Next Article in Journal
Slip Effects on MHD Squeezing Flow of Jeffrey Nanofluid in Horizontal Channel with Chemical Reaction
Previous Article in Journal
A Shuffle-Based Artificial Bee Colony Algorithm for Solving Integer Programming and Minimax Problems
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

A Two-Parameter Model: Properties and Estimation under Ranked Sampling

1
Deanship of Scientific Research, King Abdulaziz University, Jeddah 21589, Saudi Arabia
2
Deanship of Information Technology, King Abdulaziz University, Jeddah 21589, Saudi Arabia
3
Faculty of Graduate Studies for Statistical Research, Cairo University, Giza 12613, Egypt
4
The Higher Institute of Commercial Sciences, Al Mahalla Al Kubra, Algharbia 31951, Egypt
5
Department of Mathematics, Adamas University, Kolkata 700020, India
6
Department of Mathematics, LMNO, University of Caen, CEDEX, 14032 Caen, France
7
Department of Statistics, The Islamia University of Bahawalpur, Punjab 63100, Pakistan
*
Author to whom correspondence should be addressed.
Mathematics 2021, 9(11), 1214; https://doi.org/10.3390/math9111214
Submission received: 7 April 2021 / Revised: 22 May 2021 / Accepted: 24 May 2021 / Published: 27 May 2021

Abstract

:
This study introduces a flexible model with two parameters by combining the type II half-logistic-G family with the inverted Topp–Leone distribution. The proposed model is referred to as the half logistic inverted Topp–Leone (HLITL) distribution. The associated probability density function can be considered a mixture of the inverted Topp–Leone distributions. The proposed model can be deemed an acceptable model for fitting the right-skewed, reversed J-shaped, and unimodal data. The statistical properties, including the moments, Bonferroni and Lorenz curves, Rényi entropy, and quantile function, are derived. Additionally, the plots of the skewness and kurtosis measures are plotted based on the quantiles. The parameter estimators are implemented using the maximum likelihood method based on two sampling schemes: the simple random sample method and the ranked set sampling method. The proposed method is evaluated by using simulations. The results show that the maximum likelihood estimates of the parameters under ranked set sampling are more accurate than those under simple random sampling. Generally, there is good agreement between the theoretical and empirical results. Two real datasets are used to compare the HLITL model with the following models: alpha power exponential, alpha power Lindley, odd Fréchet inverse exponential, and odd Fréchet inverse Rayleigh models. The comparison results show that the HLITL model represents a better alternative lifetime distribution than the other competitive distributions.

1. Introduction

The Topp–Leone (TL) distribution, also known as the J-shaped distribution, is a significant model that possesses a bathtub-shaped hazard rate. The probability density function (PDF) and cumulative distribution function (CDF) of the TL distribution with a shape parameter, respectively, are given by
f ( y ; θ ) = 2 θ y θ - 1 ( 1 y ) ( 2 y ) θ 1 ,   0 y 1 , θ   >   0 ,
F ( y ; θ ) = y θ ( 2 y ) θ .
Significant properties of the TL distribution were discussed in [1]. The TL distribution was analyzed in many studies [2,3,4]. The inverted (or inverse) distributions are formally regarded as inverse transforms of random variables. The inverted distributions have different densities and hazard rate shapes from the associated noninverted distributions. The usefulness and applicability of inverted distributions have been used in various fields, such as survey sampling, biological sciences, life testing problems, and engineering sciences. In the literature, various inverted distributions and their applications in many areas have been studied [5,6,7,8,9,10].
In [11], the inverted Topp–Leone (ITL) was proposed using the transformation X = 1 Y 1 , where Y denotes a random variable with the CDF and PDF that are, respectively, defined as
G ( x ; θ ) = 1 ( 1 + 2 x ) θ ( 1 + x ) 2 θ   ,   x ,   θ > 0 ,
g ( x ; θ ) = 2 θ x ( 1 + 2 x ) θ 1 ( 1 + x ) 2 θ + 1   ,   x ,   θ > 0 ,
where θ is the shape parameter and θ > 0. In [11], several properties and the estimation of the parameter of the ITL distribution were discussed based on censored samples. Recently, serious efforts have been put into developing new families of continuous probability distributions that extend the classical ones. These extended distributions have more additional parameters as well as enhanced flexibility in modeling a variety of data (see [12,13,14,15,16,17]). In [18], a highly flexible extended family, called Marshall–Olkin-G (MO-G), was proposed, and its PDF is expressed as
f ( x ; δ ) = δ g ( x ) { 1 δ ¯ ( G ¯ ( x ) ) } 2 ,
where δ ¯ = 1 δ and G ¯ ( x ) is the survival function (SF) of a baseline distribution with CDF G ( x ) , and PDF g ( x ) . Several new families that are related to the MO-G family have been considered in recent studies. This study considers the univariate family generated by a half-logistic random variable, as presented in [19]. The CDF of the HL-G family approaches that of the MO-G family. It is specified as follows:
F ( x ; λ ) = 2 [ G ( x ) ] λ { 1 + [ G ( x ) ] λ } 1 , x > 0 , λ > 0 ,
where λ is the shape parameter. For λ = 1 in Equation (6) and δ = 0.5 in Equation (5), the HL-G and MO-G families coincide. The CDF given by Equation (6) extends several flexible distributions as specific cases in their skewness and kurtosis, as presented in [20,21,22,23,24].
In [25], the ranked set sampling (RSS) scheme was provided as another sampling technique for simple random sampling (SRS) in situations where the measurement of the variable of interest is costly or difficult to conduct. The RSS protocol is more suitable than the SRS protocol. To obtain an RSS sample, the following steps are performed:
  • Draw m random samples with a size n from the desired population such that m = n;
  • Without taking any measurements, rank the samples in each dataset according to the criterion determined by the experimenter;
  • Choose a sample for true judgment by involving the smallest ordered unit in the first set and the second-smallest ordered unit in the second set. The operation is continued in this manner until the largest-ranked unit is chosen from the final set;
  • To collect a sample of a size of mk units, the above operations of the “cycle” may be repeated k times. For more studies about RSS, we may refer to [26,27,28,29,30].
This paper proposes a more flexible model for the ITL distribution by inducing one extra shape parameter for enhancing the goodness-of-fit to the real data. The main motivation for introducing the half-logistic inverted Topp–Leone (HLITL) is as follows: (1) obtain a more flexible PDF for right-skewed, unimodal, and reversed J-shaped data; (2) provide decreasing, increasing, and upside-down hazard rate shapes; (3) estimate the HLITL distribution parameters on the basis of SRS and RSS; and (4) provide significant improvement in the data modeling.
The rest of the article is organized as follows. Section 2 defines the PDF, CDF, and hazard rate function (HRF) of the HLITL distribution. Section 3 derives the structural properties of the HLITL distribution. Section 4 presents the maximum likelihood (ML) parameter estimators as well as the simulation study of SRS and RSS. Section 5 describes the real data analysis. Section 6 concludes the paper.

