a topp-leone generator of exponentiated power lindley
TRANSCRIPT
Applied Mathematical Sciences, Vol. 12, 2018, no. 12, 567 - 579
HIKARI Ltd, www.m-hikari.com
https://doi.org/10.12988/ams.2018.8454
A Topp-Leone Generator of Exponentiated Power
Lindley Distribution and Its Application
Sirinapa Aryuyuen
Department of Mathematics and Computer Science, Faculty of Science and
Technology, Rajamangala University of Technology Thanyaburi
Phatum Thani 12110, Thailand
Copyright © 2018 Sirinapa Aryuyuen. This article is distributed under the Creative Commons
Attribution License, which permits unrestricted use, distribution, and reproduction in any medium,
provided the original work is properly cited.
Abstract
A new framework for generating lifetime distributions is proposed, which is
called the Topp-Leone Exponentiated Power Lindley (TL-EPL) distribution. Sub-
models of the TL-EPL distribution, such as the Topp-Leone Power Lindley,
Topp-Leone Generalized Lindley, and Topp-Leone Lindley, are introduced. Some
statistical characteristics of the distributions are investigated (i.e., mean, variance,
and functions of survival, hazard, and quantile). The maximum likelihood
estimation is used to estimate the parameters of each distribution. Some real data
sets are fitted in order to illustrate the usefulness of the proposed distribution.
Keywords: Topp-Leone, exponentiated power Lindley, quantile, hazard,
lifetime data
1 Introduction
The Exponentiated Power Lindley (EPL) distribution with three parameters, as
proposed by Warahena-Liyanage and Pararai [18], will provide many applications
in different fields, such as engineer, biology and others. All parameters of EPL
distribution indexed to this distribution makes it more flexible in describing
different types of real data than its sub-models, i.e., the power Lindley (PL)
distribution [5], the generalized Lindley (GL) distribution [11], and the Lindley
(L) distribution [9]. The EPL distribution, due to its flexibility in accommodating
different forms of the hazard function, seems to be a suitable distribution that can
be applied to a variety of problems in fitting survival data. Ashour and Eltehiwy
568 Sirinapa Aryuyuen
[2] applied the EPL distribution to real data about flood levels and the active
repair times (h) for an airborne communication transceiver. The EPL distribution
provides better fit than the other distributions (i.e., PL, GL, L, exponential,
modified Weibull, Weibull distributions). Hence the data point from the EPL
distribution has better relationship and hence this distribution is good model for
life time data.
The EPL distribution is specified in terms of the probability density function
(pdf), cumulative density function (cdf), and quantile function [18], i.e.,
respectively
12
1( ) (1 ) 1 (1 )e e ,1 1
x xxg x x x
(1)
( ) 1 (1 )e ,1
xxG x
(2)
1/
1/1 1( ) 1 ( 1)(1 )e ,x
GQ u W u
(3)
where 0,x 0, 0 and 0. When u is distributed as the uniform on the
interval (0,1), and W (.) in the Lambert W function [3]. The probability functions
of the sub-models of the EPL distribution are presented. For 1 , we have the
pdf and cdf of the PL distribution with positive parameters and , i.e.,
2
1
1( ) (1 ) e1
xg x x x
and 1( ) 1 (1 )e .1
xG x x
For 1 , we have the pdf and cdf of the GL distribution with positive parameters
and , i.e., respectively,
12
2
(1 )( ) 1 (1 )e e
1 1
x xx xg x
and
2 ( ) 1 (1 )e .1
xG x x
For 1, we have the pdf and cdf of the L distribution with positive
parameter , i.e., respectively,
2
3( ) (1 )e1
xg x x
and 3
1( ) 1 e .
1
xxG x
A Topp-Leone generator of exponentiated power Lindley distribution 569
Statistical distributions are used to model the life of an item in order to study
its important properties. Proper distribution may provide useful information that
result in sound conclusions and decisions. When there is a need for more flexible
distributions, many researchers are about to use the new one with more
generalization. Recently, applying new generators for continuous distributions
became more interesting [14]. This methodology can improve on the goodness of
fit and determine tail properties. These features have been established by the
results of many generators such as the beta distribution and Topp-Leone (TL)
distribution.
