Pak. J. Statist.
2016 Vol. 32(3), 183-206
THE TRANSMUTED WEIBULL-PARETO DISTRIBUTION
Ahmed Z. Afify1, Haitham M. Yousof1, Nadeem Shafique Butt2
and G.G. Hamedani3
1
Department of Statistics, Mathematics and Insurance
Benha University, Egypt.
Email: ahmed.afify@fcom.bu.edu.eg
haitham.yousof@fcom.bu.edu.eg
2
Department of Family and Community Medicine
Faculty of Medicine in Rabigh, King Abdulaziz University
Jeddah, Saudi Arabia. Email: nshafique@kau.edu.sa
3
Department of Mathematics, Statistics and Computer Science
Marquette University, Milwaukee, USA.
Email: gholamhoss.hamedani@marquette.edu
ABSTRACT
A new generalization of the Weibull-Pareto distribution called the transmuted
Weibull-Pareto distribution is proposed and studied. Various mathematical properties of
this distribution including ordinary and incomplete moments, quantile and generating
functions, Bonferroni and Lorenz curves and order statistics are derived. The method of
maximum likelihood is used for estimating the model parameters. The flexibility of the
new lifetime model is illustrated by means of an application to a real data set.
KEY WORDS
Transmuted family, Weibull-Pareto Distribution, Bonferroni and Lorenz curves,
Order Statistics, Likelihood Estimation.
1. INTRODUCTION
Recently, several families of distributions have been proposed via extending common
families, by adding one or more parameters to the baseline model, of continuous
distributions. These new families provide more flexibility in modeling and analyzing real
life data in many applied areas. For that reason the statistical literature contains a good
number of new families. For example, the generalized transmuted-G family proposed by
Nofal et al. (2015), the transmuted exponentiated generalized-G class defined by Yousof
et al. (2015), the transmuted geometric-G family introduced by Afify et al. (2016a) and
the Kumaraswamy transmuted-G family introduced by Afify et al. (2016b). Another
example is the Weibull-G (W-G for short) class defined by Bourguignon et al. (2014).
Using the W-G family, Tahir et al. (2015) defined and studied the Weibull-Pareto (WPa)
distribution extending the Pareto (Pa) distribution.
The cumulative distribution function (cdf) of the WPa distribution is given
(for x > > 0 ) by
© 2016 Pakistan Journal of Statistics
183
184
The Transmuted Weibull-Pareto Distribution
b
x
G ( x) = 1 exp 1 ,
(1.1)
where is a positive scale parameter and and b are positive shape parameters.
The corresponding probability density function (pdf) is given by
x
g ( x) = x 1 1
b
b 1
b
x
exp 1 .
(1.2)
In this paper, we define and study a new distribution by adding one extra shape
parameter in equation (1.2) to provide more flexibility to the generated model. In fact,
based on the transmuted-G (TG) family of distributions proposed by Shaw and Buckley
(2007), we construct a new model called the transmuted Weibull-Pareto (TWPa)
distribution and give a comprehensive descriptions of some of its mathematical
properties. We hope that the new model will attract wider applications in reliability,
engineering and other areas of applications.
Recently, many authors used the TG family to construct new distributions. For
example, Afify et al. (2015a) introduced the transmuted Marshall-Olkin Fréchet, Afify et
al. (2015b) proposed the transmuted Weibull Lomax, Afify et al. (2014) defined the
transmuted complementary Weibull geometric, Khan and King (2013) introduced the
transmuted modified Weibull and Aryal and Tsokos (2011) proposed the transmuted
Weibull distributions.
For an arbitrary baseline cdf G x , Shaw and Buckley (2007) defined the TG family
F x = G x 1 G x
with cdf and pdf given by
and
f x = g x 1 2G x ,
(1.3)
(1.4)
respectively, where 1 . The TG density is a mixture of the baseline density and the
exponentiated-G (Exp-G) density with power parameter two. For = 0 , (1.3) gives the
baseline distribution.
The rest of the paper is outlined as follows. In Section 2, we define the TWPa
distribution and provide the graphical presentation of its pdf and hazard rate function
(hrf). A useful mixture representation for its pdf and cdf is provided in Section 3.
Section 4 provides the mathematical properties including ordinary and incomplete
moments, quantile and generating functions, Bonferroni, Lorenz and Zenga curves,
moments of residual life and moments of the reversed residual life are derived. In
Section 5, the order statistics and their moments are discussed. The probability weighted
Afify, Yousof, Butt and Hamedani
185
moments (PWMs) are discussed in Section 6. Certain characterizations are presented in
Section 7. The maximum likelihood estimates (MLEs) for the model parameters are
demonstrated in Section 8. In Section 9, simulation results to assess the performance of
the proposed maximum likelihood estimation method are discussed. The TWPa
distribution is applied to a real data set to illustrate its usefulness in Section 10. Finally,
some concluding remarks are given in Section 11.
2. THE TWPa DISTRIBUTION
By inserting Equation (1.1) into Equation (1.3), the cdf of the TWPa model is given
(for x > ) by
b
b
x
x
F ( x) = 1 exp 1 1 exp 1 .
(2.1)
The corresponding pdf is obtained by
f x =
b
x
1
x
1
b 1
b
b
x
x
exp 1 1 2 exp 1 ,
(2.2)
where is a scale parameter and , b and are positive shape parameters. The
random variable X is said to have a TWPa distribution, denoted by X TWPa
(, , b, ) , if its cdf is given by Equation (2.1). It is clear that Equation (2.2) reduced to
the WPa model for = 0 . The TWPa distribution due to its flexibility in accommodating
all forms of the hrf as shown from Figure 2 seems to be an important model that can be
used.
A physical interpretation of the cdf of TWPa model is possible if we take a system
consisting of two independent components functioning independently at a given time. So,
if the two components are connected in parallel, the overall system will have the TWPa
cdf with = 1 .
Another motivation for the TWPa distribution follows by taking two iid random
variables, say Z1 and Z 2 , with cdf
b
x
G x = 1 exp 1 .
Let Z1:2 = min(Z1 , Z2 ) and Z2:2 = max(Z1 , Z2 ) . Now, consider the random variable
X defined by
186
The Transmuted Weibull-Pareto Distribution
1
;
Z1:2 , with probability
2
X =
Z , with probability 1 .
2:2
2
Then, the cdf of X is given by (2.1).
Figure 1 provide some plots of the TWPa density curves for different values of the
parameters , , and b . Plots of the hrf of TWPa for selected parameter values are
given in Figure 2. Some possible shapes for the TWPa cdf are displayed in Figure 3.
Figure 1: Plots of the TWPa Density Function for some Parameter Values
Afify, Yousof, Butt and Hamedani
Figure 2: Plots of the hrf of the TWPa for some Parameter Values
187
188
The Transmuted Weibull-Pareto Distribution
Figure 3: Plots of the cdf of the TWPa for some Parameter Values
Afify, Yousof, Butt and Hamedani
189
3. MIXTURE REPRESENTATION
The TWPa density function given in Equation (2.2) can be expressed as
b
G ( x)
f ( x) = 1 bg ( x)
exp
G ( x)b 1
G ( x)
b
b
G ( x)b 1
G ( x)b 1
G ( x)
G ( x)
2bg ( x)
exp
2
bg
(
x
)
exp
2
.
G ( x)b 1
G ( x)b 1
G ( x)
G ( x)
G ( x)b 1
1k k 1 b j 1 x
1
k!j! k 1 b 1
k , j =0
Using Equations (1.1) and (1.2) and after a power series, the we have
x
f x = 1 b
x
x
2b
x
x
2b
x
1k k 1 b j 1 x
1
k!j! k 1 b 1
k , j =0
bk b j 1
2 k k 1 b j 1 x
1
k!j! k 1 b 1
k , j =0
bk b j 1
bk b j 1
.
(3.1)
Using the generalized binomial series expansion and after some algebra, the TWPa
density can be expressed as
1k b k 1 b j 1 1 2 2k 1
f x =
k!j! k 1 b 1 k 1 b j
k , j =0
x
k 1 b j
x
or equivalently
x
1
k 1b j 1
f ( x) = k , j h,, k 1b j ( x),
(3.2)
k , j =0
where
and
1k b k 1 b j 1 1 2k 1 1
k , j =
.
k!j! k 1 b j k 1 b 1
x
h,, k 1b j ( x) = k 1 b j
x
,
x
1
k 1b j 1
190
The Transmuted Weibull-Pareto Distribution
is the exponentiated Pareto (EPa) density with parameters , and k 1 b j . This
means that the TWPa density can be expressed as a mixture of EPa densities. So, several
of its properties can be derived form those of the EPa model.
Integrating (3.2), the cdf of TWPa can be expressed as
F ( x) = k , j H ,, k 1b j ( x),
where H ,, k 1b j ( x) is the cdf of EPa with parameters , and k 1 b j .
k , j =0
4. PROPERTIES
Established algebraic expansions to determine some structural properties of the TWPa
distribution can be more efficient than computing those directly by numerical integration
of its density function. The mathematical properties of the TWPa distribution including
ordinary and incomplete moments, factorial moments, quantile and generating functions,
Bonferroni, Lorenz and Zenga curves, residual life function and reversed residual life
function are provided in this section.
4.1 Moments
Using (2.2), the r th moment of X , denoted by r , is given by
r = k , j E Yr,, k 1b j ,
k , j =0
where E Yr,, k 1b j = 0 xr h,, k 1b j ( x)dx . Then, we obtain (for r )
r
r = k 1 b j r k , j B 1 , k 1 b j ,
k , j =0
where B m, n = 0t m1 1 t
1
n 1
(4.1)
dt is the beta function.