2. HLITL Model Characterizations

The ITL distribution, having the CDF and PDF defined by Equations (3) and (4), respectively, was presented in [11]. This study considers the HLITL distribution as presented in Definition 1.
Definition 1. 
A random variable X is assumed to have the HLITL distribution if its PDF is represented as follows:
f ( x ; λ , θ ) = 4 λ θ x   ( 1 + x ) 2 θ 1 ( 1 + 2 x ) θ 1 [ 1 H ] λ 1 [ 1 + [ 1 H ] λ ] 2 , x > 0 , λ , θ > 0 ,
where H = ( 1 + x ) 2 θ ( 1 + 2 x ) θ ,   λ and θ denote the shape parameters. A random variable with the PDF given by Equation (7) is denoted by ~ ( λ ,   θ ) . For λ = 1 in Equation (7), a new distribution called the Marshall–Olkin ITL distribution with δ = 0.5 is obtained. The PDF plots of the HLITL distribution for some specific values of λ and θ are displayed in Figure 1. The shape of the HLITL distribution is unimodal, positively skewed, and reverse J-shaped.
The associated CDF and HRF, respectively, are given by
F ( x ; λ , θ ) = 2 [ 1 H ] λ { 1 + [ 1 H ] λ } 1 , x , λ , θ > 0 ,
τ ( x ; λ , θ ) = 4 λ θ x   ( 1 + x ) 2 θ 1 ( 1 + 2 x ) θ 1 [ 1 H ] λ 1 [ 1 [ 1 H ] 2 λ ] 1 .
The HRF plots for specific values of parameters with various shapes are illustrated in Figure 2.

3. Basic Properties of the HLITL Distribution

The basic properties of the HLITL distribution, including moments, skewness, kurtosis, Rényi entropy, stochastic ordering, and Bonferroni and Lorenz curves, are derived in the following sections.

3.1. Useful Representation

The properties of the HLITL distribution are deduced through simple elicitation of its PDF. According to [19], the PDF given by Equation (7) can be represented as
f ( x ; λ , θ ) = i = 0 4 λ ( 1 ) i ( i + 1 ) θ x   ( 1 + 2 x ) θ 1 ( 1 + x ) 2 θ + 1   [ 1 ( 1 + 2 x ) θ ( 1 + x ) 2 θ ] λ ( i + 1 ) 1 .
Employing the binomial expansions, the PDF of the HLITL distribution becomes
f ( x ; λ , θ ) = i , d = 0 4 θ λ ( 1 ) i + d ( i + 1 ) ( λ ( i + 1 ) 1 d ) x   ( 1 + x ) 2 ( d + 1 ) θ 1 ( 1 + 2 x ) θ ( d + 1 ) 1   ,
which further leads to
f ( x ; λ , θ ) = i , d = 0 W i , d g ( x ; θ ( d + 1 ) )   ,
where W i , d = 2 λ ( 1 ) i + d ( i + 1 ) d + 1 ( λ ( i + 1 ) 1 d ) .
The PDF g ( x ; θ ( d + 1 ) ) represents the ITL PDF with the shape parameter θ ( d + 1 ) . Thus, the PDF of the HLITL distribution can be expressed as an infinite linear combination of ITL PDFs.

3.2. HLITL Entropy

The Rényi entropy of a random variable X with the HLITL distribution for φ > 0 and φ 1 is expressed as
I e ( x ) = ( 1 φ ) 1 log ( 0 ( f ( x ; λ , θ ) ) φ d x ) .
Substituting Equation (7) into Equation (13) leads to
I e = ( 1 φ ) 1 log ( c = 0 υ c 0 x φ   ( 1 + x ) φ ( 2 θ + 1 ) ( 1 + 2 x ) φ ( θ 1 ) ( 1 H ) ϕ ( λ 1 ) + λ c d x ) ,
where υ c = ( 4 λ θ ) φ ( 1 ) c ( 2 φ + c 1 c ) . By applying the binomial expansions to Equation (14), we get
I e = ( 1 φ ) 1 log ( c , d , k = 0 υ c ( 1 ) d ( φ ( λ 1 ) + λ c d ) ( θ ( φ + d ) φ k ) B ( φ + k + 1 , θ ( φ + d ) + φ 1 ) ) .
where B(.,.) stands for the beta function. The numerical results of I e of the HLITL distribution for some φ ,   λ , and θ values are listed in Table 1.
We conclude from Table 1 that as the value of shape parameter θ increased, for the same value of parameter λ , the value of randomness decreased. In addition, as the value of φ increased, the value of randomness decreased for fixed values of θ and λ . Furthermore, the values in Table 1 are in the interval [−0.321, 0.533], meaning that  λ and θ had pivotal effects on the information amount. Thus, these parameters had a strong effect on the Rényi entropy, showing different degrees of uncertainty.

