1. Introduction
In some investigations, cost-effective sampling is a key concern, particularly when the measurement of the characteristic of interest is expensive, uncomfortable, or time-consuming. In order to increase the accuracy achieved per unit of the sample, the ranked set sampling (RSS) approach is an excellent tool for achieving observational economy. Reference [
1] proposed the RSS technique as an alternative to the frequently used simple random sample (SRS) methodology for increasing the efficiency of the sample mean. In RSS, the population is divided into
q sets of
q units each by randomly selecting
q2 units from it. Without taking actual measurements, the
q units in each set are ranked with respect to the study variable. For actual quantification, the lowest-ranked unit from the first set of
q units is chosen. The second smallest ranked unit is measured from the second set of
q units. The process is repeated; say
h times until the greatest rated unit from the last set is determined. To obtain an RSS sample of size
=
qh, the entire process can be replicated a number of times, say
h times. The following matrix notation is considered to express the RSS design.
The mathematical foundation of RSS was initially created in reference [
2]. In statistical inference, the parametric estimate approach employing the RSS sampling strategy is of utmost importance. A large number of studies have recently been conducted on the issue of RSS-based estimate for a variety of parametric models. For example, refeference [
3] used the data from the RSS technique to calculate the population variance. In subsequent years, RSS was employed to estimate the statistical distribution’s involved parameters. Reference [
4] used RSS to estimate the exponential distribution parameters, while reference [
5] investigated the estimator of the Cauchy distribution’s location parameter. Reference [
6] looked at estimating normal and exponential distribution parameters based on RSS. Reference [
7] investigated estimators for the logistic distribution’s location and scale parameters, whereas reference [
8] considered the RSS to get the geometric parameter estimators. Reference. [
9] succeeded in estimating the parameters of the half-logistic distribution, while reference. [
10] investigated estimators of logistic distribution parameters using RSS. Reference [
11] achieved estimating modified Weibull distribution parameters. Reference [
12] used RSS to derive maximum likelihood (ML) and Bayesian estimators for generalized exponential model. Reference [
13] handled with parameter estimators of Pareto distribution using RSS. The parameter estimator of the Rayleigh distribution was regarded in [
14] using different methods of estimations and ranking designs. Within the framework of RSS, reference [
15] examined the approach of the ML for the shape and scale parameters of the generalized Rayleigh distribution. Reference [
16] considered estimation of the new Weibull-Pareto distribution parameters using RSS. Reference [
17] discussed the parameter estimators of Zubair Lomax distribution using RSS. Reference [
18] investigated efficient estimation of the generalized quasi-Lindley distribution parameters under RSS. For recent results and references, see [
19,
20,
21,
22,
23,
24,
25,
26,
27].
The Kumaraswamy distribution (KD) was offered in [
28], which is one of the most significant lifetime distributions with a range of [0,1]. It cannot, however, be utilized for most lifetime data sets that theoretically have limitless support. It is regarded as a viable alternative to beta distribution since they both share the same characteristics such as being uni-modal, decreasing, increasing, or constant. The KD’s probability density function (PDF) with two positive shape parameters is defined by:
Reference [
29] proposed the inverted KD distribution (IKD) with the goal of providing a new flexible lifetime distribution for analyzing real data sets in the best situation. The following are indeed the principles of the IKD: It’s the distribution of the random variable Z = 1/
X − 1, where
X is a random variable following the KD in Equation (1). The PDF and cumulative distribution function (CDF) of the IKD are, respectively, characterized by:
Figure 1 presents some possible PDF plots of the IKD for some selected distribution parameters. It is clear that the distribution is skewed to the right.
The CDF (3) comprises several well-known distributions, including (i) the Lomax distribution for
(ii) the beta type II (inverted beta) distribution for
(iii) log-logistic distribution for
(iv) the inverse Weibull distribution as
and the generalized exponential distribution when
According to reference [
29], the IKD has a long right tail, and when compared to other distributions, the IKD gives optimistic forecasts of uncommon events happening in the right tail.
To our knowledge, there have been no published works that have utilized the RSS to estimate IKD parameters. In this study, we assumed that the ranking is perfect and we focus on several classical estimations of the IKD parameters using RSS and SRS. The considered methods are the ML, maximum product of spacing (MPS), least squares (LS), weighted least squares (WLS), Cramer–von Mises (CvM), and Anderson–Darling (AD). A simulation study compares the suggested RSS based estimators to the basic SRS based on some criteria measures.
The following is the paper’s configuration: we provide the ML estimators (MLEs) and MPS estimators (MPSEs) of the IKD parameters in
Section 2.
Section 3 obtains the IKD parameter estimators using LS and WLS approaches. Using the AD and CvM methodologies, we obtain the IKD parameter estimators in
Section 4.
Section 5 and
Section 6 describe, respectively, simulation research as well as its application to real-world situations, and comparing RSS estimators to SRS equivalents. The paper comes to a close with main findings in
Section 7.
3. LS and WLS Methods
The LS and WLS methods for estimating unknown parameters are well established in reference [
31]. Here, the LS estimators (LSEs) and WLS estimators (WLSEs) of
and
are examined using SRS and RSS design. We’ll go over the RSS framework technique first, and then obtain these estimators using SRS.
Let the ordered items
constitute a RSS of size
, from the IKD has the CDF given by Equation (3). The LSEs and WLSEs of
and
are derived by minimizing the following functions with respect to population parameters:
where
As an alternative to Equation (16), the following nonlinear equations can be used to yield the LSEs of
and
denoted by
and
respectively, as below:
Similarly, as an alternative to Equation (17), the following nonlinear equations can be used to obtain the WLSEs of
and
respectively, denoted by
and
as below:
where
and
are defined in Equations (11) and (12).
Additionally, the LSEs and WLEs, of ordered SRS
of sizes
nº, for
and
denoted by
and
are obtained, by solving numerically the following nonlinear equations
where,
and
have the same expressions as Equations (11) and (12) by replacing
with
z(i), where
z(i) is the ordered SRS of size
with CDF of IKD presented in Equation (3).
4. AD and CvM Methods
We employ two estimating methods that are based on the minimization of two well-known goodness-of-fit statistics. The two techniques are the CvM and AD, both of which are based on the difference between the CDF and empirical distribution function estimations. The CvM estimators (CvMEs) and AD estimators (ADEs) of the IKD are explored using SRS and RSS designs.
Suppose that
are OS items constitute a RSS of size
, from IKD. The CvMEs and ADEs of
and
are derived by minimizing the following functions with respect to population parameters
As an alternative to Equation (26), the following nonlinear equations can be used to get the CvMEs of
and
denoted by
and
as below:
where
and
are defined in Equations (11) and (12). Similarly, as an alternative to Equation (27), the following nonlinear equations can be used to obtain the ADEs of
and
denoted by
and
as below:
where
and
are defined in Equations (11) and (12).
Furthermore, the ADs of
and
denoted by
and
based on ordered SRS
of sizes
nº are obtained by solving numerically the following equations
where
and
are defined in Equations (11) and (12) with ordered samples
Similarly, the CvMEs of
and
denoted by
and
based on ordered SRS
of sizes
nº, are obtained by solving the following non-linear equations:
where
and
are defined in Equations (11) and (12) with ordered samples
6. Real Data Application
In order to demonstrate the utility of the suggested estimators, a real data set was taken into consideration and carefully thoroughly explained in this part. The information is based on the times between 64 consecutive eruptions of Kiama Blowhole in 1998. A popular tourist destination is the Kiama Blowhole, which is around 120 km south of Sydney. The water is forced into a cliff’s gap by the ocean’s surging. The water then bursts forth via an opening, typically dousing everyone close. Since 12 July 1998, there have been 1340 h’ worth of eruption data collected. These are the data set’s details:
83 | 51 | 87 | 60 | 28 | 95 | 8 | 27 | 15 | 10 |
18 | 16 | 29 | 54 | 91 | 8 | 17 | 55 | 10 | 35 |
47 | 77 | 36 | 17 | 21 | 36 | 18 | 40 | 10 | 7 |
34 | 27 | 28 | 56 | 8 | 25 | 68 | 146 | 89 | 18 |
73 | 69 | 9 | 37 | 10 | 82 | 29 | 8 | 60 | 61 |
61 | 18 | 169 | 25 | 8 | 26 | 11 | 83 | 11 | 42 |
17 | 14 | 9 | 12 | | | | | | |
Using the Kolmogorov-Smirnov (K–S) test, the data set is checked for such a fitted model and the estimates are obtained using the ML method. With a
p-value (PV) of 0.591, the K–S distance is 0.409. As a result, it is obvious that the IKD is a suitable model for fitting these data. Data’s estimated PDF and CDF are displayed in
Figure 10. According to this graph, the IKD seems to be a suitable model for fitting the data.
Based on the aforementioned theoretical results, actual data sets are checked using the RSS and SRS sampling techniques. The SRS and RSS estimators from the IKD are shown in
Table 13 and
Table 14 for various set sizes under five and ten cycles utilizing different estimating techniques. The R-package is used to generate the RSS and SRS observations.
We considered the K–S test for quantifying the distance between the empirical distribution function of the real data and the CDF using the estimators’ parameters in each design, based on the choices of q and h, in order to demonstrate the superiority of RSS over the SRS using various estimation methods considered in this study. Be aware that we have substituted the K–S for the mean squared in this case. Obviously, estimators that outperform the other competitors have larger PVs (greater than 5%) and lower K–S values. Remember that for data, the MLEs based on SRS are regarded as the true population parameters.
The SRS design is considered for this dataset and for each estimation method, where the estimators are obtained using a sample of size
qh = 40. Using the RSS with sample of sizes
q = 4 and
h = 10 is considered for calculating the estimators. We compare the SRS and RSS designs in terms of the K–S distance value and PVs results given in
Table 15, and the corresponding fittings are displayed in
Figure 11 and
Figure 12. Due to the smallest values of the K–S and the corresponding largest PVs, the RSS is more efficient than the SRS based on the same number of measured units for all estimators.