Biometrical Journal 41 (1999) 5, 627±636

Use of Transformed Auxiliary Variable

in Estimating the Finite Population Mean

Lakshmi N. Upadhyaya and Housila P. Singh

Indian School of Mines
Dhanbad and Vikram University
Ujjain
India

Summary

For estimating the finite population mean Y of the study character y, an estimator using a transformed
auxiliary variable has been defined. The bias and mean-squared error (MSE) of the proposed estimator
have been obtained. The regions of preference have been obtained under which it is better than usual
unbiased estimator y, the ratio estimator yR ˆ yX=  x, Sisodia and Dwivedi (1981) estimator
ys ˆ y…X ‡ Cx †=…
x ‡ Cx † and Singh and Kakran (1993) estimator yk ˆ y‰X ‡ b2 …x†Š=‰
x ‡ b2 …x†Š. An
empirical study has been carried out to demonstrate the superiority of the suggested estimator over the
others.

Key words: Finite population mean; Transformed auxiliary variable; Bias; Mean-
squared error; Coefficient of Variation; Coefficient of Kurtosis.

1. Introduction

Let U ˆ …U1 ; U2 ; . . . ; UN † be finite population of N (given) units. The variate of

interest y and the auxiliary variate x related to y assume real non-negative values
…yi ; xi † on the unit Ui …i ˆ 1; 2; . . . ; N†. This non-negativity condition is met by
almost all survey universes. From the population U, a simple random sample of
size n is drawn without replacement. Let …X;  Y†
 be the population means and
x; y† sample means for respective variates. Then the classical ratio estimator for
…
Y is defined by
 x:
yR ˆ yX= …1:1†
Here it is assumed that the population mean X of the auxiliary variate x is known.
When the population coefficient of variation Cx of x is known, Sisodia and
Dwivedi (1981) suggested a modified ratio estimator for Y as
‰X ‡ Cx Š
ys ˆ y : …1:2†
‰x ‡ Cx Š
628 L. N. Upadhyaya, H. P. Singh: Estimating the Finite Population Mean

Motivated by Sisodia and Dwivedi (1981), Singh and Kakran (1993) sug-
gested another ratio-type estimator for Y as
‰X ‡ b2 …x†Š
yk ˆ y ; …1:3†
‰x ‡ b2 …x†Š

where b2 …x†, the population coefficient of kurtosis of auxiliary variate x is as-

sumed to be known.
The variance of the usual unbiased estimator y is
…1 ÿ f † 2
V…
y† ˆ Sy …1:4†
n
and, to the first degree of approximation, the biases ans MSE's of yR , ys , and yk
are given by
…1 ÿ f †  2
yR † ˆ
B… YCx …1 ÿ K† ; …1:5†
n
…1 ÿ f †  2
ys † ˆ
B… YCx …a ÿ K† ; …1:6†
n
…1 ÿ f †  2
yk † ˆ
B… YCx …d ÿ K† ; …1:7†
n
…1 ÿ f † 2 2
yR † ˆ
MSE… Y ‰Cy ‡ Cx2 …1 ÿ 2K†Š ; …1:8†
n
…1 ÿ f † 2 2
ys † ˆ
MSE… Y ‰Cy ‡ Cx2 a…a ÿ 2K†Š ; …1:9†
n
…1 ÿ f † 2 2
yk † ˆ
MSE… Y ‰Cy ‡ Cx2 d…d ÿ 2K†Š ; …1:10†
n
where f ˆ n=N, Cy is the coefficient of variation of y, r is the coefficient of
correlation between y and x,
 X ‡ Cx † ;
a ˆ X=…  X ‡ b2 …x†Š ;
d ˆ X=‰
and
k ˆ rCy =Cx :

In many situations the values of the auxiliary variate may be available for each
unit in the population, for instance, see Das and Tripathi (1981). In such situa-
tions knowledge on X,  Cx , b2 …x† and possibly on some other parameters may be
utilised. Regarding the availability of information on Cx and b2 …x†, the researcher
may refer to Searls (1964), Sen (1978), Murthy (1967, pp. 96±99), Singh et al.
(1973) and Searls and Intarapanich (1990).
Biometrical Journal 41 (1999) 5 629
2. The Suggested Estimator

