ ⎥ (2.1)
⎣ (aX + b) + (ax + b) ⎦
where a (≠ 0) , b are either real numbers or the functions of the known parameters of the
auxiliary variable x such as coefficient of variation (C x ) , coefficient of kurtosis (β 2 ( x ))
and correlation coefficient (ρ) .
34
To, obtain the bias and MSE of t, we write
y = Y (1 + e 0 ) , x = X (1 + e1 )
such that
E(e 0 ) = E(e1 ) = 0 ,
where
N−n S2y S2
f1 = , C y = 2 , C 2x = x2 .
2
nN Y X
Expressing t in terms of e’s, we have
⎡ aX − aX (1 + e1 ) ⎤
t = Y (1 + e 0 ) exp ⎢ ⎥
⎣ aX + 2b + aX (1 + e1 ) ⎦
[
t = Y (1 + e 0 ) exp − θe1 (1 + θe1 ) −1 ] (2.2)
aX
where θ = .
2( a X + b )
Expanding the right hand side of (2.2) and retaining terms up to second power of e’s, we
have
t = Y (1 + e 0 − θe1 + θ2e12 + θe0 e1 ) (2.3)
Taking expectations of both sides of (2.3) and then subtracting Y from both sides, we get
the bias of the estimator t, up to the first order of approximation, as
B( t ) = f1Y (θ 2C 2y + θρC y C x ) (2.4)
35
Estimator Values of
a b
t0 = y 0 0
The mean per unit estimator
⎡X − x⎤ 1 1
t1 = y exp ⎢ ⎥
⎣X + x⎦
Bahl and Tuteja (1991)
estimator
⎡ X−x ⎤ 1 β2 (x )
t 2 = y exp ⎢ ⎥
⎣ X + x + 2β 2 ( x ) ⎦
⎡ X−x ⎤ 1 Cx
t 3 = y exp ⎢ ⎥
⎣ X + x + 2C c ⎦
⎡ X−x ⎤ 1 ρ
t 4 = y exp ⎢ ⎥
⎣ X + x + 2ρ ⎦
⎡ β 2 ( x )( X − x ) ⎤ β2 (x ) Cx
t 5 = y exp ⎢ ⎥
⎣ β 2 ( x )( X + x ) + 2C x ⎦
⎡ C x (X − x ) ⎤ Cx β2 (x )
t 6 = y exp ⎢ ⎥
⎣ C x ( X + x ) + 2β 2 ( x ) ⎦
⎡ Cx (X − x) ⎤ Cx ρ
t 7 = y exp ⎢ ⎥
⎣ C x ( X + x ) + 2ρ ⎦
⎡ ρ( X − x ) ⎤ ρ Cx
t 8 = y exp ⎢ ⎥
⎣ ρ( X + x ) + 2C x ⎦
⎡ β 2 ( x )( X − x ) ⎤ β2 (x ) ρ
t 9 = y exp ⎢ ⎥
⎣ β 2 ( x )( X + x ) + 2ρ ⎦
⎡ ρ( X − x ) ⎤ ρ β2 (x )
t10 = y exp ⎢ ⎥
⎣ ρ( X + x ) + 2β 2 ( x ) ⎦
36
It is observed that the expressions of the first order of approximation of bias and
MSE of the given member of the family can be obtained by mere substituting the values
of a and b in (2.4) and (2.6) respectively.
4. Modified estimators
Following Kadilar and Cingi (2006(b)), we propose modified estimator combining
estimator t1 and t i (i = 2,3,....,10) as follows
where α is a real constant to be determined such that the MSE of t *i is minimum and
t i (i = 2,3,....,10) are estimators listed in section 3.
Following the procedure of section (2), we get the MSE of t *i to the first order of
approximation as –
⎡ ⎛α ⎞
2
⎛α ⎞⎤
MSE ( t *i ) = f1Y 2 ⎢C 2y + C 2x ⎜ + θi − αθi ⎟ − 2ρC y C x ⎜ + θi − αθi ⎟⎥ (4.2)
⎢⎣ ⎝2 ⎠ ⎝2 ⎠⎥⎦
where
X X
θ2 = θ3 =
2(X + β 2 ( x ) ) 2(X + C x )
, ,
X β2 ( x )X
θ4 = θ5 =
2(X + ρ) ) 2(β2 ( x ) X + C x ) )
, ,
Cx X Cx X
θ6 = θ7 =
2(C x X + β 2 ( x ) ) 2(C x X + ρ)
, ,
ρX β2 (x )X
θ8 = θ9 =
2(ρX + C x )
, ,
2(β 2 ( x ) + ρ)
ρX
θ10 = .
2(ρ + β 2 ( x ) )
Substitution of (4.3) in (4.10) gives optimum estimator t *o (say) , with minimum MSE as
37
min MSE ( t *i ) = f1Y 2C 2y (1 − ρ2 ) = MSE ( t * ) o (4.4)
The min MSE ( t *i ) at (4.4) is same as that of the approximate variance of the usual linear
regression estimator.
5. Efficiency comparison
It is well known that under SRSWOR the variance of the sample mean is
Var ( y) = f1Y 2 C 2y (5.1)
we first compare the MSE of the proposed estimators, given in (2.6) with the variance of
the sample mean, we have the following condition:
θi
K≤ , i = 2,3,.....,10 (5.2)
2
When this condition is satisfied, proposed estimators are more efficient than the sample
mean.
Next we compare the MSE of proposed estimators t *i ( i = 2,3,.....,10 ) in (4.4) with
the MSE of estimators listed in section 3. We obtain the following condition
(θC x − ρC y ) 2 ≥ 0 , i = 2,3,.....,10 . (5.3)
We can infer that all proposed estimators t *i , ( i = 2,3,.....,10 ) are more efficient than
estimators proposed in section 3 in all conditions, because the condition given in (5.1) is
always satisfied.
6. Numerical illustration
To illustrate the performance of various estimators of Y , we consider the data given in
Murthy (1967 pg-226). The variates are:
y : Output, x: number of workers
X =283.875, Y =5182.638, C y = 0.3520, C x = 0.9430, ρ = 0.9136, β 2 ( x ) = 3.65.
38
Table 6.1: PRE of different estimators of Y with respect to y
Estimator PRE
y 100
t1 366.96
t2 385.72
t3 368.27
t4 371.74
t5 386.87
t6 368.27
t7 372.03
t8 372.05
t9 368.27
t10 386.91
t *o 877.54
7. Conclusion
We have developed some exponential ratio type estimators using some known value
of the population parameter(s), listed in section 3. We have also suggested modified
estimators t *i ( i = 2,3,.....,10 ). From table 6.1 we conclude that the proposed estimators are
better than Bahl and Tuteja (1991) estimator t1 . Also, the modified estimator t *i
( i = 2,3,.....,10 ) under optimum condition performs better than the estimators proposed and
listed in section 3 and than the Bahl and Tuteja (1991) estimator t1 . The choice of the
estimator mainly depends upon the availability of information about known values of the
parameter(s) ( C x , ρ , β 2 ( x ) , etc.).
39
References
Bahl, S. and Tuteja, R.K. (1991): Ratio and Product type exponential estimator,
Information and Optimization sciences, Vol.XII, I, 159-163.
Cochran, W.G.(1977): Sampling techniques. Third U.S. edition. Wiley eastern limited,
325.
Kadilar, C. and Cingi, H. (2006(a)): A new ratio estimator using correlation coefficient.
Inter Stat, 1-11.
Murthy, M.N., (1967): Sampling Theory and Methods, Statistical Publishing Society,
Calcutta.
Sen, A.R., (1978) : Estimation of the population mean when the coefficient of variation is
known. Comm. Stat.-Theory Methods, A7, 657-672.
Singh, H.P. and Kakran, M.S. (1993): A modified ratio estimator using coefficient of
variation of auxiliary character. Unpublished.
Singh, H.P. and Tailor, R. (2003): Use of known correlation coefficient in estimating the
finite population mean. Statistics in Transition, 6,4,555-560.
Singh, J. Pandey, B.N. and Hirano, K. (1973): On the utilization of a known coefficient
of kurtosis in the estimation procedure of variance. Ann. Inst. Stat. Math.,
25, 51-55.
Sisodiya, B.V.S. and Dwivedi, V.K. (1981): A modified ratio estimator using coefficient
of variation of auxiliary variable. Jour. Ind. Soc. Agri. Stat., 33, 13-18.
Sukhatme, P.V. and Sukhatme, B.V., (1970): Sampling theory of surveys with
applications. Iowa State University Press, Ames, U.S.A.
Upadhyaya, L.N. and Singh, H.P.,(1984): On the estimation of the population mean with
known coefficient of variation. Biometrical Journal, 26, 915-922.
40
Almost Unbiased Exponential Estimator for the Finite Population Mean
(rsinghstat@yahoo.com)
Florentin Smarandache
(smarand@unm.edu)
Abstract
In this paper we have proposed an almost unbiased ratio and product type
exponential estimator for the finite population mean Y . It has been shown that Bahl and
Tuteja (1991) ratio and product type exponential estimators are particular members of the
proposed estimator.
1. Introduction
It is well known that the use of auxiliary information in sample surveys results in
substantial improvement in the precision of the estimators of the population mean. Ratio,
product and difference methods of estimation are good examples in this context. Ratio
41
method of estimation is quite effective when there is a high positive correlation between
study and auxiliary variables. On other hand, if this correlation is negative (high), the
Consider a finite population with N units ( U 1 , U 2 ,...., U N ) for each of which the
mean of character y under study. Let ( y, x ) be the sample mean estimator of ( Y, X ) the
character y, assuming the knowledge of the population mean X of the auxiliary character
x, Bahl and Tuteja (1991) suggested ratio and product type exponential estimator
⎛X−x⎞
t 1 = y exp⎜⎜ ⎟⎟ (1.1)
⎝X+x⎠
⎛x−X⎞
t 2 = y exp⎜⎜ ⎟⎟ (1.2)
⎝x+X⎠
Up to the first order of approximation, the bias and mean-squared error (MSE) of
⎛ N − n ⎞ Cx ⎛1 ⎞
2
B(t 1 ) = ⎜ ⎟Y ⎜ − K⎟ (1.3)
⎝ nN ⎠ 2 ⎝2 ⎠
⎛ N−n⎞ 2⎡ 2 2⎛1 ⎞⎤
MSE (t 1 ) = ⎜ ⎟ Y ⎢C y + C x ⎜ − K ⎟ ⎥ (1.4)
⎝ nN ⎠ ⎣ ⎝4 ⎠⎦
⎛ N − n ⎞ Cx ⎛ 1 ⎞
2
B(t 2 ) = ⎜ ⎟Y ⎜ + K⎟ (1.5)
⎝ nN ⎠ 2 ⎝ 2 ⎠
42
⎛ N−n⎞ 2⎡ 2 2⎛1 ⎞⎤
MSE (t 2 ) = ⎜ ⎟ Y ⎢C y + C x ⎜ + K ⎟ ⎥ (1.6)
⎝ nN ⎠ ⎣ ⎝4 ⎠⎦
(y i − Y )2 , (x i − X )2 ,
1 N
1 N Sy Sx
where S 2y = ∑
(N − 1) i=1
S 2x = ∑
(N − 1) i=1
Cy =
Y
, Cx =
X
,
⎛ Cy ⎞
(y − Y )(x i − X ) .