3.3. Quantile Function

The quantile function of X ~ H L I T L ( λ ,   θ ) is given as q in (0, 1) as
Q ( q ; λ , θ ) = V + V 2 V , V = 1 [ 1 { q 2 q } 1 λ ] 1 θ .
Setting q = 0.5 in Equation (16), we get the median M of the HLITL distribution. Furthermore, by setting q = 0.25 and 0.75 in Equation (16), the 25th and 75th percentiles are obtained, respectively. Based on the quantiles, Bowley’s skewness is given by
B o = Q ( 3 4 ; λ , θ ) 2 Q ( 1 2 ; λ , θ ) + Q ( 1 4 ; λ , θ ) Q ( 3 4 ; λ , θ ) Q ( 1 2 ; λ , θ ) .
Moor’s kurtosis is given by
M o = Q ( 7 8 ; λ , θ ) Q ( 5 8 ; λ , θ ) Q ( 3 8 ; λ , θ ) + Q ( 1 8 ; λ , θ ) Q ( 6 8 ; λ , θ ) Q ( 2 8 ; λ , θ )
The Bo and Mo plots of the HLITL distribution were obtained based on the quantile function, shown in Figure 3.
Bowley’s skewness was plotted for increasing values of λ   and   θ . From the plot of the skewness, it is clear that the distribution was positively skewed, and the skewness increased with the decrease in θ and increase in λ and vice versa.
Moor’s kurtosis was plotted for increasing values of λ   and   θ . From the plot, it is clear that the kurtosis of the distribution increased with the increase in λ   and   θ .

3.4. Moments and Related Measures

If X has the PDF given by Equation (12), then its nth moment can be calculated as
μ n = i , d = 0 W i , d 0 2 θ ( d + 1 ) x n + 1 ( 1 + x ) 2 θ ( d + 1 ) 1 ( 1 + 2 x ) θ ( d + 1 ) 1   d x .
Using the binomial expansion, the nth moment of the HLITL distribution can be obtained as follows:
μ n = i , d , k = 0 W i , d 2 θ ( d + 1 ) ( θ ( d + 1 ) 1 k ) 0 x k + n + 1 ( 1 + x ) θ ( d + 1 ) k 2   d x = i , d , k = 0 ψ i , d , k B ( k + n + 2 , θ ( d + 1 ) n ) ,
where ψ i , d , k = 2 θ ( d + 1 ) ( θ ( d + 1 ) 1 k ) W i , d .
Further, the nth central moment of X is defined as
μ n = E ( X μ 1 ) n = i = 0 n ( 1 ) i ( n i ) ( μ 1 ) i μ n i .
The numerical values for specific values of the parameters of the first four ordinary moments μ 1 , μ 2 , μ 3 , and μ 4 , the variance ( σ 2 ) , the coefficient of variation ( CV = σ μ 1 ), the coefficient of skewness ( SK = μ 3 ( μ 2 ) 1.5 ), and the coefficient of kurtosis ( KU = μ 4 ( μ 2 ) 2 ) of the HLITL distribution are listed in Table 2 and Table 3.
As shown in Table 2 and Table 3, for the same value of θ and for the increasing value of λ , the values of the moments increased, but the values of SK, KU, and CV decreased. At the same value of λ , the moment values increased with the value of θ , while the values of SK, KU, and CV decreased. As the value of θ increased, all measures decreased with the value of λ . Thus, the HLITL distribution was positively skewed and leptokurtic.

3.5. Bonferroni and Lorenz Curves

Bonferroni and Lorenz curves have been applied to several fields, such as economics and engineering, and they are respectively defined by
L o ( x ) = ( E ( X ) ) 1 0 x y f ( y ; λ , θ ) d y ,
B e ( x ) = ( F ( x ; λ , θ ) ) 1 L o ( x ) ,
where 0 x y f ( y ; λ , θ ) d y , denotes the lower incomplete moment. For the HLITL distribution, the Bonferroni and Lorenz curves are respectively defined as follows:
L o ( x ) = i , d , k = 0 ψ i , d , k B ( k + 3 , θ ( d + 1 ) 1 , x / ( 1 + x ) ) i , d , k = 0 ψ i , d , k B ( k + 3 , θ ( d + 1 ) 1 ) ,
B e ( x ) = { 1 + [ 1 H ] λ } 2 [ 1 H ] λ [ i , d , k = 0 ψ i , d , k Β ( k + 3 , θ ( d + 1 ) 1 , x / ( 1 + x ) ) i , d , k = 0 ψ i , d , k Β ( k + 3 , θ ( d + 1 ) 1 ) ] ,
where ψ i , d , k is defined above and Β ( . , . , x / ( 1 + x ) ) represents an incomplete beta function.

4. Parameters Estimation

In this section, the estimation of λ and θ of the HLITL distribution under the SRS and RSS are discussed. Additionally, we examine the behavior of the estimates, namely in the simulation study.

4.1. Parameter Estimation under SRS

In this section, the HLITL parameters are estimated by the ML method. Let X 1 ,   X 2 , ,   X n be SRS obeying the HLITL distribution with observed values x 1 ,   x 2 , , x n , respectively. The log-likelihood function of the HLITL distribution, denoted by l n , is obtained as follows:
l n = n   l n 4 λ + n l n θ + i = 1 n l n x i ( 2 θ + 1 ) i = 1 n l n ( 1 + x i ) + ( θ 1 ) i = 1 n l n ( 1 + 2 x i ) + ( λ 1 ) i = 1 n l n [ 1 H i ] 2 i = 1 n l n [ 1 + [ 1 H i ] λ ] ,
where H i = ( 1 + 2 x i ) θ ( 1 + x i ) 2 θ . The partial derivatives of the log-likelihood function associated with λ and θ are respectively given by
l n λ = n λ + i = 1 n ln [ 1 H i ] 2 i = 1 n ln [ 1 H i ] { [ 1 H i ] λ + 1 } 1 ,
l n θ = n θ 2 i = 1 n ln [ 1 + x i ] + i = 1 n ln [ 1 + 2 x i ] ( λ 1 ) i = 1 n h i ( 1 H i ) 1 + i = 1 n   2 λ h i [ 1 H i ] λ 1 1 + [ 1 H i ] λ ,
where h i = ( ( 1 + 2 x i ) ( 1 + x i ) 2 ) θ l n ( ( 1 + 2 x i ) ( 1 + x i ) 2 ) .
By solving the nonlinear equations l n / λ = 0 and l n / θ = 0 with respect to λ and θ , the ML estimators could be obtained.

