Journal of Physics: Conference

Series

PAPER • OPEN ACCESS You may also like

- Estimation of the Shape Parameter for
Parameter estimation for the Lomax distribution Lomax Distribution Under Bilateral Type-I
Censored Sample
using the E-Bayesian method Hua Liu and Chang Xin Xi

- Inferences of inverted gompertz

parameters from an adaptive type-ii
To cite this article: A Fitrilia et al 2018 J. Phys.: Conf. Ser. 1108 012081 progressive hybrid censoring
Heba S Mohammed, Ahmed Elshahhat
and Refah Alotaibi

- Experimental verification of IMPT

treatment plans in an anthropomorphic
View the article online for updates and enhancements. phantom in the presence of delivery
uncertainties
F Albertini, M Casiraghi, S Lorentini et al.

This content was downloaded from IP address 169.224.81.34 on 11/05/2024 at 16:16

MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

Parameter estimation for the Lomax distribution using the E-

Bayesian method

A Fitrilia1,2, I Fithriani1,3 and S Nurrohmah1,4

1
Department of Mathematics, Faculty of Mathematics and Natural Sciences, Universitas
Indonesia, Depok 16424, Indonesia.

Email : [email protected], [email protected], [email protected]

Abstract. In this paper, we will find the parameter estimator from Lomax distribution on one
of censored data, that is right censored data type II. Lomax distribution is also called Pareto
type II distribution. The parameter will be estimate is the shape parameter with the assumption
of scale parameters (𝛽) has been known. The parameter estimation method used in this study is
E-Bayesian estimation method. E-Bayesian estimation is an expectation of Bayes estimation, in
order to obtain Bayes estimation expectations is by calculating the mean of Bayes estimators.
E-Bayesian estimation method is used to estimate failure rate. The estimation will use prior
Gamma as conjugate prior distribution from Lomax distribution and Loss function will be used
is balanced squared error loss function (BSELF). Thus, the main purpose of this study is to find
the parameter estimator of the Lomax distribution on the right censored data type II using the
E-Bayesian method. The final result of this study, we get the likelihood function from Lomax
distribution on the right censored data type II and the parameter estimator from Lomax
distribution on the right censored data type II using the E-Bayesian method.

1. Introduction
The Lomax distribution was first introduced by K. S. Lomax in 1954 [1]. Lomax distribution is a
probability distribution used in business, economics, and actuarial science. The Lomax distribution has
two parameters and denoted by Lomax(𝛼, 𝛽) where 𝛼 is a form parameter and 𝛽 is the scale
parameter. Lomax distribution has probability density function (pdf) that defined as

𝛼 𝑡 −(𝛼+1)
𝑓(𝑥; 𝛼, 𝛽) = (1 + ) , 𝑥 > 0, (𝛼, 𝛽 > 0) (1)
𝛽 𝛽

and cumulative distribution function (cdf) that defined as

𝑥 −𝛼
𝐹(𝑥; 𝛼, 𝛽) = 1 − (1 + ) , 𝑥 > 0, (𝛼, 𝛽 > 0) (2)
𝛽

The survival function or reliability function of the Lomax distribution defined as the probability of an
object can survive until time 𝑡 is

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

𝑡 −𝛼 𝑡 −𝛼
𝑆(𝑡) = 1 − 𝐹(𝑡) = 1 − (1 − (1 + ) ) = (1 + ) , 𝑡>0 (3)
𝛽 𝛽

The hazard function of the Lomax distribution represents the probability of a component failing at
time 𝑡 is

𝛼 𝑡 −(𝛼+1)
𝑓(𝑡) 𝛽 (1 + ) 𝛼 𝑡 −1
𝛽
ℎ(𝑡) = = = (1 + )
𝑆(𝑡) 𝑡 −𝛼 𝛽 𝛽
(1 + )
𝛽
𝛼
ℎ(𝑡) = , 𝑡>0 (4)
𝛽+𝑡

Mean and variance of Lomax distribution is

𝛽
𝜇= , 𝛼>1 (5)
𝛼−1

and

𝛼𝛽 2
𝜎2 = , 𝛼>2 (6)
(𝛼 − 1)2 (𝛼 − 2)

Since the parameters of 𝛼 and 𝛽 are unknown, parameter estimation is necessary, one of the
paramereter estimation that can be used is the E-Bayesian estimation method. E-Bayesian estimation
method are used to estimate failure rates, furtheremore they are suitable for censored data with small
sample sizes and high reliability [2]. Estimated parameters using the Bayesian approach, which to
obtain E-Bayesian estimate of 𝛼 (i.e. expectation of the Bayes estimates of 𝛼) is obtained by
calculating the mean of the Bayes estimator.

2. Balanced square error loss function (BSELF)

The estimator of a function, using BSEL is actually a combination of the maximum likelihood
estimator of a function and the Bayes estimator using the SEL. Ahmadi et al. [3] introduced a balanced
loss function (BLF) derived from Zellner [4] in the form
𝒒
𝑳𝝆,𝝎,𝜹𝟎 (𝚲(𝜶), 𝜹) = 𝝎𝒒(𝜶)𝝆(𝜹𝟎 , 𝜹) + (𝟏 − 𝝎)𝒒(𝜶)𝝆(𝚲(𝜶), 𝜹) (𝟕)

where 𝒒(⋅) is a suitable positive weight function and 𝝆(𝚲(𝜶), 𝜹) is any Loss function when
estimating 𝚲(𝜶) with 𝜹. The parameter 𝜹𝟎 is the selected prior estimator of 𝚲(𝜶), obtained for
example from the maximum likelihood criterion, and 𝟎 ≤ 𝝎 < 𝟏. If 𝝆(𝚲(𝜶), 𝜹) = (𝜹 − 𝚲(𝜶))𝟐
and 𝒒(𝜶) = 𝟏 are substituted in equation (5), obtained BSELF, given as

𝟐
𝑳𝝎,𝜹𝟎 (𝚲(𝜶), 𝜹) = 𝝎(𝜹 − 𝜹𝟎 )𝟐 + (𝟏 − 𝝎)(𝜹 − 𝚲(𝜶)) (𝟖)

The corresponding Bayes estimator of the function 𝚲(𝜶) is given as

𝜹𝝎,𝚲,𝜹𝟎 (𝐱) = 𝝎𝜹𝟎 + (𝟏 − 𝝎)𝑬(𝚲(𝜶)|𝐱) (𝟗)

2
MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

3. Likelihood function from Lomax distribution on the right censored data type II
In the right censored data type II, there are 𝑟 observations of 𝑛 samples observed, and the study is
stopped after the 𝑟th failure that occurred. In this censored, 𝑟 is determined before the data is
collected. The joint probability density function of right censored data type II of 𝑋1 , 𝑋2 , … , 𝑋𝑟 for 𝑟 <
𝑛 is
𝑟
𝑛!
𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑟 ) = [∏ 𝑓(𝑥𝑖 )] [𝑆(𝑥𝑟 )]𝑛−𝑟 (10)
(𝑛 − 𝑟)!
𝑖=1

then the joint probability density function of Lomax distribution on the right censored data type II is

𝑛! 1 1 1
𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑟 ) = 𝛼𝑟 𝛼
(𝑛 − 𝑟)! 𝑥 𝛼(𝑛−𝑟)
𝛽 𝑟 ∏𝑟𝑖=1 (1 + 𝑖 ) ∏𝑟𝑖=1 (1 + 𝑥𝑖 ) (1 + 𝑥𝑟 )
𝛽 𝛽 𝛽
𝑛!
𝑓(𝑥1 , 𝑥2 , … , 𝑥𝑟 ) = 𝛼 𝑟 𝑣(x; 𝛽)𝑒 −𝑀𝛼 (11)
(𝑛 − 𝑟)!

Thus the likelihood function of the Lomax distribution on the right censored data type II can be
written as

𝑛!
𝐿(x|𝛼, 𝛽) = 𝛼 𝑟 𝑣(x; 𝛽)𝑒 −𝑀𝛼 (12)
(𝑛 − 𝑟)!

𝟏
where 𝐱 = (𝒙𝟏 , 𝒙𝟐 , … , 𝒙𝒓 ), 𝒗(𝐱; 𝜷) =
𝒙𝒊
𝜷𝒓 ∏𝒓𝒊=𝟏 (𝟏 +
)
𝜷
𝒓
𝒙𝒊 𝒙𝒓
and 𝑴 ≡ 𝑴(𝐱; 𝜷) = ∑ 𝐥𝐧 (𝟏 + ) + (𝒏 − 𝒓) 𝐥𝐧 (𝟏 + )
𝜷 𝜷
𝒊=𝟏

4. Bayesian estimation of Lomax distribution on the right censored data type II

Bayes estimation is estimation method that combines prior information with information obtained
from sample data. Prior information is obtained from the distribution of parameters, while the
information from the data is summarized in the likelihood function. The combination of prior
information and information from the data will produce posterior information (posterior distribution).
The results are expressed in terms of this posterior distribution which will be the basis in the Bayes
method for estimating the parameters.
From (12), assuming 𝛽 is known, the maximum likelihood estimator of the parameter 𝛼 is given
as

𝑛!
ln 𝐿(x|𝛼) = ln + 𝑟 ln 𝛼 + ln 𝑣(x; 𝛽) + (−𝑀𝛼)
(𝑛 − 𝑟)!
𝑑
ln 𝐿(x|𝛼) = 0
𝑑𝛼

so that the maximum likelihood estimator of the parameter α is obtained

𝑟
𝛼̂𝑀𝐿 = (13)
𝑀

3
MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

To search for Bayes estimator, we need to find the prior distribution first. According to Box
and Tiao [5], the conjugate prior distribution, refers to the reference model analysis, especially in the
creation of likelihood function. So in the defining of conjugate prior always thought about the
defining of the prior distribution pattern that has the conjugate form with the probability density
function of the likelihood construction. According to Castillo [6], If ℱ is the class of the sampling
distribution 𝑝(𝑦|𝜃) and 𝒫 is the class of the prior distribution for 𝜃, 𝑝(𝜃), then class 𝒫 is said to
conjugate for ℱ if the posterior probability 𝑝(𝜃|𝑦) has the same distribution form with the prior
probability function 𝑝(𝜃) for all 𝑝(𝑦|𝜃) ∈ ℱ. In this study, Gamma conjugate prior is used to find
Bayes estimators. Likelihood function of the Lomax distribution on the right censored data type II is

𝑛!
𝑓(x|𝛼, 𝛽) = 𝛼 𝑟 𝑣(x; 𝛽)𝑒 −𝑀𝛼 (14)
(𝑛 − 𝑟)!

Assuming 𝛽 is known, we can use the Gamma distribution as an conjugate prior distribution of 𝛼 with
shape and scale parameter 𝑎 and 𝑏 respectively and its probability density function given by

𝑏 𝑎 𝑎−1 −𝑏𝛼
𝑔(𝛼|𝑎, 𝑏) = 𝛼 𝑒 , 𝛼>0 (15)
Γ(𝑎)

The likelihood function and the probability density function of Gamma distribution has the same
functional form where 𝑟 = 𝑎 − 1 and 𝑏 = 𝑀. Therefore, the Gamma Distribution is used as the
conjugate prior distribution of the Lomax distribution on the right censored data type II. Furtheremore,
we will find the posterior distribution density function

𝑓(x|𝛼, 𝛽)𝑔(𝛼|𝑎, 𝑏) 𝛼𝑟+𝑎−1 𝑒−(𝑏+𝑀)𝛼 (𝑏 + 𝑀)𝑟+𝑎

𝑞(𝛼|x) = ∞ =
∫0 𝑓(x|𝛼, 𝛽)𝑔(𝛼|𝑎, 𝑏)𝑑𝛼 Γ(𝑟 + 𝑎)
𝑞(𝛼|x) = 𝜅𝛼 𝑟+𝑎−1 𝑒 −(𝑏+𝑀)𝛼 , 𝛼 > 0 (16)

(𝑏+𝑇)𝑟+𝑎
with 𝜅 = Γ(𝑟+𝑎)
. Posterior probability density function in (16) is the probability density function of
Gamma distribution with parameters 𝑟 + 𝑎 and 𝑏 + 𝑇. Since the resulting posterior distribution is
Gamma distributed, just like the prior distribution which Gamma distributed, it is proven that the
Gamma distribution is the conjugate prior of the Lomax distribution on the right censored data type II.
The Bayes estimator of 𝛼 with balanced square error loss function (BSELF) can be given as

𝛼̂𝐵 (𝑎, 𝑏) = 𝜔𝛿0 + (1 − 𝜔)𝐸(Λ(𝛼)|𝐱)

with 𝛿0 is maximum likelihood estimator and 𝐸(Λ(𝛼)|𝐱) is Bayes estimator with square error loss
function.
∞ (𝑟 + 𝑎)
𝐸(Λ(𝛼)|𝐱) = ∫ Λ(𝛼) 𝑞(Λ(𝛼)|𝐱)𝑑𝛼 = (17)
0 (𝑏 + 𝑀)

Then the Bayes estimator for 𝛼 is obtained as

𝑟 (𝑟 + 𝑎)
𝛼̂𝐵 (𝑎, 𝑏) = 𝜔 ( ) + (1 − 𝑤) (18)
𝑀 (𝑏 + 𝑀)

4
MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

5. E-Bayesian estimation of Lomax distribution on the right censored data type II

E-Bayesian estimation methods are used to estimate failure rates. E-Bayesian estimation is an
expectation of Bayes estimation, in order to obtain Bayes estimation expectations is by calculating the
mean of Bayes estimators.
Suppose that life an object is Lomax distributed on the right censored data type II with a joint
probability density function is (11). Let the prior distribution of 𝛼 be the conjugate
−Γ(𝑎, 𝑏) distribution with the probability density function is (15) where
∞
Γ(𝑎) = ∫0 𝑥 𝑎−1 𝑒 −𝑥 𝑑𝑥 is a function of Gamma, 𝑎 > 0, 𝑏 > 0, and 𝑎 and 𝑏 are hyperparameters.
According to Han [7], the hyperparameters 𝑎 and 𝑏 must be choose to guarantee that 𝑔(𝛼|𝑎, 𝑏) is a
decreasing function of 𝛼. The derivative of 𝑔 (𝛼│𝑎, 𝑏) with respect to 𝛼 is

𝑑𝑔(𝛼|𝑎, 𝑏) 𝑏𝑎
= [(𝑎 − 1)𝛼 𝑎−2 𝑒 −𝑏𝛼 + 𝛼 𝑎−1 (−𝑏)𝑒 −𝑏𝛼 ]
𝑑𝛼 Γ(𝑎)
𝑑𝑔(𝛼|𝑎, 𝑏) 𝑏 𝑎 𝑎−2 −𝑏𝛼
= 𝛼 𝑒 [(𝑎 − 1) − 𝑏𝛼] (19)
𝑑𝛼 Γ(𝑎)

Thus, for 0 < 𝑎 < 1, 𝑏 > 0, the prior 𝑔 (𝛼│𝑎, 𝑏) is a decreasing function of 𝛼. Assuming that the
hyperparameters 𝑎 and 𝑏 in 𝑔 (𝛼│𝑎, 𝑏) are independent and 𝜋(𝑎, 𝑏) = 𝜋1 (𝑎)𝜋2 (𝑏). The E-Bayesian
estimator of 𝛼 (the expectation of the Bayes estimator of 𝛼) can be written as

𝛼̂𝐸𝐵 = 𝐸(𝛼|𝑋) = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋(𝑎, 𝑏)𝑑𝑎𝑑𝑏 (20)

𝜚

where 𝜚 is the set of all possible values of 𝑎 and 𝑏 where the prior density function is a derivative of
𝛼. 𝛼̂𝐵 (𝑎, 𝑏) is the Bayes estimator of 𝛼 given by (18). For more detail see Han [2].
The E-Bayesian estimates of 𝛼 are derived depending on three different distributions of the
hyperparameters 𝑎 and 𝑏. These distribution are used to investigate the impact of the different prior
distributions on E-Bayesian estimation of 𝛼. The prior distribution used must be conjugate to make it
easier to find the posterior distribution. If the prior distribution is not conjugate, we will have difficulty
to find the posterior distribution.The following distributions of 𝑎 and 𝑏 may be used

2(𝑐 − 𝑏)
𝜋1 (𝑎, 𝑏) = , 0 < 𝑎 < 1, 0 < 𝑏 < 𝑐 (21)
𝑐
1
𝜋2 (𝑎, 𝑏) = , 0 < 𝑎 < 1, 0 < 𝑏 < 𝑐 (22)
𝑐
2𝑏
𝜋3 (𝑎, 𝑏) = 2 , 0 < 𝑎 < 1, 0 < 𝑏 < 𝑐 (23)
𝑐

We can obtained the E-Bayesian estimate of 𝛼 based on 𝜋1 (𝑎, 𝑏), 𝜋2 (𝑎, 𝑏), and 𝜋3 (𝑎, 𝑏) by using (18)
and (21) in (20), (18) and (22) in (20), and (18) and (23) in (20) respectively with 𝛼̂𝐸𝐵 is the E-
Bayesian estimator of 𝛼, 𝛼̂𝐵 (𝑎, 𝑏) is the Bayes estimator of 𝛼 given by (18), and 𝑎, 𝑏 is
hyperparameters. The corresponding E-Bayesian estimate of 𝛼 are respectively.

For 𝜋1 (𝑎, 𝑏),

1 𝑐
𝛼̂𝐸𝐵1 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋1 (𝑎, 𝑏)𝑑𝑏𝑑𝑎 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋1 (𝑎, 𝑏)𝑑𝑏𝑑𝑎
𝜚 0 0
1 𝑐 (𝑟 + 𝑎)
𝑟 2(𝑐 − 𝑏)
= ∫ ∫ [𝜔 ( ) + (1 − 𝑤) ]× 𝑑𝑏𝑑𝑎
0 0 𝑀 (𝑏 + 𝑀) 𝑐2

5
MISEIC 2018 IOP Publishing
IOP Conf. Series: Journal of Physics: Conf. Series 1108 (2018)
1234567890 ‘’“” 012081 doi:10.1088/1742-6596/1108/1/012081

𝑟 2 1 𝑐+𝑀 𝑐+𝑀
𝛼̂𝐸𝐵1 = 𝜔 ( ) + (1 − 𝑤) (𝑟 + ) ( ln ( ) − 1) (24)
𝑀 𝑐 2 𝑠 𝑠

For 𝜋2 (𝑎, 𝑏),

1 𝑐
𝛼̂𝐸𝐵2 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋1 (𝑎, 𝑏)𝑑𝑏𝑑𝑎 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋2 (𝑎, 𝑏)𝑑𝑏𝑑𝑎
𝜚 0 0
1 𝑐 (𝑟 + 𝑎)
𝑟 1
= ∫ ∫ [𝜔 ( ) + (1 − 𝑤) ] × 𝑑𝑏𝑑𝑎
0 0 𝑀 (𝑏 + 𝑀) 𝑐
𝑟 (1 − 𝑤) 1 𝑐+𝑀
𝛼̂𝐸𝐵2 = 𝜔( )+ (𝑟 + ) ln ( ) (25)
𝑀 𝑐 2 𝑀

For 𝜋3 (𝑎, 𝑏),

1 𝑐
𝛼̂𝐸𝐵3 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋1 (𝑎, 𝑏)𝑑𝑏𝑑𝑎 = ∫ ∫ 𝛼̂𝐵 (𝑎, 𝑏)𝜋3 (𝑎, 𝑏)𝑑𝑏𝑑𝑎
𝜚 0 0
1 𝑐 (𝑟 + 𝑎)
𝑟 2𝑏
= ∫ ∫ [𝜔 ( ) + (1 − 𝑤) ] × 2 𝑑𝑏𝑑𝑎
0 0 𝑀 (𝑏 + 𝑀) 𝑐
𝑟 2 1 𝑀 𝑐+𝑀
𝛼̂𝐸𝐵3 = 𝜔 ( ) + (1 − 𝑤) (𝑟 + ) (1 − ln ( )) (26)
𝑀 𝑐 2 𝑐 𝑐

6. Conclusion
In this paper, the E-Bayesian method is used to estimate the 𝛼 parameter of the Lomax distribution in
the right censored data type II. Since E-Bayesian estimation is an expectation of Bayes estimation, in
order to obtain Bayes estimation expectations is by calculating the mean of Bayes estimators, it is
necessary to find the Bayes estimation first. Loss function used in this study is balanced square error
loss function (BSELF) which is a combination of the maximum likelihood estimator of a function and
the Bayes estimator using the square error loss function (SELF).

7. References
[1] K. S. Lomax 1954 J. Am. Stat. Association 49 847
[2] Ming Han 2008 App. Math. Model. 33 1915
[3] Jafar Ahmadi, M. J. Jozani, Eric Marchand & Ahmad Parsian 2008 J. Stat. Plan. Inference 1180
[4] Arnold Zellner 1994 Statistical Desicion Theory and Methods V 337-390
[5] G. E. P. Box & G. C. Tiao 1973 Bayesian inference in statistical analysis (Philippines: Addison-
Wesley Publishing Company, Inc)
[6] Enrique D. Castillo 2007 Process optimization a statistical approach (New York: Springer
Science + Business Media, LCC)
[7] Ming Han 1997 The structure of hierarchical prior distribution and its applications 31-40
[8] Essam K. Al-Hussaini & Mohamed Hussein 2011 Open J. of Stat 28-38
[9] Hassan M. Okasha 2013 J. Egyptian Math. Soc. 489
[10] Hesham M. Reyad & Soha O. Ahmed 2014 Int. J. Adv. Math. Sci. 3 108-120

Acknowledgments
We would like to thank the DRPM Universitas Indonesia for supporting in our research with Hibah
PITTA UI 2018.