Open Access

The Marshall-Olkin extended Weibull family of distributions

Journal of Statistical Distributions and Applications20141:9

https://doi.org/10.1186/2195-5832-1-9

Received: 28 November 2013

Accepted: 22 April 2014

Published: 16 June 2014

Abstract

We introduce a new class of models called the Marshall-Olkin extended Weibull family of distributions based on the work by Marshall and Olkin (Biometrika 84:641–652, 1997). The proposed family includes as special cases several models studied in the literature such as the Marshall-Olkin Weibull, Marshall-Olkin Lomax, Marshal-Olkin Fréchet and Marshall-Olkin Burr XII distributions, among others. It defines at least twenty-one special models and thirteen of them are new ones. We study some of its structural properties including moments, generating function, mean deviations and entropy. We obtain the density function of the order statistics and their moments. Special distributions are investigated in some details. We derive two classes of entropy and one class of divergence measures which can be interpreted as new goodness-of-fit quantities. The method of maximum likelihood for estimating the model parameters is discussed for uncensored and multi-censored data. We perform a simulation study using Markov Chain Monte Carlo method in order to establish the accuracy of these estimators. The usefulness of the new family is illustrated by means of two real data sets.

Mathematics Subject Classification (2010)

60E05; 62F03; 62F10; 62P10

Keywords

Extended Weibull distribution Hazard rate function Marshall-Olkin distribution Maximum likelihood estimation Survival function

1 Introduction

The Weibull distribution has assumed a prominent position as statistical model for data from reliability, engineering and biological studies (McCool 2012). This model has been exaustively used for describing hazard rates – an important quantity of survival analysis. In the context of monotone hazard rates, some results from the literature suggest that the Weibull law is a reasonable choice due to its negatively and positively skewed density shapes. However, this distribution is not a good model for describing phenomenon with non-monotone failure rates, which can be found on data from applications in reliability and biological studies. Thus, extended forms of the Weibull model have been sought in many applied areas. As a solution for this issue, the inclusion of additional parameters to a well-defined distribution has been indicated as a good methodology for providing more flexible new classes of distributions.

Marshall and Olkin (1997) derived an important method of including an extra shape parameter to a given baseline model thus defining an extended distribution. The Marshall and Olkin (“ O ” for short) transformation furnishes a wide range of behaviors with respect to the baseline distribution. The geometrical and inferential properties associated with the generated distribution depend on the values of the extra parameter. These characteristics provide more flexibility to the O generated distributions. Considering the proportional odds model, Sankaran and Jayakumar (2008) presented a detailed discussion about the physical interpretation of the O family.

This family has a relationship with the odds ratio associated with the baseline distribution. Let X be a distributed O random variable which describes the lifetime relative to each individual in the population with a vector of p-covariates z=(z1,…,z p ), where (·) denotes the transposition operator. Then, the cumulative distribution function (cdf) of X is given by
F ¯ ( x ; z ) = k ( z ) G ¯ ( x ) 1 [ 1 k ( z ) ] G ¯ ( x ) ,
(1)

where k(z)=λ G (x)/λ F (x ; z) is a non-negative function such that z is independent of the time x, λ F (x ; z) is the proportional odds model [for a discussion about such modeling, see Sankaran and Jayakumar (2008)] and λ G ( x ) = G ( x ) / G ¯ ( x ) represents an arbitrary odds for the baseline distribution.

In this paper, we consider k(z)=δ. Before, however, it is important to highlight two important properties of the O transformation: (i) the stability and (ii) geometric extreme stability (Marshall and Olkin 1997). In other words, the O distribution possesses a stability property in the sense that if the method is applied twice, it returns to the same distribution. In addition, the following stochastic behavior can also be verified: let {X1,…,X N } be a random sample from the population random variable equipped with the survival function (1) at k(z)=δ. Suppose that N has the geometric distribution with probability p and that this quantity is independent of X i , for i=1,…,N. Then, U=m i n(X1,…,X N ) and V=m a x(X1,…,X N ) are random variables having survival functions (1) such that k(z) can be equal to p and p−1, respectively, i.e., the O transform satisfies the geometric extreme stability property.

Due to these advantages, many papers have employed the O transformation. In Marshall and Olkin work, the exponential and Weibull distributions were generalized. Subsequently, the O extension was applied to several well-known distributions: Weibull (Ghitany et al.2005, Zhang and Xie 2007), Pareto (Ghitany 2005), gamma (Ristić et al.2007), Lomax (Ghitany et al.2007) and linear failure-rate (Ghitany and Kotz 2007) distributions. More recently, general results have been addressed by Barreto-Souza et al. (2013) and Cordeiro and Lemonte (2013). In this paper, we aim to apply the O generator to the extended Weibull ( E W ) class of distributions to obtain a new more flexible family to describe reliability data. The proposed family can also be applied to other fields including business, environment, informatics and medicine in the same way as it was originally done with the Birnbaum-Saunders and other lifetime distributions.

Let G ¯ ( x ) = 1 G ( x ) and g(x)=d G(x)/d x be the survival and density functions of a continuous random variable Y with baseline cdf G(x). Then, the O extended distribution has survival function given by
F ¯ ( x ; δ ) = δ G ¯ ( x ) 1 δ ¯ G ¯ ( x ) = δ G ¯ ( x ) G ( x ) + δ G ¯ ( x ) , x X R , δ > 0 ,
(2)

where δ ¯ = 1 δ.

Clearly, δ=1 implies F ¯ ( x ) = G ¯ ( x ) . The family (2) has probability density function (pdf) given by
f ( x ; δ ) = δg ( x ) [ 1 δ ¯ G ¯ ( x ) ] 2 , x X R , δ > 0 .
Its hazard rate function (hrf) becomes
τ ( x ; δ ) = g ( x ) G ¯ ( x ) [ 1 δ ¯ G ¯ ( x ) ] , x X R , δ > 0 .
Further, the class of extended Weibull ( E W ) distributions pioneered by Gurvich et al. (1997) has achieved a prominent position in lifetime models. Its cdf is given by
G ( x ; α , ξ ) = 1 exp [ αH ( x ; ξ ) ] , x D R + , α > 0 ,
(3)
where H(x;ξ) is a non-negative monotonically increasing function which depends on the parameter vector ξ. The corresponding pdf is given by
g ( x ; α , ξ ) = α exp [ αH ( x ; ξ ) ] h ( x ; ξ ) ,
(4)

where h(x;ξ) is the derivative of H(x;ξ).

Different expressions for H(x;ξ) in Equation (3) define important models such as:
  1. (i)

    H(x;ξ)=x gives the exponential distribution;

     
  2. (ii)

    H(x;ξ)=x 2 leads to the Rayleigh (Burr type-X) distribution;

     
  3. (iii)

    H(x;ξ)= log(x/k) leads to the Pareto distribution;

     
  4. (iv)

    H(x;ξ)=β −1[ exp(β x)−1] gives the Gompertz distribution.

     
In this paper, we derive a new family of distributions by compounding the O and E W classes. We define a new generated family in order to provide a “better fit” in certain practical situations. The compounding procedure follows by taking the E W class (3) as the baseline model in Equation (2). The Marshall-Olkin extended Weibull ( O E W ) family of distributions contains some special models as those listed in Table 1 with the corresponding H(·;·) and h(·;·) functions and the parameter vectors.
Table 1

Special models and the corresponding functions H ( x ; ξ ) and h ( x ; ξ )

Distribution

H( x; ξ)

h( x; ξ)

α

ξ

References

Exponential (x≥0)

x

1

α

Johnson et al. (1994)

Pareto (xk)

log(x/k)

1/x

α

k

Johnson et al. (1994)

Burr XII (x≥0)

log(1+x c )

c xc−1/(1+x c )

α

c

Rodriguez (1977)

Lomax (x≥0)

log(1+x)

1/(1+x)

α

Lomax (1954)

Log-logistic (x≥0)

log(1+x c )

c xc−1/(1+x c )

1

c

Fisk (1961)

Rayleigh (x≥0)

x 2

2x

α

Rayleigh (1880)

Weibull (x≥0)

x γ

γ x γ−1

α

γ

Johnson et al. (1994)

Fréchet (x≥0)

x γ

γ x−(γ+1)

α

γ

Fréchet (1927)

Linear failure rate(x≥0)

a x+b x2/2

a+b x

1

[a,b]