S yx 1 N
K = ρ⎜⎜ ⎟⎟ , ρ =
(S y S x ) , S yx = ∑
(N − 1) i=1 i
⎝ Cx ⎠
From (1.3) and (1.5), we see that the estimators t 1 and t 2 suggested by Bahl and
Following Singh and Singh (1993) and Singh and Singh (2006) we have proposed almost
unbiased estimators of Y .
⎛X−x⎞ ⎛x−X⎞
Suppose t 0 = y , t 1 = y exp⎜⎜ ⎟⎟ , t 2 = y exp⎜⎜ ⎟⎟
⎝X+x⎠ ⎝x+X⎠
such that t 0 , t 1 , t 2 ∈ H , where H denotes the set of all possible estimators for estimating
2
th = ∑ hiti ∈ H (2.1)
i =0
2
for ∑h
i =0
i = 1, hi ∈ R (2.2)
where h i (i = 0,1,2 ) denotes the statistical constants and R denotes the set of real numbers.
y = Y (1 + e 0 ) , x = X (1 + e1 ) .
such that
E (e0)=E (e1)=0.
43
⎛N−n⎞ 2 ⎛N−n⎞ 2 ⎛N−n⎞
E(e 02 ) = ⎜ ⎟C y , E ( e 1 ) = ⎜ ⎟C x , E(e 0 e1 ) = ⎜ ⎟ρC y C x .
2
⎝ Nn ⎠ ⎝ Nn ⎠ ⎝ Nn ⎠
⎡ ⎛ − e1 ⎞ ⎛ e ⎞⎤
t h = Y (1 + e 0 )⎢h 0 + h 1 exp⎜⎜ ⎟⎟ + h 2 exp⎜⎜ 1 ⎟⎟⎥ (2.3)
⎣ ⎝ 2 + e1 ⎠ ⎝ 2 + e1 ⎠ ⎦
Expanding the right hand side of (2.3) and retaining terms up to second powers of e’s, we
have
⎡ e e2 e2 e e e e ⎤
t h = Y ⎢1 + e 0 − 1 (h 1 − h 2 ) + h 1 1 + h 2 1 − h 1 0 1 + h 2 0 1 ⎥ (2.4)
⎣ 2 8 8 2 2 ⎦
Taking expectations of both sides of (2.4) and then subtracting Y from both sides, we get
⎛ N − n ⎞ Cx ⎡1 ⎤
2
B( t h ) = ⎜ ⎟Y ⎢ 4 (h 1 + h 2 ) − K (h 1 − h 2 )⎥ (2.5)
⎝ Nn ⎠ 2 ⎣ ⎦
⎡ e ⎤
(t h − Y ) ≅ Y ⎢e 0 − h 1 ⎥ (2.6)
⎣ 2⎦
Squaring both the sides of (2.7) and then taking expectations, we get MSE of the
⎛ N−n⎞ 2⎡ 2 2 ⎛h ⎞⎤
MSE ( t h ) = ⎜ ⎟ Y ⎢C y + C x h ⎜ − K ⎟ ⎥ (2.8)
⎝ Nn ⎠ ⎣ ⎝4 ⎠⎦
h = 2K. (2.9)
44
Thus the minimum MSE of th is given by
⎛N−n⎞ 2 2
min .MSE ( t h ) = ⎜ ⎟Y C y 1 − ρ
2
( ) (2.10)
⎝ Nn ⎠
h1-h2 = h = 2K . (2.11)
From (2.2) and (2.11), we have only two equations in three unknowns. It is not possible
to find the unique values for hi’s, i=0,1,2. In order to get unique values of hi’s, we shall
∑ h B(t ) = 0.
i =0
i i (2.12)
Equations (2.2), (2.11) and (2.12) can be written in the matrix form as
⎡1 1 1 ⎤ ⎡h 0 ⎤ ⎡ 1 ⎤
⎢0 1 − 1 ⎥⎥ ⎢⎢ h 1 ⎥⎥ = ⎢⎢2K ⎥⎥ (2.13)
⎢
⎢⎣0 B( t 1 ) B( t 2 )⎥⎦ ⎢⎣h 2 ⎥⎦ ⎢⎣ 0 ⎥⎦
h 0 = 1 − 4K 2 ⎫
⎪
h 1 = K + 2K 2 ⎬ (2.14)
⎪
h 2 = − K + 2K 2 ⎭
Use of these hi’s (i=0,1,2) remove the bias up to terms of order o(n-1) at (2.1).
preliminary large sample on which only the auxiliary characteristic is observed. The
45
value of population mean X of the auxiliary character x is then replaced by this estimate.
procedure for finding the reliable estimate in first phase sample for the unknown
parameters of the auxiliary variable x and hence has eminent role to play in survey
of size n ′ on which only the characteristic x is measured. Then a second phase sample of
size n (n < n ′) is drawn on which both y and x characteristics are measured. Let
1 n′
x = ∑ x i denote the sample mean of x based on first phase sample of size n ′ ;
n ′ i =1
1 n 1 n
y= ∑
n i =1
y i and x = ∑ x i be the sample means of y and x respectively based on
n i =1
⎛ x′ − x ⎞
t 1d = y exp⎜ ⎟ (3.1)
⎝ x′ + x ⎠
⎛ x − x′ ⎞
t 2 d = y exp⎜ ⎟ (3.2)
⎝ x + x′ ⎠
y = Y (1 + e 0 ) , x = X (1 + e1 ) , x ′ = X (1 + e1′ )
such that
46
and
E(e 0 e1 ) = f 1ρC y C x ,
E(e 0 e1′ ) = f 2 ρC y C x ,
E(e1e1′ ) = f 2 C 2x .
⎛1 1 ⎞ ⎛1 1⎞
where f 1 = ⎜ − ⎟ , f 2 = ⎜ − ⎟ .
⎝n N⎠ ⎝ n′ N ⎠
⎡ C 2x 1 ⎤
B( t 1d ) = Yf 3 ⎢ − ρC y C x ⎥ (3.3)
⎣ 8 2 ⎦
⎡C2 1 ⎤
B( t 2 d ) = Yf 3 ⎢ x + ρC y C x ⎥ (3.4)
⎣ 8 2 ⎦
⎡ 2 ⎛ C 2x ⎞⎤
MSE ( t 1d ) = Y ⎢f 1C y + f 3 ⎜⎜
2
− ρC x C y ⎟⎟⎥ (3.5)
⎣⎢ ⎝ 4 ⎠⎦⎥
⎡ ⎛ C2 ⎞⎤
MSE ( t 2d ) = Y 2 ⎢f 1C 2y + f 3 ⎜⎜ x + ρC x C y ⎟⎟⎥ (3.6)
⎢⎣ ⎝ 4 ⎠⎥⎦
⎛1 1 ⎞
where f3 = ⎜ − ⎟ .
⎝ n n′ ⎠
From (3.3) and (3.4) we observe that the proposed estimators t 1d and t 2d are biased,
47
Suppose t 0 = y , t 1d and t 2d as defined in (3.1) and (3.2) such that
t0, t 1d , t 2d ∈ W , where W denotes the set of all possible estimators for estimating the
2
tW = ∑ witi ∈ W . (4.1)
i =0
2
for ∑w
i =1
i = 1, wi ∈ R . (4.2)
where w i (i = 0,1,2 ) denotes thee statistical constants and R denotes the set of real
numbers.
To obtain the bias and MSE of t w , using notations of section 3 and expressing t w in
⎡ ⎛ e′ − e ⎞ ⎛ e − e′ ⎞⎤
t w = Y (1 + e 0 )⎢ w 0 + w 1 exp⎜ 1 1 ⎟ + w 2 exp⎜ 1 1 ⎟⎥ (4.3)
⎣ ⎝ 2 ⎠ ⎝ 2 ⎠⎦
t w = Y[1 + e 0 −
w
(e1 − e1′ ) + w 1 (e12 + e1′ 2 ) + w 2 (e12 + e1′ 2 ) − ⎛⎜ w 1 + w 2 ⎞⎟e1e1′
2 8 8 ⎝ 4 4 ⎠
+
w
(e 0 e1′ − e 0 e1 )] (4.4)
2
where w = w 1 − w 2 . (4.5)
Taking expectations of both sides of (4.4) and then subtracting Y from both sides, we get
⎡⎛ w + w 2 ⎞ 2 w ⎤
Bias( t w ) = Yf 3 ⎢⎜ 1 ⎟C x − ρC y C x ⎥ (4.6)
⎣⎝ 8 ⎠ 2 ⎦
48
⎡ ⎤
t w ≅ Y ⎢e 0 − (e1 − e1′ )⎥
w
(4.7)
⎣ 2 ⎦
Squaring both sides of (4.7) and then taking expectation, we get MSE of the estimator
⎡ ⎛w⎞ ⎤
MSE (t w ) = Y 2 ⎢f1C 2y + f 3 wC 2x ⎜ ⎟ − K ⎥ (4.8)
⎣ ⎝4⎠ ⎦
w = 2K. (4.9)
[
min .MSE ( t w ) = Y 2 C 2y f 1 − f 3 ρ 2 ] (4.10)
which is same as that of two-phase linear regression estimator. From (4.5) and (4.9), we
have
w 1 − w 2 = w = 2K (4.11)
From (4.2) and (4.11), we have only two equations in three unknowns. It is not
possible to find the unique values for w i ' s(i = 0,1,2 ) . In order to get unique values of
∑ w B(t
i =0
i id )=0 (4.12)
Equations (4.2), (4.11) and (4.12) can be written in the matrix form as
⎡1 1 1 ⎤ ⎡w 0 ⎤ ⎡ 1 ⎤
⎢0 1 − 1 ⎥⎥ ⎢⎢ w 1 ⎥⎥ = ⎢⎢2K ⎥⎥ (4.13)
⎢
⎢⎣0 B( t 1d ) B( t 2 d )⎥⎦ ⎢⎣ w 2 ⎥⎦ ⎢⎣ 0 ⎥⎦
49
w 0 = 1 − 8K 2 ⎫
⎪
w 1 = K + 4K 2 ⎬ (4.14)
⎪
w 2 = − K + 4K 2 ⎭
Use of these w i ' s(i = 0,1,2 ) removes the bias up to terms of order o(n −1 ) at (4.1).
5. Empirical study
The data for the empirical study are taken from two natural population data sets
Scalars Population
I II
h0 -2.2065 -20.93
h1 2.4985 8.62
h2 0.7079 13.30
Using these values of hi’s (i = 0,1,2) given in the table 5.1, one can reduce the bias
50
In table 5.2, Percent relative efficiency (PRE) of y , t1, t2 and th (in optimum case) are
Population I Population II
y 100 100
t1 272.75 32.55
t2 47.07 126.81
Table 5.2 clearly shows that the suggested estimator th in its optimum condition is
better than usual unbiased estimator y , Bahl and Tuteja (1991) estimators t1 and t2.
populations:
y : Output
x : Number of workers
C y = 0.3542 , C x = 0.9484 , ρ = 0.9150 , N = 80, n ′ = 20 , n = 8.
In table 5.3 the values of scalars w i ' s(i = 0,1,2 ) are listed.
51
Table 5.3: Values of w i ' s(i = 0,1,2 )
w0 0.659 0.2415
w1 0.808 0.0713
w2 0.125 0.6871
Using these values of w i ' s(i = 0,1,2 ) given in table 5.3 one can reduce the bias
Population I Population II
y 100 100
t 1d 128.07 74.68
t 2d 41.42 103.64
tw 138.71 106.11
52
References
Bahl, S. and Tuteja, R.K. (1991): Ratio and product type exponential estimator.
Cochran, W. G. (1977): Sampling techniques. Third edition Wiley and Sons, New York.
Hidiroglou, M. A. and Sarndal, C.E. (1998): Use of auxiliary information for two-phase
Murthy, M..N (1967): Sampling Theory and Methods. Statistical Publishing Society,
Calcutta, India.
Rao, T. J. (1983): A new class of unbiased product estimators. Tech. Rep. No. 15183,
Singh, R. and Singh, J. (2006): Separation of bias in the estimators of population mean
332.
Singh, S. and Singh, R. (1993): A new method: Almost separation of bias precipitates in
53
Almost Unbiased Ratio and Product Type Estimator of Finite
(rsinghstat@yahoo.com)
Florentin Smarandache
(smarand@unm.edu)
Abstract
It is well recognized that the use of auxiliary information in sample survey design
Out of many ratio, product and regression methods of estimation are good examples in
this context. Using the knowledge of kurtosis of an auxiliary variable Upadhyaya and
Singh (1999) have suggested an estimator for population variance. In this paper,
following the approach of Singh and Singh (1993), we have suggested almost unbiased
1. Introduction
54
Let U = (U1 ,U 2 ,......,U N ) denote a population of N units from which a simple random
denote the study and the auxiliary variables respectively. The problem is to estimate the
parameter
N
S y2 = σ y2 (1.1)
N −1
with σ y2 =
1 N
∑ ( yi − Y )2 of the study variate y when the parameter
N i =1
N
S x2 = σ x2 (1.2)
N −1
with σ x2 =
1 N
∑ (xi − X )2 of the auxiliary variate x is known,
N i =1
N N
yi x
where Y = ∑ and X = ∑ i ; are the population means of y and x respectively.
i =1 N i =1 N
∑ (y − Y )
n
2
i
s y2 = i =1
(1.3)
(n − 1)
n
yi
where y = ∑ is the sample mean of y.
i =1 n
S x2
t1 = s 2 2
y (1.4)
sx
1 n
where s x =
2
∑ (xi − x )2 is unbiased estimator of S x2 .
(n − 1) i=1
55
In many survey situations the values of the auxiliary variable x may be available
for each unit in the population. Thus the value of the kurtosis β 2 ( x ) of the auxiliary
⎡ S x2 + β 2 ( x )⎤
t2 = s ⎢ 2 2
⎥ (1.5)
⎣ s x + β 2 (x ) ⎦
y
For simplicity suppose that the population size N is large enough relative to the sample
size n and assume that the finite population correction (fpc) term can be ignored. Up to
the first order of approximation, the variance of s y2 , and t1 and bias and variances of t2
S 4y
( ) = n {β (y) − 1}
var s 2
y 2 (1.6)
S 4y
var(t 1 ) = [{β 2 (y ) − 1} + {β 2 (x ) − 1}(1 − 2C)] (1.7)
n
S y2
B(t 2 ) = [{β (x ) − 1}θ (θ − C )]
2 (1.8)
n
S y4
var(t 2 ) = [{β ( y ) − 1} + θ {β (x ) − 1}(θ − 2C )]
2 2 (1.9)
n
where θ =
S x2 μ
; β 2 ( y ) = 40
μ
; β 2 ( x ) = 042 ; h =
μ 22
;C=
(h − 1) and
S x + β 2 (x )
2
μ 20
2
μ02 (μ20 .μ02 ) β (x ) − 1
2
μ rs = ∑ ( yi − Y ) (xi − X ) .
s
1 N r
N i =1
56
From (1.8), we see that the estimator t2 suggested by Upadhyay and Singh (1999) is a
biased estimator. In some application bias is disadvantageous. This led authors to suggest
⎛ S 2 + β 2 (x ) ⎞
i
⎝ x
s + β 2
y
( x ) ⎠
denotes the set of all possible ratio-type estimators for estimating the population variance
3
tr = ∑ wit Ri ∈ R, (2.1)
i =1
3
where ∑w
i =1
i = 1 and wi are real numbers. (2.2)
For simplicity we assume that the population size N is large enough so that the
s y2 = S y2 (1 + e0 ), s x2 = S x2 (1 + e1 )
1 n
Noting that for large N, ≅ 0 and ≅ 0 , and thus to the first degree of approximation,
N N
β 2 ( y) − 1 β 2 ( x) − 1 (h − 1) [β 2 ( x) − 1]C
E (e02 ) = , E (e12 ) = , E (e0 e1 ) = = .
n n n n
t r = S (1 + e0 )∑ ai (1 + θe1 )
3
2 −i
y (2.3)
i =1
57
θe1 < 1 so that (1 + θe1 ) is expandable. Thus expanding the right hand side
i
Assume that
of the above expression (2.3) and retaining terms up to second power of e’s , we have
⎡ 3
⎛ ⎛ i + 1 ⎞ 2 2 ⎞⎤
t r = S ⎢1 + e0 − ∑ ai i⎜θe1 + θe0 e1 − ⎜
2
⎟θ e1 ⎟⎥
⎝ 2 ⎠
y
⎣ i =1 ⎝ ⎠⎦
or
⎡ 3
⎛ ⎛ i + 1 ⎞ 2 2 ⎞⎤
t r − S = S ⎢e0 − ∑ ai i⎜θe1 + θe0 e1 − ⎜
2 2
⎟θ e1 ⎟⎥ (2.4)
⎝ 2 ⎠
y y
⎣ i =1 ⎝ ⎠⎦
Taking expectation of both sides of (2.3) we get the bias of tr , to the first degree of
approximation, as
S2y ⎡ 3
⎤
B(t r ) = ⎢{β 2 ( x ) − 1}∑ ia i θ(θi − 2C + θ)⎥ (2.5)
2n ⎣ i =1 ⎦
Squaring both sides of (2.4), neglecting terms involving power of e’s greater than two
and then taking expectation of both sides, we get the mean-squared error of tr to the first
degree of approximation, as
S y4
MSE (t r ) = [{β ( y ) − 1} + R {θβ (x ) − 1}{θR
2 1 2 1 − 2C}] (2.6)
n
3
where R1 = ∑ i.wi (2.7)
i =1
Minimizing the MSE of tr in (2.7) with respect to R1 we get the optimum value of R1 as
C
R1 = (2.8)
θ
Thus the minimum MSE of tr is given by
min .MSE (t r ) =
S 4y
n
[{β 2 (y ) − 1} − {β 2 (x ) − 1}C 2 ]
58
=
n
[{β (y) − 1}(1 − ρ )]
S y4
2
2
1 (2.9)
where ρ1 =
(h − 1) is the correlation coefficient between ( y − Y )
2
{β (x ) − 1}{β ( y ) − 1}
2 2
and (x − X ) .
2
∑w
i =1
i =1 (2.10)
1
3
C ρ ⎧ β ( y) − 1 ⎫ 2
and ∑ iw i = = 1 ⎨ 2 ⎬ (2.11)
i =1 θ θ ⎩ β 2 ( x ) − 1⎭
From (2.10) and (2.11) we have three unknown to be determined from two equations
only. It is therefore, not possible to find a unique value of the constants wi ' s(i = 1,2,3) .
Thus in order to get the unique values of the constants wi ' s(i = 1,2,3) , we shall impose a
linear constraint as
B (tr ) = 0 (2.12)
(θ − C )a + (3θ − 2C )a
1 2 + (6θ − 3C )a3 = 0 (2.13)
Equation (2.10), (2.11) and (2.13) can be written in the matrix form as
⎡ 1 1 1 ⎤ ⎡ w1 ⎤
⎡ 1 ⎤
⎢ 1 ⎥ ⎢w ⎥ = ⎢C /θ ⎥
⎢
2 3
⎥ ⎢ 2⎥ ⎢ ⎥ (2.14)
⎢⎣(θ − C ) (3θ − 2C ) (6θ − 3C )⎥⎦ ⎢⎣ w3 ⎥⎦ ⎢⎣ 0 ⎥⎦
59
⎫
w1 =
1
[3θ ⎪
2
− 3θC + C 2 ]
θ 2
⎪
2 ⎪
w2 = 2 [− 3θ + 5θC − 2C ]⎬
1 2
(2.15)
θ ⎪
⎪
w3 = 2 [θ − 2θC + C 2 ]
1 2
θ ⎪
⎭
Use of these wi ' s(i = 1,2,3) remove the bias up to terms of order o(n −1 ) at (2.1).
Substitution of (2.14) in (2.1) yields the almost unbiased optimum ratio-type estimator of
⎡ s x2 + β 2 ( x ) ⎤
i
⎣ S x + β 2 ( x )⎦
y
the set of all possible product-type estimators for estimating the population variance S y2 .
3
t P = ∑ ki t Pi ∈ P , (3.1)
i =1
∑k
i =1
i = 1 and ki are real numbers.
S y2 ⎡
B(t P ) = {β 2 (x ) − 1}∑ iaiθ (θi + 2C − θ )⎤⎥
3
(3.2)
2n ⎣ ⎢ i =1 ⎦
60
S y4
MSE (t P ) = [{β ( y ) − 1} + R θ {(β (x ) − 1)}(θR
2 2 2 2 + 2C )] (3.3)
n
3
where, R2 = ∑ iki (3.4)
i =1
Minimizing the MSE of t P in (3.4) with respect to R2 , we get the optimum value of R2 as
C
R2 = − (3.5)
θ
Thus the minimum MSE of t P is given by
S y4
min .MSE (t P ) = {β ( y ) − 1}(1 − ρ )
2 1
2
(3.7)
n
⎫
k1 =
1
[3θ 2
+ 2θC + C 2 ]
⎪
θ 2
⎪
2 ⎪
k 2 = − 2 [3θ + 3θC + 2C ]⎬
1 2
(3.8)
θ ⎪
⎪
k 3 = 2 [θ 2 + θC + C 2 ]
1
θ ⎪
⎭
Use of these ki’s (i=1,2,3) removes the bias up to terms of order O (n-1) at (3.1).
4. Empirical Study
The data for the empirical study are taken from two natural population data sets
61
Y: number of agricultural laborers for 1971.
X: number of households
In table 4.1 the values of scalars wi’s (i=1,2,3) and ki’s (i=1,2,3) are listed.
I II I II
Using these values of wi’s and ki’s (i=1,2,3) given in table 4.1,one can reduce the bias to
the order O(n-1) respectively, in the estimators tr and tp at (2.1) and (3.1).
2
In table 4.2 percent relative efficiency (PRE) of s y ,t1,t2,tr (in optimum case) and tp
2
(in optimum case) are computed with respect to s y .
62
2
Table 4.2: PRE of different estimators of S y2 with respect to s y
Population I Population II
s y2 100 100
t1 223.14 228.70
t2 235.19 228.76
Table 4.2 clearly shows that the suggested estimators tr and tp in their optimum
case are better than the usual unbiased estimator s y2 , Isaki’s (1983) estimator t1 and
References
Ahmed, M.S., Abu Dayyeh, W. and Hurairah, A. A. O. (2003): Some estimators for finite
143-150.
Das, A.K. (1988): Contributions to the theory of sampling strategies based on auxiliary
Bengal, India.
Statistical Association.
63
Singh, S. and Singh, R. (1993): A new method: Almost Separation of bias precipitates in
Upadhyaya, L.N. and Singh, H. P. (1999): An estimator for population variance that
64
A General Family of Estimators for Estimating Population Variance
Using Known Value of Some Population Parameter(s)
Florentin Smarandache
Department of Mathematics, University of New Mexico, Gallup, USA
(fsmarandache@yahoo.com)
Abstract
variable under study, which make use of known value of certain population parameter(s),
is proposed. Some well known estimators have been shown as particular member of this
family. It has been shown that the suggested estimator is better than the usual unbiased
estimator, Isaki’s (1983) ratio estimator, Upadhyaya and Singh’s (1999) estimator and
Kadilar and Cingi (2006). An empirical study is carried out to illustrate the performance
1. Introduction
researchers are interested in the variation of their produce or yields (Ahmed et.al. (2003)).
65
Let ( U = U1 , U 2 ,......., U N ) denote a population of N units from which a simple random
sample without replacement (SRSWOR) of size n is drawn. Further let y and x denote the
population mean and population variance of the study character y. Assume that
population size N is very large so that the finite population correction term is ignored. It
is established fact that in most of the survey situations, auxiliary information is available
(or may be made to be available diverting some of the resources) in one form or the other.
If used intelligibly, this information may yield estimators better than those in which no
Assume that a simple random sample of size n is drawn without replacement. The
1 n
s 2y = ∑
n − 1 i =1
(y i − y )2 (1.1)
∑y
1
where y = i is the sample mean of y.
n i =1
⎛ s 2y ⎞
t1 = ⎜ 2 ⎟S2x (1.2)
⎜ sx ⎟
⎝ ⎠
1 n
where s 2x = ∑
n − 1 i =1
(x i − x )2 is an unbiased estimator of S2x .
66
Several authors have used prior value of certain population parameter(s) to find
more precise estimates. The use of prior value of coefficient of kurtosis in estimating the
population variance of study character y was first made by Singh et. al. (1973). Kadilar
and Cingi (2006) proposed modified ratio estimators for the population variance using
variation.
for estimating the population variance S2y . The expressions of bias and mean-squared
error (MSE), up to the first order of approximation, have been obtained. Some well
t = s 2y
(aS 2
x −b ) (2.1)
[α (as 2x − b) + (1 − α )(aS 2x − b)]
where (a ≠ 0), b are either real numbers or the function of the known parameters of the
(β 2 ( x )) .
The following scheme presents some of the important known estimators of the
67
Table 2.1 : Some members of the proposed family of the estimators ‘t’
Estimator Values of
α a b
t 0 = s 2y 0 0 0
s 2y 1 1 0
t1 = S2x Isaki (1983)
s 2x
estimator
s 2y 1 1 Cx
t2 = [S2x − C x ] Kadilar
s 2x − Cx
s 2y 1 1 β2 (x)
t3 = [S2x − β 2 ( x )]
s 2x − β 2 ( x )
s 2y 1 β2 (x) Cx
t4 = [S2xβ 2 ( x ) − C x ]
s 2xβ 2 ( x ) − C x
s 2y 1 Cx β2 (x)
t5 = [S2x C x − β 2 ( x )]
s 2x C x − β2 (x)
s 2y 1 1 - β2 ( x )
t6 = [S 2x + β 2 ( x )]
s 2x + β 2 ( x )
The MSE of proposed estimator ‘t’ can be found by using the firs degree approximation
MSE( t ) ≅ d ∑ d ′ (2.2)
68
where
∂h (a , b) ∂h (a , b)
h= [ ]
∂a S2y ,S2x ∂b S2y ,S2x
⎡ V(s 2y ) Cov(s 2y , s 2x )⎤
∑=⎢ 2 2 ⎥.
⎢⎣Cov(s x , s y ) V(s 2x ) ⎥⎦
Here h(a,b) = h( s 2y , s 2x ) = t. According to this definition, we obtain ‘d’ for the proposed
estimator, t, as follows:
αaS 2y
d= [1 - ]
aS 2x + b
⎛ aS 2y ⎞ ⎛ αaS 2y ⎞
MSE ( t ) ≅ V(s 2y ) − 2α⎜ 2 ⎟Cov(s 2 , s 2 ) + ⎜ ⎟V(s 2 ) (2.3)
⎜ aS x − b ⎟ y x
⎜ aS 2x − b ⎟ x
⎝ ⎠ ⎝ ⎠
where
V(s 2y ) = λS4y [β 2 ( y) − 1] ⎫
⎪⎪
V(s 2x ) = λS4y [β 2 ( x ) − 1] ⎬ (2.4)
⎪
Cov(s 2y , s 2x ) = λS2yS2x (h − 1)⎪⎭
1 μ 40 μ 04 μ 22
where λ = , β 2 ( y) = , β2 (x ) = , h= ,
n μ 220 μ 02
2
μ 20μ 02
∑ (y
1
μ rs = i − Y ) r ( x i − X )s , (r, s) being non negative integers.
N i =1
{
MSE ( t ) ≅ λS4y β 2 ( y) − 1 − 2αθ( h − 1) + α 2θ 2 (β 2 ( x ) − 1) } (2.5)
aS 2x
where θ = .
aS 2x − b
69
{ }
MSE ( t i ) ≅ λS4y β 2 ( y) − 1 − 2αθi (h − 1) + α 2 θi2 (β 2 ( x ) − 1) , i = 2,3,4,5 ,6 (2.6)
where
S2x S2x
θ2 = , θ3 = ,
S2x − C x S2x − β 2 ( x )
S 2x β 2 ( x ) S 2x C x S 2x
θ4 = , θ5 = , θ6 = .
S 2x β 2 ( x ) − C x S 2x C x − β 2 ( x ) S 2x + β 2 ( x )
C
α= = α opt (2.7)
θ
(h − 1)
where C = .
{β2 (x ) − 1}
3. Efficiency comparisons
provided C ≠ θi .
70
Thus it follows from (3.3) and (3.4) that the suggested estimator under ‘optimum’
condition is (i) always better then s 2y , (ii) better than Isaki’s (1983) estimator t1 except
when C = 1 in which both are equally efficient, and (iii) Kadilar and Cingi (2006)
equally efficient.
4. Empirical study
Table 4.1: Data statistics of the population for the simple random sampling
The percent relative efficiencies of the estimators s 2y , t i (i = 2,3,4,5,6) and min .MSE ( t )
with respect to s 2y have been computed and presented in Table 4.2 below.
71
Table 4.2: Relative efficiencies (%) of s 2y , t i (i = 2,3,4,5,6) and min .MSE ( t ) with respect
to s 2y .
t 0 = s 2y 100
t1 201.6564
t2 201.6582
t3 201.6782
t4 201.6565
t5 201.6672
t6 201.6347
5. Conclusion
infer that the proposed estimator ‘t’ under optimum condition performs better than usual
estimator s 2y , Isaki’s (1983) estimator t1, Kadilar and Cingi’s (2006) estimators (t2, t3, t4,
References
Ahmed, M.S., Dayyeh, W.A. and Hurairah, A.A.O. (2003): Some estimators for finite
population variance under two-phase sampling. Statistics in Transition, 6, 1,
143-150.
72
Isaki, C.T. (1983): Variance estimation using auxiliary information. Jour. Amer. Stat.
Kadilar, C. and Cingi, H. (2006): Ratio estimators for the population variance in simple
Singh, J., Pandey, B.N. and Hirano, K. (1973): On the utilization of a known coefficient
25, 51-55.
Upadhyaya, L.N. and Singh, H. P. (1999): An estimator for population variance that
73
This volume is a collection of six papers on the use of auxiliary
information and a priori values in construction of improved estimators.
The work included here will be of immense application for researchers
and students who employ auxiliary information in any form.
ISBN 1-59973-046-4
53995>
9 781599 730462
74