The TL distribution is one of the continuous distributions that is attractive as a
generator. This distribution was proposed by Topp and Leone [16] for empirical
data with J-shaped histograms, such as powered band tool and automatic
calculating machine failures. Let T be a random variable which is distributed as
the TL distribution with parameter , which is denoted by T ~TL( ), the cdf,
and pdf of T respectively,
TL ( ) (2 )F t t t and
1 1
TL
2 (1 )(2 )( ) ,
1 (2 )
t t tf t
t t
(4)
where 0 1t and 0. Solving for the quadratic equation of 2 1/2 0t t u
(see [7]) obtains the quantile function of TL distribution, i.e.,
1 1/
TL TL( ) ( ) 1 1 ,Q t F u u (5)
where u is distributed as the uniform on the interval (0,1).
The TL distributions as a generator for continuous distributions are proposed
such as two-sided generalized Topp and Leone distribution [17], Topp-Leone
generalized exponential distribution [14], Topp-Leone Gumbel distribution [19],
and Generalized Topp-Leone family of distributions [10]. Generating a new
family of distribution requires two principal components, which are a generator
and a parent distribution (see [1, 7, 14]). Indeed, the pdf of a generator is
transformed into a new pdf through the cdf ( )G of a parent distribution. The TL
distribution is a continuous unimodal distribution with a wide range of
applications in reliability fields and is used for modeling lifetime phenomena,
which has a J-shaped density function with a bathtub-shaped hazard function [15].
In this article is proposed a new Topp-Leone generated family distribution
where the parent distribution is the EPL distribution. Sub-models of the proposed
distribution are studied. Some statistical characteristics of the distributions are
investigated. The maximum likelihood estimation (MLE) is used to estimate the
parameters of each distribution. Some real data sets are presented in order to
illustrate that the data fits by using the proposed distribution.
2 A Topp-Leone generated family distribution
Alzaatreh et al. [1] presented a method for generating a new family of distribution
570 Sirinapa Aryuyuen
with the following definition of a random variable 1 2[ , ],T c c
1 2 ,c c
and a random variable X with cdf ( ).G x Let W[ ( )]G x be a function of ( )G x and
satisfy these conditions: a) 1 2W[ ( )] [ , ],G x c c b) W[ ( )]G x is differentiable and
monotonically non-decreasing, and c) W[ ( )]G x a as x and W[ ( )]G x
b as .x Let T be a random variable of a generator distribution with pdf ( )r t
defined as 1 2[ , ].c c Let X be a continuous random variable with cdf ( ).G x Thus,
the cdf and pdf of a new family of distributions are given, respectively, as,
W[ ( )]
TL-G0
( ) ( ) d G x
F t r t t and TL-G ( ) {W[ ( )]} W[ ( )] .
df t r G x G x
dx
If a random variable T is distributed as the TL and bounded on [0,1]. Let X
be a continuous random variable with the TL-G distribution. The cdf can written
as,
TL-G ( ) ( ) [2 ( )] ,F x G x G x (6)
where 0 is a shape parameter. The associated pdf is,
1 1
TL-G ( ) 2 ( )[1 ( )] ( ) [2 ( )] ,f x g x G x G x G x (7)
where ( ) ( ) .g x dG x dx In addition, a TL random variable with finite support has
the same bounds as the cdf ( )G x of any other random variable. Therefore, the
relation of a random variable X having the TL distribution is 1( )X G T where
T ~ TL( ). Let G ( )Q be the quantile function of a parent distribution, by which
can be simulated the TL-G random variate from
1/
G (1 1 ),x Q u (8)
where 1/1 1 u is the quantile function of the TL distribution in (5). The
results obtained in Section 3 can be a new TL-G distribution.
The moment of the TL-G distribution can be computed from the probability
weighted moments order ( ,s r ) of the parent distribution. Let ( )G x be the cdf of
the parent distribution, then the ( , )s r th probability weighted moment of X (see
[12, 14]) will be,
1
, G
0
( ) ( ) ( )d ( ) d .s r s r s r
s r E X G X x G X g x x Q u u x
The moment of the TL-G distribution is,
G , 1
1
E ( ) ( 1) 2 ( ) .s i i
s i
i
X ii
(9)
A Topp-Leone generator of exponentiated power Lindley distribution 571
3 A new TL-G distribution
In this section, we introduce a new distribution, which is called the Topp-Leone
exponential power Lindley (TL-EPL) distribution.
3.1 The TL-EPL distribution
From the cdf in (6) and pdf in (7) of the TL-G distribution, let ( )g x and ( )G x be
the pdf and cdf of the EPL distribution (see, [2, 16]). Consequently, a random
variable X of the TL-EPL distribution, X ~TL-EPL( , , , ), has the cdf and
pdf as follows;
TL-EPL ( ) 1 (1 )e 2 1 (1 )e ,1 1
x xx xF x
(10)
1
21
TL-EPL
2( ) (1 ) 1 (1 )e e
1 1
x xxf x x x
1
1 1 (1 )e 2 1 (1 )e .1 1
x xx x
(11)
We define the hazard function of the TL-EPL distribution as follows,
1
TL-EPL( ) 1 (1 )e e 1 1 (1 )e1 1
x x xx xH x
12
1
TL-EPL
2(1 ) 2 1 (1 )e ( ) ,
1 1
xxx x S x
where TL-EPL ( )S x is the survival function of the TL-EPL distribution, i.e.,
TL-EPL ( ) 1 1 (1 )e 2 1 (1 )e .1 1
x xx xS x
The quantile function of the TL-EPL distribution is obtained by substituting
(3) for (8), i.e.,
1/
1/ 1/ 1
TL-EPL
1 1( ) 1 ( 1) 1 (1 1 ) e ,Q u W u
(12)
where u is distributed as the uniform on the interval (0,1), and ( )W is the
Lambert W function [3]. From the moment of the TL-G distribution in (9) and the
572 Sirinapa Aryuyuen
quantile function of the EPL distribution in (3), we obtain the mean and variance
of ,X i.e., respectively,
TL-EPL 1, 1
1
E ( ) ( 1) 2 ( ) ,i i
i
i
X ii
and
2
TL-EPL 2, 1 1, 1
1 1
( ) ( 1) 2 ( ) ( 1) 2 ( ) ,i i i i
i i
i i
V X i ii i
where
1/1
1/ 1
1, 1
0
1 11 ( 1)(1 )e d ,x i
i W u u x
and
2/1
1/ 1
2, 1
0
1 11 ( 1)(1 )e d .x i
i W u u x
Some pdf plots of the TL-EPL distribution are shown in Figure 1 and Figure 2.
Figure 1: Plots of the pdf of the TL-EPL distribution with (a) different values
of and (b) different values of
Figure 2: Plots of the pdf of the TL-EPL distribution with (a) different values
of and (b) different values of
A Topp-Leone generator of exponentiated power Lindley distribution 573
The pdf of the TL-EPL distribution is unimodal. It increases or decreases for
various values of the parameters giving the shapes obtained in Figure 2(a). In
Figure 1(b), for values of 1 and 1 the pdf seems almost symmetric.
3.2 Sub-models
The EPL distribution with three parameters , , and has three sub-models
(see [2, 16]). Thus, the TL-EPL distribution has three sub-models, which are
presented as follows.
3.2.1 The Topp-Leone power Lindley distribution
For X ~TL-EPL( , , , ), when 1 , we obtain the Topp-Leone power
Lindley (TL-PL) distribution, which is denoted by X ~TL-PL( , , ). The TL-
PL distribution has the cdf and pdf, i.e., respectively,
2 2
TL-PL ( ) 1 (1 ) e ,1
xxF x
(13)
12
1 2 2 2
TL-PL
2( ) (1 ) (1 ) 1 (1 ) e e .
1 1 1
x xx xf x x x
(14)
3.2.2 The Topp-Leone generalized Lindley distribution
For X ~TL-EPL( , , , ), when 1 , we obtain the Topp-Leone generalized
Lindley (TL-GL) distribution, which is denoted by X ~ TL-EL ( , , ). The cdf
and pdf of the TL-GL distribution are
TL-GL ( ) 1 (1 )e 2 1 (1 )e ,1 1
x xx xF x
(15)
12
TL-GL
2( ) (1 ) 1 (1 )e e
1 1
x xxf x x
1
1 1 (1 )e 2 1 (1 )e .1 1
x xx x
(16)
3.2.3 The Topp-Leone Lindley distribution
For X ~TL-EPL( , , , ), when 1 , we obtain the Topp-Leone Lindley
(TL-L) distribution, which is denoted by X ~TL-L( , ). The TL-L distribution
has the cdf and pdf in (17) and (18), respectively,
2 2
TL-PL ( ) 1 (1 ) e ,1
xxF x
(17)
574 Sirinapa Aryuyuen
121 2 2 2
TL-PL
2( ) (1 ) (1 ) 1 (1 ) e e .