Let ui ˆ b2 …x† xi ‡ Cx ; …i ˆ 1; 2; . . . ; N†; so that ui ˆ b2 …x† x ‡ Cx is the sample

mean of the transformed variate u and U ˆ b2 …x† X ‡ Cx is the corresponding
population mean. We suggest an estimator for the population mean Y as
 2 …x† ‡ Cx Š
‰Xb
Y^ ˆ y  u:
ˆ yU= …2:1†
‰
xb2 …x† ‡ Cx Š
To obtain the bias and MSE of Y, ^ we write
 ‡ e0 † ;
y ˆ Y…1  ‡ e1 †
x ˆ X…1
such that
E…ei † ˆ 0 …i ˆ 0; 1† ;
and
…1 ÿ f † 2
E…e20 † ˆ Cy ;
n
…1 ÿ f † 2
E…e21 † ˆ Cx ; …2:2†
n
…1 ÿ f †
E…e0 e1 † ˆ rCy Cx :
n
Expressing (2.1) in terms of e's, we have
 ÿ1
 2 …x†
Xb
^
 
Y ˆ Y…1 ‡ e0 † 1 ‡  e1 : …2:3†
Xb2 …x† ‡ Cx
We assume that
 
 2 …x† e1 
 Xb
 
 2 …x† ‡ Cx  < 1
Xb
so that
 ÿ1
 2 …x†
Xb
1‡  e1
Xb2 …x† ‡ Cx
is expandable. Thus, to the first degree of approximation, the bias and MSE of Y^

are, respectively, given by
…1 ÿ f †  2
^ ˆ
B…Y† YCx w…w ÿ K† ; …2:4†
n
…1 ÿ f † 2 2
^ ˆ
MSE…Y† Y ‰Cy ‡ wCx2 …w ÿ 2K†Š ; …2:5†
n
 2 …x†
Xb
where w ˆ  :
Xb2 …x† ‡ Cx
630 L. N. Upadhyaya, H. P. Singh: Estimating the Finite Population Mean

3. Efficiency Comparisons

^ < V…
We note from (1.4) and (2.5) that MSE…Y† y† if
1  2 …x† Cx
Xb
r>  2 …x† ‡ Cx Š Cy : …3:1†
2 ‰Xb
It follows from (1.8) and (2.5) that the estimator y^ is more efficient than the usual
ratio estimator yR if
 
1 Cx  2 …x†
Xb
r< 1‡  : …3:2†
2 Cy Xb2 …x† ‡ Cx
Therefore the range of r for which Y^ is better than y and yR is
 
1  2 …x† Cx
Xb 1 Cx  2 …x†
Xb
 2 …x† ‡ Cx Š Cy < r < 2 Cy 1 ‡ Xb
2 ‰Xb  2 …x† ‡ Cx : …3:3†

From (1.9) and (2.5), we note that the estimator Y ^ would be more precise than the
estimator ys proposed by Sisodia and Dwivedi (1981), if
 
1 Cx fb2 …x† ÿ 1g Cx X
r> 2‡   : …3:4†
2 Cy Xb2 …x† ‡ Cx X ‡ Cx
As the left hand side of (3.4) is greater than that of (3.1), the range of superiority
over y; yR and ys is given by
   
1 Cx fb2 …x† ÿ 1g Cx X 1 Cx  2 …x†
Xb
2‡  <r< 1‡  :
2 Cy Xb2 …x† ‡ Cx X ‡ Cx 2 Cy Xb2 …x† ‡ Cx
…3:5†
It follows from (1.10) and (2.5) that the estimator y ^ is more efficient than Singh

and Kakran (1993) estimator yk, if
" #
1 Cx X fb2z …x† ÿ Cx g
r<  2‡  ; Cx < b22 …x† : …3:6†
2 Cy X ‡ b2 …x† fXbz …x† ‡ Cx g

4. Another Estimator

Using the transformation zi ˆ xi Cx ‡ b2 …x†; …i ˆ 1; 2; . . . ; N†; we define another

ratio-type estimator for Y as
 x ‡ b2 …x†
XC
Y^1 ˆ y  z
ˆ yZ= …4:1†
xCx ‡ b2 …x†
where z ˆ xCx ‡ b2 …x† is an unbiased estimator (based on n observations) of popu-
lation mean Z ˆ XC
 x ‡ b2 …x† of z.
Biometrical Journal 41 (1999) 5 631
The bias and MSE of Y^1 , to the first degree of approximation, are respectively
given by
…1 ÿ f † 
B…y^
1 † ˆ Yq…q ÿ K† Cx2 ; …4:2†
n
…1 ÿ f † 2 2
MSE…y^1 † ˆ Y ‰Cy ‡ q…q ÿ 2K† Cx2 Š ; …4:3†
n
 x =fXC
where q ˆ XC  x ‡ b2 …x†g:
From (1.4) and (4.3), we have that MSE…Y^1 † < V… y†, if
1 Cx2 X
r>  x ‡ b2 …x†Š : …4:4†
2 Cy ‰XC
From (1.8) and (4.3), we note that the estimator Y^
1 has smaller MSE than that
of the usual ratio estimator yR if
 
1 Cx  x
XC
r< 1‡  : …4:5†
2 Cy XCx ‡ b2 …x†
Thus, from (4.4) and (4.5), it follows that the estimator Y^
1 is more efficient than y
and yR if
   
1 Cx  x
XC 1 Cx  x
XC
 x ‡ b2 …x† < r < 1 ‡  x ‡ b2 …x† : …4:6†
2 Cy XC 2 Cy XC
From (1.9) and (4.3) we note that the estimator Y^
1 will dominate over Sisodia and
Dwivedi (1981) estimator if
  
1 Cx fCx2 ÿ b2 …x†g X
r< 2‡  ; Cx2 < b2 …x† : …4:7†
2 Cy fXCx ‡ b2 …x†g X ‡ Cx
Now combining (4.6) and (4.7) we get the range of r for which Y^
1 is more effi-
cient than y; yR and ys as
    
1 Cx ^ x
XC 1 Cx fCx2 ÿ b2 …x†g X 2
2 Cy XC  x ‡ b2 …x† < r < 2 Cy 2 ‡ fXC
 x ‡ b2 …x†g X ‡ Cx ; Cx < b2 …x† :
…4:8†
^1 will dominate over
Further we note from (1.10) and (4.3) that the estimator Y
Singh and Kakran (1993) estimator yk if
 
1 Cx X fX ‡ b2 …x†g Cx
r< 1‡  ; Cx < 1 : …4:9†
2 Cy fX ‡ b2 …x†g fXCx ‡ b2 …x†g
Remark 4.1: When the study variate y and auxiliary variate x is negatively
correlated, one may consider the following product-type estimators for Y as
632 L. N. Upadhyaya, H. P. Singh: Estimating the Finite Population Mean

‰
xb2 …x† ‡ Cx Š
Y^
* ˆ y
 2 …x† ‡ Cx Š ; …4:10†
‰Xb
‰
xCx ‡ b2 …x†Š
Y^
1* ˆ y
 x ‡ b2 …x†Š : …4:11†
‰XC

5. Unbiased Versions of the Suggested Estimators

We find the unbiased versions of the suggested estimators using two well known
procedures: (i) Interpenetrating subsamples design, (ii) Jack-knife technique.

(i) Interpenetrating subsamples design

Let the sample in the form of n independent interpenetrating subsamples be drawn.

Let yi and xi be unbiased estimates of the population totals Y and X respectively
based on the ith independent interpenetrating subsample, i ˆ 1; 2; . . . ; n. We,
now, consider two ratio-type estimators:
Y^ ˆ yU=
 u …5:1†
and
U Pn
Y^
n ˆ …yi =ui † …5:2†
n iˆ1
Pn Pn
 where y ˆ
as estimators for the population mean Y, yi =n, u ˆ ui =n, and
ui ˆ xi b2 …x† ‡ Cx . iˆ1 iˆ1
^n † ˆ n
B…Y†
Following Murthy (1964), we can show that B…Y ^
 and hence that

nY^
 ÿ Y^n
Y^u ˆ …5:3†
…n ÿ 1†
is unbiased for Y.
Similarly, the unbiased versions of Y^1 , Y^* and Y^
1* can be obtained as:

nY^1 ÿ Y^1n
Y^1u ˆ ; …5:4†
…n ÿ 1†
…nY^* ÿ Y^n*†
Y^*u ˆ …5:5†
…n ÿ 1†
and
^* ^*
* ˆ …nY 1 ÿ Y1n † ;
Y^1u …5:6†
…n ÿ 1†
Biometrical Journal 41 (1999) 5 633

Z Pn
Y^  z† ;
1 ˆ y…Z= Y^1n ˆ …yi =zi † ; zi ˆ xi Cx ‡ b2 …x† ;
n iˆ1
P P
* ˆ 1
n n
Y^* ˆ yu=U ; Y^n* ˆ  ;
yi ui =…nU† Y^
1* ˆ y…  ;
z=Z† Y^
1n yi zi :
iˆ1 nZ iˆ1
The properties of these unbiased estimators can be studied on the lines of
Murthy and Nanjamma (1959).

(ii) Jack-knife technique

We may take n ˆ 2m and split the sample at random into two subsamples of m
units each. Let yi , xi …i ˆ 1; 2† be unblased estimators of population means Y and
X based on the subsamples and y, x the means based on the entire sample.
Take
ui ˆ xi b2 …x† ‡ Cx ; zi ˆ xi Cx ‡ b2 …x† ; …i ˆ 1; 2†
and
u ˆ xb2 …x† ‡ Cx ; z ˆ xCx ‡ b2 …x† :
Thus the unbiased versions of ratio-type estimators Y^
 and Y^
1 are respectively given by
…2N ÿ n† ^ …N ÿ n† ^…1† ^…2†
Y^J ˆ
…u†
Y ÿ fY ‡ Y g ; …5:7†
N 2N
…2N ÿ n† ^ …N ÿ n† ^…1† ^…2†
Y^1J ˆ
…u†
Y1 ÿ fY1 ‡ Y1 g ; …5:8†
N 2N
…2N ÿ n† ^* …N ÿ n† ^*…1† ^*…2† g ;
Y^j ˆ
…u†
Y ÿ fY ‡ Y …5:9†
N 2N
…2N ÿ n† ^* …N ÿ n† ^*…1† ^*…2† g ;
Y^*1j ˆ
…u†
Y1 ÿ fY1 ‡ Y 1 …5:10†
N 2N
^ Y^1 , Y^
where Y, *, Y^1* are defined in (2.1), (4.1), (4.10), (4.11); and

Y^…i† ˆ yi …U= Y^1 ˆ yi …Z= Y^

*…i† ˆ yi ui =U ;
 ui † ; …i†  zi † ;

Y^1* ˆ yi zi =Z ;

…i†
…i ˆ 1; 2† :
The variance expressions of these unbiased estimators Y^ J…u†, Y^…u† ^…u†
1J , Yj and Y^ *1j…u† can
be derived on the lines similar to those of Sukhatme and Sukhatme (1970,
pp. 161±165). It is interesting to note that the variance expressions of Y^ J…u†, Y^ …u†
1J ,
^
…u† ^
*…u† ^
 ^
 ^
* ^
*
Yj and Y 1j and the MSE expressions of Y, Y1 , Y , Y 1 are respectively equal upto
the first order of approximation. Hence one can prefer the estimators Y^ J…u†, Y^ …u† ^…u†
1J , Yj
^ …u† ^
and Y*1j as compared to Y, ^ ^ ^
 Y1 , Y*, Y1* as they are unbiased.
634 L. N. Upadhyaya, H. P. Singh: Estimating the Finite Population Mean

6. Empirical Study

In this section we show the performance of various estimators of the population

mean Y through two natural population data.

Population I: [Source: Das (1988)]

The population consists of 278 villages towns/wards under Gajole Police station
of Malda district of West Bengal, India. The variates considered are:
y: the number of agricultural labourers in 1971 ;
x: the number of agricultural labourers in 1961 :
Values of different required parameters are:
Y ˆ 39:0680 ; X ˆ 25:1110 ; Cy ˆ 1:4451 ; Cx ˆ 1:6198 ;
r ˆ 0:7213 ; b2 …x† ˆ 38:8898 :
The percent relative efficiencies (PRE's) of various estimators are presented in
Table 1.

Table 1
^ Y^
Percent relative efficiencies of y, yR , ys , yk , Y, 1 with respect to y

Estimator y yR ys yk Y^

 Y^
1


PRE…: ; Y† 100.00 156.39 118.54 178.90 157.00 199.31

From the Table 1 it can be concluded that the proposed estimator Y^1 is better
than the sample mean, the ratio estimator and the estimators proposed by Sisodia
and Dwivedi (1981), and Singh and Kakran (1993).

Population II: [Source: Cochran (1977, pp. 325)]

The variates are defined as follows:

y: Number of persons per block ;
x: Number of rooms per block :
Values of different required parameters are:
Y ˆ 101:1 ; X ˆ 58:80 ; Cy ˆ 0:144 50 ; Cx ˆ 0:128 1 ;
r ˆ 0:650 0 ; b2 …x† ˆ 2:238 7 :

Table 2 presents the PRE's of different estimators of the population mean Y.
Biometrical Journal 41 (1999) 5 635
Table 2
^ Y^
Percent Relative Efficiencies of y, yR , ys , yk , Y, 1 with respect to y

Estimator y yR ys yk Y^

 Y^
1


PRE…:; Y† 100.00 158.23 158.45 161.83 158.33 172.87

From Table 2 there is a clear indication of the superiority of the proposed esti-
mator Y^1 over the sample mean, the ratio estimator and the estimators proposed by
Sisodia and Dwivedi (1981), and Singh and Kakran (1993), while the proposed
estimator Y^ is marginally inferior to the estimators proposed by Sisodia and Dwi-
vedi (1981), and Singh and Kakran (1993).
It may also be concluded from Tables 1 and 2 that the transformation
zi ˆ xi Cx ‡ b2 …x†, i ˆ 1; 2; . . . ; N; is more appropriate than that of ui ˆ b2 …x†
 xi ‡ Cx , i ˆ 1; 2; . . . ; N.

Acknowledgements

The authors are thankful to the referee for his constructive suggestions for the
improvement of the paper.

References

Cochran, W. G., 1977: Sampling Techniques. John Wiley and Sons, New York.
Das, A. K. and Tripathi, T. P., 1981: A class of sampling strategies for population mean using infor-
mation on mean and variance of an auxiliary character. Proceedings of the I.S.I. Golden Jubilee
International Conference on Statistics: Applications and New directions. Calcutta, 16±19 Dec.,
1981, 174±181.
Das, A. K., 1988: Contributions to the theory of sampling strategies based on auxiliary information.
Ph.D. thesis submitted to BCKV, Mohanpur, Nadia, West Bengal, India.
Murthy, M. N., 1964: Product method of estimation. Sankhya 26 A, 294±307.
Murthy, M. N., 1967: Sampling theory and methods. Statistical Publishing Society, Calcutta.
Murthy, M. N. and Nanjamma, N. S., 1959: Almost unbiased ratio estimates based on interpenetrat-
ing subsample estimates. Sankhya 21, 381±392.
Quenouille, M. H., 1956: Notes on bias in estimation. Biometrika 43, 353±360.
Searls, D. T., 1964: The utilization of a known coefficient of variation in the estimation procedure.
Jour. Amer. Stat. Assoc. 59, 1225±1226.
Searls, D. T. and Intarapanich, P., 1990: A note on an estimator for the variance that utilizes the
kurtosis. The American Statistician 44, 295±296.
Sen, A. R., 1978: Estimation of the population mean when the coefficient of variation is known. Com-
mun. Statist., Theory-Meth. A 7, 657±672.
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. Statist. Math. 25, 51±55.
Singh, H. P. and Kakran, M. S., 1993: A modified ratio estimator using known coefficient of kurtosis
of an auxiliary character. Revised version submitted to Journal of Indian Society of Agricultural
Statistics. New Delhi, India.
636 L. N. Upadhyaya, H. P. Singh: Estimating the Finite Population Mean

Sisodia, B. V. S. and Dwivedi, V. K., 1981: A modified ratio estimator using coefficient of variation of
auxiliary variable. Jour. Ind. Soc. Agril. Statist. 33, 13±18.
Sukhatme, P. V. and Sukhatme, B. V., 1970: Sampling Theory of Surveys with Applications, 2nd
edition. Iowa State University Press, Ames.

Dr. Lakshmi N. Upadhyaya Received, July 1998

Department of Applied Mathematics Revised, March 1999
Indian School of Mines Accepted, March 1999
Dhanbad ±± 826 004
India

Dr. Housila P. Singh

School of Studies in Statistics
Vikram University
Ujjain ±± 456 010
India