4.2. ML Estimation under RSS

Suppose that X ( i ) i c ,   i = 1 , 2 , , m , c = 1 , 2 , , k denotes a selected RSS from the HLITL distribution, where the sample size is n = mk, m denotes the set size, and k is the number of cycles. For simplicity, assume that Y i c = X ( i ) i c . Then, the fixed   c , Y i c is independent and has a PDF equal to the PDF of the ith order statistics. The likelihood function L R of the RSS of samples   y 1 c , y 2 c , , y m c can be expressed as
                                                                        L R = c = 1 k i = 1 m ( m i )     [ F ( y i c ; λ , θ ) ] i 1   f ( y i c ; λ , θ )   [ 1 F ( y i c ; λ , θ ) ] m i L R = c = 1 k i = 1 m ( m i )   [ 2 [ 1 H i c ] λ 1 + [ 1 H i c ] λ ] i 1 4 λ θ y i c ( 1 + y i c ) 2 θ 1 ( 1 + 2 y i c ) θ 1 [ 1 H i c ] λ 1 [ 1 + [ 1 H i c ] λ ] 2   [ 1 [ 1 H i c ] λ 1 + [ 1 H i c ] λ ] m i ,
where H i c = ( 1 + 2 y i c ) θ ( 1 + y i c ) 2 θ . The log-likelihood function of the HLITL distribution is given by
l n L R = C + m k ln 4 λ + m k ln θ + c = 1 k i = 1 m ( i 1 ) ln [ 2 [ 1 H i c ] λ 1 + [ 1 H i c ] λ ] + c = 1 k i = 1 m ln [ y i c ] ( 2 θ + 1 ) c = 1 k i = 1 m ln [ 1 + y i c ] + ( θ 1 ) c = 1 k i = 1 m ln [ 1 + 2 y i c ] + ( λ 1 ) c = 1 k i = 1 m ln [ 1 H i c ] 2 c = 1 k i = 1 m ln [ 1 + [ 1 H i c ] λ ] + i = 1 m ( m i ) ln [ 1 [ 1 H i c ] λ 1 + [ 1 H i c ] λ ] ,
where C = c = 1 k i = 1 m ln [ ( m i ) ] . The partial derivatives of l n L R associated with unknown parameters are expressed as follows:
l n L R λ = m k λ c = 1 k i = 1 m ( i 1 ) ln [ 1 H i c ] 1 + [ 1 H i c ] λ + c = 1 k i = 1 m i ln [ 1 H i c ] 2 c = 1 k i = 1 m ln [ 1 H i c ] [ 1 H i c ] λ 1 + [ 1 H i c ] λ   2 c = 1 k i = 1 m ( m i ) ln [ 1 H i c ] ( 1 H i c ) λ 1 ( 1 H i c ) 2 λ ,
l n L R θ = m k θ + c = 1 k i = 1 m ( i 1 ) λ ( 1 H i c ) λ 1 h i c ( 1 + ( 1 H i c ) λ ) 2 c = 1 k i = 1 m ln [ 1 + y i c ] c = 1 k i = 1 m ( λ i 1 ) h i c 1 H i c + c = 1 k i = 1 m ln [ 1 + 2 y i c ] + 2 c = 1 k i = 1 m λ ( 1 H i c ) λ 1 h i c 1 + ( 1 H i c ) λ + 2 λ c = 1 k i = 1 m ( m i ) h i c ( 1 H i c ) λ 1 1 ( 1 H i c ) 2 λ ,
where h i c = ( 1 + 2 y i c ( 1 + y i c ) 2 ) θ l n ( 1 + 2 y i c ( 1 + y i c ) 2 ) . In solving the nonlinear equations l n L R / λ = 0 and   l n L R / θ = 0 with respect to λ and   θ numerically, the ML estimators can be obtained.

4.3. Simulation Procedures

This section presents the numerical method to obtain the ML estimates (MLEs) of λ and   θ for the HLITL distribution under the RSS and SRS. A comparison study was conducted based on the mean square error (MSE) value, biases, and relative efficiency (RE). The following algorithm was performed using MATHEMATICA software [31] to obtain the MLEs and the suggested criteria measures.
Step 1: A random sample of size n = 30, 50, 100, 200, or 300 with a set size of m and number of cycles k where n = m k is generated from the HLITL distribution.
Step 2: The chosen values of the parameters are as follows: ( λ = 2 , θ = 5 ) ,   ( λ = 3 , θ = 5 ) , ( λ = 1 , θ = 8 ) , ( λ = 2 , θ = 8 ) , and ( λ = 3 , θ = 8 ) .
Step 3: For each n and specified parameter value, the MLE values under SRS and RSS are obtained.
Step 4: Repeat steps 1–3 N = 2000 times. Then, the bias, MSE, and RE values, where RE = MSE (SRS)/MSE (RSS), are computed.
Step 5: The outcomes of this study are given in Table 4, Table 5, Table 6, Table 7 and Table 8.
Based on the numerical study, the following conclusions can be drawn.
As presented in Table 4 and Table 8, for a fixed value of θ , the MSE of the estimates of λ increased with the value of λ , while the MSE of the estimates of θ decreased under SRS and RSS with n. Furthermore, the MSE values of the estimates of λ were smaller than those of the estimates of θ for all values of n under both sampling schemes.
As λ increased from two to three and θ increased from five to eight, the MSE values of both estimates increased under both schemes. In addition, the MSE values of the estimates of θ were smaller than those of the estimates of λ for all sample sizes, as shown in Table 4 and Table 8.
When the value of λ increased and the value of θ decreased, the MSE values of the estimates of λ increased, while the MSE values of the estimates of θ decreased under both sampling schemes (see Table 4 and Table 6).
Generally, the MLE values of the HLITL distribution under RSS were smaller than those under the SRS. Additionally, when the MSE values of both estimates decreased, the RE increased. When n increased, the RE values increased, and the MSE values decreased.