1 1 1
x xx xf x x x
(18)
3.4 Maximum likelihood estimation
In this section, we describe the MLE procedure to obtain the estimated value of
the parameters of the TL-EPL based on the random sample 1 2( , ,..., )nx x x x of
size .n Let iX for 1,2,...,i n be independent and identically distributed. The
log-likelihood function of iX ~TL-EPL( , , , ), on the observed sample x is
log ( , , , | ) ( | )L x x given by,
1 1 1
( | ) log(1 ) ( 1) log log 2 log logn n n
i i i
i i i
x x x x n n n
( , ; )
1
2 log log log( 1) ( 1) log 2i
n
x
i
n n n
( , ; ) ( , ; )
1 1
( 1) log log 1 ,i i
n n
x x
i i
(19)
where ( , ; ) 1 (1 ( 1))e .i
i
x
x ix
By differentiating ( | ),x the partial derivatives of ( | )x with respect to
, , and are given by,
( , ; ) ( , ; )
1 1
( | )log 2 log ,
i i
n n
x x
i i
x n
(20)
( , ; )
1 1
( | ) 2( 1) log 2
1 i
n n
x i
i i
x n nx
( , ; ) ( , ; )
1 1
log 1 ( 1) log ,i i
n n
x x
i i
(21)
1 1 1
( | )log(1 ) log
n n n
i i i
i i i
xx x x
( , ; ) ( , ; )
1 1
( 1) log log 1 ,i i
n n
x x
i i
n
(22)
( , ; ) ( , ; )
1 1
( | )log log 1
i i
n n
x x
i i
x
( , ; )
1
( 1) log 2 .i
n
x
i
n
(23)
A Topp-Leone generator of exponentiated power Lindley distribution 575
The expression of these differential equations in (19)-(23) are not in the closed
form. In this study, the parameter estimates of ˆ ˆˆ , , and ˆ , can be obtained by
using the numerical optimization with the nlm function in the R language [13],
which R code for the MLE of the TL-EPL distribution are as follows:
#The TL exponentiated power Lindley (TL-EPL) distribution x <- c(...,…,…) logL_TL_EPL <- function(theta0,x) { beta <- (theta0[1]) lambda <-(theta0[2]) omega <- (theta0[3]) alpha <- (theta0[4]) G <- (1-(1+(beta*x^lambda)/(beta+1))*exp(-beta*x^lambda))^omega g1 <- ((lambda*beta^2*omega)/(beta+1))*(1+x^lambda) *(x^(lambda-1))*exp(-beta*x^lambda) g2 <- (1-(1+(beta*x^lambda)/(beta+1)) *exp(-beta*x^lambda))^(omega-1) g <- g1*g2 logL <- -log(2*alpha)-log(g)-log(1-G)-(alpha-1)*log(G)-(alpha-1) *log(2-G) return(sum(logL)) } theta0 <- c(...,...,...,...) Est<-nlm(logL_TL_EPL,theta0,x) 3.5 Application study In this section we provide a data analysis in order to assess the goodness-of-fit of
the TL-EPL model with two real data sets. In addition, the sub-models of the TL-
EPL distribution (i.e., TL-PL, TL-GL, and TL-L distributions), and the EPL
distribution and its sub-models (e.g., PL, GL, L distribution) are considered. The
parameter (s) of each distribution are estimated by the MLE method. To verify
which distribution fits better with real data sets, the Komogorov-Smirnov test (KS
test) will be employed. Other criteria including the Akaike Information Criterion
(AIC) and Bayesian Information Criterion (BIC) are considered, i.e., AIC
2 2log L p and BIC 2log( ) log( ),L p n where n is the sample size and
p is the number of parameters of each distribution.
The first dataset consists of 20 observations with the respect to maximum
flood level data to see how the new model works in practice. The data has been
obtained from Dumonceaux and Antle (see [2, 4]), as shown in Table 1. For the
results of Table 2 , the KS test indicates that the TL-EPL distribution is a strong
competitor compared to other distributions.
576 Sirinapa Aryuyuen
Table 1. Maximum flood levels data from Dumonceaux and Antle [4].
0.654, 0.613, 0.315, 0.449, 0.297, 0.402, 0.379, 0.423, 0.379, 0.3235
0.269, 0.740, 0.418, 0.412, 0.494, 0.416, 0.338, 0.392, 0.484, 0.265
Table 2. Values of parameter estimates and Statistical criteria concerning
maximum flood level data.
Distribution Parameter estimate values Statistical values
-Log AIC BIC KS p-value
TL-EPL 4.5801 7.7322 0.6639 21.0802 16.1471 40.2942 47.0497 0.1171 0.9466
TL-PL 547.9893 6.7696 0.6472 - 16.2989 38.5978 43.6644 0.1247 0.9148
TL-GL 109.269 6.0083 - 0.6412 16.1725 38.3450 43.4116 0.1171 0.9467
TL-L 53.0705 6.1417 - - 16.1462 36.2924 39.6702 0.1228 0.9236
EPL - 13.2699 0.6129 847.2759 16.3058 38.6116 43.6782 0.1284 0.8968
PL - 15.2883 3.5207 - 13.2487 30.4974 33.8752 0.1991 0.4060
GL - 11.6817 - 54.9888 16.1547 36.3094 39.6872 0.1223 0.9258
L - 2.9602 - - 2.1792 6.3584 8.0473 0.4532 0.0005
For a second application, we analyze a real data set on the active repair times
(h) for an airborne communication transceiver. The data is given in Table 3, and
its source is Jorgensen (see [2, 8]). For the KS test in Table 4, the TL-EPL
distribution is the best distribution corresponding to a high p-value of the KS test.
Table 3. Active repair time (h) for an airborne communication transceiver [8].
0.50, 0.60, 0.60, 0.70, 0.70, 0.70, 0.80, 0.80, 1.00, 1.00, 1.00, 1.00, 1.10, 1.30,
1.50, 1.50, 1.50, 1.50, 2.00, 2.00, 2.20, 2.50, 2.70, 3.00, 3.00, 3.30, 4.00, 4.00,
4.50, 4.70, 5.00, 5.40, 5.40, 7.00, 7.50, 8.80, 9.00, 10.20, 22.00, 24.50
Table 4. Values of parameter estimates and Statistical criteria of the active repair
time data.
Distribution Parameter estimate values Statistical values
-Log AIC BIC KS p-value
TL-EPL 9.1321 7.0096 0.0966 300.5382 89.6808 187.3616 194.1171 0.0950 0.8630
TL-PL 2,755.4010 4.3244 0.1390 - 89.5073 185.0146 190.0812 0.0955 0.8592
TL-GL 1,025.1370 0.1403 - 0.0109 94.0351 194.0702 199.1368 0.1525 0.3097
TL-L 0.6692 0.2074 - - 98.9111 201.8222 205.2000 0.1678 0.2100
EPL - 8.9099 0.1293 6,196.2090 89.4702 184.9404 190.007 0.0959 0.8552
PL - 0.5867 0.7988 - 95.9425 195.885 199.2628 0.1346 0.4637
GL - 0.3588 - 0.7460 97.9109 199.8218 203.1996 0.166 0.2201
L - 0.4242 - - 98.7913 199.5826 201.2715 0.2157 0.0484
A Topp-Leone generator of exponentiated power Lindley distribution 577
4 Conclusion
The Topp-Leone Exponentiated Power Lindley (TL-EPL) distribution is
proposed, which has the Topp-Leone Power Lindley, Topp-Leone Generalized
Lindley, and Topp-Leone Lindley, are sub-model. Some statistical characteristics
of the distributions are investigated. The maximum likelihood estimation is used
to estimate the parameters of each distribution. We provide a data analysis in
order to assess the goodness-of-fit of the TL-EPL model with two real data sets.
The results of KS test indicates that the TL-EPL distribution is a strong
competitor compared to other distributions (i.e., TL-PL, TL-GL, TL-L, PL, GL,
and L distributions).
Acknowledgements. We wish to gratefully acknowledge the referee of this paper
who helped to clarify and improve its presentation.
References
[1] A. Alzaatreh, C. Lee and F. Famoye, A new method for generating families
of continuous distributions, METRON, 71 (2013), 63 - 79.
https://doi.org/10.1007/s40300-013-0007-y
[2] S.K. Ashour and M.A. Eltehiwy, Exponentiated power Lindley distribution,
Journal of Advanced Research, 6 (2015), 895 – 905.
https://doi.org/10.1016/j.jare.2014.08.005
[3] R.M. Corless, G.H. Gonnet, D.E.G. Hare, D.J. Jeffrey and D.J. Knuth, On
the Lambert W function, Adv. Comput. Math., 5 (1996), 329 – 359.
https://doi.org/10.1007/bf02124750
[4] R. Dumonceaux and C. Antle, Discrimination between the Log-Normal and
the Weibull distributions, Technometrics, 15 (2012), 923 – 926.
https://doi.org/10.1080/00401706.1973.10489124
[5] M.E. Ghitany, D.K. Al-Mutairi, N. Balakrishnan and LJ. Al-Enezi, Power
Lindley distribution and associated inference, Comput. Stat. Data Anal., 64
(2013), 20 – 33. https://doi.org/10.1016/j.csda.2013.02.026
[6] J.A Greenwood, J. Landwehr, N. Matalas and J. Wallis, Probability weighted
moments: Definition and relation to parameters of several distributions
expressible in inverse form, Water Resources Research, 15 (1979), 1049 -
1054. https://doi.org/10.1029/wr015i005p01049
[7] M.C. Jones, Families of distributions arising from distributions of order
statistics. TEST, 13 (2004), 1 – 43. https://doi.org/10.1007/bf02602999
578 Sirinapa Aryuyuen
[8] B. Jorgensen, Statistical Properties of the Generalized Inverse Gaussian
Distribution, New York: Springer-Verlag, 1982.
https://doi.org/10.1007/978-1-4612-5698-4
[9] D.V. Lindley, Fiducial distributions and Bayes’ theorem, JR Stat. Soc. Ser.
A, 20 (1958), 102 – 107.
[10] A. Mahdavil, Generalized Topp-Leone family of distributions, J. Biostat.
Epidemiol., 3 (2017). 65 – 75.
[11] S. Nadarajah, H.S. Bakouch and R.A. Tahmasbi, А Generalized Lindley
distribution, Sankhya B, 73 (2011), 331 - 359.
https://doi.org/10.1007/s13571-011-0025-9
[12] S. Nadarajah and S. Kotz, Moments of some J-shaped distributions, Journal
of Applied Statistics, 30 (2003), 311 – 317.
https://doi.org/10.1080/0266476022000030084
[13] R Core Team, A Language and environment for Statistical computing, R
Foundation for Statistical Computing, Vienna, Austria, (2016).
https://www.R-project.org/
[14] Y. Sangsanit and W. Bodhisuwan, The Topp-Leone generator of
distributions: properties and inferences, Songklanakarin J. Sci. Technol., 38
(2016), 537 – 548.
[15] H. Sultan and S.P. Ahmad, Bayesian Analysis of Topp-Leone Distribution
under different loss functions and different priors, J. Stat. Appl. Pro. Lett., 3
(2016), 109 – 118. https://doi.org/10.18576/jsapl/030302
[16] C.W. Topp and F.C. Leone, A family of J-shaped frequency functions,
Journal of the American Statistical Association, 50 (1995), 209 – 219.
https://doi.org/10.1080/01621459.1955.10501259
[17] D. Vicaria, J.R.V Dorpb and S. Kotzb, Two-sided generalized Topp and
Leone (TS-GTL) distributions, Journal of Applied Statistics, 35 (2018),
1115 – 1129. https://doi.org/10.1080/02664760802230583
[18] G. Warahena-Liyanage and M. Pararai, A Generalized power Lindley
distribution with applications, ASIAN Journal of Mathematics and
Applications, (2014), Article ID ama0169, 23.
http://scienceasia.asia
A Topp-Leone generator of exponentiated power Lindley distribution 579
[19] W. Bodhisuwan, The Topp-Leone Gumbel Distribution, 12th International
Conference on Mathematics, Statistics, and Their Applications (ICMSA),
Banda Aceh, Indonesia. (2016). https://doi.org/10.1109/icmsa.2016.7954316
Received: April 16, 2018; Published: May 23, 2018