See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/348418521

Computing the Effect of Measurement Errors on the Use of Auxiliary

Information under Systematic Sampling

Article in Communication in Statistics- Theory and Methods · January 2021

DOI: 10.1080/03610926.2020.1868513

CITATION READS

1 185

2 authors:

Neha Singh Gajendra Kumar Vishwakarma

Amrita Vishwa Vidyapeetham Indian Institute of Technology (ISM) Dhanbad
16 PUBLICATIONS 51 CITATIONS 113 PUBLICATIONS 752 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

COVID-19 View project

sampling technique View project

All content following this page was uploaded by Neha Singh on 12 February 2021.

The user has requested enhancement of the downloaded file.

 Communications in Statistics - Theory and Methods

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/lsta20

Computing the effect of measurement errors on

the use of auxiliary information under systematic
sampling

Neha Singh & Gajendra K. Vishwakarma

To cite this article: Neha Singh & Gajendra K. Vishwakarma (2021): Computing the effect
of measurement errors on the use of auxiliary information under systematic sampling,
Communications in Statistics - Theory and Methods, DOI: 10.1080/03610926.2020.1868513

To link to this article: https://doi.org/10.1080/03610926.2020.1868513

Published online: 11 Jan 2021.

Submit your article to this journal

Article views: 21

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=lsta20
COMMUNICATIONS IN STATISTICS—THEORY AND METHODS
https://doi.org/10.1080/03610926.2020.1868513

Computing the effect of measurement errors on the use

of auxiliary information under systematic sampling
Neha Singh and Gajendra K. Vishwakarma
Department of Applied Mathematics, Indian Institute of Technology (ISM) Dhanbad, Dhanbad,
Jharkhand, India

ABSTRACT ARTICLE HISTORY

The ratio, product, difference estimators, and unbiased estimator Received 15 January 2020
under systematic sampling scheme has been studied in the presence Accepted 19 December 2020
of measurement error. The methods of estimation have been pro-
KEYWORDS
posed for the estimation of the finite population mean. To exhibit
Systematic sampling;
the effect of measurement error, the study variable and auxiliary measurement error; ratio
variable are supposed to be observed with measurement error. The estimator; product
properties of estimators’ viz. bias and MSE have been obtained. The estimator; difference
simulation study has been conducted to compute the effect of estimator; mean
measurement error on the MSE for the different levels of correlation square error
coefficient and different levels of error variance.
MATHEMATICAL SUBJECT
CLASSIFICATION
62D05

1. Introduction
The importance of systematic sampling is emphasized as one of the sampling schemes
widely used in practice due to its simplicity. Systematic sampling has fine features of
selecting samples by a fixed pattern with just one random start. Apart from its simpli-
city, this method’s yield in certain situations is more efficient than simple random sam-
pling and stratified random sampling as discussed by Cochran (1946) and Gautschi
(1957). Madow (1949, 1953), Lahiri (1954), and Williams (1956) have done pioneered
work in systematic sampling. In survey sampling literature, the auxiliary information is
substantially used at the estimation stage to improve the efficiency of the estimators.
Ratio, regression, and product estimators are more prevalent in practice. Murthy (1964)
envisaged specific conditions under which any of the three estimations can be efficiently
used, namely ratio, product, and unbiased estimators. Hasel (1942) and Griffth (1945)
found systematic sampling to be efficient and convenient in sampling from certain nat-
ural populations like forest areas for estimating the volume of timber. In the case of
estimating the volume of timber, Swain (1964) suggested the leaf area or the girth of
the tree may be taken as the auxiliary variable and proposed ratio estimator under sys-
tematic sampling. Shukla (1971) proposed product estimators in the context of system-
atic sampling.

CONTACT Neha Singh nehaismstats@gmail.com Department of Applied Mathematics, Indian Institute of

Technology (ISM) Dhanbad, Dhanbad 826004, Jharkhand, India.
ß 2021 Taylor & Francis Group, LLC
2 N. SINGH AND G. K. VISHWAKARMA

The survey of some natural populations like forest areas or other areas where system-
atic sampling is used as discussed by Gregoire and Valentine Harry (2008). During
these surveys, the study variable and auxiliary variable may be observed with some error
e.g., the volume of timber, the leaf area or the girth of the tree may be observed with
some error. To obtain the behavior of customers, survey conducted by supermarket or
to collect information from individuals, systematic sampling is widely used. In study
related to customers or information collected from individuals during sample surveys,
the data may be obtained with some error with true value of the variable. Cochran
(1968) and Murthy (1967) discussed the real-life situations when the data is obtained
with the error. In general, the measurement error is defined as the difference between
the true and observed values of the variable. The impacts of measurement error com-
mingle with data on the statistical properties of estimators of parameters are generally
not discussed in the textbooks. Fuller (1987), Cheng and Van Ness (1994) and Carroll
et al. (2006) have discussed the impacts of measurement error in regression modeling
but not in the context of sample surveys. Inference in the presence of measurement
error can be misleading, as elucidated by Biemer and Stokes (1991).
By using auxiliary information, many authors have done major contribution to the lit-
erature related to ratio estimator. Goodman and Hartley (1958), Hartley and Ross (1954)
have done premier contribution and studied unbiased ratio estimators.
The study of measurement error in the context of ratio estimators was encountered by
Shalabh (1997). The study of measurement error related to survey sampling was further
extended by many authors. Maneesha and Singh (2002) have studied the regression estima-
tor for the mean in presence of measurement error. Allen, Singh, and Smarandache (2003)
used multi-auxiliary variable and proposed a general method of estimation of population
mean. Sahoo, Sahoo, and Senapati (2006) studied ratio and regression estimator in the
presence of measurement error. Gregoire and Salas (2009) have studied ratio estimator
with measurement error in auxiliary variate using several probability density function.
Salas and Gregoire (2010) have done pioneered work and studied random and systematic
measurement errors in ratio estimators. Shalabh and Tsai (2017) proposed a class of ratio
and product method of estimation in the presence of correlated measurement error. Singh,
Vishwakarma, and Kim (2019) have studied the effect of measurement error on ratio, prod-
uct, and mean estimator simultaneously. Singh and Vishwakarma (2019) studied the simul-
taneous effect of measurement error and non response on mean estimation. These all
works used simple random sampling but not in the context of systematic sampling.
The theory of parametric estimation is related to estimation of population parameters
using sample estimate. Parametric estimation is about to provide inference about par-
ameter using sample values. There are two approach design-based and model-based
inference in survey sampling. S€arndal, Swensson, and Wretman (1992) and Gregoire
(1998) have studied in detail about the differences between the above two models. In
studies related to design-based parametric estimation in systematic sampling, the effect
of measurement errors has not yet been considered according to our knowledge. But
during survey related to systematic sampling, data may be obtained with some contam-
ination. In this manuscript, we propose the three well-existing estimators namely ratio,
product, difference estimators and unbiased mean estimator in systematic sampling
under measurement errors.
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 3

Suppose, the population consists of N units u ¼ ðu1 , u2 , :::, uN Þ from a finite popula-
tion. The population size is divided into k intervals such that N ¼ nk: To select a sam-
ple, the first unit is selected at random from the first k units. This sampling method is
similar to that of selecting a cluster at random out of k cluster (each cluster containing
n units), made such that ith cluster contains serially numbered units i, i þ k, i þ
2k, :::, i þ ðn  1Þk: After sampling of n units, we observe both the study and auxiliary
variables. We consider a situation where each data value may be observed with error. It
is assumed that ðxij , yij Þ are observed values instead of their true values ðXij , Yij Þ for
every ijthði ¼ 1, 2, :::, k, j ¼ 1, 2, :::, nÞ i.e., jth sampling unit of ith cluster. In such a way,
these values are expressible in additive form as, xij ¼ Xij þ Vij and yij ¼ Yij þ Uij : The
errors ðU, VÞ are normally distributed with mean zero and variance ðr2U , r2V Þ: We
assume that the error variables U and V are uncorrelated to each other as well as
uncorrelated to all combinations with X and Y, respectively. This implies
CovðX, UÞ ¼ CovðX, VÞ ¼ CovðY, VÞ ¼ CovðY, UÞ ¼ CovðU, VÞ ¼ 0 and CovðX, YÞ 6¼ 0:
Let lYsy , lXsy be the population mean and r2Ysy , r2Xsy be the population variance of the
study and the auxiliary variables, respectively. q is the correlation coefficient between
study and auxiliary variable. Y  sy and X  sy are the sample means of the true values of the
study and auxiliary variable. Further, the sample mean of the observed data are
unbiased estimators of the population mean lXsy and lYsy respectively.
1X n
^ Ysy ¼ y sy ¼
l yij , ði ¼ 1, 2, :::, kÞ (1)
n j¼1

1X n
^ Xsy ¼ x sy ¼
l xij , ði ¼ 1, 2, :::, kÞ (2)
n j¼1

For determining bias and variance, it is expressed by means of error terms eY and
eX , which are defined as
y sy  lYsy x sy  lXsy
eY ¼ ) y sy ¼ lYsy ð1 þ eY Þ, eX ¼ ) x sy ¼ lXsy ð1 þ eX Þ (3)
lYsy lXsy

We can write
Pn
EðeY Þ ¼ EðeX Þ ¼ 0,  i: ¼ U
U  sy ¼ 1 Uij : and
n j¼1

  1=k n 2 o   1=k n o 1=kqrXsy rYsy

E e2X ¼ 2 rXsy þ r2Vsy , E e2Y ¼ 2 r2Ysy þ r2Usy , EðeY eX Þ ¼ ,
lXsy lYsy lYsy lXsy
lXsy
R¼
lYsy
k  2 k  2
1X  sy lYsy , r2Xsy ¼ 1
X
 sy lXsy , r2Usy ¼ 1
Xk
 i: Þ2 ,
r2Ysy ¼ Y X ðU
k i¼1 k i¼1 k i¼1

1X k
 i:: Þ2
r2Vsy ¼ ðV
k i¼1
4 N. SINGH AND G. K. VISHWAKARMA

2. The proposed estimator

Considering the severity of misleading inference based on data obtained with measure-
ment error, we propose ratio, product, difference, and mean estimators when the data
are observed with measurement errors under systematic sampling. The ratio estimator
and product estimators are bias estimators of population mean. If h is the value of a
parameter, then bias of the estimator can be defined as the difference between estimated
values and actual value.
biasð^hÞ ¼ Eð^hÞ  h (4)
The mean square error is also another vitally used in the efficiency of the estimator. The
mean square error of an estimator measures the average of the squared of the errors that
is, the average squared difference between the estimated values and actual value.
 
MSEð^hÞ ¼ E ð^h  hÞ
2
(5)
The variance of the systematic sample mean without measurement error is
k  2
1X  sy  lYsy
r2Ysy ¼ Y (6)
k i¼1

Considering that the observations are recorded with measurement error, then the
variance in the presence of measurement error is
k  2
  1X k
  1X  sy  lYsy
 sy þ U
V y sym ¼ r2Ysym ¼ y sy  lYsy 2 ¼ Y (7)
k i¼1 k i¼1
 
V y sym ¼ r2Ysy þ r2Usy (8)

where the term r2Usy is the variance of the measurement error.

The classical ratio due to Swain (1964) under the systematic random sampling is given as
l
 sy Xsy
y Rsy ¼ Y (9)
X sy

The mean square error of the estimator is given as

   
MSE y Rsy ¼ 1=k r2Ysy þ R2 r2Xsy  2qRrYsy rXsy (10)

We consider the situations when both study variables and the auxiliary variables are
observed with measurement error. In that case, we propose the ratio estimator as
lXsy
y Rsym ¼ y sy (11)
x sy
In order to obtain the bias and mean square error we can rewrite (11) as
lXsy
y Rsym ¼ lYsy ð1 þ eY Þ (12)
lXsy ð1 þ eX Þ

y Rsym ¼ lYsy ð1 þ eY Þð1 þ eX Þ1 (13)

 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 5

Expanding (13) up-to first order and simplifying to obtain the bias of the estimator,
we can write    
y Rsym ¼ lYsy ð1 þ eY Þ 1  eX þ e2X ¼ lYsy 1 þ e2X  eY eX (14)
Taking expectation of (14), we get the bias of the estimator as
  1=k n 2  2  o
bias y Rsym ¼ R rXsy þ r2Vsy  qRrXsy rYsy (15)
lYsy

Expanding (13) up-to first order and ignoring higher order, we can write
y Rsym ¼ lYsy ð1 þ eY  eX Þ (16)

To obtain the MSE, by squaring (16) we have

   
y Rsym  lYsy 2 ¼ l2Ysy e2Y þ e2X  2eY eX (17)

Taking expectation of (17), we get the mean square error as

  h   i
MSE y Rsym ¼ 1=k r2Ysy þ r2Usy þ R2 r2Xsy þ r2Vsy  2qRrYsy rXsy (18)

We can obtain the result under no measurement error by putting r2Usy and r2Vsy equal
to zero. This will give the same result as obtained by Swain (1964). From (10) and (18),
we can write that MSE in the presence of measurement error is always high.
The product estimators due to Shukla (1971) under systematic random sampling is
defined as

X
 sy sy
y Psy ¼ Y (19)
lXsy
The mean square error of the estimator is obtained by Shukla (1971) as
   
MSE y Psy ¼ 1=k r2Ysy þ R2 r2Xsy þ 2qRrYsy rXsy (20)

Under the circumstances when both study variable and the auxiliary variable are
observed with measurement error in that case, we define the product estimator as
x sy
y Psym ¼ y sy (21)
lXsy
To obtain the bias and mean square error, we can write (21) as
y Psym ¼ lYsy ð1 þ eY Þð1 þ eX Þ (22)

For the bias, by taking the expectation of (22), we get

 
bias y Psym ¼ 2qRrYsy rXsy =k (23)

For the mean square error, we can write from (22)

   
y Psym  lYsy 2 ¼ l2Ysy e2Y þ e2X þ 2eY eY (24)

Taking the expectation of (24), we get the mean square error as

  h   i
MSE y Psym ¼ 1=k r2Ysy þ r2Usy þ R2 r2Xsy þ r2Vsy þ 2qRrYsy rXsy (25)
6 N. SINGH AND G. K. VISHWAKARMA

By substituting the value r2Usy and r2Vsy equal to zero, we can obtain the MSE without
measurement error which is the same as obtained by Shukla (1971). From (20) and
(25), we can conclude that MSE is always higher in the presence of measurement error.
The difference estimator under systematic sampling is defined as,
 
y ¼ Y  sy þ b lXsy  X sy (26)
dsy

where b is a constant to be chosen such that the variance of the estimator

Vðy dsy Þ is minimum.
The minimum variance of difference estimator is given as
  h  i
V y dsym ¼ 1=k r2Ysy q2  1 (27)

Considering the situation when both the study variable and the auxiliary variable are
commingled with measurement error, in that case the difference type estimator is
defined as
 
y dsym ¼ y sy þ b lXsy  x sy (28)

In order to obtain variance, we can rewrite (28) as

 
y dsym ¼ lYsy ð1 þ eY Þ þ b lXsy  lXsy ð1 þ eX Þ (29)
   
y dsym  lYsy ¼ lYsy eY  beX lXsy (30)

Squaring both sides of (30) and we get

  h i
y dsym  lYsy 2 ¼ E l2Ysy e2Y þ b2 e2X l2Xsy  2blYsy lXsy eY eX (31)

By taking the expectation of (31), we get the variance of the estimator as

  h    i
V y dsym ¼ 1=k r2Ysy þ r2Usy þ b2 r2Xsy þ r2Vsy  2bqrYsy rXsy (32)

To obtain minimum variance differentiate (32) with respect to b and equate it to

zero, we get
  h    i
V y dsym ¼ 1=k r2Ysy þ r2Usy þ b2 r2Xsy þ r2Vsy  2bqrYsy rXsy (33)
qrYsy rXsy
b¼  (34)
r2Xsy þ r2Vsy

By substituting the value of b in (33), we get the minimum variance of the

estimator as
2 3
    q r
2 2
Ysy r 2
Xsy
V y dsym ¼ 1=k4 r2Ysy þ r2Usy   5 (35)
r2Xsy þ r2Vsy

From (27) and (35), we can write that MSE in the presence of measurement error is
always high. By putting r2Usy and r2Vsy equal to zero, we can obtain the MSE under no
measurement error which is same as given in (27).
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 7

Remark: If N is not a multiple of n, i.e., N 6¼ nk, then the sampling interval k cannot
be uniquely defined. In such a case, take k to be an integer nearest to ðN=nÞ:
Depending on the choice of the starting number i, sometimes we may have i þ ðn 
1Þk > N and so, the sample size instead of n reduces to ðn  1Þ:
The problem of variable sample size when N 6¼ nk can be overcome by adopting a
modified version of the above scheme introduced by Lahiri (1954) and known as circu-
lar systematic sampling (CSS). The procedure consist in selecting the unit i by random
start from 1 to N, and hence select every kth unit in a circular way, here k being an
integer nearest to ðN=nÞ: The systematic sample corresponding to the numbers:
)
i þ jk, if i þ jk  N
, j ¼ 0, 1, :::, ðn  1Þ: (36)
i þ jk  N, if i þ jk > N

The unbiased estimate of mean in case of CSS is given as

1X n
^ Ysycs ¼ y sycs ¼
l yij , ði ¼ 1, 2, :::, kÞ (37)
n j¼1

n is the size of the expected units selected as sample in CSS.

Considering the case of measurement error, the unbiased estimator of mean will be
same as above.
The variance of the circular systematic sample mean without measurement error is
N  2
1X  sy  lYsy
r2Ysycs ¼ Y (38)
N i¼1

Considering that the observations are recorded with measurement error, then the
variance in the presence of measurement error for the CSS is
N  2
  1X N
  1X  sy  lYsy
 sy þ U
V y symcs ¼ r2Ysymcs ¼ y sy  lYsy 2 ¼ Y (39)
N i¼1 N i¼1
 
V y symcs ¼ r2Ysycs þ r2Usycs (40)

1X N
 2
 sy
r2Usycs ¼ U (41)
N i¼1

The proposed methods of estimation in the presence of measurement error for the
CSS will change according to change in the sample size. The formulae of the bias and
mean square error will change in similar way as changes occur in the formula
of variance.

3. Simulation study and results

Simulation study has been carried out for the validation of results using R studio for
large sample properties of the estimator. The data matrices on X, Y, u, and v have been
generated using multivariate normal distribution for four variables with mean
8 N. SINGH AND G. K. VISHWAKARMA

Table 1. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 30, 20Þ, ðr2U , r2V ¼ 0, 0Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
0 0 –0.9 y dsym 0.0909 526.32 0.5607 526.32
y Rsym 1.6506 29.33 10.4063 28.46
y Psym 0.1152 411.94 0.6267 471.92
y sym 0.4785 100.00 2.9508 100.00
0 0 –0.5 y dsym 0.3595 133.33 2.2281 133.33
y Rsym 1.2658 37.60 8.1050 36.80
y Psym 0.4914 95.59 2.8694 103.25
y sym 0.4794 100.00 2.9707 100.00
0 0 –0.1 y dsym 0.4734 101.01 2.9533 101.01
y Rsym 0.9601 49.06 6.0552 49.05
y Psym 0.8101 58.04 5.0133 59.17
y sym 0.4782 100.00 2.9831 100.00
0 0 –0.1 y dsym 0.4760 101.01 2.9300 101.01
y Rsym 0.8088 58.08 4.9412 59.36
y Psym 0.9580 49.10 5.9694 49.19
y sym 0.4809 100.00 2.9596 100.00
0 0 –0.5 y dsym 0.3598 133.33 2.1916 133.33
y Rsym 0.4924 95.52 2.8181 102.70
y Psym 1.2717 37.55 7.9570 36.58
y sym 0.4798 100.00 2.9221 100.00
0 0 –0.9 y dsym 0.0900 526.32 0.5454 526.32
y Rsym 0.1139 412.48 0.6013 476.77
y Psym 1.6369 29.29 10.0286 28.62
y sym 0.4739 100.00 2.8703 100.00

vectorð lY lX 0 0 Þ and covariance matrix

0 1
r2Y qrX rY 0 0
B qrX rY r2X 0 0 C
B C
@ 0 0 r2U 0 A
0 0 0 r2V
We generate the data for N ¼ 1000 by using different combination of parameters

lX ¼ 16, lY ¼ 18, r2X ¼ ð40, 20Þ, r2Y ¼ ð50, 30Þ

q¼ (–0.9, –0.5, –0.1, 0.9, 0.5, 0.1), r2U ¼ (0, 4), and r2V ¼ (0, 4).

We select the random sample using systematic sampling method, for the two sets of
sample for n ¼ 250 (k ¼ 4) and n ¼ 100 (k ¼ 10).
The mean and variances for the systematic samples have been calculated. The mean
square errors and percentage relative efficiency are obtained. Using simulation, 5000
replication has been taken.
For plotting the graph, the values of variables are ðr2Y , r2X Þ ¼ ð20, 30Þ, q ¼ 0:9, k ¼ 4
lX ¼16, lY ¼18, and r2U ¼ (0, 4) and r2V ¼ (0, 4).
The percent relative efficiency (PRE) of an estimator /ð¼ y dsym , y Rsym , y Psym , y sym Þ
with respect to usual unbiased estimator y sym is calculated by
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 9

Table 2. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 30, 20Þ, ðr2U , r2V ¼ 4, 4Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
4 4 –0.9 y dsym 0.2141 245.25 1.3688 246.90
y Rsym 1.8144 31.14 11.4683 30.21
y Psym 0.2597 206.49 1.5499 220.54
y sym 0.5469 100.00 3.4332 100.00
4 4 –0.5 y dsym 0.4411 119.61 2.7466 121.35
y Rsym 1.4053 38.33 8.8619 37.87
y Psym 0.6308 82.23 3.7145 89.05
y sym 0.5356 100.00 3.3498 100.00
4 4 –0.1 y dsym 0.5318 100.64 3.3815 100.71
y Rsym 1.0957 48.21 6.8731 49.21
y Psym 0.9466 55.50 5.8382 57.80
y sym 0.5355 100.00 3.4061 100.00
4 4 0.1 y dsym 0.5264 100.65 3.2847 100.71
y Rsym 0.9408 55.26 5.7992 56.61
y Psym 1.0889 48.00 6.8231 48.25
y sym 0.5300 100.00 3.3084 100.00
4 4 0.5 y dsym 0.4512 119.66 2.7650 121.28
y Rsym 0.6434 82.28 3.7461 88.76
y Psym 1.4309 38.47 8.8873 37.94
y sym 0.5478 100.00 3.3689 100.00
4 4 0.9 y dsym 0.2135 245.10 1.3454 246.81
y Rsym 0.2597 205.98 1.5098 221.54
y Psym 1.8069 31.09 11.2391 30.20
y sym 0.5439 100.00 3.3609 100.00

 
V y sym
PREð/, y sym Þ ¼  100 (42)
MSEð/Þ
Using simulation, graph and eight numbers of tables have been formed by using four
levels of measurement errors. We also obtained the impact of two levels of variances of
study and auxiliary variable on the MSE in the estimation procedure of mean.
From the graph which is shown below, we can infer that with the increase in the
value of measurement error, the values of mean square errors are increasing for all the
estimators. The mean square error of estimators is highest for the highest degree of
commingle of measurement error. Graph is also showing the efficiency of difference
estimator over other estimators. Graph has been plotted for positive correlation coeffi-
cient thus we can conclude that the properties of the estimators are not changing in the
presence of measurement error as ratio estimator is more efficient than prod-
uct estimator.
The results of the simulation study are extensively shown in tables. Table 1 shows the
MSE and PRE of the proposed estimators for all the proposed estimators for the differ-
ent levels of the correlation coefficient ðq ¼ 0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ between
study and auxiliary variable. From Table 1, we infer that difference estimator has higher
efficiency than ratio, product, and mean estimators for the different levels of correlation
coefficient under no measurement error. Table 2 shows the contamination of measure-
ment error (r2U ¼ 4,r2V ¼ 4) in MSE and PRE for all estimators. From Table 2, we
find that MSE of all the estimators is higher in the presence of measurement error
and is highest when the error is observed in both study as well as auxiliary variables.
10 N. SINGH AND G. K. VISHWAKARMA

Table 3. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 30, 20Þ, ðr2U , r2V ¼ 4, 0Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
4 0 –0.9 y dsym 0.1549 339.80 0.9526 347.41
y Rsym 1.7065 33.97 10.7264 31.86
y Psym 0.1802 292.91 1.0181 325.85
y sym 0.5424 100.00 3.3456 100.00
4 0 –0.5 y dsym 0.4170 126.20 2.6007 127.61
y Rsym 1.3100 41.80 8.3943 40.10
y Psym 0.5506 95.71 3.2318 102.76
y sym 0.5346 100.00 3.3369 100.00
4 0 –0.1 y dsym 0.5480 100.84 3.3158 100.87
y Rsym 1.0240 53.96 6.2882 53.35
y Psym 0.8729 62.93 5.2759 63.43
y sym 0.5529 100.00 3.3453 100.00
4 0 0.1 y dsym 0.5392 100.83 3.2836 100.87
y Rsym 0.8710 62.04 5.3157 62.30
y Psym 1.0201 53.25 6.3381 52.36
y sym 0.5440 100.00 3.3128 100.00
4 0 0.5 y dsym 0.4231 126.32 2.6287 127.59
y Rsym 0.5551 96.60 3.2768 102.24
y Psym 1.3289 41.84 8.5503 39.82
y sym 0.5429 100.00 3.3723 100.00
4 0 0.9 y dsym 0.1554 338.22 0.9580 343.58
y Rsym 0.1804 291.99 1.0122 325.38
y Psym 1.6905 34.11 10.5726 32.11
y sym 0.5386 100.00 3.3351 100.00

Table 4. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 30, 20Þ, ðr2U , r2V ¼ 0, 4Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
0 4 –0.9 y dsym 0.1506 311.51 0.9527 307.51
y Rsym 1.7416 27.00 10.7147 27.35
y Psym 0.1969 241.25 1.1215 263.37
y sym 0.4825 100.00 2.9481 100.00
0 4 –0.5 y dsym 0.3821 124.39 2.3209 125.72
y Rsym 1.3472 34.54 8.4743 34.29
y Psym 0.5745 79.47 3.3109 86.94
y sym 0.4768 100.00 2.9193 100.00
0 4 –0.1 y dsym 0.4742 100.78 2.8995 100.82
y Rsym 1.0438 44.29 6.4199 45.16
y Psym 0.8936 51.55 5.4033 53.57
y sym 0.4779 100.00 2.9231 100.00
0 4 0.1 y dsym 0.4750 100.78 2.8970 100.83
y Rsym 0.8858 51.68 5.3981 53.62
y Psym 1.0346 44.40 6.4255 45.12
y sym 0.4787 100.00 2.9209 100.00
0 4 0.5 y dsym 0.3749 124.30 2.3316 125.74
y Rsym 0.5650 78.76 3.3139 86.82
y Psym 1.3268 34.24 8.5098 34.11
y sym 0.4676 100.00 2.9371 100.00
0 4 0.9 y dsym 0.1487 309.32 0.9669 308.47
y Rsym 0.1950 239.06 1.1360 264.43
y Psym 1.7135 26.71 10.9902 27.13
y sym 0.4728 100.00 3.0185 100.00
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 11

Table 5. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 50, 40Þ, ðr2U , r2V ¼ 0, 0Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
0 0 –0.9 y dsym 0.1526 526.32 0.9300 526.32
y Rsym 3.0466 26.68 18.5360 26.50
y Psym 0.2072 390.95 1.1068 450.73
y sym 0.8031 100.00 4.8945 100.00
0 0 –0.5 y dsym 0.6019 133.33 3.6373 133.33
y Rsym 2.2935 35.01 14.5312 33.64
y Psym 0.8862 90.50 5.1093 95.55
y sym 0.8025 100.00 4.8498 100.00
0 0 –0.1 y dsym 0.8021 101.01 4.8763 101.01
y Rsym 1.7638 45.67 10.9054 45.21
y Psym 1.4864 54.20 9.0182 54.68
y sym 0.8102 100.00 4.9256 100.00
0 0 0.1 y dsym 0.7904 101.01 4.8505 101.01
y Rsym 1.4679 54.71 8.9441 54.89
y Psym 1.7425 46.08 10.8066 45.40
y sym 0.7984 100.00 4.8995 100.00
0 0 0.5 y dsym 0.5920 133.33 3.6950 133.33
y Rsym 0.8988 88.78 5.2067 94.64
y Psym 2.3025 34.72 14.7890 33.32
y sym 0.7894 100.00 4.9267 100.00
0 0 0.9 y dsym 0.1472 526.32 0.9507 526.32
y Rsym 0.2009 387.29 1.0957 457.96
y Psym 2.9144 26.79 18.9414 26.48
y sym 0.7745 100.00 5.0038 100.00

Table 6. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 50, 40Þ, ðr2U , r2V ¼ 4, 4Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
4 4 –0.9 y dsym 0.2697 295.50 1.6541 306.08
y Rsym 3.0993 28.09 19.1890 27.36
y Psym 0.3515 232.80 1.9683 261.47
y sym 0.8393 100.00 5.1811 100.00
4 4 –0.5 y dsym 0.6852 123.32 4.2263 125.66
y Rsym 2.4339 36.11 15.6814 34.51
y Psym 1.0296 82.89 6.1171 87.61
y sym 0.8576 100.00 5.3347 100.00
4 4 –0.1 y dsym 0.8449 100.76 5.1768 100.82
y Rsym 1.8516 45.96 11.6698 44.94
y Psym 1.5834 53.54 9.8178 53.37
y sym 0.8516 100.00 5.2197 100.00
4 4 0.1 y dsym 0.8778 100.76 5.1969 100.82
y Rsym 1.6224 54.02 9.7475 54.01
y Psym 1.8974 46.38 11.5884 45.48
y sym 0.8849 100.00 5.2401 100.00
4 4 0.5 y dsym 0.7059 123.73 4.1575 125.64
y Rsym 1.0608 82.94 5.9926 87.23
y Psym 2.5115 35.81 15.3753 34.23
y sym 0.8857 100.00 5.2466 100.00
4 4 0.9 y dsym 0.2746 297.65 1.7036 307.71
y Rsym 0.3506 238.35 1.9912 264.64
y Psym 3.1680 27.85 19.7950 27.24
y sym 0.8618 100.00 5.3603 100.00
12 N. SINGH AND G. K. VISHWAKARMA

Table 7. MSE and PRE of various estimator of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 50, 40Þ, ðr2U , r2V ¼ 4, 0Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
4 0 –0.9 y dsym 0.2129 373.81 1.3611 392.37
y Rsym 3.0382 29.63 19.5498 28.45
y Psym 0.2712 302.95 1.5483 350.67
y sym 0.8490 100.00 5.4510 100.00
4 0 –0.5 y dsym 0.6556 128.32 4.1526 129.66
y Rsym 2.3718 37.21 15.1878 35.99
y Psym 0.9555 90.49 5.6208 96.88
y sym 0.8530 100.00 5.4068 100.00
4 0 –0.1 y dsym 0.8389 100.44 5.2289 100.92
y Rsym 1.8144 48.59 11.2343 47.34
y Psym 1.5393 57.10 9.3650 56.75
y sym 0.8269 100.00 5.2777 100.00
4 0 0.1 y dsym 0.8571 100.88 5.3197 100.92
y Rsym 1.5313 57.34 9.3904 57.53
y Psym 1.8033 48.85 11.2727 47.98
y sym 0.8651 100.00 5.3695 100.00
4 0 0.5 y dsym 0.6588 128.21 4.0573 129.53
y Rsym 0.9474 91.13 5.5256 96.12
y Psym 2.3569 37.45 14.9174 35.73
y sym 0.8569 100.00 5.2801 100.00
4 0 0.9 y dsym 0.2153 380.75 1.3312 391.23
y Rsym 0.2678 312.84 1.4839 353.05
y Psym 3.0759 29.66 19.2053 27.90
y sym 0.8609 100.00 5.3232 100.00

Table 8. MSE and PRE of various estimators of y sym for q ¼ ð0:9,  0:5,  0:1, 0:1, 0:5, 0:9Þ when
ðr2Y , r2X ¼ 50, 40Þ, ðr2U , r2V ¼ 0, 4Þ, and ðk ¼ 4, 10Þ:
k¼4 k ¼ 10
r2U r2V q Estimator MSE PRE MSE PRE
0 4 –0.9 y dsym 0.2091 365.25 1.2633 374.69
y Rsym 3.1113 25.42 19.0479 25.40
y Psym 0.2894 268.58 1.6083 300.74
y sym 0.7989 100.00 4.8144 100.00
0 4 –0.5 y dsym 0.6149 127.52 3.7565 128.70
y Rsym 2.3759 32.96 14.9476 32.47
y Psym 0.9698 79.81 5.6185 86.19
y sym 0.7866 100.00 4.8371 100.00
0 4 –0.1 y dsym 0.7995 100.86 4.8819 100.90
y Rsym 1.8350 43.49 11.4335 43.05
y Psym 1.5599 51.07 9.5583 51.48
y sym 0.8065 100.00 4.9258 100.00
0 4 –0.1 y dsym 0.7869 100.86 4.8558 100.90
y Rsym 1.5426 50.72 9.4814 51.54
y Psym 1.8136 43.21 11.3524 43.07
y sym 0.7936 100.00 4.8996 100.00
0 4 –0.5 y dsym 0.6203 127.39 3.8128 128.79
y Rsym 0.9668 81.08 5.7078 85.83
y Psym 2.3723 33.55 15.2434 32.24
y sym 0.7921 100.00 4.9164 100.00
0 4 –0.9 y dsym 0.2087 364.25 1.3126 372.93
y Rsym 0.2826 270.75 1.5973 305.94
y Psym 3.0454 25.62 19.3430 25.58
y sym 0.7917 100.00 4.9633 100.00
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 13

Figure 1. Figures showing the MSE of different estimators in the presence of measurement error.

Figure 1 is showing the impact of measurement error on the mean square error of the
different estimators. From Tables 3 and 4, we find that MSE of all proposed estimators
is increasing when measurement error is observed either in study or in auxiliary vari-
able. In case while studying measurement error, one thing is considerable that PRE may
not be a suitable parameter to study the impact of error. Analyzing Tables 1 and 5, we
find that an increase in variance of study variable as well as auxiliary variable is a large
factor to increase the MSE of estimators. From Tables 5–8, we can find the similar
trend as obtained in Tables 1–4. Study with respect to correlation coefficient, from
Tables 1–8, MSE is always minimum for difference estimators and for ratio and product
14 N. SINGH AND G. K. VISHWAKARMA

Figure 1. (Continued).

estimator is minimum for positive and negative correlation among study and auxiliary
variable, respectively. Form Tables 1–8, we can also infer that MSE is highly influencing
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 15

with the increase in sample size. MSE is less for large sample size n ¼ 250 and PRE is
more efficient as compared to sample size 100.

4. Conclusions
In this manuscript, four well-known forms of estimators, namely ratio, product, differ-
ence, and mean estimator under systematic sampling have been proposed when study
variable and auxiliary variable are observed with measurement error. From our study,
we can conclude that MSE has always been larger when study and auxiliary variables
are commingled with measurement error. Measurement error highly affects the MSE as
well as PRE of the estimator when its value is high, but the properties of estimators do
not change in the presence of measurement error. Also, PRE does not precisely show
the effect of measurement error as PRE is the ratio of two parameters, variance of y
and MSE of estimators. By simulation study of the tables, we can infer that MSE is min-
imum for difference estimator and is more efficient than ratio and product estimator
for all correlation coefficient. Measurement error is inherent in data observation during
systematic sampling, and inference based on these data may be misleading. The infer-
ence based on data when the amount of measurement error is high may ruin the pur-
pose of the study. In the presence of measurement error, the difference estimator is
most efficient among all estimators; thus, it can be used in the estimation of parameters
of mean when the data under study are observed with measurement error.

Acknowledgements
Authors are deeply indebted to the editor-in-chief Prof. N. Balakrishnan and learned referees for
their valuable suggestions leading to improving the quality of contents and presentation of the
original manuscript.

ORCID
Neha Singh http://orcid.org/0000-0001-6719-6771
Gajendra K. Vishwakarma http://orcid.org/0000-0002-2804-4334

References
Allen, J., H. P. Singh, and F. Smarandache. 2003. A family of estimators of population mean
using multi auxiliary information in presence of measurement error. International Journal of
Social Economics 30 (7):837–48. doi: 10.1108/03068290310478775.
Biemer, P., and S. L. Stokes. 1991. Approaches to the modeling of measurement errors. In
Measurement errors in surveys, eds. P.P. Biemer, R.M. Groves, L.E. Lyberg, N.A. Mathiowetz,
and S. Sudman, 487–517. New York: Wiley.
Carroll, R. J., D. Ruppert, L. A. Stefansky, and C. M. Crainiceanu. 2006. Measurement error in
nonlinear models: A modern perspective. London: Chapman & Hall.
Cheng, C.-L., and J. W. Van Ness. 1994. On estimating linear relationships when both variables
are subject to error. Journal of the Royal Statistical Society: Series B (Methodological) 56 (1):
167–83. doi: 10.1111/j.2517-6161.1994.tb01969.x.
16 N. SINGH AND G. K. VISHWAKARMA

Cochran, W. G. 1946. Relative accuracy of systematic and stratified random samples for a certain
class of populations. The Annals of Mathematical Statistics 17 (2):164–77. doi: 10.1214/aoms/
1177730978.
Cochran, W. G. 1968. Errors of measurement in statistics. Technometrics 10 (4):637–66. doi: 10.
2307/1267450.
Fuller, W. A. 1987. Measurement error models. New York: John Wiley & Sons.
Gautschi, W. 1957. Some remarks on systematic sampling. The Annals of Mathematical Statistics
28 (2):385–94. doi: 10.1214/aoms/1177706966.
Goodman, L. A., and H. O. Hartley. 1958. The precision of unbiased ratio-type estimators.
Journal of the American Statistical Association 53 (282):491–508. doi: 10.1080/01621459.1958.
10501454.
Gregoire, T. G. 1998. Design-based and model-based inference in survey sampling: Appreciating
the difference. Canadian Journal of Forest Research 28 (10):1429–47. doi: 10.1139/x98-166.
Gregoire, T. G., and C. Salas. 2009. Ratio estimation with measurement error in the auxiliary
variate. Biometrics 65 (2):590–98. doi: 10.1111/j.1541-0420.2008.01110.x.
Gregoire, T. G., and T. Valentine Harry. 2008. Sampling Strategies for Natural Resources and the
Environment. Boca Raton, FL; London: Chapman & Hall/CRC.
Griffth, A. L. 1945. The efficiency of enumerations. In Indian forest leaflets, 83–93. Dehradun:
Forest-Research Institute.
Hartley, H. O., and A. Ross. 1954. Unbiased ratio estimators. Nature 174 (4423):270–71. doi: 10.
1038/174270a0.
Hasel, H. A. 1942. Estimation of volume of timber stands by strip sampling. The Annals of
Mathematical Statistics 13 (2):179–206. doi: 10.1214/aoms/1177731605.
Lahiri, D. B. 1954. On the question of bias of systematic sampling. Proceedings of World
Population Conference 6:349–62.
Madow, W. G. 1949. On the theory of systematic sampling, II. The Annals of Mathematical
Statistics 20 (3):333–54. doi: 10.1214/aoms/1177729988.
Madow, W. G. 1953. On the theory of systematic sampling, III. Comparison of centered and ran-
dom start systematic sampling. The Annals of Mathematical Statistics 24 (1):101–106. doi: 10.
1214/aoms/1177729087.
Maneesha, S, and R. K. Singh. 2002. Role of regression estimator involving measurement errors.
Brazilian Journal of Probability and Statistics 16:39–46.
Murthy, M. N. 1964. Product method of estimation. Sankhya. The Indian Journal of Statistics,
Series A 26 (1):69–74.
Murthy, M. N. 1967. Sampling theory and methods. Calcutta, India: Statistical Publishing Society.
Sahoo, L. N., R. K. Sahoo, and S. C. Senapati. 2006. An empirical study on the accuracy of ratio
and regression estimators in the presence of measurement error. Monte Carlo Methods and
Applications 12 (5):495–501. doi: 10.1515/156939606779329026.
Salas, C., and T. G. Gregoire. 2010. Statistical analysis of ratio estimators and their estimators of
variances when the auxiliary variate is measured with error. European Journal of Forest
Research 129 (5):847–61. doi: 10.1007/s10342-009-0277-3.
S€arndal, C. E., B. Swensson, and J. Wretman. 1992. Model assisted survey sampling. New York:
Springer-Verlag.
Shalabh, S. 1997. Ratio method of estimation in the presence of measurement error. Journal of
the Indian Society of Agricultural Statistics 50 (2):150–55.
Shalabh, S., and J. R. Tsai. 2017. Ratio and product methods of estimation of population mean in
the presence of correlated measurement error. Communications in Statistics-Simulation and
Computation 46 (7):5566–93.
Shukla, N. D. 1971. Systematic sampling and product method of estimation. Paper presented at
the all India Seminar on Demography and Statistics, BHU, Varanasi, India.
Singh, N., and G. K. Vishwakarma. 2019. A generalised class of estimator of population mean
with the combined effect of measurement errors and non-response in sample survey. Revista
Investigacion Operacional 40 (2):275–85.
 COMMUNICATIONS IN STATISTICS—THEORY AND METHODS 17

Singh, N., G. K. Vishwakarma, and J. M. Kim. 2019. Computing the effect of measurement error
on efficient variant of ratio and product estimator using auxiliary variable. Communications in
Statistics - Simulation and Computation. Advance online publication. doi: 10.1080/03610918.
2019.1656742.
Swain, A. K. P. S. 1964. The use of systematic sampling in ratio estimate. Journal of Indian
Statistical Association 2 (213):160–64.
Williams, R. M. 1956. The variance of the mean of systematic samples. Biometrika 43 (1–2):
137–48. doi: 10.1093/biomet/43.1-2.137.

View publication stats