5. Applications to Real Data

The experiments on two real datasets were conducted to assess the flexibility of the proposed HLITL model. The first dataset included the relief times of 20 patients receiving an analgesic [32]. The second dataset was taken from [33], which included the failure times of the air conditioning system of an airplane. The HLITL model was compared with four competing models: AP exponential (APE) [34], alpha power (AP) Lindley (APL) [35], Fréchet inverse exponential (OFIE) [36], and odd Fréchet inverse Rayleigh (OFIR) [37]. Some descriptive statistics of the two datasets are given in Table 9, and the boxplot is shown in Figure 4.
The MLEs, along with their standard errors (SEs) of the model parameters, are provided in Table 10 and Table 11. Table 12 and Table 13 give the values of the Anderson–Darling statistic (A*), Cramér–von Mises statistic (W*), Akaike information criterion (AIC), corrected AIC (CAIC), Bayesian information criterion (BIC), and Hannan–Quinn information criterion (HQIC).
The values in Table 12 and Table 13 show that the HLITL distribution had the smallest values among all fitted distributions. Consequently, this distribution can be considered more suitable than the others for the considered data. The estimated PDF, CDF, SF, and PP plots of the HLITL model are presented in Figure 5 and Figure 6, where it can be seen that the HLITL model exhibits a close fit to both datasets.
The author of [38] mentioned that the total time test (TTT) represents a useful graphical approach to confirm whether the data can be applied to a specific distribution or not. The empirical version of the TTT plot is expressed as
T ( r n ) = i = 1 n [   y i : n + ( n r ) y r : n ] i = 1 n y i : n ,
where r = 1 , 2 , , n and y i : n denote the ith order statistics of the sample. In [38], it was shown that when the TTT plot represents a straight line, then the HRF is constant, and when the TTT plot is concave (or convex), the HRF increases (or decreases). Additionally, when the TTT plot is first convex and then concave, the HRF is U-shaped (bathtub); otherwise, the HRF is unimodal. The TTT plots presented in Figure 7 show that the estimated HRFs of the first and second datasets increased and decreased. This claim is also confirmed in Figure 8.
For both real datasets, the MLE values of both parameters under RSS were obtained. The observed RSS of the first dataset with the set size of four were 1.1, 2.7, 3, and 4.1, and those of the second dataset with the set size of five were 1, 5, 7, 11, and 261. The MLE values of the HLITL distribution for the first dataset and their SE values were λ ^ = 6.653 ( 5.558 ) and θ ^ = 2.766(1.348), and those of the second dataset were λ ^ = 2.239(1.3448) and θ ^ = 0.844(0.071).

6. Conclusions

This paper proposes an HLITL distribution with two parameters. It is demonstrated that the HLITL distribution can be regarded as a mixture of ITL distributions. Its statistical properties, including the moments, quantile function, skewness, kurtosis, Bonferroni and Lorenz curves, and stochastic ordering, are defined. The maximum likelihood estimators of the HLITL distribution parameters are derived under both SRS and RSS schemes. It was shown empirically that the maximum likelihood estimates taken from the observed RSS were more efficient than the corresponding SRS. The merit of the HLITL distribution was evaluated experimentally using two real datasets. The results show that it performed better than the other fitted distributions. Based on the obtained results, the proposed distribution can be useful for modeling data in various fields.

Author Contributions

Conceptualization, F.J.; Data curation, D.S.; Formal analysis, R.B.; Investigation, C.C.; Methodology, M.E. (Mahmoud Elsehetry); Project administration, A.S.H.; Software, M.E. (Mohammed Elgarhy). All authors have read and agreed to the published version of the manuscript.

Funding

This work was funded by the Deanship of Scientific Research (DSR) at King Abdulaziz University, Jeddah, under grant No. FP-188-42.

Institutional Review Board Statement

Not applicable.

Informed Consent Statement

Not applicable.

Data Availability Statement

Datasets are provided in the paper.

Acknowledgments

We would like to thank all four reviewers and the academic editor for their valuable and helpful comments on this manuscript, which greatly improved it in this regard. This project was funded by the Deanship of Scientific Research (DSR) at King Abdulaziz University, Jeddah, under grant No. FP-188-42. The authors, therefore, acknowledge with thanks the DSR’s technical and financial support.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Nadarajah, S.; Kotz, S. Moments of some J-shaped distributions. J. Appl. Stat. 2003, 30, 311–317. [Google Scholar] [CrossRef]
  2. Ghitany, M.E.; Kotz, S.; Xie, M. On some reliability measures and their stochastic orderings for the Topp–Leone distribution. J. Appl. Stat. 2005, 32, 715–722. [Google Scholar] [CrossRef]
  3. Zghoul, A.A. Record values from a family of J-shaped distributions. Statistica 2011, 71, 355–365. [Google Scholar]
  4. Genc, A.I. Estimation of P(X>Y) with Topp–Leone distribution. J. Stat. Comput. Simul. 2013, 83, 326–339. [Google Scholar] [CrossRef]
  5. Keller, A.Z.; Kamath, A.R.R. Alternative reliability models for mechanical systems. In Proceedings of the 3rd International Conference on Reliability and Maintainability, Toulouse, France, 26 February 2015. [Google Scholar]
  6. Sharma, V.K.; Singh, S.K.; Singh, U.; Agiwal, V. The inverse Lindley distribution: A stress-strength reliability model with application to head and neck cancer data. J. Ind. Prod. Eng. 2015, 32, 162–173. [Google Scholar] [CrossRef]
  7. Abd AL-Fattah, A.M.; EL-Helbawy, A.A.; AL-Dayian, G.R. Inverted Kumaraswamy distribution: Properties and estimation. Pak. J. Stat. 2017, 33, 37–61. [Google Scholar]
  8. Tahir, M.H.; Cordeiro, G.M.; Ali, S.; Dey, S.; Manzoor, A. The inverted Nadarajah–Haghighi distribution: Estimation methods and applications. J. Stat. Comput. Simul. 2018, 88, 2775–2798. [Google Scholar] [CrossRef]
  9. Hassan, A.S.; Abd-Allah, M. On the Inverse Power Lomax Distribution. Ann. Data Sci. 2018, 6, 259–278. [Google Scholar] [CrossRef]
  10. Hassan, A.S.; Mohamed, R.E. Parameter estimation for inverted exponentiated Lomax distribution with right censored data. Gazi Univ. J. Sci. 2019, 32, 1370–1386. [Google Scholar]
  11. Hassan, A.S.; Elgarhy, M.; Ragab, R. Statistical properties and estimation of inverted Topp-Leone distribution. J. Stat. Appl. Probab. 2020, 9, 319–331. [Google Scholar]
  12. Hassan, A.S.; Elgarhy, M. Kumaraswamy Weibull-generated family of distributions with applications. Adv. Appl. Stat. 2016, 48, 205–239. [Google Scholar] [CrossRef]
  13. Cordeiro, G.M.; Alizadeh, M.; Marinho, P.R.D. The type I half-logistic family of distributions. J. Stat. Comput. Simul. 2015, 86, 707–728. [Google Scholar] [CrossRef]
  14. Al-Shomrani, A.; Arif, O.; Hanif, S.; Shahbaz, M.Q.; Shawky, A. Topp–Leone Family of Distributions: Some Properties and Application. Pak. J. Stat. Oper. Res. 2016, 12, 443. [Google Scholar] [CrossRef] [Green Version]
  15. Hassan, A.S.; Hemeda, S.E.; Maiti, S.S.; Pramanik, S. The generalized additive Weibull-G family of probability distributions. Int. J. Stat. Probab. 2017, 6, 65–83. [Google Scholar] [CrossRef] [Green Version]
  16. Hassan, A.S.; Nassr, S.G. Power Lindley-G family of distributions. Ann. Data Sci. 2019, 6, 189–210. [Google Scholar] [CrossRef]
  17. Hassan, A.S.; Elgarhy, M.; Ahmad, Z. Type II generalized Topp–Leone family of distributions: Properties and applications. J. Data Sci. JDS 2019, 17, 638–659. [Google Scholar] [CrossRef]
  18. Marshall, A.W.; Olkin, I. A new method for adding a parameter to a family of distributions with applications to the exponential and Weibull families. Biometrika 1997, 84, 641–652. [Google Scholar] [CrossRef]
  19. Hassan, A.S.; Elgarhy, M.; Shakil, M. Type II half logistic family of distributions with applications. Pak. J. Stat. Oper. Res. 2017, 13, 245–264. [Google Scholar]
  20. Hassan, A.S.; Elgarhy, M.; Haq, M.A.; Alrajhi, S. On type II half logistic Weibull distribution with applications. Math. Theory Modeling 2019, 19, 49–63. [Google Scholar]
  21. Haq, M.A.; Almarashi, A.M.; Hassan, A.S.; Elgarhy, M. Type II half logistic Rayleigh: Properties & estimation based on censored samples. J. Adv. Math. Comput. Sci. 2018, 19, 1–19. [Google Scholar]
  22. Hassan, A.S.; Elgarhy, M.; Mohamed, R.E. Statistical properties and estimation of type II half logistic Lomax distribution. Thail. Stat. 2020, 18, 290–305. [Google Scholar]
  23. Zeineldin, R.A.; Haq, M.A.U.; Hashmi, S.; Elsehety, M.; Elgarhy, M. Type II half logistic Kumaraswamy distribution with applications. J. Funct. Spaces 2020, 2020, 1–15. [Google Scholar] [CrossRef]
  24. Bantan, R.; Hassan, A.S.; Elsehetry, M.; Golam Kibria, B.M. Half-logistic xgamma distribution: Properties and estimation under censored samples. Discret. Dyn. Nat. Soc. 2020. [Google Scholar] [CrossRef]
  25. McIntyre, G. A method for unbiased selective sampling, using ranked sets. Aust. J. Agric. Res. 1952, 3, 385–390. [Google Scholar] [CrossRef]
  26. Zamanzade, E. New ranked set sampling for estimating the population mean and variance. Hacet. J. Math. Stat. 2015, 46, 1891–1905. [Google Scholar] [CrossRef]
  27. Haq, A.; Brown, J.; Moltchanova, E.; Al-Omari, A.I. Paired Double Ranked Set Sampling. Commun. Stat. Theory Methods 2015, 45, 2873–2889. [Google Scholar] [CrossRef]
  28. Al-Omari, A.I.; Almanjahie, I.M.; Hassan, A.S.; Nagy, H.F. Estimation of the stress-strength reliability for exponentiated pareto distribution using median and ranked set sampling methods. Comput. Mater. Contin. 2020, 64, 835–857. [Google Scholar] [CrossRef]