Bain (1974)

Modified Weibull (x≥0)

x γ exp(λ x)

xγ−1 exp(λ x)(γ+λ x)

α

[γ, λ]

Lai et al. (2003)

Weibull extension (x≥0)

λ[ exp(x/λ) β −1]

β exp(x/λ) β (x/λ)β−1

α

[γ, λ, β]

Xie et al. (2002)

Phani (0<μ<x<σ<)

[(xμ)/(σx)] β

β[(xμ)/(σx)]β−1[(σμ)/(σt)2]

α

[μ, σ, β]

Phani (1987)

Weibull Kies (0<μ<x<σ<)

( x μ ) β 1 / ( σ x ) β 2

( x μ ) β 1 1 ( σ x ) β 2 1 [ β 1 ( σ x ) + β 2 ( x μ ) ]

α

[μ, σ, β1, β2]

Kies (1958)

Additive Weibull (x≥0)

( x / β 1 ) α 1 + ( x / β 2 ) α 2

( α 1 / β 1 ) ( x / β 1 ) α 1 1 + ( α 2 / β 2 ) ( x / β 2 ) α 2 1

1

[α1, α2, β1, β2]

Xie and Lai (1995)

Traditional Weibull (x≥0)

x b [ exp(c x d −1)]

b xb−1[ exp(c x d )−1]+c d xb+d−1 exp(c x d )

α

[b, c, d]

Nadarajah and Kotz (2005)

Gen. power Weibull (x≥0)

[ 1 + ( x / β ) α 1 ] θ 1

( θ α 1 / β ) [ 1 + ( x / β ) α 1 ] θ 1 ( x / β ) α 1

1

[α1, β, θ]

Nikulin and Haghighi (2006)

Flexible Weibull extension(x≥0)

exp(γ xβ/x)

exp(γ xβ/x)(γ+β/x2)

1

[γ, β]

Bebbington et al. (2007)

Gompertz (x≥0)

β−1[ exp(β x)−1]

exp(β x)

α

β

Gompertz (1825)

Exponential power (x≥0)

exp[(λ x) β ]−1

β λ exp[(λ x) β ](λ x)β−1

1

[λ, β]

Smith and Bain (1975)

Chen (x≥0)

exp(x b )−1

b xb−1 exp(x b )

α

b

Chen (2000)

Pham (x≥0)

(a x ) β −1

β(a x ) β log(a)

1

[a, β]

Pham (2002)

The paper unfolds as follows. Section 2 presents the cdf and pdf of the proposed distribution and some expansions for the density function. The main statistical properties of the new family are derived in Section 3 including the moments, moment generating function (mgf) and incomplete moments, quantile function (qf), random number generator, skewness and kurtosis measures, order statistics, mean deviations and average lifetime functions. In Section 4, we derive four measures of information theory: Shannon and Rényi entropies, cross entropy and Kullback-Leibler divergence. The maximum likelihood method to estimate the model parameters is adopted in Section 5. Two special models are studied in some details in Section 6. We perform a simulation study using Monte Carlo’s experiments in order to assess the accuracy of the maximum likelihood estimators (MLEs) in Section 7.1 and two applications to real data in Section 7.2. Conclusions and some future lines of research are addressed in Section 8.

2 The O E W family

The cdf of the new family of distributions is given by
F ( x ; δ , α , ξ ) = 1 exp [ αH ( x ; ξ ) ] 1 δ ¯ exp [ αH ( x ; ξ ) ] , x D ,
(5)
where α>0 and δ>0. Using (5), we can express its survival function as
F ¯ ( x ; δ , α , ξ ) = δ exp [ αH ( x ; ξ ) ] 1 δ ¯ exp [ αH ( x ; ξ ) ] , x D
(6)
and the associated hrf reduces to
τ ( x ; δ , α , ξ ) = α h ( x ; ξ ) 1 δ ¯ exp [ αH ( x ; ξ ) ] , x D.
(7)
The corresponding pdf is given by
f ( x ; δ , α , ξ ) = δ α h ( x ; ξ ) exp [ αH ( x ; ξ ) ] { 1 δ ¯ exp [ αH ( x ; ξ ) ] } 2 ,
(8)

where H(x;ξ) can be any special distribution listed in Table 1.

Hereafter, let X be a random variable having the O E W pdf (8) with parameters δ,α and ξ, say X O E W ( δ , α , ξ ) . Equation (8) extends several distributions which have been studied in the literature.

The O Pareto (Ghitany 2005) is obtained by taking H(x; ξ)= log(x/k)(xk). Further, for H(x; ξ)=x γ we obtain the O Weibull (Ghitany et al. 2005, Zhang and Xie 2007). The O Lomax (Ghitany et al. 2007) and O log-logistic are derived from (8) by taking H(x; ξ)= log(1+x c ) with c=1 and H(x; ξ)= log(1+x c ) with α=1, respectively. For H(x; ξ)=a x+b x2/2 and α=1, Equation (8) reduces to the O linear failure rate (Ghitany and Kotz 2007). In the same way, for H(x; ξ)= log(1+x c ), we have the O Burr XII (Jayakumar and Mathew 2008). Finally, we obtain the O Fréchet (Krishna et al. 2013) from Equation (8) by setting H(x; ξ)=xγ. Table 1 displays some useful quantities and corresponding parameter vectors for special distributions.

A general approximate goodness-of-fit test for the null hypothesis H0:X1,…,X n with X i following F(x;θ), where the form of F is known but the p-vector θ=(δ,α,ξ) is unknown, was proposed by Chen and Balakrishnan (1995). This method is based on the Cramér-von Mises (CM) and Anderson-Darling (AD) statistics and, in general, the smaller the values of these statistics, the better the fit. In this paper, such methodology is applied to provide goodness-of-fit tests for the distributions under study.

Some results in the following sections can be obtained numerically in any software such as MAPLE (Garvan 2002), MATLAB (Sigmon and Davis 2002), MATHEMATICA (Wolfram 2003), Ox (Doornik 2007) and R (R Development Core Team 2009). The Ox (for academic purposes) and R are freely available at http://www.doornik.com and http://www.r-project.org, respectively. The results can be computed by taking in the sums a large positive integer value in place of .

2.1 Expansions for the density function

For any positive real number a, and for |z|<1, we have the generalized binomial expansion
( 1 z ) a = k = 0 ( a ) k k ! z k ,
(9)
where (a) k =Γ(a+k)/Γ(a)=a(a+1)…(a+k−1) is the ascending factorial and Γ(·) is the gamma function. Applying (9) to (8), for 0<δ<1, gives
f ( x ; δ , α , ξ ) = j = 0 η j g ( x ; ( j + 1 ) α , ξ ) ,
(10)
where η j = δ δ ¯ j and g(x;(j+1)α,ξ) denotes the E W density function with parameters (j+1)α and ξ. Otherwise, for δ>1, after some algebra, we can express (8) as
f ( x ; δ , α , ξ ) = g ( x ; α , ξ ) δ 1 1 1 / δ [ 1 exp ( αH ( x ; ξ ) ) ] 2 .
(11)
In this case, we can verify that |(1−1/δ)[1− exp(−α H(x;ξ))]|<1. Then, applying twice the expansion (9) in Equation (11), we obtain
f ( x ; δ , α , ξ ) = j = 0 ν j g ( x ; ( j + 1 ) α , ξ ) ,
(12)
where
ν j = ν j ( δ ) = ( 1 ) j δ ( j + 1 ) ! k = j ( k + 1 ) ! ( 1 1 / δ ) k .
We can verify that j = 0 η j = j = 0 ν j = 1 . Then, the O E W density function can be expressed as an infinite linear combination of E W densities. Equations (10) and (12) have the same form except for the coefficients η js in (10) and ν js in (12). They depend only on the generator parameter δ. For simplicity, we can write
f ( x ; δ , α , ξ ) = j = 0 w j g ( x ; ( j + 1 ) α , ξ ) ,
(13)
where
w j = η j , if 0 < δ < 1 , ν j , if δ > 1 ,

and η j and ν j are given by (10) and (12), respectively. Thus, some mathematical properties of (13) can be obtained directly from those E W properties. For example, the ordinary, incomplete, inverse and factorial moments and the mgf of X follow immediately from those quantities of the E W distribution.

3 General properties

3.1 Moments, generating function and incomplete moments

The n th ordinary moment of X can be obtained from (13) as
E ( X n ) = j = 0 w j E ( Y j n ) ,

where from now on Y j E W ( ( j + 1 ) α , ξ ) denotes a random variable having the E W density function g(y;(j+1)α,ξ).

The mgf and the k th incomplete moment of X follow from (13) as
M X ( t ) = E e tX = j = 0 w j M j ( t )
and
T k ( z ) = j = 0 w j T k ( j ) ( z ) ,
(14)

where M j (t) is the mgf of Y j and T k ( j ) ( z ) = z x k g ( x ; ( j + 1 ) α , ξ ) d x comes directly from the E W model.

3.2 Quantile function and random number generator

The qf of X follows by inverting (5) and it can be expressed in terms of H−1(·) as
Q ( u ) = H 1 1 α log 1 δ ¯ u 1 u , ξ .
(15)
In Table 2, we provide the function H−1(x;ξ) for some special models.
Table 2

The H −1 ( x ; ξ ) function

Distribution

H−1( x; ξ)

Exponential power

log x + 1 1 / β λ

Chen

[log(x+1)]1/β

Weibull extension

λ log x λ + 1 1 / β

Log-Weibull

σ log(x)+μ

Kies

x 1 / β σ + μ x 1 / β + 1

Gen. Power Weibull

β ( x + 1 ) 1 / θ 1 1 / α 1

Gompertz

log β x + 1 β

Pham

log x + 1 log ( a ) 1 / β

Hence, the generator for X can be given by the algorithm:

The O E W distributions can be very useful in modeling lifetime data and practitioners may be interested in fitting one of these models. We provide a script using the R language to generate the density, distribution function, hrf, qf, random numbers, Anderson-Darling test, Cramer-von Mises test and likelihood ratio (LR) tests. This script can be be obtained from the authors upon requested.

3.3 Mean deviations

The mean deviations of X about the mean and the median are given by
δ 1 = D | x μ | f ( x ; δ , α , ξ ) d x and δ 2 = D | x M | f ( x ; δ , α , ξ ) d x ,
respectively, where μ=E(X) denotes the mean and M=M e d i a n(X) the median. The median follows from the nonlinear equation F(M;δ,α,ξ)=1/2. So, these quantities reduce to
δ 1 = 2 μ F ( μ ; δ , α , ξ ) 2 T 1 ( μ ) and δ 2 = μ 2 T 1 ( M ) ,
where T1(z) is the first incomplete moment of X obtained from (14) as
T 1 ( z ) = j = 0 w j T 1 ( j ) ( z ) ,

and T 1 ( j ) ( z ) = z x g ( x ; ( j + 1 ) α , ξ ) d x is the first incomplete moment of Y j .

An important application of the mean deviations is related to the Bonferroni and Lorenz curves. These curves are useful in economics, reliability, demography, medicine and other fields. For a given probability p, they are defined by B(p)=T1(q)/(p μ) and L(p)=T1(q)/μ, respectively, where q=Q(p) is the qf of X given by (15) at u=p.

3.4 Average lifetime and mean residual lifetime functions

The average lifetime is given by
t m = 0 1 F ( x ; δ , α , ξ ) d x = j = 0 w j 0 G ¯ ( x ; ( j + 1 ) α , ξ ) d x.
In fields such as actuarial sciences, survival studies and reliability theory, the mean residual lifetime has been of much interest; see, for a survey, Guess and Proschan (1988). Given that there was no failure prior to x0, the residual life is the period from time x0 until the time of failure. The mean residual lifetime is given by
m ( x 0 ; δ , α , ξ ) = E X x 0 | X x 0 ; δ , α , ξ = { x : x > x 0 } ( x x 0 ) f ( x ; δ , α , ξ ) Pr ( X > x 0 ) d x = Pr ( X > x 0 ) 1 0 y f ( x 0 + y ; δ , α , ξ ) d y = F ¯ ( x 0 ; δ , α , ξ ) 1 j = 0 w j 0 y g ( x 0 + y ; ( j + 1 ) α , ξ ) d y.

The last integral can be computed from the baseline E W distribution. Further, m(x0;δ,α,ξ)→E(X) as x0→0.

4 Information theory measures

The seminal idea about information theory was pioneered by Hartley (1928), who defined a logarithmic measure of information for communication. Subsequently, Shannon (1948) formalized this idea by defining the entropy and mutual information concepts. The relative entropy notion (which would later be called divergence) was proposed by Kullback and Leibler (1951). The Kullback-Leibler’s measure can be understood like a comparison criterion between two distributions. In this section, we derive two classes of entropy measures and one class of divergence measures which can be understood as new goodness-of-fit quantities such those discussed by Seghouane and Amari (2007). All these measures are defined for one element or between two elements in the O E W family.

4.1 Rényi entropy

The Rényi entropy of X with pdf (8) is given by
H R s ( X ) = 1 1 s log D f ( x ; δ , α , ξ ) s d x ,

where s(0,1)(1,).

It is a difficult problem to obtain H R s ( X ) in closed-form for the O E W family. So, we derive an expansion for this quantity.

By using (9), f(x;δ,α,ξ) s can be expanded as
f ( x ; δ , α , ξ ) s = j = 0 w j exp [ ( j + s ) αH ( x ; ξ ) ] h ( x ; ξ ) s ,
(16)
where
w j = η j ( α , δ ) = α s δ s ( 2 s ) j δ ¯ j j ! , for 0 < δ < 1 , ν j ( α , δ ) = α s δ s j ! k = 0 ( 2 s ) k ( k ) j k ! ( 1 1 / δ ) k , for δ > 1 .

The proof of this expansion is given in Appendix 8.

Finally, based on Equation (16), the Rényi entropy can be expressed as
H R s ( X ) = 1 1 s log j = 0 w j D exp [ ( j + s ) αH ( x ; ξ ) ] h ( x ; ξ ) s d x .

An advantage of this expansion is its dependence of an integral which has closed-form for some E W distributions.

4.2 Shannon entropy

The Shannon entropy of X is given by
H S ( X ) = E X log [ f ( X ; δ , α , ξ ) ] ,
where the log-likelihood function corresponding to one observation follows from (8) as
log [ f ( x ; δ , α , ξ ) ] = log ( δα ) + log [ h ( x ; ξ ) ] αH ( x ; ξ ) 2 log 1 δ ¯ exp [ αH ( x ; ξ ) ] .
Thus, it can be reduced to
H S ( X ) = log ( αδ ) + 2 E log 1 δ ̄ G ̄ ( X ; ξ ) E log h ( X ; ξ ) + α E H ( X ; ξ ) .

4.3 Cross entropy and Kullback-Leibler divergence and distance

Let X and Y be two random variables with common support R + whose densities are f X (x;θ1) and f Y (y;θ2), respectively. Cover and Thomas (1991) defined the cross entropy as
C X ( Y ) = E X log f Y ( X ; θ 2 ) = 0 f X ( z ; θ 1 ) log f Y ( z ; θ 2 ) d z.
We consider that X O E W ( δ x , α x , ξ x ) and Y O E W ( δ y , α y , ξ y ) . After some algebraic manipulations, we obtain
C X ( Y ) = D f X ( z ; δ x , α x , ξ x ) log f Y ( z ; δ y , α y , ξ y ) d z = log δ y α y E X log h ( X ; ξ y ) + α y E X H ( X ; ξ y ) + 2 E X log 1 δ ̄ G ̄ ( X ; ξ y ) .
(17)
An important measure in information theory is the Kullback-Leibler divergence given by
D ( X | | Y ) = C X ( Y ) H S ( X ) = E X log f X ( X ; δ x , α x , ξ x ) f Y ( X ; δ y , α y , ξ y ) .
(18)
Applying (4.2) and (17) in Equation (18) gives
D ( X | | Y ) = log δ x α x δ y α y + E X log h ( X ; ξ x ) h ( X ; ξ y ) + 2 E X log 1 δ ̄ G ̄ ( X ; ξ y ) 1 δ ̄ G ̄ ( X ; ξ x ) + α y E X [ H ( X ; ξ y ) ] α x E X [ H ( X ; ξ x ) ] .
(19)

According to Cover and Thomas (1991), the Kullback-Leibler measure D(X||Y) is the quantification of the error considering that the Y model is true when the data follow the X distribution. For example, this measure has been proposed as essential parts of test statistics, which has seen strongly applied to contexts of radar synthetic aperture image processing in both univariate (Nascimento et al. 2010) and polarimetric (or multivariate) (Nascimento et al. 2014) perspectives.

In order to work with measures that satisfied the non-negativity, symmetry and definiteness properties, Nascimento et al. (2010) considered the symmetrization of (19)
d KL ( X , Y ) = 1 2 D ( X | | Y ) + D ( Y | | X ) = D f X x ; δ x , α x , ξ x f Y x ; δ y , α y , ξ y log f X ( x ; δ x , α x , ξ x ) f Y ( x ; δ x , α x , ξ x ) IntegrandKL ( x , y ) d x ,
which is given by
2 d KL ( X , Y ) = α y E X H ( X ; ξ y ) E Y H ( Y ; ξ y ) + α x E Y H ( Y ; ξ x ) ] E X [ H ( X ; ξ x ) + E X log h X ; ξ x h X ; ξ y + E Y log h Y ; ξ y h Y ; ξ x + 2 E X log 1 δ ̄ G ̄ X ; ξ y 1 δ ̄ G ̄ X ; ξ x + 2 E Y log 1 δ ̄ G ̄ Y ; ξ x 1 δ ̄ G ̄ Y ; ξ y .
(20)

Although this measure does not satisfy the triangle inequality, it is usually called the Kullback-Leibler distance (Jensen-Shannon divergence). The new measure can be used to answer questions like “how could one quantify the difference in selecting the Phani model with three parameters as the baseline distribution instead of the Weibull Kies distribution which has four parameters?”.

As an illustration for (20), we initially consider two distinct elements of the generated special model from the specifications: H(x;β)=β−1[ exp(β x)−1] and h(x;β)= exp(β x) in (8). This model will be presented with more details in future sections and its parametric space is represented by the vector (δ,α,β). Suppose that we are interested in quantifying the influence of a nuisance degree ε in the parameter α over the distance between two distinct elements, (2,1,3) and (2,1+ε,3), at such parametric space. Figure 1(a) displays the integrand of (20) for ε=0.1, 1, 2 and 4 for which the distances (or areas) associated with dKL(X,Y) are 6.50×10−3, 3.56×10−1, 9.46×10−1 and 2.25, respectively. It is notable that dKL(X,Y) takes smaller values for more closer points (or, equivalently, for more closer fits) and, therefore, (20) consists of new goodness-of-fit measures. In Figures 1(b) and 1(c), we show the influence of η=α/β on dKL([δ,α,β],[δ,α,β+ε]) (for β=δ=3 and α{1,3,9}) and of δ on dKL([δ,α,β],[δ+ε,α,β]) (for β=α=3 and δ{3,4,5}). For all cases, the contamination ε takes values in the interval (−2.9,2.9).
Figure 1

MSE curves for δ {0 . 3,0 . 8,1,2,4}, η =0 . 5,1,2 and n =200. (a) Behavior of the function IntegrandKL. (b) Influence of η=α/β under β=δ=3 and α{1,3,9}. (c) Influence of δ under (α,β)=(3,3).

5 Estimation

Here, we present a general procedure for estimating the O E W parameters from one observed sample and from multi-censored data. Additionally, we provide a discussion about how one can test the significance of additional parameter at the proposed class. Let x1,…,x n be a sample of size n from X. The log-likelihood function for the vector of parameters θ=(δ,α,ξ) can be expressed as
( θ ) = n log ( δα ) + i = 1 n log h ( x i ; ξ ) α i = 1 n H ( x i ; ξ ) 2 i = 1 n log 1 δ ¯ exp [ αH ( x i ; ξ ) ] .
From the above log-likelihood, the components of the score vector, U ( θ ) = ( U δ , U α , U ξ ) , are given by
U δ ( θ ) = ∂ℓ ( θ ) ∂δ = n δ 2 i = 1 n exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] , U α ( θ ) = ∂ℓ ( θ ) ∂α = n α i = 1 n H ( x i ; ξ ) 2 δ ¯ i = 1 n H ( x i ; ξ ) exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] and U ξ k ( θ ) = ∂ℓ ( θ ) ξ k = i = 1 n 1 h ( x i ; ξ ) ∂h ( x i ; ξ ) ξ k α i = 1 n ∂H ( x i ; ξ ) ξ k 2 δ ¯ α i = 1 n ∂H ( x i ; ξ ) ξ k exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] .
Finally, the partitioned observed information matrix for the O E W family is
whose elements are
U δδ ( θ ) = n δ 2 , U δα ( θ ) = 2 i = 1 n H ( x i ; ξ ) exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] 2 , U δ ξ k ( θ ) = 2 α i = 1 n ∂H ( x i ; ξ ) ξ k exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] 2 , U αα ( θ ) = n α 2 + 2 δ ¯ i = 1 n H ( x i ; ξ ) 2 exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] 2 , U α ξ k ( θ ) = 2 δ ¯ i = 1 n ∂H ( x i ; ξ ) ξ k exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] 1 αH ( x i ; ξ ) 1 δ ¯ exp [ αH ( x i ; ξ ) ] + i = 1 n ∂H ( x i ; ξ ) ξ k and U ξ k ξ j ( θ ) = i = 1 n 1 h ( x i ; ξ ) 2 h ( x i ; ξ ) ξ k ξ j 1 h ( x i ; ξ ) ∂h ( x i ; ξ ) ξ k ∂h ( x i ; ξ ) ξ j α i = 1 n 2 H ( x i ; ξ ) ξ k ξ j 2 α δ ¯ i = 1 n exp [ αH ( x i ; ξ ) ] 1 δ ¯ exp [ αH ( x i ; ξ ) ] 2 H ( x i ; ξ ) ξ k ξ j ∂H ( x i ; ξ ) ξ k αH ( x i ; ξ ) 1 δ ¯ exp [ αH ( x i ; ξ ) ] .

When some standard regularity conditions are satisfied (Cox and Hinkley 1974), one can verify that n α ̂ , δ ̂ , ξ ̂ α , δ , ξ converges in distribution to the multivariate N p + 2 0 , K ( [ α , δ , ξ ] ) 1 distribution, where p denotes the dimension of ξ and K ( [ α , δ , ξ ] ) is the expected information matrix for which the limit identity lim n J n ( [ α , δ , ξ ] ) = K ( [ α , δ , ξ ] ) is satisfied. Based on this result, one can compute confidence regions for the O E W parameters. Such regions can be used as decision criteria in several practical situations.

For checking if δ is statistically different from one, i.e. for testing the null hypothesis H0:δ=1 against H1:δ≠1, we use the LR statistic given by LR = 2 ( θ ̂ ) ( θ ~ ) , where θ ̂ is the vector of unrestricted MLEs under H1 and θ ~ is the vector of restricted MLEs under H0. Under the null hypothesis, the limiting distribution of LR is a χ 1 2 distribution. If the test statistic exceeds the upper 100(1−α)% quantile of the χ 1 2 distribution, then we reject the null hypothesis.

Censored data occur very frequently in lifetime data analysis. Some mechanisms of censoring are identified in the literature as, for example, types I and II censoring (Lawless 2003). Here, we consider the general case of multi-censored data: there are n=n0+n1+n2 subjects of which n0 is known to have failed at the times x 1 , , x n 0 , n1 is known to have failed in the interval [ si−1,s i ], i=1,…,n1, and n2 survived to a time r i , i=1,…,n2, but not observed any longer. Note that type I censoring and type II censoring are contained as particular cases of multi-censoring. The log-likelihood function of θ=(δ,α,ξ) for this multi-censoring data reduces to
( θ ) = n 0 log ( δα ) + i = 1 n 0 log h ( x i ; ξ ) α i = 1 n 0 H ( x i ; ξ ) 2 i = 1 n 0 log 1 δ ¯ exp [ αH ( x i ; ξ ) ] + i = 1 n 1 log 1 exp [ α H ( s i ; ξ ) ] 1 δ ¯ exp [ α H ( s i ; ξ ) ] 1 exp [ α H ( s i 1 ; ξ ) ] 1 δ ¯ exp [ α H ( s i 1 ; ξ ) ] + n 2 log ( δ ) α i = 1 n 2 H ( r i ; ξ ) 2 i = 1 n 2 log 1 δ ¯ exp [ αH ( r i ; ξ ) ] .
(21)

The score functions and the observed information matrix corresponding to (21) is too complicated to be presented here.

6 Two special models

In this section, we study two special O E W models, namely the Marshall-Olkin modified Weibull ( O W ) and Marshall-Olkin Gompertz ( O G ) distributions. We provide plots of the density and hazard rate functions for some parameters to illustrate the flexibility of these distributions.

6.1 The O W model

For H(x;λ,γ)=x γ exp(λ x) and h(x;λ,γ)=xγ−1 exp(λ x)(γ+λ x), we obtain the O W distribution. Its density function is given by
f ( x ; α , δ , λ , γ ) = δα ( γ + λx ) x γ 1 exp [ λx α x γ exp ( λx ) ] 1 δ ¯ exp [ α x γ exp ( λx ) ] 2 , x > 0 ,
where λ,γ≥0. If δ=1, it leads to the special case of the modified Weibull ( W ) distribution (Lai et al.2003). In addition, when λ=0, it gives the Weibull distribution. Its cdf and hrf are given by
F ( x ; α , δ , λ , γ ) = 1 exp [ α x γ exp ( λx ) ] 1 δ ¯ exp [ α x γ exp ( λx ) ]
and
τ ( x ; α , δ , λ , γ ) = α x γ 1 exp ( λx ) ( γ + λx ) 1 δ ¯ exp [ α x γ exp ( λx ) ] ,
respectively. In Figures 2(a), 2(b), 2(c) and 2(d), we note some different shapes of the O W pdf. Further, Figures 3(a), 3(b), 3(c) and 3(d) display plots of the O W hrf, which can have increasing, decreasing, non-monotone and bathtub forms.
Figure 2

The O W density functions. (a) For α=0.5,λ=2.0,γ=0.5. (b) For δ=2.0,λ=2.0,γ=0.5. (c) For δ=5.0,δ=2.0,γ=0.5. (d) For α=0.5,δ=2.0,λ=2.0.

Figure 3

The O W N hrfs. (a) For α=0.5,λ=2.0,γ=0.5. (b) For δ=2.0,λ=2.0,γ=0.5. (c) For δ=5.0,δ=2.0,γ=0.5. (d) For α=0.5,δ=2.0,λ=2.0.

The r th raw moment of the O W distribution comes from (13) as
E X r = j = 1 w j μ r ( j ) ,
(22)
where μ r ( j ) = 0 x r g ( x ; ( j + 1 ) α , γ , λ ) ) d x denotes the r th raw moment of the W distribution with parameters (j+1)α,γ and λ. Carrasco et al. (2008) determined an infinite representation for μ r (j) given by
μ r ( j ) = i 1 , , i r = 1 A i 1 , , i r Γ ( s r / γ + 1 ) [ ( j + 1 ) α ] s r / γ ,
(23)
where
A i 1 , , i r = a i 1 , , a i r and s r = i 1 , , i r ,
and
a i = ( 1 ) i + 1 i i 2 ( i 1 ) ! λ γ i 1 .

Hence, the O W moments can be obtained directly from (22) and (23).

Let x1,…,x n be a sample of size n from X O W ( α , δ , λ , γ ) . The log-likelihood function for the vector of parameters θ=(α,δ,λ,γ) can be expressed as
( θ ) = n log ( δα ) + i = 1 n log ( γ + λ x i ) + ( γ 1 ) i = 1 n log ( x i ) + λ i = 1 n x i α i = 1 n x i λ exp ( λ x i ) 2 i = 1 n log 1 δ ¯ exp α x i γ exp ( λ x i ) .

6.2 The O G model

For H(x;β)=β−1[ exp(β x)−1] and h(x;β)= exp(β x), we obtain the O G distribution. Its pdf is given by
f ( x ; α , δ , β ) = δα exp βx α / β [ exp ( βx ) 1 ] 1 δ ¯ exp α / β [ exp ( βx ) 1 ] 2 , x > 0 ,
where −<β<. For δ=1, it follows the Gompertz distribution as a special case. The O G model is a special case of the Marshall-Olkin Makeham distribution (EL-Bassiouny and Abdo 2009). The cdf and hrf of the O G distribution are given by
F ( x ; α , δ , β ) = 1 exp α / β [ exp ( βx ) 1 ] 1 δ ¯ exp α / β [ exp ( βx ) 1 ]
and
τ ( x ; α , δ , β ) = α exp ( βx ) 1 δ ¯ exp α / β [ exp ( βx ) 1 ] .
Figures 4(a), 4(b) and 4(c) display some plots of the density functions for some values of α, δ and β. The hrf of the Gompertz distribution is increasing (β>0) and decreasing (β<0). Besides these two forms, Figures 5(a), 5(b) and 5(c) indicate that the O G hrf can be bathtub shaped.
Figure 4

The O G density functions. (a) For α=0.5,β=0.7. (b) For δ=2.0,β=2.0. (c) For δ=5.0,α=1.5.

Figure 5

The O G hrfs. (a) For α=25,β=2.0. (b) For δ=0.2,β=0.5. (c) For α=0.01,δ=0.5.

From Equation (15), the O G qf becomes
Q ( u ) = β 1 log β α log 1 δ ¯ u 1 u + 1 .
Let x1,…,x n be a sample of size n from the O G model. The log-likelihood function for the vector of parameters θ=(δ,α,β) can be expressed as
( θ ) = n log ( δα ) + β i = 1 n x i α β i = 1 n exp ( β x i ) 1 2 i = 1 n log 1 δ ¯ exp α [ exp ( β x i ) 1 ] / β .

7 Simulation and applications

This section is divided in two parts. First, we perform a simulation study in order to assess the performance of the MLEs on some points at the parametric space of one of the special models. Second, an application to real data provides evidence in favor of one distribution in the O E W class.

7.1 Simulation study

We present a simulation study by means of Monte Carlo’s experiments in order to assess the performance of the MLEs described in Section 5. To that end, we work with the O G distribution. One of advantages of this model is that its cdf has tractable analytical form. This fact implies in a simple random number generation (RNG) determined by the O G qf given in Section 6.2. The O G generator is illustrated in Figure 6.
Figure 6

Illustration of the O G generator for two points at the parametric space.

The simulation study is conducted in order to quantify the influence of η=α/β over the estimation of the extra parameter δ. It is known that η>1 gives the Gompertz distribution which presents mode at zero or, for η<1, having their modes at x=β−1 [1 − log(η)]. An initial discussion using the Kullback-Leibler distance derived in Section 4.3 points out that increasing the contamination (or the bias of the estimates) can affect the quality of fit.

In this study, the following scenarios are taken into account. For the sample size n=50,100,150,200, we adopt as the true parameters the following cases:

  • Scenario η<1: (α,β)=(1,2) and δ{0.3,1,4};

  • Scenario η=1: (α,β)=(2,2) and δ{0.3,1,4};

  • Scenario η>1: (α,β)=(4,2) and δ{0.3,1,4}.

Also, we use 10,000 Monte Carlo’s replications and, at each one of them, we quantify (i) the average of the MLEs and (ii) the mean square error (MSEs).

Table 3 gives the results of the simulation study. In general, the MLEs present smaller values of the biases and MSEs when the sample size increases. It is important to highlight the following atypical case: for the MLEs of α at the scenarios (α,δ,β){(1,4,2),(2,1,2),(4,0.3,2),(4,1,2)} and of δ at (4,0.3,2), the associated biases do not have an inverse monotonic relationship with sample sizes, as expected.However, based on the fact that their MSEs tend to zero, we can expect that there exists a sample size n0 such that biases of the MLEs decrease when the sample sizes increase from n0.
Table 3

Performance of the MLEs for the O G distribution

( α, δ, β)

n

θ ̂ i ¯ ( MSE ( θ ̂ i ) )

  

α ̂ ¯ (MSE ( α ̂ ))

δ ̂ ¯ (MSE ( δ ̂ ))

β ̂ ¯ (MSE ( β ̂ ))

For η<1

(1,0.3,2)

50

1.201

(2.837)

0.478

(0.883)

2.502

(1.698)

·

100

1.181

(1.745)

0.406

(0.290)

2.320

(1.238)

·

150

1.156

(1.299)

0.385

(0.195)

2.249

(1.015)

·

200

1.103

(1.008)

0.358

(0.134)

2.244

(0.899)

(1,1,2)

50

1.202

(1.965)

1.620

(5.938)

2.425

(1.630)

·

100

1.134

(1.199)

1.361

(2.690)

2.305

(1.145)

·

150

1.079

(0.884)

1.231

(1.638)

2.288

(0.979)

·

200

1.063

(0.735)

1.180

(1.244)

2.250

(0.845)

(1,4,2)

50

0.965

(0.810)

4.764

(26.798)

2.544

(1.561)

·

100

0.958

(0.544)

4.398

(14.813)

2.390

(1.025)

·

150

0.959

(0.443)

4.283

(11.454)

2.328

(0.831)

·

200

0.970

(0.369)

4.246

(8.953)

2.262

(0.653)

For η=1

(2,0.3,2)

50

2.246

(7.571)

0.426

(0.473)

2.787

(3.543)

·

100

2.137

(4.502)

0.361

(0.172)

2.561

(2.473)

·

150

2.073

(3.279)

0.341

(0.116)

2.471

(1.981)

·

200

2.011

(2.596)

0.324

(0.083)

2.434

(1.698)

(2,1,2)

50

2.161

(5.462)

1.481

(4.886)

2.687

(3.051)

·

100

2.012

(3.115)

1.199

(1.798)

2.543

(2.157)

·

150

1.947

(2.277)

1.100

(1.062)

2.483

(1.763)

·

200

1.923

(1.874)

1.056

(0.787)

2.430

(1.507)

(2,4,2)

50

1.805

(2.404)

4.534

(21.279)

2.785

(2.783)

·

100

1.817

(1.681)

4.202

(12.456)

2.572

(1.869)

·

150

1.828

(1.390)

4.097

(9.474)

2.487

(1.527)

·

200

1.861

(1.153)

4.075

(7.495)

2.388

(1.184)

For η>1

(4,0.3,2)

50

3.770

(13.137)

0.336

(0.191)

3.400

(6.701)

·

100

3.737

(8.129)

0.304

(0.072)

2.951

(4.152)

·

150

3.731

(6.119)

0.298

(0.051)

2.764

(3.184)

·

200

3.685

(4.865)

0.289

(0.038)

2.676

(2.613)

(4,1,2)

50

3.845

(13.615)

1.272

(3.153)

3.149

(6.239)

·

100

3.735

(7.757)

1.076

(1.043)

2.833

(4.060)

·

150

3.717

(5.760)

1.024

(0.634)

2.689

(3.150)

·

200

3.721

(4.759)

1.000

(0.472)

2.588

(2.601)

(4,4,2)

50

3.608

(8.172)

4.605

(21.140)

3.036

(5.150)

·

100

3.677

(5.234)

4.262

(11.467)

2.668

(2.989)

·

150

3.737

(4.039)

4.172

(8.228)

2.510

(2.169)

·

200

3.796

(3.247)

4.138

(6.370)

2.389

(1.588)

The results provide evidence that the scenarios under the condition η>1 yield a hard estimation (having larger variation ranges of the MSEs than those obtained for the cases when η<1) for α and β parameters, and that the MLEs present smaller values of the MSEs under such conditions. Figure 7 illustrates the above behavior for the cases δ{0.3,0.8,1,2,4} and n=200. In summary, the scenario with less numerical problems is (η,δ)=(2,0.1), whereas that one which requires more attention for estimating the O G parameters is (η,δ)=(0.5,4).
Figure 7

MSE curves for δ {0 . 3,1,4}, η =0 . 5,1,2 and n =200. (a) MSE ( α ̂ ) , (b) MSE ( β ̂ ) , (c) MSE ( δ ̂ ) .

7.2 Applications

Here, the usefulness of the O E W distribution is illustrated by means of two real data sets.

7.2.1 Uncensored data

Here, we compare the fits of some special models of the O E W family using a real data set. The estimation of the model parameters is performed by the maximum likelihood method discussed in Section 5. We use the maxLik function of the maxLik package in R language. In this function, if the argument “method” is not specified, a suitable method is selected automatically. For this application, we use the Newton-Raphson method. The data represent the percentage of body fat determined by underwater weighing for 250 men. For more details about the data see http://lib.stat.cmu.edu/datasets/bodyfat.

Table 4 provides some descriptive measures. They suggest an empirical distribution which is slightly asymmetric and platykurtic.
Table 4

Descriptive statistics

Mean

Mode

Median

Std. Desv.

Skewness

Kurtosis

Min

Max

19.30

20.40

19.25

8.23

0.19

2.62

3.00

47.50

We compare the classical models and generalized models within the O family. The null hypothesis H0:δ=1 is tested against H1:δ≠1 using the LR statistic. The comparisons are presented in Table 5. For the O W and O E P models, one cannot say that the parameter δ is statistically different from one at the 10% significance level. Based on this result, we fit the , exponential power ( E P ), O G and Marshall-Olkin flexible Weibull extension ( O F W E ) models to the current data (see Table 1). These models are compared with two other three-parameter models, namely: the modified Weibull ( W ) and generalized Birnbaum-Saunders ( G S ) (Owen 2006) distributions. The G S density is given by
f ( x ; ϕ , η , κ ) = 1 ϕ 2 πη x κ 1 κ + ηκ x exp 1 2 ϕ 2 ( x η ) 2 η x 2 κ , x > 0 .
Table 5

Comparison of fitted models using the LR test

Null hypothesis

Models

LR statistic

p-value

 

G × O G

11.2963

0.0008

 

W × O W

0.7638

0.3822

  δ=1

E P × O E P

2.1959

0.1384

 

F W E × O F W E

12.3659

0.0004

In Table 6, we present the MLEs (standard errors in parentheses) of the parameters of the fitted O F W E , O G , E P , , W and G S distributions. Also, we provide the goodness-of-fit measures (p-values in parentheses). Thus, these values indicate that the null models are strongly rejected for the O F W E and O G distributions, since the associated p-values are much lower than 0.001.
Table 6

MLEs and goodness-of-fit statistics

Model

Estimates (standard errors)

Goodness-of-fit ( p-value)

 

α ̂ (or ϕ ̂ )

δ ̂ (or η ̂ )

γ ̂ (or κ ̂ )

β ̂ (or λ ̂ )

AD

CM

O F W E

1

2.9136

0.0552

14.3666

0.1082

0.0115

 

×

(1.1321)

(0.0022)

(3.7615)

(0.9939)

(0.9987)

O G

0.1289

18.8183

0.0183

0.6825

0.0938

 

(0.0151)

(2.9308)

×

(0.0063)

(0.0739)

(0.1361)

E P

1

1

0.0359

1.7778

0.2537

0.0273

 

×

×

(0.0008)

(0.0870)

(0.7301)

(0.8800)

0.0004

1

2.5373

0.4344

0.0667

 

(0.0002)

×

(0.1434)

×

(0.2985)

(0.3079)

W

0.0007

1

2.2292

0.0149

0.2761

0.0384

 

(0.0007)

×

(0.4384)

(0.0191)

(0.6546)

(0.7094)

G S

1.3189

18.7623

0.1328

×

0.5672

0.0876

 

(0.1847)

(0.5784)

(0.0513)

×

(0.1404)

(0.1642)

Table 7 gives the values of the Akaike information criterion (AIC), Bayesian information criterion (BIC), consistent Akaike information criterion (CAIC) and Hannan-Quinn information criterion (HQIC). Since the values of the AIC, CAIC and HQIC are smaller for the O F W E distribution compared to those values of the other fitted models. Thus, this new distribution seems to be a very competitive model to explain the current data.
Table 7

Statistics AIC, BIC, CAIC and HQIC

Models

AIC

BIC

CAIC

HQIC

O F W E

1753.989

1764.553

1754.087

1758.241

O G

1767.305

1777.870

1767.403

1771.557

E P

1764.136

1771.178

1764.184

1766.970

1756.843

1763.886

1756.892

1759.678

W

1757.997

1768.561

1758.094

1762.248

G S

1761.136

1771.701

1761.234

1765.388

Figures 8(a) and 8(b) display the estimated density and survival functions of the O F W E distribution. The plots confirm the excellent fit of this distribution to the data. Figure 8(c) shows that the estimated O F W E hrf is an increasing curve.
Figure 8

Plots of the estimated. (a) density, (b) survivor function (c) and hrf of the O F W E model.

7.2.2 Censored data

Now, we consider a set of remission times from 137 cancer patients [Lee and Wang (2003), pag. 231]. Lee and Wang (2003) showed that the log-logistic ( ) model provides a good fit to the data. Ghitany et al. (2005) compared the fits of the O W and models to these data. Now, we present a more detailed study by comparing the fitted , , E P , O W , Marshall-Olkin log-logistic ( O ), O E P and G S models to these data. The functions H(x;γ,c)= log(1+γ x c ) and h(x;γ,c)=γ c xc−1/(1+γ x c ) are associated with the model.

The hypothesis that the underlying distribution is (or E P ) versus the alternative hypothesis that the distribution is the O W (or O E P ) is rejected with p-value = 0.0055 (or p-value = <0.0001). Further, the hypothesis test that the underlying distribution is versus the O distribution yields the p-value =1.0000. Thus, we compare the O W , O E P , and G S models to determine which model gives the best fit to the current data.

Table 8 lists the MLEs (and corresponding standard errors in parentheses) of the parameters and the values of the AD and CM statistics (their p-values in parentheses). The figures in this table, specially the p-values, suggest that the O W distribution yields a better fit to these data than the other three distributions.
Table 8

MLEs and goodness-of-fit statistics

Model

Estimates (standard errors)

Goodness-of-fit ( p-value)

 

α ̂ (or ϕ ̂ )

δ ̂ (or η ̂ )

γ ̂ (or κ ̂ or β ̂ )

c ̂ (or λ ̂ )

AD

CM

O W

0.0037

0.0736

1.5719

0.1889

0.0264

 

(0.0043)

(0.0727)

(0.1616)

×

(0.8994)

(0.8908)

O E P

0.0233

0.0144

1.6012

0.2057

0.0279

 

×

(0.0165)

(0.1423)

(0.0042)

(0.8686)

(0.8733)

G S

1.6313

7.1422

0.3356

1.2753

0.2116

 

(0.1226)

(0.7374)

(0.0314)

×

(0.0025)

(0.0038)

0.0427

1.6900

0.2891

0.0380

 

×

×

(0.0118)

(0.1249)

(0.6101)

(0.7164)

Table 9 lists the values of the AIC, BIC, CAIC and HQIC statistics. The figures in this table indicate that there is a competitiveness among the O W , O E P and models. However, if we observe the Figures 9(a), 9(b) and 9(c), we note that the O W and O E P models present better fits to the current data.
Table 9

Statistics AIC, BIC, CAIC and HQIC

Models

AIC

BIC

CAIC

HQIC

O W

843.1171

851.8770

843.2975

846.6769

O E P

843.1898

851.9498

843.3703

846.7497

G S

858.3686

867.1285

858.5490

861.9284

843.7586

849.5986

843.8481

846.1318

Figure 9

Plots of the estimated. (a) Q-Q plot of O W , (b) , (c) O E P distributions and (d) Kaplan-Meier curve estimated survival and upper and lower 95% confidence limits for the cancer patients.

Figure 9(d) really shows that the O W and O E P distributions present good fits to the current data. We can conclude that the O W and O E P distributions are excellent alternatives to explain this data set.

8 Conclusion

In this paper, the Marshall-Olkin extended Weibull family of distributions is proposed and some of its mathematical properties are studied. The maximum likelihood procedure is used for estimating the model parameters. Two special models in the family are described with some details. In order to assess the performance of the maximum likelihood estimates, a simulation study is performed by means of Monte Carlo experiments. Special models of the proposed family are compared (through goodness-of-fit measures) with other well-known lifetime models by means of two real data sets. The proposed model outperforms classical lifetime models to these data.

Appendix: An expansion for f(x;δ,α,ξ)F(x;δ,α,ξ) c

Here, we obtain an expansion for the quantity f(x;δ,α,ξ)F(x;δ,α,ξ) c . First, we consider an expansion for F(x;δ,α,ξ) c . Based on (5), the power of the cdf can be expressed as
F ( x ; δ , α , ξ ) c = { 1 exp [ αH ( x ; ξ ) ] } c A { 1 δ ¯ exp [ αH ( x ; ξ ) ] } c B .
Applying expansion (9), we have
A = k = 0 ( 1 ) k c k exp [ kαH ( x ; ξ ) ] .
Now, we expand the quantity B. Equation (9) under the restriction δ<1 (implying that δ ¯ exp [ αH ( x ; ξ ) ] < 1 ) yields
B = j = 0 ( c ) j j ! δ ¯ j exp [ jαH ( x ; ξ ) ] .
Moreover, it is clear that δ=1 implies B=1. Finally, for δ>1 (i.e., { 1 δ ¯ exp [ αH ( x ; ξ ) ] } > 1 ), the quantity B can be rewritten as
B = 1 [ 1 { 1 δ ¯ exp [ αH ( x ; ξ ) ] } 1 ] c .
Using the binomial expansion, we have
B = j = 0 ( 1 ) j c j 1 1 δ ¯ exp αH ( x ; ξ ) 1 j .
Thus,
F ( x ; δ , α , ξ ) c = I ( δ < 1 ) j , k = 0 ( 1 ) k ( c ) j j ! c k δ ¯ j exp [ ( j + k ) αH ( x ; ξ ) ] + I ( δ = 1 ) k = 0 ( 1 ) k c k exp [ kαH ( x ; ξ ) ] + I ( δ > 1 ) j , k = 0 ( 1 ) j + k c k c j exp [ kαH ( x ; ξ ) ] × [ 1 { 1 δ ¯ exp [ αH ( x ; ξ ) ] } 1 ] j .
Hence, based on Equation (13), the following expansion holds
f ( x ; δ , α , ξ ) F ( x ; δ , α , ξ ) c = v = 0 w v g x ; ( v + 1 ) α , ξ F ( x ; δ , α , ξ ) c = I ( δ < 1 ) j , k , v = 0 ( 1 ) k × w v ( c ) j j ! c k δ ¯ j exp [ ( j + k ) αH ( x ; ξ ) ] g ( x ; ( v + 1 ) α , ξ ) + I ( δ = 1 ) k , v = 0 ( 1 ) k w v c k exp [ kαH ( x ; ξ ) ] g ( x ; ( v + 1 ) α , ξ ) + I ( δ > 1 ) j , k , v = 0 ( 1 ) j + k w v c k c j exp [ kαH ( x ; ξ ) ] × 1 { 1 δ ¯ exp [ αH ( x ; ξ ) ] } 1 j g ( x ; ( v + 1 ) α , ξ ) .
(24)

Declarations

Acknowledgements

The authors gratefully acknowledge financial support from CAPES and CNPq. The authors are also grateful to three referees and an associate editor for helpful comments and suggestions.

Authors’ Affiliations

(1)
Departamento de Estatística, Universidade Federal de Campina Grande
(2)
Departamento de Estatística, Universidade Federal do Piauí
(3)
Departamento de Estatística, Universidade Federal do Rio Grande do Norte
(4)
Departamento de Estatística, Universidade Federal de Pernambuco, Cidade Universitária

References

  1. Bain LJ: Analysis for the linear failure-rate life-testing distribution. Technometrics 16: 551–559.MathSciNetView ArticleGoogle Scholar
  2. Barreto-Souza W, Lemonte AJ, Cordeiro GM: General results for the Marshall and Olkin’s family of distributions. An. Acad. Bras. Cienc 85: 3–21.MathSciNetView ArticleGoogle Scholar
  3. Bebbington M, Lai CD, Zitikis R: A flexible Weibull extension. Reliability Eng. Syst. Saf 92: 719–726.View ArticleGoogle Scholar
  4. Carrasco JMF, Ortega EMM, Cordeiro GM: A generalized modified Weibull distribution for lifetime modeling. Comput. Stat. Data Anal 53: 450–462.MathSciNetView ArticleGoogle Scholar
  5. Chen Z: A new two-parameter lifetime distribution with bathtub shape or increasing failure rate function. Stat. Probability Lett 49: 155–161.MathSciNetView ArticleGoogle Scholar
  6. Chen G, Balakrishnan N: A general purpose approximate goodness-of-fit test. J. Qual. Technol 27: 154–161.Google Scholar
  7. Cordeiro GM, Lemonte AJ: On the Marshall-Olkin extended Weibull distribution. Stat. Paper 54: 333–353.MathSciNetView ArticleGoogle Scholar
  8. Cover TM, Thomas JA: Elements of Information Theory. John Wiley & Sons, New York;View ArticleGoogle Scholar
  9. Cox DR, Hinkley DV: Theoretical Statistics. Chapman and Hall, London;View ArticleGoogle Scholar
  10. Doornik J: Ox 5: object-oriented matrix programming language. Timberlake Consultants, London; (2007)Google Scholar
  11. EL-Bassiouny AH, Abdo NF: Reliability properties of extended makeham distributions. Comput. Methods Sci. Technol 15: 143–149. (2009)View ArticleGoogle Scholar
  12. Fisk PR: The graduation of income distributions. Econometrica 29: 171–185.View ArticleGoogle Scholar
  13. Fréchet M: Sur la loi de probabilite de l’écart maximum.́. Ann. Soc. Polon. Math 6: 93–93. (1927)Google Scholar
  14. Garvan F: The Maple Book. Chapman and Hall/CRC, London; (2002)Google Scholar
  15. Gompertz B: On the nature of the function expressive of the law of human mortality and on the new model of determining the value of life contingencies. Philos. Trans. R. Soc. Lond 115: 513–585. (1825)View ArticleGoogle Scholar
  16. Guess F, Proschan F: Mean residual life: Theory and applications. In Handbook of Statistics, vol. 7. Edited by: Krishnaiah PR, Rao CR. Elsevier; http://dx.doi.org/10.1016/S0169-7161(88)07014-2Google Scholar
  17. Ghitany ME: Marshall-Olkin extended Pareto distribution and its application. Int. J. Appl. Math 18: 17–31.(2005)MathSciNetGoogle Scholar
  18. Ghitany ME, Kotz S: Reliability properties of extended linear failure-rate distributions. Probability Eng. Informational Sci 21: 441–450. (2007)MathSciNetView ArticleGoogle Scholar
  19. Ghitany ME, AL-Hussaini EK, AL-Jarallah: Marshall-Olkin extended Weibull distribution and its application to Censored data. J. Appl. Stat 32: 1025–1034. (2005)MathSciNetView ArticleGoogle Scholar
  20. Ghitany ME, AL-Awadhi FA, Alkhalfan LA: Marshall-Olkin extended Lomax distribution and its applications to censored data. Comm. Stat. Theor. Meth 36: 1855–1866. (2007)MathSciNetView ArticleGoogle Scholar
  21. Gurvich M, DiBenedetto A, Ranade S: A new statistical distribution for characterizing the random strength of brittle materials. J. Mater. Sci 32: 2559–2564. (1997)View ArticleGoogle Scholar
  22. Hartley RVLL: Transmission of information. Bell Syst. Techn. J 7: 535–563. (1928)View ArticleGoogle Scholar
  23. Jayakumar K, Mathew T: On a generalization to Marshall-Olkin scheme and its application to Burr type XII distribution. Stat. Paper 49: 421–439. (2008)MathSciNetView ArticleGoogle Scholar
  24. Johnson NL, Kotz S, Balakrishnan N: Continuous Univariate Distributions. Wiley, New York; (1994)Google Scholar
  25. Kies JA: The Strength of Glass, NRL Report 5093. Naval Research Lab., Washington, DC (1958)Google Scholar
  26. Krishna E, Jose KK, Ristić M: Applications of Marshal-Olkin Fréchet distribution. Comm. Stat. Simulat. Comput 42: 76–89. (2013)MathSciNetGoogle Scholar
  27. Kullback S, Leibler RA: On information and sufficiency. Ann. Math. Stat 22: 79–86. 1951MathSciNetView ArticleGoogle Scholar
  28. Lai CD, Xie M, Murthy DNP: A modified Weibull distribution. Trans. Reliab 52: 33–37. 2003View ArticleGoogle Scholar
  29. Lawless JF: Statistical Models and Methods for Lifetime Data. Wiley, New York; 2003Google Scholar
  30. Lee ET, Wang JW: Statistical Methods for Survival Data Analysis. Wiley, New York; 2003View ArticleGoogle Scholar
  31. Lomax KS: Business failures; another example of the analysis of failure data. J. Am. Stat. Assoc 49: 847–852. 1954View ArticleGoogle Scholar
  32. Marshall A, Olkin I: A new method for adding a parameter to a family of distributions with application to the exponential and Weibull families. Biometrika 84: 641–652. 1997MathSciNetView ArticleGoogle Scholar
  33. McCool JI: Using the Weibull Distribution: Reliability, Modeling and Inference. John Wiley & Sons, New Jersey; 2012View ArticleGoogle Scholar
  34. Nadarajah S, Kotz S: On some recent modifications of Weibull distribution. IEEE Trans. Reliab 54: 561–562. 2005View ArticleGoogle Scholar
  35. Nascimento ADC, Cintra RJ, Frery AC: Hypothesis testing in speckled data with stochastic distances. IEEE Trans. Geosci. Remote Sensing 48: 373–385. 2010View ArticleGoogle Scholar
  36. Nascimento ADC, Horta MM, Frery AC, Cintra RJ: Comparing edge detection methods based on stochastic entropies and distances for PolSAR imagery. IEEE J. Selected Topics Appl. Earth Observations Remote Sensing 7: 648–663. 2014View ArticleGoogle Scholar
  37. Nikulin M, Haghighi F: A chi-squared test for the generalized power Weibull family for the head-and-neck cancer censored data. J. Math. Sci 133: 1333–1341. 2006MathSciNetView ArticleGoogle Scholar
  38. Owen WJ: A new three-parameter extension to the Birnbaum-Saunders distribution. IEEE Trans. Reliab 55: 475–479. 2006View ArticleGoogle Scholar
  39. Pham H: A vtub-shaped hazard rate function with applications to system safety. Int. J. Reliab. Appl 3: 1–16. 2002Google Scholar
  40. Phani KK: A new modified Weibull distribution function. Commun. Am. Ceramic Soc 70: 182–184. 1987Google Scholar
  41. R Development Core Team: R: A Language and Environment for Statistical Computing. R Foundation for Statistical Computing, Vienna; 2009Google Scholar
  42. Rayleigh JWS: On the resultant of a large number of vibrations of the same pitch and of arbitrary phase. Phil. Mag 10: 73–78. 1880View ArticleGoogle Scholar
  43. Ristić MM, Jose KK, Ancy J: A Marshall-Olkin gamma distribution and minification process. STARS: Stress Anxiety Res. Soc 11: 107–117. 2007Google Scholar
  44. Rodriguez N: A guide to the Burr type XII distributions. Biometrika 64: 129–134. 1977MathSciNetView ArticleGoogle Scholar
  45. Sankaran PG, Jayakumar K: On proportional odds model. Stat. Paper 49: 779–789. 2008MathSciNetView ArticleGoogle Scholar
  46. Seghouane A-K, Amari S-I: The AIC criterion and symmetrizing the Kullback–Leibler divergence. IEEE Trans. Neural Netw 18: 97–106. 2007View ArticleGoogle Scholar
  47. Shannon CE: A mathematical theory of communication. Bell Syst. Techn. J 27: 379–423. 1948MathSciNetView ArticleGoogle Scholar
  48. Sigmon K, Davis TA: MATLAB Primer. Chapman and Hall/CRC, London; 2002Google Scholar
  49. Smith RM, Bain LJ: An exponential power life testing distribution. Comm. Stat. Theor. Meth 4: 469–481. 1975View ArticleGoogle Scholar
  50. Wolfram S: The Mathematica Book. Wolfram Media, Cambridge; 2003Google Scholar
  51. Xie M, Lai D: Reliability analysis using additive Weibull model with bathtub-shaped failure rate function. Reliab. Eng. Syst. Saf 52: 87–93. 1995View ArticleGoogle Scholar
  52. Xie M, Tang Y, Goh TN: A modified Weibull extension with bathtub-shaped failure rate function. Reliab. Eng. Syst. Saf 76: 279–285. 2002View ArticleGoogle Scholar
  53. Zhang T, Xie M: Failure data analysis with extended Weibull distribution. Comm. Stat. Simulat. Comput 36: 579–592. 2007MathSciNetView ArticleGoogle Scholar

Copyright

© Santos-Neto et al.; licensee Springer. 2014

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited.