7.2
CiteScore
3.7
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
ABUNDANCE ESTIMATION IN AN ARID ENVIRONMENT
Case Study
Correspondence
Corrigendum
Editorial
Full Length Article
Invited review
Letter to the Editor
Original Article
Retraction notice
REVIEW
Review Article
SHORT COMMUNICATION
Short review
7.2
CiteScore
3.7
Impact Factor
Generic selectors
Exact matches only
Search in title
Search in content
Post Type Selectors
Search in posts
Search in pages
Filter by Categories
ABUNDANCE ESTIMATION IN AN ARID ENVIRONMENT
Case Study
Correspondence
Corrigendum
Editorial
Full Length Article
Invited review
Letter to the Editor
Original Article
Retraction notice
REVIEW
Review Article
SHORT COMMUNICATION
Short review
View/Download PDF

Translate this page into:

04 2022
:35;
102536
doi:
10.1016/j.jksus.2022.102536

Exponential ratio estimator of the median: An alternative to the regression estimator of the median under stratified sampling

Department of Mathematics, (SSBSR), Sharda University, Greater Noida, UP 201306, India
Department of Basic Sciences, College of Science and Theoretical Studies, Saudi Electronic University, Jeddah Campus, Riyadh (11673), Saudi Arabia
Department of Statistics, Faculty of Science, King Abdulaziz University, Jeddah 21551, Saudi Arabia
Department of Statistics, Faculty of Science, King Abdulaziz University, Jeddah 21551, Saudi Arabia
Division of Agricultural Economics & Statistics, FoA, Wadura, SKUAST-Kashmir, India

⁎Corresponding author. subzarstat@gmail.com (Mir Subzar)

Disclaimer:
This article was originally published by Elsevier and was migrated to Scientific Scholar after the change of Publisher.

Peer review under responsibility of King Saud University.

Abstract

This article develops statistical inference about the population median under the stratified sampling method. An exponential class of ratio estimators of the median was suggested using the combination of scalars and known supplementary information on the population median. Mean square error and bias expressions were derived theoretically and also the AOE (asymptotic optimum estimator) conditions were obtained with its mean square error and bias expressions. From both the empirical evidence and analytical approach evaluations of the AOE with other obtainable members of the suggested class of estimators show that the AOE performs better than its competitors in the literature and is also the alternative to the regression estimator of the median under a stratified random sampling scheme.

Keywords

Population Median
AOE
Bias
MSE
Efficiency
Stratified Sampling
62D05
PubMed
1

1 Introduction

Auxiliary information is commonly used in survey sampling in order to increase precision while estimating the population parameters. So, whenever this information is available and every researcher wants to utilize it in order to get more precise results. However various authors have put their sincere efforts to do the same for details see Kadilar and Cingi (2003) who amended the estimators in Upadhyaya and Singh (1999) to the sampling design stratified random sampling. Also, Singh and Vishwakarma (2008), Sharma and Singh (2015), Verma et al. (2015) suggested a new family of estimators in Stratified random sampling and Subzar et al (2018) have anticipated different estimators using Non-Conventional measures of dispersion for estimating finite population mean in stratified random sampling. Recently Lone et al (2021) have also proposed a general class of ratio estimator for estimating the finite population variance. So this type of sampling procedure is used when the population under study is heterogeneous, it is then usually used to make substrata, which are homogeneous within and heterogeneous between, with a view to producing smaller bound on the error of estimation for a fixed cost of the survey. Having been motivated by the above research works, the present study focuses on how we will get precise results from a heterogonous population and even that is skewed. However, we will get by using the linear regression estimator, but one thing is clear while using OLS (ordinary least square) our results are not precise because it is sensitive to extreme values. But our present study focuses on obtaining reliable results from the data having skewed distribution.

Consider a finite population with auxiliary variate U t and study variable V t which is divided into M strata containing D t units in each t th stratum such that t = 1 M D t = D . Let u ti and v ti represent the sample medians corresponding to the population medians Q ut and with Q vt correlation coefficient between Q ̂ ut and Q ̂ vt as ρ ct . Let f ut u t and f vt v t be the marginal densities; f ut Q ut and f vt Q vt the probability density functions of the variables. Let Q ̂ ust = t = 1 M W t Q ̂ ut and Q ̂ vst = t = 1 M W t Q ̂ vt be the respective weighted sample medians corresponding to the population medians Q u = Q ust = t = 1 M W t Q ut and Q v = Q vst = t = 1 M W t Q vt where W t = D t / D , f t = d t / d where d t is the sample size from stratum t = 1 , 2 , , M and d is the total sample size. Also, let

(1.1)
ϕ 0 t = Q ^ vt - Q vt / Q vt , ϕ 1 t = Q ^ ut - Q ut / Q ut

Such that

(1.2)
E ϕ 0 t = E ϕ 1 t = 0 , E ϕ 0 t 2 = ζ t C Qvt 2 , E ϕ 1 t 2 = ζ t C Qut 2 , E ϕ 0 t ϕ 1 t = ζ t ρ ct C Qvt C Qut C Qvt = Q vt f vt Q vt - 1 , C Qut = Q ut f ut Q ut - 1 , ζ t = 1 - f t / d t , f t = d t / D t
(1.3)
K t = ρ ct C Qvt / C Qut

2

2 Review of some existing estimators for population median in stratified sampling scheme

For the above described population different authors have proposed different estimators for population median in stratified sampling in different years whose literature is mentioned in this section, given as under.

  1. The usual unbiased sample median estimator (Gross; 1980) given by Q ̂ vst with variance

(2.1)
V a r Q ̂ vst = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2
(2.2)
Q ^ RS = t = 1 M W t Q ^ vt Q ut / Q ^ ut w i t h b i a s B Q ^ RS = t = 1 M W t ζ t Q vt C Qut 2 1 - K t a n d M e a n s q u a r e e r r o r M S E Q ^ RS = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 1 - 2 K t
(2.3)
Q ^ PS = t = 1 M W t Q ^ vt Q ^ ut / Q ut w i t h b i a s B Q ^ PS = t = 1 M W t ζ t Q vt C Qut 2 K t a n d m e a n s q u a r e e r r o r M S E Q ^ PS = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 1 + 2 K t
Q ̂ exp R S = t = 1 M W t Q ̂ vt exp Q ut - Q ̂ ut / Q ut + Q ̂ ut

With bias B Q ̂ exp P S = t = 1 M W t ζ t Q vt C Qut 2 3 - 4 K t / 8 and Mean square error

(2.4)
M S E Q ̂ exp R S = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 / 4 1 - 4 K t Q ̂ exp P S = t = 1 M W t Q ̂ vt exp Q ̂ ut - Q ut / Q ̂ ut + Q ut

With bias B Q ̂ exp P S = t = 1 M W t ζ t Q vt C Qut 2 4 K t - 1 / 8 and Mean square error

(2.5)
M S E Q ̂ exp P S = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 / 4 1 + 4 K t
(2.6)
Q ^ CRS = t = 1 M W t Q ^ vt Q ut / Q ^ ut 2 w i t h b i a s B Q ^ CRS = t = 1 M W t ζ t Q vt C Qut 2 1 + 2 K t m e a n s q u a r e e r r o r M S E Q ^ CRS = t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + 4 C Qut 2 1 + K t

3

3 The proposed median estimator in stratified random sampling scheme

While taking the motivation from the existing estimators and suggesting the new class of Exponential Ratio estimator of the median using ancillary information and the combination of scalars. The obtainable estimators are the members of the suggested class of estimator and is given as

(3.1)
ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = i = 1 M W t Q ̂ vt ħ ι t - θ ι t Q ̂ ut Q ut ϑ ι t exp τ ι t Q ̂ ut - Q ut Q ̂ ut + Q ut

Where ħ ι t , θ ι t , ϑ ι t , τ ι t are the scalars chosen suitably, such that ħ ι t and θ ι t fulfil the condition

(3.2)
ħ ι t = 1 + θ ι t ; - < λ ι t <

In order to derive the estimated expressions of bias and mean square error for the suggested class of estimator we have expressed (3.1) in terms of (1.1), then we obtain ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = i = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t Q ut 1 + ϕ 1 t Q ut ϑ ι t exp τ ι t Q ut 1 + ϕ 1 t - Q ut 2 Q ut 1 + ϕ 1 t / 2 = i = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t 1 + ϕ 1 t ϑ t exp τ ι t Q ut ϕ 1 t 2 Q ut 1 + ϕ 1 t / 2 = i = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t 1 + ϕ 1 t ϑ ι t exp τ ι t ϕ 1 t 2 1 + ϕ 1 t / 2 - 1

Assuming that ϕ 1 t < 1 and expanding 1 + ϕ 1 t ϑ ι t , τ ι t ϕ 1 t 2 1 + ϕ 1 t / 2 - 1 and 1 + ϕ 1 t / 2 - 1 , we have ψ ħ ι t , θ ι t , ϑ ι t , ι t = t = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t 1 + ϑ ι t ϕ 1 t + ϑ ι t ϑ ι t - 1 2 ϕ 1 t 2 + × 1 + τ ι t ϕ 1 t 2 1 + ϕ 1 t / 2 - 1 + τ ι t 2 ϕ 1 t 2 8 1 + ϕ 1 t / 2 - 2 ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = t = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t 1 + ϑ ι t ϕ 1 t + ϑ ι t ϑ ι t - 1 2 ϕ 1 t 2 + × 1 + τ ι t ϕ 1 t 2 1 - ϕ 1 t 2 + ϕ 1 t 2 4 - + τ ι t 2 ϕ 1 t 2 8 1 + ϕ 1 t ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = t = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ t 1 + ϑ ι t ϕ 1 t + ϑ ι t ϑ ι t - 1 2 ϕ 1 t 2 + × 1 + τ ι t ϕ 1 t 2 - τ ι t ϕ 1 t 2 4 + τ ι t 2 ϕ 1 t 2 8 t = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t - λ ι t ϕ 1 t 2 2 ϑ ι t + τ ι t - λ ι t ϕ 1 t 2 8 4 ϑ ι t 2 - 4 ϑ ι t - 2 τ ι t + τ ι t 2 + 4 ϑ ι t τ ι t t = 1 M W t Q vt 1 + ϕ 0 t ħ ι t - θ ι t - θ ι t 2 ϑ ι t + τ ι t ϕ 1 t 2 - θ ι t 2 ϑ ι t + τ ι t 2 ϑ ι t + τ ι t - 2 ϕ 1 t 2 8

Neglecting terms of ϕ it i = 0 o r 1 having power greater than two, we have

(3.3)
t = 1 M W t Q vt ħ ι t - θ ι t - θ ι t 2 ϑ ι t + τ ι t ϕ 1 t 2 - θ ι t 2 ϑ ι t + τ ι t 2 ϑ ι t + τ ι t - 2 ϕ 1 t 2 8 + ħ ι t - θ ι t ϕ 0 t θ ι t 2 ϑ ι t + τ ι t ϕ 0 t ϕ 1 t 2 t = 1 M W t Q vt ħ ι t - θ ι t - θ ι t 2 ϑ ι t + τ ι t 2 ϕ 1 t + 2 ϑ ι t + τ ι t - 2 ϕ 1 t 2 4 + ϕ 0 t ϕ 1 t + ħ ι t - θ ι t ϕ 0 t

Therefore,

(3.4)
ψ ħ ι t , θ ι t , ϑ ι t , τ ι t - Q vt t = 1 M W t Q vt ħ ι t - θ ι t - 1 - θ t 2 ϑ ι t + τ ι t 2 ϕ 1 t + 2 ϑ ι t + τ ι t - 2 ϕ 1 t 2 4 + ϕ 0 t ϕ 1 t + ħ ι t - θ ι t ϕ 0 t

Then, taking expectation of (3.4) and using (1.1), we obtain the bias for the proposed class of ratio median estimators ψ ħ ι t , θ ι t , ϑ ι t , τ ι t to the first degree of approximation as B ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t - Q vt t = 1 M W t ħ ι t - θ ι t - 1 Q vt + 1 - f t d t Q vt - θ ι t 2 ϑ ι t + τ ι t 2 2 ϑ ι t + τ ι t - 2 4 C Qut 2 + ρ ct C Qv C Qu

(3.5)
t = 1 M W t ħ ι t - θ ι t - 1 Q vt + 1 - f t d t Q vt C Qut 2 - θ ι t 2 ϑ ι t + τ ι t 2 2 ϑ ι t + τ ι t - 2 4 + K t

While squaring equation (3.4) on both sides, ignoring terms of ϕ it i = 0 o r 1 having power greater than two, taking expectation, and using (1.1), we obtain the mean square error for the suggested class of ratio median estimator to the first degree of approximation as M S E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t - Q vt 2 E t = 1 M W t Q vt ħ ι t - θ ι t - 1 - θ ι t 2 ϑ ι t + τ ι t 2 + ħ ι t - θ ι t ϕ 1 t 2

(3.6)
t = 1 M W t 2 E Q vt 2 ħ ι t - θ ι t - 1 2 + θ ι t 2 ϕ 1 t 2 2 ϑ ι t + τ ι t 2 4 + ħ ι t - θ ι t 2 ϕ 0 t 2 + 2 ħ ι t - θ ι t - 1 ħ ι t - θ ι t ϕ 0 t - ħ ι t - θ ι t - 1 2 ϑ ι t + τ ι t θ ι t ϕ 1 t - θ ι t 2 ϑ ι t + τ ι t ħ ι t - θ ι t ϕ 0 t ϕ 1 t t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 + θ ι t 2 2 ϑ ι t + τ ι t 2 4 C Mut 2 - θ ι t 2 ϑ ι t + τ ι t ħ ι t - θ ι t ρ ct C Mvt C Mut t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 + 2 ϑ ι t + ι t 4 C Mut 2 θ ι t 2 2 ϑ ι t + τ ι t - 4 K t θ ι t ħ ι t - θ ι t

Now to examine the optimality condition for the suggested class of ratio median estimator, let M S E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t θ ι t = 0

So that θ ι t 2 C Qut 2 2 ϑ ι t + τ ι t = 2 K ι t θ ι t ħ ι t - θ ι t C Qut 2 2 ϑ ι t + τ ι t = 2 K t θ ι t ħ ι t - θ ι t θ ι t

(3.7)
2 ϑ ι t + τ ι t = 2 K t θ ι t S i n c e ħ ι t = 1 + θ ι t

Putting (3.7) in (3.5), we have.

2 K t θ ι t = 2 ϑ ι t + τ ι t and ħ ι t = 1 + θ ι t .

Then B ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t - Q vt t = 1 M W t 1 + θ ι t - θ ι t - 1 Q vt + 1 - f t d t Q vt C Qut 2 - θ ι t 2 K t θ ι t 2 2 K t θ ι t - 2 4 + K t t = 1 M W t 1 - f t d t Q vt C Qut 2 - 2 K t 2 2 K t - 2 θ ι t 4 θ ι t + K t

(3.8)
t = 1 M W t 1 - f t d t Q vt C Qut 2 - K t 4 θ ι t 2 K t - 2 θ ι t + 4 K t θ ι t

Hence substituting (3.7) in (3.6), we obtain M S E ψ ħ ι t , θ ι t , ϑ ι t , τ ι t t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 + 2 K t ħ ι t - θ ι t 4 λ ι t C Mut 2 × 2 K t θ ι t 2 ħ ι t - θ ι t 4 θ ι t - 4 K t θ ι t ħ ι t - θ ι t t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 + K t ħ ι t - θ ι t 2 θ ι t C Mut 2 - 2 K t θ ι t ħ ι t - θ ι t t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 - ħ ι t - θ ι t 2 K t 2 C Mut 2 t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 - K t 2 C Mut 2

(3.9)
t = 1 M W t 2 Q vt 2 ħ ι t - θ ι t - 1 2 + 1 - f t d t Q vt 2 ħ ι t - θ ι t 2 C Mvt 2 1 - ρ ct 2

Thus substituting (3.2) in (3.9) we obtain the asymptotic optimum MSE for the Proposed class of ratio median estimator ψ ħ ι t , θ ι t , ϑ ι t , τ ι t as MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t t = 1 M W t 2 1 - f t d t Q vt 2 C Mvt 2 1 - ρ ct 2

(3.10)
t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2

Thus the mean square error of the suggested class of exponential ratio estimator of the median in stratified random sampling scheme at optimal condition has the same proficiency as the unbiased linear regression estimator as the mean square error expression at the optimal condition of the suggested estimator is the same as the expression of unbiased linear regression estimator.

4

4 Some existing members of the proposed class of estimators

How the existing estimators which are mentioned in this study fit into the suggested class of exponential ratio estimator of the median are presented in this section.

Note. As ħ ι t , θ ι t , ϑ ι t , τ ι t take on definite distinctive values, the suggested class of exponential ratio estimator of the median giving the same expressions of bias and mean square error which the estimators give mentioned in section 2. Thus, the study provides unified treatment towards the properties of the existing members of the suggested class of estimators point out in this study (see for details Table 1).

Table 1 Some Existing members of the suggested class of estimators.
ħ ι t θ ι t ϑ ι t τ ι t Estimators
1 0 ϑ ι t τ ι t Q ̂ vst Usual Median Unbiased estimator
0 −1 −1 0 Q ̂ RS = t = 1 M W t Q ̂ vt Q ut / Q ̂ ut Classical ratio median estimator
0 −1 1 0 Q ̂ PS = t = 1 M W t Q ̂ vt Q ̂ ut / Q ut Product type median estimator
0 −1 0 −1 Q ̂ exp R S = t = 1 M W t Q ̂ vt exp Q ut - Q ̂ ut / Q ut + Q ̂ ut
Bahl and Tuteja exponential ratio median estimator
0 −1 0 1 Q ̂ exp P S = t = 1 M W t Q ̂ vt exp Q ̂ ut - Q ut / Q ̂ ut + Q ut
Bahl and Tuteja exponential product median estimator
0 −1 2 0 Q ̂ CRS = t = 1 M W t Q ̂ vt Q ut / Q ̂ ut 2 Chain ratio median estimator

5

5 Efficiency comparisons among members of the proposed class of estimators

Here, the mean square error of the suggested exponential ratio estimator of the median at optimum condition is equated with the mean square error of some existing median estimators.

  1. Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with sample median estimator Q ̂ vst .

Comparing (2.1) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ vst t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

t = 1 M W t 2 θ t Q vt 2 C Mvt 2 ρ ct 2 > 0 , Which is always true.

  • b)

    Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with classical ratio median estimator Q ̂ RS .

Comparing (2.2) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ RS t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 1 - 2 K t - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

1 - K t 2 > 0 , Which is always true.

  • c)

    Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with classical product median estimator Q ̂ PS .

Comparing (2.3) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ PS t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 1 + 2 K t - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

1 + K t 2 > 0 , Which is always true.

  • d)

    Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with classical exponential ratio median estimator Q ̂ exp R S .

Comparing (2.4) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ exp R S t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 4 1 - 4 K t - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

1 - 2 K t 2 > 0 , Which is always true.

  • e)

    Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with classical exponential product median estimator Q ̂ exp P S .

Comparing (2.5) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ exp P S t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + C Qut 2 4 1 + 4 K t - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

1 + 2 K t 2 > 0 , Which is always true.

  • f)

    Comparison of MS E opt ψ ħ ι t , θ ι t , ϑ ι t , τ ι t with classical chain ratio median estimator Q ̂ CRS .

Comparing (2.6) and (3.9), optimum ψ ħ ι t , θ ι t , ϑ ι t , τ ι t will be more efficient than Q ̂ CRS t = 1 M W t 2 ζ t Q vt 2 C Qvt 2 + 4 C Qut 2 1 + K t - t = 1 M W t 2 ζ t Q vt 2 C Mvt 2 1 - ρ ct 2 > 0

K t 2 + 4 K t + 4 > 0 , Which is always true.

6

6 Empirical study

In this present study, for verifying the general results and also check the optimality performance i.e., (asymptotic optimum estimator) for the suggested class of exponential ratio estimator of the median ψ ħ ι t , θ ι t , ϑ ι t , τ ι t over the existing members of the suggested class of estimators cited in this study, we the data of the population [Source: PDS(2012)] and the summary of the population is given in Table 2.

Table 2 Summary Statistics of the population.
D = 144 d = 20
D 1 36 D 2 36 D 3 36 D 4 36
d 1 5 d 2 5 d 3 5 d 4 5
W 1 0.25 W 2 0.25 W 3 0.25 W 4 0.25
Q u 1 1480 Q u 2 127,289 Q u 3 54,559 Q u 4 71,615
Q v 1 38 Q v 2 3058 Q v 3 2033 Q v 4 2382
ρ c 1 0.7776 ρ c 2 0.8888 ρ c 3 0.8888 ρ c 4 0.8888
f v 1 Q v 1 0.007056 f v 2 Q v 2 0.00032023 f v 3 Q v 3 0.0004219 f v 4 Q v 4 0.0003012
f u 1 Q u 1 0.0001641 f u 2 Q u 2 0.000007827 f u 3 Q u 3 0.0000141 f u 4 Q u 4 0.00001026
C M v 1 3.729562 C M v 2 1.021175775 C M v 3 1.16587797 C M v 4 1.393809035
C M u 1 4.117462984 C M u 2 1.003722805 C M u 3 1.299913595 C M u 4 1.360970285
ζ 1 0.043056 ζ 2 0.043056 ζ 3 0.043056 ζ 4 0.043056
K 1 0.70434 K 2 0.90425 K 3 0.79715 K 4 0.91025

Considering the above population, setting θ ι 1 = 2 satisfies the condition in (3.2), implies ħ 1 = 3 . Then set ϑ ι 1 = 177387 / 1100000 and τ ι 1 = 21 / 55 so that 2 ϑ ι 1 + τ ι 1 = 2 K 1 / θ ι 1 , K 1 = 0.70434 fulfils the condition in (3.7). Hereafter setting ħ ι 1 , θ ι 1 , ϑ ι 1 , τ ι 1 = 3 , 2 , 177387 / 1100000 , 21 / 55 in (3.1) we find an asymptotic optimum estimator (AOE) for the suggested class of estimators for the population median for strata 1 as

(6.1)
ψ ħ ι 1 , θ ι 1 , ϑ ι 1 , τ ι 1 = 3 , 2 , 177387 1100000 , 21 55 = Q ̂ v 1 3 - 2 Q ̂ u 1 Q u 1 177387 1100000 exp 21 Q ̂ u 1 - Q u 1 55 Q ̂ u 1 + Q u 1

and the first degree approximation of bias and mean square error given respectively in Table 3,

(6.2)
B ψ 3 , 2 , 177387 / 1100000 , 21 / 55 1 - f 1 d 1 W 1 Q v 1 C M u 1 2 - K 1 4 θ ι 1 2 K 1 - 2 θ ι 1 + 4 K 1 θ ι 1
(6.3)
M S E ψ 3 , 2 , 177387 1100000 , 21 55 = 1 - f 1 Q v 1 d 1 C M v 1 2 + 35216590909 200000000000 C M u 1 2 2 10216590909 12500000000 - 8 K 1

For Strata 2 we have Set θ ι 2 = 3 satisfies the condition in (3.2), implies ħ ι 2 = 4 . Then set ϑ ι 2 = 20283 / 2000000 and τ ι 2 = 12 / 30 so that 2 ϑ ι 2 + τ ι 2 = 2 K 2 / θ ι 2 , K 2 = 0.90425 fulfils the condition in (3.7). Hereafter setting ħ ι 2 , θ ι 2 , ϑ ι 2 , τ ι 2 = 4 , 3 , 20283 / 200000 , 12 / 30 in (3.1) find an asymptotic optimum estimator (AOE) for the suggested class of estimators for the population median for strata 2 as

(6.4)
ψ ħ ι 2 , θ ι 2 , ϑ ι 2 , τ ι 2 = 4 , 3 , 20283 200000 , 12 30 = Q ̂ v 2 4 - 3 Q ̂ u 2 Q u 2 20283 200000 exp 12 Q ̂ u 2 - Q u 2 30 Q ̂ u 2 + Q u 2

and the first degree approximation of bias and mean square error given respectively in Table 3,

(6.5)
B ψ 4 , 3 , 20283 / 200000 , 12 / 30 1 - f 2 d 2 W 2 Q v 2 C M u 2 2 - K 2 4 θ ι 2 2 K 2 - 2 θ ι 2 + 4 K 2 θ ι 2
(6.6)
M S E ψ 4 , 3 , 20283 200000 , 12 30 = 1 - f 2 Q v 2 d 2 C M v 2 2 + 60283 400000 C M u 2 2 5 42547 100000 - 12 K 2

For Strata 3, we have set θ ι 3 = 4 fulfils the condition in (3.2), implies ħ ι 3 = 5 . Then set ϑ ι 3 = 3943 / 80000 and τ ι 3 = 15 / 50 so that 2 ϑ ι 3 + τ ι 3 = 2 K 3 / θ ι 3 , K 3 = 0.79715 fulfils the condition in (3.7). Hereafter setting ħ ι 3 , θ ι 3 , ϑ ι 3 , τ ι 3 = 5 , 4 , 3943 / 80000 , 15 / 50 in (3.1) we find an asymptotic optimum estimator (AOE) for the suggested class of estimators for the population median for strata 3 as

(6.7)
ψ ħ ι 3 , θ ι 3 , ϑ ι 3 , τ ι 3 = 5 , 4 , 3943 80000 , 15 50 = Q ̂ v 3 5 - 4 Q ̂ u 3 Q u 3 3943 80000 exp 15 Q ̂ u 3 - Q u 3 50 Q ̂ u 3 + Q u 3

and the first degree approximation of bias and mean square error given respectively in Table 3,

(6.8)
B ψ 5 , 4 , 3943 / 80000 , 15 / 50 1 - f 3 d 3 W 3 Q v 3 C M u 3 2 - K 3 4 θ ι 3 2 K 3 - 2 θ ι 3 + 4 K 3 θ ι 3
(6.9)
M S E ψ 5 , 4 , 3943 80000 , 15 50 = 1 - f 3 Q v 3 d 3 C M v 3 2 + 15943 160000 C M u 3 2 6 943 2500 - 16 K 3

and for strata 4, we have set θ ι 4 = 4 fulfils the condition in (3.2), implies ħ ι 4 = 5 . Then set ϑ ι 4 = 13179 / 304000 and τ ι 4 = 35 / 95 so that 2 ϑ ι 4 + τ ι 4 = 2 K 4 / θ ι 4 , K 4 = 0.91025 fulfils the condition in (3.7). Hereafter setting ħ ι 4 , θ ι 4 , ϑ ι 4 , τ ι 4 = 5 , 4 , 13179 / 304000 , 35 / 95 in (3.1) we find an asymptotic optimum estimator (AOE) for the suggested class of estimators for the population median for strata 4 as

(6.10)
ψ ħ ι 4 , θ ι 4 , ϑ ι 4 , τ ι 4 = 5 , 4 , 13179 304000 , 35 95 = Q ̂ v 4 5 - 4 Q ̂ u 4 Q u 4 13179 304000 exp 35 Q ̂ u 4 - Q u 4 95 Q ̂ u 4 + Q u 4

and the first degree approximation of bias and mean square error given respectively in Table 3,

(6.11)
B ψ 5 , 4 , 13179 / 304000 , 35 / 95 1 - f 4 d 4 W 4 Q v 4 C M u 4 2 - K 4 4 θ ι 4 2 K 4 - 2 θ ι 4 + 4 K 4 θ ι 4
(6.12)
M S E ψ 5 , 4 , 13179 304000 , 35 95 = 1 - f 4 Q v 4 d 4 C M v 4 2 + 11274 100000 C M u 4 2 7 67372368417 312500000000 - 16 K 4
Table 3 Bias and Mean Square error (MSE) values of members of the suggested class of estimators.
Estimators Bias MSE %RE
Q ̂ vst 0.00 284303.40 100.00
Q ̂ RS 16.99 64711.48 439.34
Q ̂ PS 107.58 1083839.42 26.23
Q ̂ exp R S −7.08 102014.42 278.69
Q ̂ exp P S 38.22 611578.41 46.49
Q ̂ CRS 339.72 2463319.41 11.54
ψ ħ ι 1 , θ ι 1 , ϑ ι 1 , τ ι 1 = 3 , 2 , 177387 / 1100000 , 21 / 55 −6.47 341.89 83156.59
ψ ħ ι 2 , θ ι 2 , ϑ ι 2 , τ ι 2 = 4 , 3 , 20283 / 200000 , 12 / 30 −57.98 88186.16 322.39
ψ ħ ι 3 , θ ι 3 , ϑ ι 3 , τ ι 3 = 5 , 4 , 3943 / 80000 , 15 / 50 −40.69 50804.88 559.60
ψ ħ ι 4 , θ ι 4 , ϑ ι 4 , τ ι 4 = 5 , 4 , 13179 / 304000 , 35 / 95 −78.95 99681.49 285.21
ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = 17 , 13 , 1490967 2000000 , 1 4502311 10000000 −184.08 59753.60 475.79

Then, when combining all the strata, from the (6.1) to (6.12) we came to conclude that setting.

θ ι t = θ ι 1 + θ ι 2 + θ ι 3 + θ ι 4 = 2 + 3 + 4 + 4 = 13 , implies ħ ι t = 3 + 4 + 5 + 5 = 17 . Then set ϑ ι t = 177387 1100000 + 20283 200000 + 3945 80000 + 13179 304000 = 1490967 2000000 and τ ι t = 21 55 + 12 30 + 15 50 + 35 95 = 1 4502311 10000000 so that 2 ϑ ι t + τ ι t = 2 K t / θ ι t , K t = 3.39599 fulfils the condition in (3.7). Hereafter setting ħ ι t , θ ι t , ϑ ι t , τ ι t = 17 , 13 , 1490967 2000000 , 1 4502311 10000000 in (3.1) we find an asymptotic optimum estimator (AOE) for the suggested class of estimators for the population median as

(6.13)
ψ ħ ι t , θ ι t , ϑ ι t , τ ι t = 17 , 13 , 1490967 2000000 , 1 4502311 10000000 = t = 1 M W t 2 Q ̂ vt 17 - 13 Q ̂ ut Q ut 1490967 2000000 exp 1 4502311 Q ̂ ut - Q ut 10000000 Q ̂ ut + Q ut

and the first degree approximation of bias and mean square error given respectively in Table 3,

(6.14)
B ψ 15 , 13 , 1490967 2000000 , 1 4502311 100000000 1 - f t d t W t Q vt C Mut 2 - K t 4 θ ι t 2 K t - 2 θ ι t + 4 K t θ ι t
(6.15)
M S E ψ 15 , 13 , 1490967 2000000 , 1 4502311 10000000 = t = 1 M W t 2 1 - f t Q vt d t C Mvt 2 + 294199 4000000 C M u 4 2 4971 10 - 52 K t

7

7 Conclusion

In this study, we have suggested an exponential ratio estimator of the median in stratified random sampling, and its asymptotic conditions were also derived which shows our suggested estimator has the same expression as the unbiased linear regression estimator for mean square error. Thus our suggested exponential ratio estimator of the median using ancillary information and the combination of scalars is the general class of estimator and the existing estimators are actually the members of a proposed class of estimator and performs better at optimal conditions which we have revealed from Table 3. Thus our suggested estimator is preferred over existing estimators and is an alternative to regression estimator for practical applications in case of skewed data.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

References

  1. , , . Ratio and product type exponential estimator. J. Information Optimiz. Sci.. 1991;12(1):159-164.
    [Google Scholar]
  2. , . Median estimation in sample surveys. Proc. Surv. Res. Meth. Secl. Amer. Statist. Assoc. 1980:181-184.
    [Google Scholar]
  3. , , . Ratio estimator in stratified sampling. Biomet. J.. 2003;45:218-225.
    [Google Scholar]
  4. , , . A study on the chain ratio type estimator. Hacettepe J. Math. Statist.. 2003;32:105-108.
    [Google Scholar]
  5. , , . Median estimation in the presence of auxiliary information. J. Royal Statist. Soc., B. 1989;51:261-269.
    [Google Scholar]
  6. , , , . Enhanced estimators of population variance with the use of supplementary information in survey sampling. Math. Probl. Eng.. 2021;1:1-8.
    [Google Scholar]
  7. , . Product method of estimation. Sankhya: Indian J. Statist. Ser. A. 1964;26:69-74.
    [Google Scholar]
  8. PDS (2012): Punjab Development Statistics, Bureau of Statistics, Government of the Punjab, Lahore, Pakistan
  9. , . Application of multivariate Polykays to the theory of unbiased ratio type estimation. J. Am. Stat. Assoc.. 1957;52:511-522.
    [Google Scholar]
  10. , , . Generalized class of estimators for population median using auxiliary information. Hacettepe J. Math. Statist.. 2015;44(2):443-453.
    [Google Scholar]
  11. , , . A family of estimators of population mean using auxiliary information in stratified sampling. Commun. Statist.: Theory Methods. 2008;37(7):1038-1050.
    [Google Scholar]
  12. , , , , . Estimation of finite population mean in stratified random sampling using non-conventional measures of dispersion. J. Reliab. Statist. Studies. 2018;11(1):83-89.
    [Google Scholar]
  13. , , . Use of transformed auxiliary variable in estimating the finite population mean. Biom. J.. 1999;41:627-636.
    [Google Scholar]
  14. , , , . Some families of estimators using two auxiliary variables in stratified random sampling. Rev. Investig. Operacional. 2015;36(2):140-150.
    [Google Scholar]

Appendix A

Supplementary material

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jksus.2022.102536.

Appendix A

Supplementary material

The following are the Supplementary data to this article:

Supplementary Data 1

Show Sections