  29. Bantan, R.; Hassan, A.S.; Elsehetry, M. Zubair Lomax distribution: Properties and estimation based on ranked set sampling. Comput. Mater. Contin. 2020, 65, 2169–2187. [Google Scholar] [CrossRef]
  30. Hassan, A.S.; Al-Omari, A.; Nagy, H.F. Stress-strength reliability for the generalized inverted exponential distribution using MRSS. Iran. J. Sci. Technol. Trans. A Sci. 2021, 45, 641–659. [Google Scholar] [CrossRef]
  31. Mathematica, Version 9.0; Wolfram Research Inc.: Champaign, IL, USA, 2012.
  32. Lee, P.N.; Gross, A.J.; Clark, V.A. Survival distributions: Reliability applications in the biomedical sciences. J. R. Stat. Soc. Ser. C Appl. Stat. 1976, 25, 303. [Google Scholar] [CrossRef]
  33. Linhart, H.; Zucchini, W. Model Selection; Wiley: New York, NY, USA, 1986. [Google Scholar]
  34. Mahadavi, A.; Kundu, D. A new method of generating distribution with an application to exponential distribution. Commun. Stat. Theory Methods. 2017, 46, 6543–6557. [Google Scholar] [CrossRef]
  35. Dey, S.; Ghosh, I.; Kumar, D. Alpha power transformed Lindley distribution: Properties and associated inference with application to earthquake data. Ann. Data Sci. 2019, 6, 623–650. [Google Scholar] [CrossRef]
  36. Alrajhi, S.H. The odd Fréchet inverse exponential distribution with application. J. Nonlinear Sci. Appl. 2019, 8, 535–542. [Google Scholar] [CrossRef] [Green Version]
  37. Elgarhy, M.; Alrajhi, S.H. The odd Fréchet inverse Rayleigh distribution: Statistical properties and applications. J. Nonlinear Sci. Appl. 2019, 12, 291–299. [Google Scholar] [CrossRef] [Green Version]
  38. Aarset, M.V. How to identify a bathtub hazard rate. IEEE Trans. Reliab. 1987, 36, 106–108. [Google Scholar] [CrossRef]
Figure 1. The PDF plots of the HLITL distribution for certain choices of parameters.
Figure 1. The PDF plots of the HLITL distribution for certain choices of parameters.
Mathematics 09 01214 g001
Figure 2. The HRF plots of the HLITL distribution for certain choices of λ and θ .
Figure 2. The HRF plots of the HLITL distribution for certain choices of λ and θ .
Mathematics 09 01214 g002
Figure 3. 3D plots of Bo and Mo.
Figure 3. 3D plots of Bo and Mo.
Mathematics 09 01214 g003
Figure 4. Box plots for both datasets.
Figure 4. Box plots for both datasets.
Mathematics 09 01214 g004
Figure 5. Empirical and estimated CDF, PDF, PP, and SF plots of the HLITL model for the first dataset.
Figure 5. Empirical and estimated CDF, PDF, PP, and SF plots of the HLITL model for the first dataset.
Mathematics 09 01214 g005
Figure 6. Empirical and estimated CDF, PDF, PP, and SF plots of the HLITL model for the second dataset.
Figure 6. Empirical and estimated CDF, PDF, PP, and SF plots of the HLITL model for the second dataset.
Mathematics 09 01214 g006
Figure 7. TTT plots for both datasets.
Figure 7. TTT plots for both datasets.
Mathematics 09 01214 g007
Figure 8. Estimated HRFs of the HLITL model for both datasets.
Figure 8. Estimated HRFs of the HLITL model for both datasets.
Mathematics 09 01214 g008
Table 1. Numerical results of Rényi entropy for the HLITL distribution.
Table 1. Numerical results of Rényi entropy for the HLITL distribution.
Parameters φ = 0.5 φ = 1.5   φ = 2.5
θ = 5 , λ = 0.5 0.245−0.193−0.321
θ = 5 , λ = 1 0.3860.06−0.022
θ = 5 , λ = 1.5 0.4490.1410.064
θ = 5 , λ = 2 0.4860.1860.111
θ = 5 , λ = 2.5 0.5130.2160.142
θ = 5 , λ = 3 0.5330.2390.165
θ = 8 , λ = 0.5 0.043−0.321−0.437
θ = 8 , λ = 1 0.167−0.095−0.167
θ = 8 , λ = 1.5 0.216−0.031−0.099
θ = 8 , λ = 2 0.2440.0022−0.065
θ = 8 , λ = 2.5 0.2630.023−0.044
θ = 8 , λ = 3 0.2770.038−0.029
θ = 10 , λ = 0.5 −0.041−0.38−0.491
θ = 10 , λ = 1 0.076−0.165−0.233
θ = 10 , λ = 1.5 0.121−0.107−0.172
θ = 10 , λ = 2 0.145−0.078−0.143
θ = 10 , λ = 2.5 0.161−0.061−0.125
Table 2. Some moments of the HLITL distribution.
Table 2. Some moments of the HLITL distribution.
μ r θ = 5.0 , λ = 0.5 θ = 5.0 , λ = 1.0 θ = 5.0 , λ = 1.5 θ = 5.0 , λ = 2.0 θ = 5.0 , λ = 2.5 θ = 5.0 , λ = 3.0
μ 1 0.3340.5550.7160.8440.9511.044
μ 2 0.3360.6460.9281.1881.4311.659
μ 3 0.8041.5972.3763.143.8894.626
μ 4 5.13710.26815.38920.49825.59430.677
σ 2 0.2250.3380.4150.4750.5260.569
SK 5.0794.3984.1764.0714.0113.973
KU 84.14666.82761.80459.54558.30657.546
CV1.4181.0480.90.8160.7620.723
Table 3. Some moments of the HLITL distribution.
Table 3. Some moments of the HLITL distribution.
μ r θ = 8.0 , λ = 0.5 θ = 8.0 , λ = 1.0 θ = 8.0 , λ = 1.5 θ = 8.0 , λ = 2.0 θ = 8.0 , λ = 2.5 θ = 8.0 , λ = 3.0
μ 1 0.2330.3810.4840.5650.630.686
μ 2 0.1370.2590.3670.4640.5520.633
μ 3 0.1480.2910.430.5630.6920.816
μ 4 0.2610.5210.7791.0341.2861.535
σ 2 0.0830.1140.1320.1450.1550.162
SK 3.2492.7322.5662.4892.4472.421
KU 23.34218.33516.95516.36716.06515.891
CV1.2320.8890.7510.6740.6240.587
Table 4. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 2 ,   θ = 5 .
Table 4. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 2 ,   θ = 5 .
n
ParametersSRSRSSRE
MLEBiasMSEMLEBiasMSE
30 λ 2.246650.246650.448142.055990.055990.072586.17398
θ 5.423950.423952.151645.099990.099990.405605.30488
50 λ 2.139330.139330.188532.027050.027050.024447.71307
θ 5.246740.246740.962275.058900.058900.128297.50083
100 λ 2.021330.021330.068652.014280.014280.0067210.21169
θ 5.128730.128730.420115.026570.026570.0383910.94189
200 λ 1.98450−0.015500.040861.99913−0.000870.0027115.06497
θ 4.96450−0.035500.191844.98999−0.010010.0145913.14884
300 λ 2.008410.008410.022861.99921−0.000790.0009923.08329
θ 5.017920.017920.143974.99748−0.002520.0056425.50732
