Engineering and Scientific International Journal (ESIJ) ISSN 2394-7187(Online)

Volume 3, Issue 2, April – June 2016 ISSN 2394 - 7179 (Print)

A Generalized Class of Jack-Knifed Estimator

for Population Mean using an
Auxiliary Variable and Attribute under Measurement Errors
Shashi Bhushan#1, Arun Kumar*2
1
Department of Mathematics and Statistics, Dr. Shakuntala Misra National Rehabilitation University, Lucknow, India
2
Department of Applied Statistics, BabasahebBhimraoAmbedkar University, India.

Abstract— In this paper, we have considered a generalized statistical inference based on data observed continue to
class of estimators using auxiliary information in both the remain valid. On the contrary, when they are not
form variable and attribute under measurement error. We substantially small and insignificant, the inferences might
also suggest a class of unbiased estimators using the Jack- not be invalid and imprecise but may often lead to
knife technique under measurement error. The bias and unforeseen, undesirable and unlucky consequences in
mean square error (MSE) for the proposed classes are Srivastava and Shalabh, (2001). Some important sources of
obtained under measurement error. Finally, some quantity errors in survey data are argued in Cochran (1968),
concluding remarks are made clearly demonstrating the Shalabh (1997), and Singh and Karpe (2008, 2010)
importance of the proposed classes. discussed some determinants of population mean under
measurement errors.
Keywords— Mean Square Error; Non-Response Errors;
For a simple random sample of size n, let   j , x j , y j  be
Generalized Estimator.
the pair of values instead of the true values
1. Introduction  j
, X j ,Y j  on the characteristics   , X,Y 
We know that efficiency of an estimator is always respectively. Let the error of the observation or
increased when we use the auxiliary information in measurement is defined as
sampling. Such kind of auxiliary information may be in the u j = yj -Yj, j = 1 , 2 , ... , n (1.1)
form of a variable or an attribute which is known in
vj = xj - X
advance or being collected while the survey is conducted j

without increasing the cost of the survey or at a little cost wj = j - (1.2)

j
to improve the efficiency of the estimators. For an example
height is closely related to the weight factor and it is also
dissimilar to male and female showing that sex is a helpful 2. Proposed Class of Generalized Estimator
attribute while estimating height, Amount of milk produced
by the cow is associated on the breed as well as on the diet, The generalized estimator of population mean μ Y , using
yield of the crop depends on the variety of the seed and
manure used as well etc. In such situation we can use the mean μ X of auxiliary variable X and the population
advantage of auxiliary information available in the form of proportion P of auxiliary attribute  is given by
variable and attribute to improve the efficiency of the  x p 
estimator. Many of the statistician have utilizes the y at  f  y , , 
(2.1)
supplementary information either in the form of variable or  x P 
in the form of attribute at a time. Here we have considered  x p 
both type of information in the present study. where f ( .) is a function of  y, , 
satisfying the
 x P 
Over the past few decades, statisticians were given their
interest towards the estimation of parameters in the following regularity conditions:
occurrence of measurement errors. In survey sampling, the (i) f   Y , 1 , 1  = 1 ,
possessions of data typically assume that the observations
are the right measurements on distinctiveness being studied. (ii) The function is continuous and bounded in the closed
However this supposition is not pleased in many functions interval R of real line,
and data is contaminated with measurement errors, like (iii) The second order of partial derivatives of the function
non-response errors, treating errors and calculating errors.  x p 
These measured errors make the result unacceptable, which f  y, ,  are exist and are continuous and bounded
meant for no quantity error case. If measured errors are  x P 
considerably very small and so we can neglect it, then the in R also the first order partial derivatives are zero;

22
Engineering and Scientific International Journal (ESIJ) ISSN 2394-7187(Online)
Volume 3, Issue 2, April – June 2016 ISSN 2394 - 7179 (Print)

where p , x and y are the sample proportion, sample therefore,

means of  , X and Y respectively for a simple random 
W u + W  W v + W x  W

1  W u + W 
2

y at - μY = 
y p y
variable of size n. f1 + f2 + f3 +  f 200
 n μX n P n 2!  n
1 
n
  
y =  y j 
n  j= 1  W v + W x  W  W 
2 2
W p u
+Wy v
+Wx
+ f 020 + f 002 + 2 f 110
1  
n 2 2
μX n P n nμ X
=   u j + Y j - μY + μY 
n  j= 1 
W v + W x  W p W u
+ W y W p



(2.5)
1 + f011 + 2 f 1 0 1  + ... 
= W u + W y  + μ Y (2.2) μX P P 
   
n
Proceeding likewise, we have,
1 3. Bias and Mean Square Error of the Proposed
x = W v + W x  + μx (2.3) Class of Generalized Estimator
n
W
p= P+
p
(2.4) Taking expectation on both side of (2.5), we have
n
Where,
B ia s  y a t  = E y at
- μy 
 
W u + W y  1  W u + W y 
n 2
1 W v + W x  Wp
Wu =  ui , = E 

f1 + f2 + f3 +  f 200
n i 1 n μX n P n 2!  n
 
n n
1 1
W y
  Y i
 μY  , Wv =  vi ,
W v + W x 
2
W
2
W u
+Wy  W v
+Wx 
n i 1 n i 1 + f020 +
p
f002 + 2 f 110
2 2
1
n
1
n μX n P n nμ X
Wx   X i
 μX  and W p    i
 P  . It is
n i 1 n i 1
W v + W x  W p W u
+ W y W p



also noteworthy that in most practical situations the + f011 + 2 f 1 0 1  + ... 
attribute is less prone to measurement error than a variable μX P P  

and therefore, we assume that the measurement error
1  σv + σ x 
2 2
  σ  y, x  σ  x , p  f 0 1 1 σ  y, p  
2
σp
w j  0 and hence W w  0 so the resultant is (2.4). From 
1
(3.1)
= 2
f 020
+ f
2 002  +  f 100
+ + f101 
generalized estimator (2.3) expanding by Taylor’s series 2n  μX P   μ X μX P P 

and use the results (2.2), (2.3) and (2.4), we have Now, squaring (2.5) on both sides and then take
 x p expectation, we have
y at = f  y, , 
 μX P   y at  =  y at 
2
M SE E - μY

 f  x  f  p  f
=  f  μY , 1, 1  + y - μY  +  -1 +  -1  W + W  W v + W x  WP 
2
j u y
 u μ
 X   v  P  w = E  f1 + f2 + f3 
 n μX n P n 
1 
2 2
 f  x   f  p   f  x   f
2 2 2 2

  y - μY  + 2  y - μY  
2
+ -1 +  -1 -1
W 
+  W +W
2

 u 
2

W v + W x 
2
u    uv
2 2 2
2!  μ
 X  v  P  w μ
 X = E 
y v2
 f1 +
2
f2 +
2
f3
2


2 2
n μX n P n
 x  p   
  f 
2 2
 p  f
+2  -1 -1 + 2 -1 y - μY   + ...
μ  P   v  w  P  wu 
 X    W + W  W + W  W v + W x  W p W p W u + W Y  
 u y v x

+2  f1 f2 + f2 f3 + f3 f1  

W u + W y  1  W u + W 
2

W v + W x  Wp
 n μX n μX n P n P n n  
=  μY +
y
f1 + f2 + f3 +  f 200
 n μX n P n 2!  n
  2
 σ  y, x  σ  x, p  σ  y, p   
1  2 2 σx 2 σp 2
2

M S E  y at  = σ f
 y 1 + f + f +2 f1 f 2 + f2 f3 + f3 f1  
W v + W x  W  W 
2 2 2 2 2 3
W p u
+Wy v
+Wx n μX P  μ X μX P P  
+ 2
f020 + 2
f002 + 2 f 110
μX n P n nμ X

1  2 2 
2
W v + W x  W p  W +Wy  W  

 σv 2
+ f011 + 2
u p
f 1 0 1  + ... +  σ u f1 + 2 f 2
1
 (3.2)
μX P P   n  μX 
  

23
Engineering and Scientific International Journal (ESIJ) ISSN 2394-7187(Online)
Volume 3, Issue 2, April – June 2016 ISSN 2394 - 7179 (Print)

4. Proposed Class of Generalized Unbiased 2 1

xm = W  2  + W  2   + μ (4.11)
m  
Estimator v x X

Now, let us consider a simple random sample of size 2 1 2

pm = WP + P (4.12)
n=2mbe split randomly into two sub-samples each of size m
m each. Let  y 2 m , x 2 m , p 2 m  be the sample mean of On expanding (4.1), (4.2) and (4.3) in third order Taylor ’s
values  Y , X ,   respectively for the entire sample of size 2m series similarly as expand (2.1), we get
 
W u + W y  1  W u + W y 
2

and  y m , x m , p m  W v + W x 
i i i Wp
y a t,2 m =  μ Y +
3
be the sample means of values of f1 + f2 + f3 +  f 200
 2m μX 2m P 2m 2!  2m
 
 Y , X ,   respectively for  i = 1 , 2  sub-sample of size
th
i
W v + W x  W  W 
2 2
Wp +Wy +Wx
3 + f020 + f002 + 2
u v
f 110
m. Let y a is a generalized estimator for the entire sample 2 2
2μ X m 2P m 2m μ X
1 2
of size 2m. y a and y a be the generalized estimators for
W v + W x W W u
+W y W p



(4.13)
f 1 0 1  + ... 
p
the two randomly splited sub-samples of size m each so + f011 + 2
μX P P  
that,  

 
  W   + W     
2
1 1 1 1
 x 2m p 2m  1
1 1
Wu +W y 1  Wu +W y
3 1  v x Wp
ya = f  y 2m , , (4.1) y a t,m = μY + f1 + f2 + f3 +  f 200
 
m μX m P m 2!  m
 μX P  
 

W   + W    W   + W    W   + W   
2

 1 x 1 p 1  1  2

1 1 1 1 1 1

1 m m
v x Wp u y v x
ya = f  ym , ,  (4.2) + 2
f 020 + 2
f 002 + 2 f 110
μ P μX m P m mμX
 X 
W  W W W 
1 1 1 1 1
+Wx +W  (4.14)
and 
f 1 0 1  + ...
v p u y p
+ f011 + 2
μX P P 

2 2  
 pm 
2 2 xm
ya = f  ym , ,  (4.3)  
  W   + W     
2
2 2 2 2 2
2 2
Wu +W y 1  Wu +W y
 μX P  2  v x Wp
 y a t,m = μY +

f1 + f2 + f3 +  f 200
m μX m P m 2!  m
satisfying the regularity conditions mentioned in (2.1).  

Also, we can easily define the following terms mentioned
W   + W    W   + W    W   + W   
2
in (2.1), (2.3) and (2.4) on similar lines so that
2 2
22 2 2 2 2
v x Wp u y v x
+ 2
f 020 + 2
f 002 + 2 f 110
1 μX m P m mμX
y 2m = W u + W y  + μ Y (4.4)
 
W  W W  W  
2 2 2 2 2
2m +Wx +Wy 
 (4.15)
f 1 0 1  + ...
v p u p
+ f011 + 2
1 μX P P 

x 2m = W v + W x  + μX (4.5)  
2m Thus, on the lines of Sukhatme and Sukhatme (Chapter
IV, page 162), for N large, the Jack-knifed generalized
1
p 2m = WP + P (4.6) estimator y a t j be defined as
2m

 
3 1   1 2
1 1 1 1
y a tj  2 y a t , 2 m  y  y at , m (4.16)
W + μ
at , m
ym = +W (4.7) 2
m  
u y Y

 
W u + W y  1  W u + W y 
2

W v + W x  W
= 2  μY +
p
f1 + f2 + f3 +  f 200
1 1 
xm = W  1  + W  1   + μ (4.8) 
2m μX 2m P 2m 2! 

2m

m  
v x X

W v + W x  W u + W Y  W v + W x 
2 2
W
1 1 1  + f020 +
p
f002 + 2 f 110
pm = WP + P (4.9) 2 2
2μ X m 2P m 2m μ X
m

2 1 W v + W x  W p W + W y W 


W  2  + W 2
u p

ym = + μ (4.10) + f011 + 2 f 1 0 1  + ... 

m  
u y Y μX P P  


24
Engineering and Scientific International Journal (ESIJ) ISSN 2394-7187(Online)
Volume 3, Issue 2, April – June 2016 ISSN 2394 - 7179 (Print)

 1
Wu +W y
1
  W   + W   
1 1
1  1
 1

2
5. Bias and Mean Square Error of the Proposed
1  v x W p 1  Wu +W y
-
2 
μ
 Y
+
m
f1 +
μX m
f2 +
P m
f3 +
2 !  m
f 200 Unbiased Class of Generalized Estimator
 

On taking expectation on both sides of (4.17), we get the
W   + W    W   + W    W   + W   
2
1 1
1 2 1 1 1 1
v x Wp u y v x
+ 2
f 020 + 2
f002 + 2 f110 bias of y a t j given by
μX m P m mμX

B ia s  y a tj   E  y a tj - μ Y 
W   + W    W W   + W    W    
1 1 1 1 1

 + ...
v x p u y p
+ f011 + 2 f101   
W u + W y  W v + W x  1  W u + W y 
2
μX P P  Wp
 
  = 2E f1 + f2 + f3 +  f 200
 2m μ X 2m P 2m 2!  2m
 
 
     
2
2 2 2 2 2 2 2
 Wu +W y Wv +Wx Wp 1  Wu +W y
+  μY + f1 + f2 + f3 + f 200
W v + W x  W  W +Wx 
2 2
2 !  Wp +Wy
 m μX m P m m u v


+ f020 + f002 + 2 f 110
 2 2
2μ X m 2P m 2m μ X

W   + W    W   + W    W   + W   
2
2 2
22 2 2 2 2

+
v x
f 020 +
Wp
f 002 + 2
u y v x
f 110 W v + W x  W p W u
+ W y W p 


2 2 + f011 + 2 f 1 0 1  + ...
μX m P m mμX μX P P  

 
W u + W y  W v + W x  1  W u + W y 
2
Wp
y a tj
- μY  = 2 

f1 + f2 + f3 +  f 200
 
  W   + W     
2
2m μ X 2m P 2m 2!  2m 1 1 1 1 1
1 1
  1  Wu + W y v x Wp 1  Wu +W y
- E  f1 + f2 + f3 + f 200

2 ! 
W v + W x  W u + W Y  W v + W x 
2 2 2 m μX m P m m
Wp  
+ f020 + f002 + 2 f 110 
2 2
2μ X m 2P m 2m μ X

W   + W    W   + W    W   + W   
2
1 1
1 2 1 1 1 1
W v + W x W p W u + WY W p 
  v x Wp u y v x
+ f011 + 2 f 1 0 1  + ...  + f 020 + f002 + 2 f110
μX P P 
  2 2
μX m P m mμX

 
  W   + W     
2

W   + W    W W   + W    W  
1 1 1 1
1
 
1 1
1 1 1 1 1
1  Wu + W y Wp 1  Wu +W y
 + ...
v x
v x p u y p

- f1 + f2 + f3 + f 200
2  m μX m P m 2 !  m + f011 + 2 f101 
  μX P P 
 
 

W   + W    W   + W    W   + W   
2
1 1
1 2 1 1 1 1
 
  W   + W     
v x W u y v x 2 2 2 2
2
p 2 2
2
+ 2
f 020 + 2
f002 + 2 f110  Wu + W y v x Wp 1  Wu +W y
μX m P m mμX + f1 + f2 + f3 + f 200
 m μX m P m 2 !  m
 
W   + W    W W   + W    W    
1 1 1 1 1


f 1 0 1  + ...
v x p u y p
+ f011 + 2
μX P P 

W   + W    W   + W    W   + W   
2
  2 2
22 2 2 2 2
v x Wp u y v x
+ 2
f 020 + 2
f 002 + 2 f 110
  μX m P m mμX
  W   + W     
2
2 2 2 2
2 2 2
 Wu + W y v x Wp 1  Wu +W y
+ f1 + f2 + f3 + f 200
2 ! 
W   + W    W W   + W    W    
2 2 2 2 2
 m μX m P m m

f 1 0 1  + ...
v x p u y p
  + f011 + 2
μX P P  
W   + W    W   + W    W   + W   
2
2 2
22 2 2 2 2

v x Wp u y v x   
+ 2
f 020 + 2
f 002 + 2 f 110

W   + W    W W   + W    W  
μX m P m mμX 2 2 2 2 2
 

f 1 0 1  + ... = 0
v x p u y p
(5.1)
W v
2
+Wx
2
W p W u
2
+Wy
2
W 2
p
 
 + ... 
+
μX P
f011 + 2
P 


+ f011 + 2 f101 

  
μX P P 

   Therefore, y a t j is an unbiased estimator of population
W v
2
+Wx
2
W p W u
2
+Wy
2
W p
2  
 + ...  (4.17) mean  Y . On squaring (4.17) on both sides and then take
+ f011 + 2 f101 
μX P P  
 expectation we get mean square error of y a t j given by
  

25
Engineering and Scientific International Journal (ESIJ) ISSN 2394-7187(Online)
Volume 3, Issue 2, April – June 2016 ISSN 2394 - 7179 (Print)

σ 2 + σ 2
 σ 2
+ σx
2
 σ
2
 presence of measurement errors. Furthermore, since the
y =
u y v p

2 2 2
M SE a tj
f1 + 2
f2 + 2
f3  attribute auxiliary information is less prone to measurement
 2m 2m μ X 2m P 
error this estimator has more practical utility than other
1  σ  y,x  σ  x, p  σ  y, p   estimators like ratio estimator where such information is
+  f1 f 2 + f2 f3 + f3 f1  lacking.
m  μ X μX P P 
References
 σ  y , x  σ  x, p  σ  y, p   
2
1  2 2 σx 2 σp 2
2

= σ f
 y 1 + 2
f 2
+ 2
f3 + 2  f1 f2 + f2 f3 + f3 f1  
2m  μX P  μ X μX P P   [1] Bhushan, S., “Improved Sampling Strategies in Finite Population”,
Scholars Press, Germany, 2013.
[2]` Bhushan S., “Some Efficient Sampling Strategies based on Ratio
1  2 2 
σv
2 Type Estimator”, Electronic Journal of Applied Statistical Analysis,
+  σ u f1 + 2 f2
1 2
 (5.2) Vol. 5(1), 2012, pp.74 – 88.
2m  μX  [3] Bhushan S. and Katara, S., “On Classes of Unbiased Sampling
 Strategies”, Journal of Reliability and Statistical Studies, Vol. 3(2),
2010, pp.93-101.
6. Conclusion [4] Bhushan S. and Pandey A., “Modified Sampling Strategies using
Correlation Coefficient for Estimating Population Mean”, Journal
of Statistical Research of Iran, Vol. 7(2), 2010, pp.121- 131.
We can easily see that bias and mean square error of [5] Bhushan S., Singh, R. K. and Katara, S., “Improved Estimation
y a t and y a t j are under Midzuno – Lahiri – Sen-type Sampling Scheme”, Journal of
Reliability and Statistical Studies, Vol. 2(2), 2009, pp.59 – 66.
Bhushan S., Masaldan R. N. and Gupta P. K., “Improved Sampling
1  σv + σ x 
2 2 [6]
  σ  y, x  σ  x , p  f 0 1 1 σ  y, p  
2
σp
B ia s  y a t  = 
1 (6.1) Strategies based on Modified Ratio Estimator”, International
2 002 
f 020
+ f + f 100 + + f 101 
2
μX P Journal of Agricultural and Statistical Sciences, Vol. 7(1), 2011,
2n 
   μ X μX P P  pp.63-75.
[7] Abu-Dayyeh W. A. et al, “Some estimators of a finite population
 σ  y , x  σ  x, p  σ  y, p   
2
1  2 2 σx 2 σp 2
2

M S E (y a t ) =  σ y f1 + 2 f 2 + 2 f3 + 2  f1 f2 + f2 f3 + f3 f1   mean using auxiliary information”, Applied Mathematics and

n μX P  μ X μX P P   Computation, Vol. 139, 2003, pp.287-298.
[8] Allen, J., Singh, H. P. and Smarandache, F. , “A family of
estimators of population means using multi auxiliary information in
presence of measurement errors”, International Journal of Social
Economics, Vol. 30(7), 2003, pp. 837-849.
1  2 2 
2
σv 2 (6.2) [9] Kumar, M., Singh, R., Sawan, N. and Chauhan, P., “Exponential
 σ u f1 + 2 f 2 
1
+
n  μX 
ratio method of estimators in the presence of measurement errors”,
 Int. J. Agricult. Stat. Sci., Vol. 7(2), 2011, pp. 457-461.
B ia s  y a tj  = 0 [10] Kumar, M., Singh, R., Singh, A. K. and Smarandache, F., “Some
ratio type estimators under measurement errors”, WASJ, Vol. 14(2),
(6.3) 2011, pp.272-276.
[11] Manisha and Singh, R. K., “An estimation of population mean in
 σ  y , x  σ  x, p  σ  y, p   
2
1  2 2 σx 2 σp 2
2

M S E  y a tj  = σ f
 y 1 + 2
f 2
+ 2
f3 + 2  f1 f2 + f2 f3 + f3 f1   the presence of measurement errors”, Journal of Indian Society of
2m  μX P  μ X μX P P   Agricultural Statistics, Vol. 54(1), 2001, pp.13-18.
[12] Manisha and Singh, R. K., “Role of regression estimator involving
measurement errors”, Brazilian Journal of Probability Statistics,
Vol.16, 2002, pp.39-46.
1  2 2 
2
σv 2 (6.4) [13] Shalabh, “Ratio method of estimation in the presence of
 σ u f1 + 2 f2 
1
+
2m  μX  measurement errors”, Journal of Indian Society of Agricultural

Statistics, Vol. 50(2), 1997, Pp.150-155.
From (6.2) and (6.4), we see that both estimators y a t and [14] Singh, H. P. and Karpe, N., “Ratio product estimator for
population mean in presence of measurement errors”, Journal of
y a t j have the same mean square error but from (6.1) bias Applied Statistical Sciences, Vol.16, 2008, pp.49-64.
[15] Singh, H. P. and Karpe, N., “On the estimation of ratio and product
of y a t is not zero whereas from (6.3) bias of the jack- of two populations mean using supplementary information in
presence of measurement errors”, Department of Statistics,
knifed estimator y a t j is zero. Hence, both the estimators University of Bologna, Vol. 69(1), 2009, pp.27-47.
[16] Srivastava, A. K. and Shalabh, “Effect of measurement errors on
y a t and y a t j having the same mean square error, but the the regression method of estimation in survey sampling”, J. Statist.
Res., Vol. 35(2), 2001, pp.35-44.
jack-knifed estimator y a tj may be preferred to the [17] Sukahtme, P. V. and Sukhatme, B. V., “Sampling Theory of
Surveys with Applications”, Iowa State University Press, Ames,
estimator y a t in the sense of unbiasedness under the U.S.A., 1970.

26