The skewness and kurtosis measures can be calculated from the ordinary moments
using well-known relationships.
Corollary 1
Using the relation between the central moments and non-centeral moments, we can
obtain the n th central moment of a TWPa random variable, denoted by M n , as follows
n n
n
n r
M n = E X = 1 E X r ,
r =0 r
then,
Afify, Yousof, Butt and Hamedani
191
n n
nr
nr
M n = 1 1 r
r =0 r
and cumulants ( n ) of X are obtained from (4.1) as
n 1 n 1
n = n
r n r ,
r =0 r 1
where 1 = 1 hence 2 = 2 12 , 3 = 3 321 13 etc. The skewness and kurtosis
measures can be calculated from the ordinary moments using well-known relationships.
TWPa (, , , b) , is obtained by
4.2 Quantile and Generating Functions
The quantile function (qf) of X , where X
inverting (2.1) to obtain xu = F 1 u as
1
xu = 1 log
1
2
2
1/
1/ b
4u
.
Simulating the TWPa random variable is straightforward. If U is a uniform variate
on the unit interval (0,1) , then the random variable X = xU follows (2.2).
Here, we will provide two formulas for the moment generating function (mgf) of the
TWPa distribution. The mgf is defined by M X t = r =0
M X t =
t r
tr
E X r . Then
r!
k , j B 1 , k 1 b
k 1 b j
r!
k , j , r =0
r
j .
The second can be computed using Maple. For t 0 , p > 0 and q > 0 Let
J q, p, t = q x p etx dx . Using this software, we can obtain
csc p p p
p p, tq
etq
p
J q, p, t = 1 q
qt
tq
tq p
tq p 1
and
M X t = dk , j J , m 1 1, t ,
m =0
where ( z, s) = s y z 1 e y dy is the complementary incomplete gamma function and
d k , j =
m 1
k 1 b j 1
k , j k 1 b j
.
m
k , j =0
1m
192
The Transmuted Weibull-Pareto Distribution
4.3 Incomplete Moments
The s th incomplete moments, say s t , is given by
s t = 0x s f x dx.
t
Using equation (3.2) and the lower incomplete beta function, we obtain (for s )
s
s t = s k , j k 1 b j Bt
, k 1 b
k , j =0
where
Bz a, b = 0 wa 1 1 w
b 1
z
j ,
(4.2)
dw is the incomplete beta function. The first
incomplete moment of the TWPa distribution can be obtained by setting s = 1 in (4.2).
The main application of the first incomplete moment refers to Bonferroni and Lorenz
curves. These curves are very useful in economics, reliability, demography, insurance
and medicine. The answers to many important questions in economics require more than
just knowing the mean of the distribution, but its shape as well. This is obvious not only
in the study of econometrics but in other areas as well.
The important application of the first incomplete moment is related to the Lorenz and
Bonferroni curves. These curves are very useful in economics, reliability, demography,
insurance and medicine. The Lorenz curve, say LF x , and Bonferroni curve, say
B F x , are defined, respectively, by
LF x =
LF x
1
1
x
x
.
0 tf t dt and B F x =
0 tf t dt =
EX
E X F x
F x
Then,
1
, k 1 b j
k , j =0
LF x =
r
k , j B 1 , k 1 b j
k , j =0
and
1
, k 1 b j
k , j =0
B F x =
.
r
F x k , j B 1 , k 1 b j
k , j =0
k , j Bt 1
k , j Bt 1
Another application of the first incomplete moment is related to mean residual life and
mean waiting time given by m1 t = 1 1 t / R(t ) t and M1 t = t 1 t / F t ,
respectively.
Afify, Yousof, Butt and Hamedani
193
4.4 Residual Life Function
Several functions are defined related to the residual life. The failure rate function,
mean residual life function and the left censored mean function, also called vitality
function. It is well known that these three functions uniquely determine F ( x) .
Moreover, the moments of the residual life, mn (t ) = E[( X t )n | X > t ] , n = 1 , 2 ,...,
uniquely determine F ( x) . The n th moment of the residual life of X is given by
mn (t ) =
1
n
t ( x t ) dF ( x).
1 F (t )
Then, we can write (for r < )
n n
1
nr r
mn (t ) =
t t x dF ( x),
1 F (t ) r =0 r
mn (t ) =
n
1 n
r n
k 1 b j t r k , j
R(t ) r =0 k , j =0
r
r
B
, k 1 b
r
j Bt
, k 1 b
j .
Another interesting function is the mean residual life (MRL) function or the life
expectation at age x defined by m1 ( x) = E ( X x) | X > x , which represents the
expected additional life length for a unit which is alive at age x . The MRL of the TWPa
distribution can be obtained by setting n = 1 in the last equation.
The moments of the reversed residual life, M n (t ) = E (t X )n | X t for t > 0 ,
n = 1 , 2 ,... uniquely determine F ( x) . We obtain
4.5 Reversed Residual Life Function
M n (t ) = 0(t x)n dF ( x).
t
Therefore, the n th moment of the reversed residual life of X given that r <
becomes
1 n n
r nr t r
M n (t ) =
1 t 0x dF ( x),
F (t ) r =0 r
1 n n
r
M n (t ) =
r n k 1 b j k , j Bt 1 , k 1 b j .
F (t ) r =0k , j =0 r t
r
The mean inactivity time (MIT) or mean waiting time (MWT) also called the mean
reversed residual life function is defined by M1 (t ) = E[(t X ) | X t ] , and it represents
the waiting time elapsed since the failure of an item on condition that this failure had
occurred in (0, x) . The MIT of X can be obtained by setting n = 1 in the last equation.
194
The Transmuted Weibull-Pareto Distribution
5.ORDER STATISTICS
If X1 , X 2 ,..., X n is a random sample of size n from the TWPa distribution and
X 1 , X 2 ,..., X n are the corresponding order statistics, then the pdf of i th order
statistic denoted by fi:n x is given by
fi:n x =
n 1
f ( x)
j n 1 i j 1
( x).
1
F
B i, n i 1 j =0
j
(5.1)
Using Equations (2.1) (2.2) and (5.1), we can write
f x F ( x)i j 1 = tk ,l h,, k 1b l ( x).
k ,l =0
where
t k ,l
1k m w b m 1 i j m1 k 1 b l 1
=
k
m, w=0
k!l! k 1 b l w 1 k 1 b 1
i j 1 i j m 1
k
k
1 w 1 2 w 2 .
w
m
The q th moment of X i:n is
1 j
n 1
q
tk ,l E Y,, k 1b l .
k ,l =0 j =0 B i, n i 1 j
n 1
E X iq:n =
Using (4.1), we can write
(5.2)
n 1
j
1 j
q
k 1 b l q B 1 , k 1 b l .
k ,l =0 j =0 B i, n i 1
n 1
E X iq:n =
Based upon the moments in Equation (5.2), we can derive explicit expressions for the
L-moments of X as infinite weighted linear combinations of the means of suitable
TWPa distribution. They are linear functions of expected order statistics defined by
r =
r 1
1 r 1
1k E X r k:r , r 1.
r k =0
k
The first four L-moments are given by: 1 = E X1:1 , 2 =
1
E X 2:2 X1:2 ,
2
1
1
E X 3:3 2 X 2:3 X1:3 and 4 = E X 4:4 3 X 3:4 3 X 2:4 X1:4 . One simply
4
3
can obtain the 's for X from equation (5.2) with q = 1 .
3 =
Afify, Yousof, Butt and Hamedani
195
6. PROBABILITY WEIGHTED MOMENTS
The PWMs are expectations of certain functions of a random variable. They can be
derived for any random variable whose ordinary moments exist. The PWM approach can
be used for estimating parameters of any distribution whose inverse form cannot be
expressed explicitly.
The ( s, r ) th PWM of X , say s , r , is defined by
s,r = E X s F ( X )r = x s F ( x)r f x dx.
w 1k k 1 b j 1 m
r m r r m
1
k!j! k 1 b 1 k 1 b j
m w
Using Equations (2.1) and (2.2), we can write
f x F ( x) r =
b 1
m, w, k , j =0
k m w
x
k
k
1 w 1 2 w 2 k 1 b j
x
Then, we have
x
1
k 1b j 1
.
f x F ( x)r = d k , j h,, k 1b j ( x),
k , j =0
where
dk , j
1k m w b m 1 r m k 1 b j 1
=
k
m, w=0 k!j! k 1 b j w 1
k 1 b 1
r r m
k
k
1 w 1 2 w 2 .
m w
Then, the ( s, r ) th PWM of X can be expressed as
s ,r = dk , j E Ys,, k 1b j .
k , j =0
Using (4.1), we can write
s
s ,r = dk , j k 1 b j s B 1 , k 1 b j .
k , j =0
7. CHARACTERIZATIONS
The problem of characterizing a distribution is an important problem in various fields
which has recently attracted the attention of many researchers. These characterizations
have been established in many different directions. This section deals with two
characterizations of TWPa distribution. These characterizations are based on (i) a simple
relationship between two truncated moments and (ii) on conditional expectation of a
196
The Transmuted Weibull-Pareto Distribution
function of the random variable. It should be mentioned that for our characterization (i),
the cdf need not have a closed form. We believe, due to the nature of the cdf of TWPa,
there may not be other possible characterizations of this distribution than the ones
presented here.
7.1 Characterizations Based on Two Truncated Moments
In this subsection we present characterizations of TWPa distribution in terms of a
simple relationship between two truncated moments. Our first characterization result
borrows from a theorem due to Glänzel (1987), see Theorem A below. We refer the
interested reader to Glänzel (1990) for a proof of Theorem A. Note that the result holds
also when the interval H is not closed. Moreover, as mentioned above, it could be also
applied when the cdf F does not have a closed form. As shown in Glänzel (1990), this
characterization is stable in the sense of weak convergence.
Theorem 7.1.
Let , , P be a given probability space and let H = d , e be an interval for some
d < e ( d = , e = might as well be allowed). Let X : H be a continuous
random variable with the distribution function F and let g and h be two real functions
defined on H such that
E g X | X x = E h X | X x x , x H ,
is defined with some real function . Assume that g , h C1 H , C 2 H and F
is twice continuously differentiable and strictly monotone function on the set H . Finally,
assume that the equation h = g has no real solution in the interior of H . Then F is
uniquely determined by the functions g , h and , particularly
F x = aC
x
u
u h u g u
exp s u du,
where the function s is a solution of the differential equation s' = ' h / h g and
C is the normalization constant, such that HdF = 1 .
Proposition 7.1.
Let X : , be a continuous random variable and let
b
x
h x = 1 2 exp 1
and
1
Afify, Yousof, Butt and Hamedani
197
b
x
g x = h x exp 1 for x > .
The random variable X belongs to TWPa family (2.2) if and only if the function
defined in Theorem 7.1 has the form
b
1
x
x = exp 1 , x > .
2
(7.1)
Proof.
Let X be a random variable with density (2.2), then
b
x
1 F x E h x | X x = exp 1 , x > ,
and
b
1
x
1 F x E g x | X x = 2 exp 2 1 , x > ,
and finally
b
1
x
x h x g x = h x exp 1 < 0, for x > .
2
Conversely, if is given as above, then
s x =
x h x
x h x g x
= b x
1
x
1
b 1
, x > ,
and hence
x
s x = 1 , x > .
b
Now, in view of Theorem 7.1, X has density (2.2).
Corollary 7.1.
Let X : , be a continuous random variable and let h x be as in
Proposition 7.1. The pdf of X is (6) if and only if there exist functions g and defined
in Theorem 7.1 satisfying the differential equation
198
x h x
x h x g x
The Transmuted Weibull-Pareto Distribution
= b x
1
x
1
b 1
, x > .
(7.2)
The general solution of the differential equation in Corollary 7.1 is
b 1
b x1 x 1
b
x
x = exp 1
,
b
1
x
exp
1 h x g x dx D
where D is a constant. Note that a set of functions satisfying the differential equation
(7.2) is given in Proposition 7.1 with D = 0. However, it should be also noted that there
are other triplets h, g , satisfying the conditions of Theorem 7.1.
7.2 Characterizations Based on Conditional Expectation
of a Function of the Random Variable
In this subsection we present a characterization result in terms of a function of the
random variable X . The following Proposition has appeared in our previous work which
will be used to characterize TWPa distribution
Proposition 7.2.
Let X : a, b be a continuous random variable with cdf F and corresponding
Let x be a differentiable function greater than 1 on
that lim xa x = 1 and lim xb x = 1 c . Then, for 0 < c < 1,
pdf
f .
a, b
E X | X x = c 1 c x ,
if and only if
x 1
F x =
c
such
(7.3)
1 c
c
.
(7.4)
Remark 7.1.
Taking, e.g., a, b = , and
b
b 1 c
x
x
x = 1 c 1 exp 1 1 exp 1 .
c
Proposition 7.3 gives a characterization of TWPa distribution.
Afify, Yousof, Butt and Hamedani
199
8. ESTIMATION
The maximum likelihood estimators (MLEs) for the parameters of the TWPa
distribution are discussed in this section. Let x1 ,..., xn be a random sample of of size n
from the TWPa( x; ) distribution, where is the unknown parameter vector
= , , b, .
T
Then, the log-likelihood function for the parameters vector , say
expressed as
=
, can be
n ln b n ln n ln 1 in=1 ln xi in=1sib
b 1 in=1 ln si in=1 ln 1 2pi ,
sb
x
where si = i 1 and pi = e i .
Assuming
U =
known,
therefore
are given by
= ,
,
b
the
score
vector
components,
T
z
z p sb 1
n
= n ln in=1 ln xi b 1 in=1 i b in=1zi sib 1 2b in=1 i i i
,
si
1 2pi
p sb ln si
n
= in=1 ln si in=1sib ln si in=1 i i
b b
1 2pi
and
2 pi 1
= in=1
,
1 2pi
x
x
where zi = i ln i
.
We can find the estimates of the unknown parameters by setting the score vector to
ˆ = 0, and solving them simultaneously to obtain the ML estimators ˆ , bˆ and ̂ .
zero, U
These equations cannot be solved analytically and statistical software can be used to solve
them numerically by means of iterative techniques such as the Newton-Raphson algorithm.
For the four-parameter TWPa distribution all the second order derivatives exist.
For interval estimation of the model parameters, we require the 3 3 observed
information matrix, J = J rs for r, s = , b, , whose elements are given by
200
The Transmuted Weibull-Pareto Distribution
J =
n
2
xi
b 1
b
n
1 b i =1 b 1 zi si si ln
x
pi b 1 zi sib 1 sib ln i bzi pi si2b 1
,
1 2pi 2
b 1 in=1
2b in=1
J b = in=1
J = 2b in=1
J bb =
b
1 b ln s bs
1 2pi
i
b
i
ln si
,
2zi pi2 sib 1 1 2pi zi pi sib 1
n sb
2 i =1 i
J b = in=1
xi
ln
zi
bz p 2 s 2b 1 ln si
2
in=1zi sib 1 1 b ln si 2 in=1 i i i
si
1 2pi 2
zi pi sib 1
n
2 i =1
n
zi
si
1 2pi 2
2
pi sib ln si 1 sib
2
n
ln si i =1
2 pi 1
2
1 2pi 2
1 2pi
and J = in=1
,
2 in=1
2 pi 1
pi2 si2b ln si 1 sib
2
1 2pi
2
,
2
1 2pi 2
.
Under standard regularity conditions, the multivariate normal
N3 0, J ˆ
1
distribution can be used to construct approximate confidence intervals for the model
parameters. Here, J ˆ is the total observed information matrix evaluated at ̂ .
Therefore, approximate 100(1 )% confidence intervals for ,b and can be
determined as:
ˆ Z Jˆ , ˆ Z Jˆ , and bˆ Z Jˆbb , where Z is the upper
th percentile of the standard normal distribution.
2
2
2
2
9. SIMULATION STUDY
Here, we assess the performance of the maximum likelihood estimation procedure for
estimating the TWPa parameters using Monte Carlo simulation. An ideal technique for
simulating from (5) is the inversion method. For different combinations of , , b and
samples of sizes n = 100, 200, 500 and 1000 are generated from the TWPa model.
We repeated the simulation k = 100 times and calculated the MLEs and the standard
deviations of the parameter estimates. We use three combinations for the parameter
values (I: = = b = 1.5 and = 1 , II: = 1.5 , = b = 2 and = 0 and
III: = = 1 , = 2 and b = 3 ). The empirical results are given in Table 1. It is evident
Afify, Yousof, Butt and Hamedani
201
that the estimates are quite stable and are close to the true value of the parameters for
these sample sizes. Additionally, as the sample size increases, the biases and the standard
deviations of the MLEs decrease as expected.
Table 1
MLEs and Standard Deviations for various Parameter Values
Estimated Values (Standard Deviations)
Sample
size ( n )
̂
b̂
̂
̂
1.498091143
1.502948857
1.501953638 0.997782415
I
(0.11582947)
(0.172767349) (0.243121613) (0.144500353)
100
1.491903012
1.48957008
1.527788397 0.999536076
(0.199914811) (0.301097609) (0.381782447) (0.199173712)
200
1.491958427
1.501891389
1.509888485 0.992963765
(0.149228829) (0.198243384) (0.309084222) (0.20302651)
500
1.502175325
1.497094523
1.499390957 0.994431539
(0.119792874) (0.179465332) (0.278695995) (0.159880471)
1000
1.497894409
1.507425395
1.499064534 1.000246218
(0.098759361) (0.151490124) (0.198275817) (0.119637727)
1.500885689
2.00032535
1.997983274 0.003727487
II
(0.121687918) (0.228695465) (0.139224288) (0.145980986)
100
1.50426267
1.974097335
1.97696601 0.014608287
(0.288560488) (0.519928598) (0.413326812) (0.521138173)
200
1.495429289
2.00483308
1.983424014 0.017654112
(0.017654112) (0.290681421) (0.305144854) (0.300017145)
500
1.502350282
1.98916705
2.00784619 0.004494032
(0.10061869)
(0.204113398) (0.097860177) (0.1008336)
1000
1.500906974
2.007625756
1.998065395
0.000529191
(0.100826821) (0.199465995) (0.099311977) (0.100231728)
1.002653476
2.004604317
2.997469654 1.009583677
III
(0.270238689) (0.175851467) (0.300490017) (0.182163635)
100
0.972887849
1.990449446
2.98928733
0.967372304
(0.490687527) (0.300195793) (0.960050826) (0.487313015)
200
0.998436465
2.058020528
2.977709622 1.112873677
(0.421358094) (0.187659576) (0.499413199) (0.403941241)
500
1.021321137
1.999576707
3.011564071
0.990865319
(0.299676141) (0.100350456) (0.301259786) (0.199220749)
1000
0.997139611
1.997850366
2.995192685 1.002505993
(0.203251199) (0.198805918) (0.194364415) (0.09876462)
The bold values are combined results of (100 + 200 + 500 + 1000 = 1800).
202
The Transmuted Weibull-Pareto Distribution
10. APPLICATION
In this section, we provide an application of the TWPa distribution to show its
flexibility and importance. We shall compare the TWPa model with other related models,
namely Weibull Pareto (WPa), McDonald Lomax (McL) (Lemonte and Cordeiro 2013),
transmuted Weibull Lomax (TWL) (Afify et al. 2015) and transmuted complementary
Weibull geometric (Afify et al. 2014) distributions. The density functions (for x > 0 )
associated to these models are given by
• The McL pdf given by
x
f ( x) =
1
1
B a , b
• The TWL pdf given by
ab x
1
f ( x) =
b1
f x = x
1 x
e
x
1 1
a 1
x
1
1
1
b 1
.
b
x
exp a 1 1
x
1 1
• The TCWG pdf given by
1
b 1
b
x
1 2 exp a 1 1 .
x
1 e
3
x
1 b 1 e
.
The parameters of the above densities are all positive real numbers except for the
TWL and TCWG distributions for which | | 1 .
We make use of the data set of gauge lengths of 10 mm from Kundu and Raqab
(2009). This data set consists of 63 observations: 1.901, 2.132, 2.203, 2.228, 2.257,
2.350, 2.361, 2.396, 2.397, 2.445, 2.454, 2.474, 2.518, 2.522, 2.525, 2.532, 2.575, 2.614,
2.616, 2.618, 2.624, 2.659, 2.675, 2.738, 2.740, 2.856, 2.917, 2.928, 2.937, 2.937, 2.977,
2.996, 3.030, 3.125, 3.139, 3.145, 3.220, 3.223, 3.235, 3.243, 3.264, 3.272, 3.294, 3.332,
3.346, 3.377, 3.408, 3.435, 3.493, 3.501, 3.537, 3.554, 3.562, 3.628, 3.852, 3.871, 3.886,
3.971, 4.024, 4.027, 4.225, 4.395, 5.020. These data have been used by Afify et al. (2015)
to fit the exponentiated transmuted generalized Rayleigh distribution.
We consider some criteria like 2 (where is the maximized log-likelihood), AIC
(Akaike information criterion) and CAIC (the consistent Akaike information criterion),
HQIC (Hannan-Quinn information criterion) and BIC (Bayesian information criterion)
in order to compare the distributions. In general, the smaller the values of these statistics,
the better the fit to the data.
Afify, Yousof, Butt and Hamedani
203
Table 2 provides the MLEs and their corresponding standard errors (SEs) of the
model parameters and the numerical values of the 2 , AIC , CAIC , HQIC and BIC .
These numerical results are obtained using the Mathcad program.
It is shown from Table 2 that the TWPa model has the lowest values for the 2 ,
AIC , CAIC , HQIC and BIC statistics among all fitted models (except for HQIC and
BIC of the WPa distribution). So, the TWPa model could be chosen as the best model.
Figure 4 displays the fitted pdf and cdf for the TWPa model and Figure 5 displays the
QQ-plot and estimated survival function of the TWPa distribution. It is clear from these
plots that the TWPa provides good fit to this data set.
Table 2
MLEs, their Standard Errors (SEs) and Goodness-of-Fit Statistics
HQIC
SEs
CAIC
AIC
BIC
Model
Estimates
2
TWPa
WPa
TWL
ˆ = 0.1885
ˆ = 0.0909
bˆ = 14.4535
ˆ = 0.7288
ˆ = 0.1834
ˆ = 0.0755
bˆ = 13.9522
ˆ = 0.3922
ˆ = 0.6603
ˆ = 0.5287
bˆ = 8.4451
ˆ = 0.7364
McL
0.065
119.282 127.282 127.972 130.654 135.855
0.115
5.252
0.284
0.102
121.790 127.790 128.197 130.319 134.219
0.159
7.705
0.339
119.688 129.688 130.741 133.903 140.404
1.174
3.32
4.397
0.286
ˆ = 45.9249 59.312 130.597 140.597 141.65 144.812 151.313
ˆ = 48.3024 63.047
8.855
ˆ = 18.1192
bˆ = 195.4633 123.217
ˆ = 353.1435 375.678
TCWG ˆ = 0.2022
ˆ = 3.3482
ˆ = 0.3876
ˆ = -0.0001
0.217
0.783
0.069
0.496
126.895 134.895 135.585 138.267 143.468
204
The Transmuted Weibull-Pareto Distribution
11. CONCLUSIONS
In this paper, We propose a new four-parameter model, called the transmuted
Weibull-Pareto (TWPa) distribution, which extends the Weibull-Pareto (WPa)
distribution introduced by Tahir et al. (2015). We provide some of its mathematical
properties. The TWPa density function can be expressed as a mixture of exponentiated
Pareto densities. We derive explicit expressions for the ordinary and incomplete
moments, quantile and generating functions, Lorenz, Bonferroni and Zenga curves,
moments of residual life and moments of the reversed residual life. We also obtain the
density function of order statistics and their moments. Further, the Probability weighted
moments are investigated and certain characterization results are provided. We discuss
the maximum likelihood estimation of the model parameters. The proposed distribution is
applied to a real data set and provides a better fit than several nested and non-nested
models.
Figure 4: Estimated pdf and cdf of the TWPa Distribution
Figure 5: QQ-Plot and Estimated Survival Function of the TWPa Distribution
Afify, Yousof, Butt and Hamedani
205
ACKNOWLEDGMENTS
The authors would like to thank the Editor and the anonymous referees for very
careful reading and valuable comments which greatly improved the paper.
REFERENCES
1. Afify, A.Z., Alizadeh, M., Yousof, H.M., Aryal, G. and Ahmad, M. (2016). The
transmuted geometric-G family of distributions: theory and applications. Pak. J.
Statist., 32(2), 139-160.
2. Afify A.Z., Cordeiro, G.M., Yousof, H.M., Alzaatreh, A. and Nofal, Z.M. (2016). The
Kumaraswamy transmuted-G family of distributions: properties and applications.
Journal of Data Science, 14, 245-270.
3. Afify, A.Z., Hamedani, G.G. and Ghosh, I. (2015). The transmuted Marshall-Olkin
Fréchet distribution: properties and applications. International Journal of Statistics
and Probability, 4, 132-184.
4. Afify, A.Z., Nofal, Z.M. and Butt, N.S. (2014). Transmuted complementary Weibull
geometric distribution. Pak. J. Stat. Oper. Res., 10, 435-454.
5. Afify, A.Z., Nofal, Z.M. and Ebraheim, A.N. (2015). Exponentiated transmuted
generalized Rayleigh distribution: a new four parameter Rayleigh distribution. Pak. J.
Stat. Oper. Res., 11, 115-134.
6. Afify, A.Z., Nofal, Z.M., Yousof, H.M., El Gebaly, Y.M. and Butt, N.S. (2015). The
transmuted Weibull Lomax distribution: properties and application. Pak. J. Stat.
Oper. Res., 11, 135-153.
7. Aryal, G.R. and Tsokos, C.P. (2011). Transmuted Weibull Distribution: A
generalization of the Weibull Probability Distribution. European Journal of Pure and
Applied Mathematics, 4, 89-102.
8. Bourguignon, M., Silva, R.B. and Cordeiro, G.M. (2014). The Weibull-G family of
probability distributions. Journal of Data Science, 12, 53-68.
9. Glanzel, W.A. (1987). Characterization theorem based on truncated moments and its
application to some distribution families. Mathematical Statistics and Probability
Theory (Bad Tatzmannsdorf, 1986), B, Reidel, Dordrecht, 75-84.
10. Glanzel, W.A. (1990). Some consequences of a characterization theorem based on
truncated moments. Statistics, 21, 613-618.
11. Huang, S. and Oluyede, B.O. (2014). Exponentiated Kumaraswamy-Dagum
distribution with applications to income and lifetime data. Journal of Statistical
Distributions and Applications, 1(8), 1-20.
12. Khan, M.S. and King, R. (2013). Transmuted modified Weibull distribution: a
generalization of the modified Weibull probability distribution. European Journal of
Pure and Applied Mathematics, 6, 66-88.
13. Kundu, D. and Raqab, M.Z. (2009). Estimation of R P Y X for three-parameter
Weibull distribution. Statistics and Probability Letters, 79, 1839-1846.
14. Lemonte, A.J. and Cordeiro, G.M. (2013). An extended Lomax distribution.
Statistics, 47, 800-816.
15. Nofal, Z.M., Afify, A.Z., Yousof, H.M. and Cordeiro, G. (2015). The generalized
transmuted-G family of distributions. Comm. Statist. Theory Methods, Forthcoming.
206
The Transmuted Weibull-Pareto Distribution
16. Shaw, W.T. and Buckley, I.R.C. (2007). The alchemy of probability distributions:
beyond gram-charlier expansions and a skew-kurtotic-normal distribution from a rank
transmutation map. Research report.
17. Tahir, M.H., Cordeiro, G.M., Alzaatreh, A., Mansoor, M. and Zubair, M. (2015). A
new Weibull-Pareto distribution: properties and applications. Comm. Statist.
Simulation and Computation. Forthcoming.
18. Tahir, M.H., Cordeiro, G.M., Mansoor, M. and Zubair, M. (2015). The WeibullLomax distribution: properties and applications. Hacettepe Journal of Mathematics
and Statistics. Forthcoming.
19. Yousof, H.M., Afify, A.Z., Alizadeh, M., Butt, N.S., Hamedani, G.G. and Ali, M.M.
(2015). The transmuted exponentiated generalized-G family of distributions. Pak. J.
Stat. Oper. Res., 11, 441-464.