Table 5. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 3 ,   θ = 5 .
Table 5. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 3 ,   θ = 5 .
n ParametersSRSRSSRE
MLEBiasMSEMLEBiasMSE
30 λ 3.498330.498331.517223.065510.065510.189987.98617
θ 5.546660.546661.875255.084730.084730.279256.71538
50 λ 3.175270.175270.395933.013430.013430.058296.79289
θ 5.234040.234040.743105.010140.010140.110536.72327
100 λ 3.065250.065250.223053.015640.015640.0215810.33606
θ 5.105400.105400.366525.018760.018760.0337510.85922
200 λ 3.061580.061580.128873.010630.010630.0059921.49613
θ 5.052760.052760.207165.012860.012860.0090122.98753
300 λ 3.045100.045100.075663.009650.009650.0024930.38488
θ 5.004450.004450.113405.013170.013170.0043226.24375
Table 6. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 1 ,   θ = 8 .
Table 6. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 1 ,   θ = 8 .
n ParametersSRSRSSRE
MLEBiasMSEMLEBiasMSE
30 λ 1.072350.072350.071011.016800.016800.011256.31288
θ 8.763510.763517.544818.276980.276981.623554.64711
50 λ 1.054110.054110.040311.001420.001420.005047.99685
θ 8.610900.610904.961548.094700.094700.679027.30688
100 λ 1.022160.022160.016260.99950−0.000500.0010715.16481
θ 8.366690.366692.782847.99541−0.004590.1510418.42416
200 λ 1.011100.011100.006691.001590.001590.0003121.29228
θ 8.159510.159510.981828.027850.027850.0467820.98708
300 λ 0.99242−0.007580.003890.99977−0.000230.0001723.37761
θ 7.99837−0.001630.526178.008090.008090.0257820.41341
Table 7. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 2 ,   θ = 8 .
Table 7. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 2 ,   θ = 8 .
n ParametersSRSRSSRE
MLEBiasMSEMLEBiasMSE
30 λ 2.114510.114510.239302.045460.045460.086262.77425
θ 8.144700.144702.722088.136910.136910.955062.85016
50 λ 2.108720.108720.179842.010790.010790.024257.41578
θ 8.309650.309652.202288.013820.013820.328926.69551
100 λ 2.073310.073310.083442.011300.011300.0075311.07905
θ 8.108550.108551.245168.042320.042320.1038111.99414
200 λ 2.003420.003420.032841.99796−0.002040.0027711.84549
θ 7.95718−0.042820.465777.99595−0.004050.0384412.11569
300 λ 2.020330.020330.025371.99699−0.003010.0010324.59269
θ 8.094560.094560.382877.98655−0.013450.0149825.55199
Table 8. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 3 ,   θ = 8 .
Table 8. MLE, bias, MSE, and RE of the HLITL distribution under SRS and RSS for λ = 3 ,   θ = 8 .
n ParametersSRSRSSRE
MLEBiasMSEMLEBiasMSE
30 λ 3.414180.414181.272293.093330.093330.156598.12492
θ 8.551750.551754.237358.202600.202600.505198.38760
50 λ 3.223760.223760.585942.99541−0.004590.061249.56734
θ 8.392820.392821.955687.99319−0.006810.250807.79780
100 λ 3.123610.123610.217253.011690.011690.0212110.24074
θ 8.190480.190481.016998.031790.031790.0858611.84500
200 λ 3.045160.045160.110873.006660.006660.0063517.44767
θ 8.051300.051300.408368.007590.007590.0284614.35016
300 λ 3.045640.045640.073223.000140.000140.0038219.15888
θ 8.082020.082020.305268.003100.003100.0160818.98614
Table 9. Descriptive statistics for the datasets.
Table 9. Descriptive statistics for the datasets.
nMinMaxMeanMedianSKKU
Data 1201.14.11.91.71.8624.185
Data 230126159.6221.7842.569
Table 10. MLEs and their SEs (in parentheses) for the first data set.
Table 10. MLEs and their SEs (in parentheses) for the first data set.
ModelMLE and SEMLE and SE
HLITL ( λ ,   θ )39.964 (25.292)7.027 (1.394)
OFIR (α, θ)1.623 (0.182)1.462 (0.265)
APL ( α ,   γ )10540 (15462.663)1.884 (0.174)
APE ( α ,   γ )16530 (23359.046)1.488 (0.161)
OFIE ( α ,   γ )1.073 (0.0618)2.929 (0.518)
Table 11. MLEs and their SEs (in parentheses) for the second dataset.
Table 11. MLEs and their SEs (in parentheses) for the second dataset.
ModelMLE and SEMLE and SE
HLITL ( λ ,   θ )4.519 (1.2539)0.612 (0.115)
OFIR ( α ,   γ )6.85 (1.761)0.229 (0.034)
APL ( α ,   γ )0.1 (0.104)0.024 (5.127 × 10−3)
APE ( α ,   γ )8.688 × 10−10 (5.698 × 10−8)8.536 × 10−4 (2.844 × 10−3)
OFIE ( α ,   γ )4.082 (0.7770)0.47 (0.069)
Table 12. Measures of goodness-of-fit statistics for the first dataset.
Table 12. Measures of goodness-of-fit statistics for the first dataset.
ModelAICCAICBICHQICA*W*
HLITL34.89435.59933.49635.2820.17020.0308
OFIR35.47636.18134.07835.8640.236350.0399
APL74.41675.12273.01874.8040.756970.09802
APE75.41176.11774.01375.80.944780.10762
OFIE35.07835.78433.6835.4670.192090.03343
Table 13. Measures of goodness-of-fit statistics for the second dataset.
Table 13. Measures of goodness-of-fit statistics for the second dataset.
ModelAICCAICBICHQICA*W*
HLITL315.459315.904314.414316.3560.83860.1152
OFIR341.187341.893340.141342.0831.531470.1613
APL370.83371.274369.784371.7271.26290.1778
APE358.775359.22357.729359.6721.58440.1978
OFIE332.232332.677331.186333.1291.43870.1817
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Bantan, R.; Elsehetry, M.; Hassan, A.S.; Elgarhy, M.; Sharma, D.; Chesneau, C.; Jamal, F. A Two-Parameter Model: Properties and Estimation under Ranked Sampling. Mathematics 2021, 9, 1214. https://doi.org/10.3390/math9111214

AMA Style

Bantan R, Elsehetry M, Hassan AS, Elgarhy M, Sharma D, Chesneau C, Jamal F. A Two-Parameter Model: Properties and Estimation under Ranked Sampling. Mathematics. 2021; 9(11):1214. https://doi.org/10.3390/math9111214

Chicago/Turabian Style

Bantan, Rashad, Mahmoud Elsehetry, Amal S. Hassan, Mohammed Elgarhy, Dreamlee Sharma, Christophe Chesneau, and Farrukh Jamal. 2021. "A Two-Parameter Model: Properties and Estimation under Ranked Sampling" Mathematics 9, no. 11: 1214. https://doi.org/10.3390/math9111214

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop