Communications in Statistics - Theory and Methods

Volume 47, 2018 - Issue 10

A reduced new modified Weibull distribution

Saad J. Almalki
School of Mathematics, University of Manchester, Manchester M13 9PL, UK
arXiv:1307.3925v1 [stat.ME] 15 Jul 2013

Abstract

In this paper, we propose a reduced version of the new modified Weibull (NMW)
distribution due to Almalki and Yuan [2] in order to avoid some estimation problems.
The number of parameters in the NMW distribution is five. The number of parameters
in the reduced version is three. We study mathematical properties as well as maximum
likelihood estimation of the reduced version. Four real data sets (two of them complete
and the other two censored) are used to compare the flexibility of the reduced version
versus the NMW distribution. It is shown that the reduced version has the same
desirable properties of the NMW distribution in spite of having two less parameters.
The NMW distribution did not provide a significantly better fit than the reduced
version for any of the four data sets.

Keywords. Hazard rate function; Maximum likelihood estimation; Weibull distribu-

tion.

1 Introduction

In reliability engineering and lifetime analysis many applications require a bathtub shaped
hazard rate function. The traditional Weibull distribution [27], which includes the expo-
nential and Rayleigh distributions as particular cases, is one of the important lifetime
distributions. Unfortunately, however, it does not exhibit a bathtub shape for its hazard

1
rate function. Many researchers have proposed modifications and generalizations of the
Weibull distribution to accommodate bathtub shaped hazard rates. Extensive reviews of
these modifications have been presented by many authors, see, for example, Rajarshi and
Rajarshi [20], Murthy et al. [15], Pham and Lai [18] and Lai et al. [11].

Although some flexible distributions among these modifications have only two or three
parameters (for example, the flexible Weibull extension [4], the modified Weibull (MW)
distribution [12] and the modified Weibull extension [29]), most modifications of the
Weibull distribution have four or five parameters. By combining two Weibull distribu-
tions, with one having an increasing hazard rate function and the other decreasing one, Xie
and Lai [28] introduced a four-parameter distribution called the additive Weibull (AddW)
distribution. Sarhan and Apaloo [22] proposed the exponentiated modified Weibull ex-
tension which exhibits a bathtub shaped hazard rate. Sarhan et al. [21] proposed the
exponentiated generalized linear exponential distribution which generalizes a large set of
distributions including the exponentiated Weibull distribution. Famoye et al. [7] proposed
the beta-Weibull (BWD) distribution with unimodal, increasing, decreasing or bathtub
shaped hazard rate functions. Another four-parameter distribution, called the general-
ized modified Weibull (GMW) distribution, was proposed by Carrasco et al. [5]. The
hazard rate function of this distribution can be increasing, decreasing, bathtub shaped or
unimodal. Cordeiro et al. [6] introduced the Kumaraswamy Weibull (KumW) distribu-
tion and studied its mathematical properties extensively. It has four parameters, three of
which are shape parameters, making it so flexible. Its hazard function can be constant,
increasing, decreasing, bathtub shaped or unimodal. A five-parameter distribution was
introduced as a modification of the Weibull distribution by Phani [19]. This distribution
generalizes the four-parameter Weibull distribution proposed by Kies [8]. Another five-
parameter distribution, the beta modified Weibull (BMW) distribution, was introduced
by Silva et al. [25]. It allows for different hazard rate shapes: increasing, decreasing,
bathtub shaped and unimodal. It was shown to fit bathtub shaped data sets very well.
Recently, a new five-parameter distribution called the beta generalized Weibull distribu-
tion was proposed by Singla et al. [26]. The hazard rate function of this distribution can
be increasing, decreasing, bathtub shaped or unimodal. It contains as sub-models some

2
well known lifetime distributions.

Almalki and Yuan [2] introduced a new modified Weibull (NMW) distribution. It
generalizes several commonly used distributions in reliability and lifetime data analysis,
including the MW distribution, the AddW distribution, the modified Weibull distribution
of Sarhan and Zaindin (SZMW) [23], the Weibull distribution, the exponential distribution,
the Rayleigh distribution, the extreme-value distribution and the linear failure rate (LFR)
[3] distribution. The cumulative distribution function (CDF) of the NMW distribution is

θ −βxγ eλx
F (x) = 1 − e−αx

for x > 0, γ > 0, θ > 0, α > 0, β > 0 and λ > 0, where γ, θ are shape parameters, α,
β are scale parameters and λ is an acceleration parameter. Almalki and Yuan [2] derived
mathematical properties of this distribution as well as estimated its parameters by the
method of maximum likelihood with application to real data sets.

The hazard rate function of the NMW distribution can be increasing, decreasing or
bathtub shaped. It has been shown to be the best lifetime distribution to date in terms of
fitting some popular and widely used real data sets like Aarset data [1] and voltage data
[13].

Although distributions with four or more parameters are flexible and exhibit bathtub
shaped hazard rates, they are also complex [16] and cause estimation problems as a con-
sequence of the number of parameters, especially when the sample size is not large. The
main purpose of this work is to reduce the number of parameters of the NMW distribution
so as to address these problems while maintaining the same flexibility to fit data so well.
1
This can be achieved by choosing the two shape parameters γ = θ = 2 as we now show.

There are various tools to assess the flexibility of a given univariate distribution. One
commonly used tool is the kurtosis-skewness plot. Values of (skewness, kurtosis) plotted
on the (x, y) plane for all possible values of the parameters of the distribution give what is
referred to as the kurtosis-skewness plot. The area or the range covered by the kurtosis-
skewness plot is a measure of flexibility of the distribution.

3
 The kurtosis-skewness plot for the NMW distribution is drawn on the left hand side
of Figure . The values of (skewness, kurtosis) were computed over θ = 0.1, 0.2, . . . , 5,
γ = 0.1, 0.2, . . . , 5, θ = γ 6= 12 , α = 0.1, 0.2, . . . , 2, β = 0.1, 0.2, . . . , 2 and λ = 0.1, 0.2, . . . , 2.
The kurtosis-skewness plot for the reduced distribution is drawn on the right hand side
of Figure . The values of (skewness, kurtosis) were computed over α = 0.1, 0.2, . . . , 2,
β = 0.1, 0.2, . . . , 2 and λ = 0.1, 0.2, . . . , 2.

1
We can see that the particular case θ = γ = 2 is a flexible member of the NMW
distribution. The range of kurtosis values is widest for the case θ = γ = 21 . The range of
skewness values is also widest for the case θ = γ = 21 , but some negative skewness values
are not accommodated by this case.

The rest of this paper is organized as follows. The reduced distribution is introduced
in Section 2. Section 3 considers the hazard rate function of the reduced distribution.
The moments, the moment generating function and the distribution of order statistics
are derived as particular cases of the NMW distribution in Section 4. Section 5 discusses
maximum likelihood estimation of the unknown parameters using complete and censoring
data. Four real data sets, uncensored and censored, are analyzed in Section 6. Finally,
Section 7 concludes the paper.

2 The reduced distribution

1
Setting γ = θ = 2 in the CDF of the NMW distribution, we obtain the CDF of the reduced
new modified Weibull (RNMW) distribution as

√ √
x−β xeλx
F (x) = 1 − e−α

for x > 0, α > 0, β > 0 and λ > 0, where α, β are scale parameters and λ is an acceleration
parameter. This reduced version of the NMW distribution has a bathtub shaped hazard
rate function, as will be shown later.

4
 The corresponding probability density function (PDF) is

1 h i √ √ λx
f (x) = √ α + β(1 + 2λx)eλx e−α x−β xe
2 x

for x > 0. The PDF of the RNMW distribution can be decreasing or decreasing then
increasing-decreasing as shown in Figure 1. It is clear that the reduced distribution is
nearly as flexible as the NMW distribution.

0.9 α=0.05, β=0.00025, λ=2.2

α=1, β=0.7, λ=0.15
0.8
α=0.5, β=0.015, λ=2.75
0.7
The probability density function

α=0.15, β=0.025, λ=1.5

0.6

0.5

0.4

0.3

0.2 α=0.75, β=0.05, λ=1

0.1

0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
X

Figure 1: PDFs of the RNMW distribution.

3 The hazard rate function

The hazard rate function of the RNMW distribution is

1 h i
h(x) = √ α + β(1 + 2λx)eλx (1)
2 x

for x > 0. To derive the shape of h(x), we obtain the first derivative of log {h(x)}:

d 1 βλ(2λx + 3)eλx
log {h(x)} = − + . (2)
dx 2x α + β(2λx + 1)eλx

Setting this to zero, we have

1 βλ(2λx + 3)eλx
− + = 0. (3)
2x α + β(2λx + 1)eλx

5
 d d
Let x0 denote the root of (3). From (2), dx log {h(x)} < 0 for x ∈ (0, x0 ), dx log {h(x0 )} =
d
0, and dx log {h(x)} > 0 for x > x0 . So, h(x) initially decreases before increasing. Hence,
we have a bathtub shape. Let x0 denote the solution of (3); that is, the solution of

4βλx2 + 4βλx − β eλx = α. (4)

The left hand side of (4) is eλx multiplied by the quadratic function (4βλx2 + 4βλx − β).
The value x0 is unique and positive as shown in Figure 2.

8 6

α=0.75 α=2
7 β=0.05
β=0.5 5 2
λ=1.25 2 λ=0.75 (4β λx +4βλx−β) exp(λ x)
6 (4β λx +4βλx−β) exp(λ x)

4
5

4 3

Y
Y

y=α
3 2

2
1
1
y=α
0
0

−1 −1
0 x 0.5 1 1.5 0 0.5 1 1.5 x 2 2.5 3 3.5 4 4.5 5
0 0
X X

(a) (b)

d
Figure 2: The root of dx log {h(x0 )} = 0.

Plots of the hazard rate function of the RNMW distribution are shown in Figure 3
(a). Many different applications in reliability and lifetime analysis require bathtub shaped
hazard rate functions with a long useful life period and the middle period of the bathtub
shape having a relatively constant hazard rate (for example, electric machine life cycles
and electronic devices [10]). The NMW distribution has this property, and so does the
RNMW distribution, as Figure 3(b) shows.

6
 −3
x 10
7 2

1.8
α=0.75, β=0.7, λ=.5
6
α=0.25, β=0.015, λ=2.5 α=0.003, β=5x10−6 , λ=0.025
1.6
α=0.002, β=2.5x10−8, λ=0.05
5 1.4 α=0.005, β=1.5x10−4, λ=0.012

The hazard function

The failure rate function

1.2
4

3
0.8
α=0.15, β=0.00015, λ1=2

α=2.5, β=0.05, λ=0.5 0.6

2

0.4
1
0.2
α=1.5, β=0.001, λ=1
0 0
0 1 2 3 4 5 6 0 50 100 150 200 250
X X

(a) (b)

Figure 3: Hazard rate function of the RNMW distribution.

4 The moments, the moment generating function and order

statistics

Let X denote a random variable having the RNMW distribution. The rth moment of X
can be derived from Section 3.2 in Almalki and Yuan [2] as

X∞
(−β)n (nλ)m Γ (n + 2(m + r))
µ′r = 2r (5)
n,m=0
n!m! αn+2(m+r)

for r = 1, 2, . . .. The moment generating function of X is

X
∞  
(−β)n (nλ)m tk+1 Γ (n + 2(m + k) + 2)
MX (t) = 1 + 2 , (6)
n,m,k=0
i!j!k! αn+2(m+k)+2

see Appendix A for a proof. Using (6), the first four moments of X are

X∞  
′ (−β)n (nλ)m Γ (n + 2m + 2)
MX (0) = =2 µ′1 ,
i!j!k! αn+2m+2
n,m=0
X∞  
′′ ′′ (−β)n (nλ)m Γ (n + 2m + 4)
MX (0) = µ2 = 4 ,
n,m=0
i!j!k! αn+2m+4
X∞  
(3) (3) (−β)n (nλ)m Γ (n + 2m + 6)
MX (0) = µ3 = 6 ,
i!j!k! αn+2m+6
n,m=0
X∞  
(4) (4) (−β)n (nλ)m Γ (n + 2m + 8)
MX (0) = µ2 = 8 n+2m+8
.
n,m=0
i!j!k! α

These expressions are consistent with the formula for moments in (5).

7
 Let X1 , . . . , Xn denote a random sample drawn from the RNMW distribution with
parameters α, β and λ. The PDF of the rth order statistic say X(r) can be derived as a
particular case from Section 3.3 of [2]:
  r−1  
n−1 X r−1 (−1)ℓ
fr:n (x) = n f (x; αℓ , βℓ , λ) ,
r−1 ℓ (n + ℓ + 1 − r)
ℓ=0

where f (x; αℓ , βℓ , λ) is the RNMW PDF with parameters αℓ = (n + ℓ + 1 − r)α, βℓ =

(n + ℓ + 1 − r)β and λ. The kth non-central moment of the rth order statistic is then
  ∞ r−1  
′ (r:n) n − 1 X X r − 1 (−1)ℓ+1 β i (iλ)j Γ (i + 2(j + k))
µk = 2nk .
r−1 ℓ
i,j=0 ℓ=0αi−1 [α(n + ℓ + 1 − r)]2(j+k)+1

5 Parameter estimation

In this section, point and interval estimators of the unknown parameters of the RNMW
distribution are derived using the maximum likelihood method. We consider both complete
data and censored data.

5.1 Complete data

The PDF of the RNMW distribution can be rewritten as

√ √
x−β xeλx
f (x) = h (x; ϑ) e−α

for x > 0, where h (x; ϑ) is the hazard rate function in (1) and ϑ = (α, β, λ) is a vector of
parameters.

Let x1 , . . . , xn denote a random sample of complete data from the RNMW distribution.
Then, the log-likelihood function is
n h
X i
√ √
ℓ (ϑ) = log (h (xi ; ϑ)) − α xi − β xi eλxi .
i=1

The likelihood equations are obtained by setting the first partial derivatives of ℓ with

8
respect to α, β and λ to zero; that is,
n
X X√ n
1
√
− xi = 0, (7)
i=1
h (xi ; α, β, λ) 2 xi i=1
Xn Xn
(0.5 + λxi ) eλxi √
√ − eλxi xi = 0, (8)
h (xi ; α, β, λ) xi
i=1 i=1
n √
n q
X xi 3 + λxi eλxi X
2 λxi
− e x3i = 0. (9)
h (xi ; α, β, λ)
i=1 i=1

5.2 Censored data

Here, we consider maximum likelihood estimation for censored data without replacement.
Let Xi and Ci denote the lifetime and the censoring time for tested individual i, i =
1, . . . , n. Suppose Xi and Ci are independent random variables. The failure times are
xi = min(Xi , Ci ), i = 1, . . . , n. Then, the log-likelihood function is
d
X h √ √ i Xh √ √ i
ℓ (ϑ) = log h (xi ; ϑ) − α xi − β xi eλxi − α xi + β xi eλxi ,
i=1 i∈C

where d is the number of failures and C indexes the censored observations.

Setting the first partial derivatives of ℓ (ϑ) with respect to α, β and λ to zero, the
likelihood equations are obtained as
d
( )
X 1 √ X√
√
− xi − xi = 0, (10)
i=1
h (xi ; α, β, λ) 2 xi i∈C
X d   X
(0.5 + λxi ) eλxi λxi √ √ λxi
√ −e xi − xi e = 0, (11)
h (xi ; α, β, λ) xi
i=1 i∈C
(√
)
X d
xi 23 + λxi eλxi q X q
λxi 3 λxi
−e xi − e x3i = 0. (12)
h (xi ; α, β, λ)
i=1 i∈C

By solving the systems of nonlinear likelihood equations, (7,8,9) and (10,11,12), nu-
merically for α, β and λ, we can obtain maximum likelihood estimates for complete and
censored data.

According to Miller [14], the MLEs (b b λ)

α, β, b of (α, β, λ) have an approximate multi-

variate normal distribution with mean (α, β, λ) and variance-covariance matrix Id

−1 ; that

9
is,
   
bλ
b, β,
α b ∼ N3 (α, β, λ), Id
−1 ,

where
 
 ℓαα ℓαβ ℓαλ 
 
I = −
 ℓββ ℓβλ 
.
 
ℓλλ
The second order partial derivatives of ℓ (ϑ) are given in Appendices B and C.

6 Applications

This section provides four applications, two of them are for complete (uncensored) data
sets and the others are for censored data sets, to show how the RNMW distribution can
be applied in practice. Almalki and Yuan [2] have shown that the NMW distribution
fits data sets better than existing modifications of the Weibull distribution like the BMW
distribution, the AddW distribution, the MW distribution and the SZMW distribution.
So, we shall compare the fits of the RNMW and NMW distributions to see if the former
can perform as well as the NMW distribution. The Kolmogorov-Smirnov (K-S) statistic
(the distance between the empirical CDFs and the fitted CDFs), the Akaike informa-
tion criterion (AIC), the Bayesian information criterion (BIC) and the consistent Akaike
information criterion (CAIC) are used to compare the candidate distributions. The log-
likelihood ratio test is used to compare the NMW and RNMW distributions by testing the
1
hypotheses: H0 : θ = γ = 2 versus H1 : H0 is false. The likelihood ratio test statistic for
testing H0 against H1 is ω = 2(LN M W − LRN M W ), which follows a χ2 distribution with
two degrees of freedom under H0 .

6.1 Complete data

In this section, we show how the RNMW distribution can be applied in practice for two
complete (uncensored) real data sets.

10
 Table 1: MLEs of parameters, standard errors, AIC, BIC and CAIC for Aarset data.

Model b
α βb b
γ θb b
λ AIC BIC CAIC

NMW 0.071 7.015 × 10−8 0.016 0.595 0.197 435.8 445.4 437.2
(0.031) (1.501 × 10−7 ) (3.602) (0.128) (0.184)
1 1
RNMW 0.102 3.644 × 10−8 2 2 0.180 433.1 439.0 433.8
(0.019) (6.089 × 10−8 ) − − (0.020)

6.1.1 Aarset data

The Aarset data [1] consisting of lifetimes of fifty devices is widely used in lifetime anal-
ysis. The data set exhibits a bathtub shaped hazard rate. Both the NMW and RNMW
distributions were fitted to this data set. Table 1 gives the MLEs of the parameters,
corresponding standard errors, AIC, BIC, and CAIC. Table 2 provides the K-S test statis-
tics. Figures 4a and 4b show the histogram of the data, PDFs of the fitted NMW and
RNMW distributions, the empirical survival function, and the survival functions of the
fitted NMW and RNMW distributions.

It is clear that both the NMW and RNMW distributions provide adequate fits. Both
have very small K-S values (0.088 and 0.092, respectively). The NMW distribution has the
larger log-likelihood of -212.9. However, the RNMW distribution has the smaller values
1
for AIC, BIC and CAIC. The likelihood ratio test statistic for testing H0 : θ = γ = 2

versus H1 : H0 is false is ω = 1.436 and the corresponding p-value is 0.488, so there is

no evidence to reject H0 . Hence, the NMW distribution does not improve significantly on
the fit of the RNMW distribution.

The plots of the empirical TTT-transform, TTT-transforms of the fitted NMW and
RNMW distributions, the nonparametric hazard rate function, and the hazard rate func-
tions of the fitted NMW and RNMW distributions are shown in Figures 4c and 4d. It is
clear that the RNMW distribution provides as good a fit as the NMW distribution.

11
 Table 2: K-S statistics for models fitted to Aarset data.

Model K-S

NMW 0.088
RNMW 0.092

The variance-covariance matrix for the fitted RNMW distribution is

 
−4 −9 −3
 5.203 × 10 −3.697 × 10 1.214 × 10 
 
−1 
I =  −3.697 × 10−9 9.078 × 10−14 −2.994 × 10−8  .
 
1.214 × 10−3 −2.994 × 10−8 9.882 × 10−3

So, approximate 95 percent confidence intervals for the parameters α, β and λ are [0.064, 0.14],
   
0, 1.558 × 10−7 and 0.219 × 10−5 , 0.142 , respectively.

6.1.2 Kumar data

Kumar et al. [9] presented data consisting of times between failures (TBF) in days of load-
haul-dump machines (LHD) used to pick up rock or waste. This data set also exhibits a
bathtub shaped hazard rate function as shown in Figures 5c and 5d.

Tables 3 and 4 show the MLEs of the parameters, corresponding standard errors, AIC,
BIC, CAIC and the K-S test statistics. Both distributions (NMW and RNMW) provide
adequate fits. The log-likelihood is larger for the NMW distribution. That for the RNMW
distribution is only slightly smaller. The K-S statistic is larger for the NMW distribution.
However, the RNMW distribution has the smaller AIC, BIC and CAIC values.

Figures 5a-d show that both distributions fit the data adequately. However, the log-
1
likelihood ratio statistic for testing H0 : θ = γ = 2 versus H1 : H0 is false is ω = 0.895
with the corresponding p-value of 0.639. Hence, again there is no evidence that the NMW
distribution provides a better fit than the RNMW distribution.

12
 0.030

1.0
Kaplan−Meier
0.025

RNMW
RNMW NMW
NMW

0.8
0.020

0.6
0.015
f(x)

S(x)

0.4
0.010
0.005

0.2
0.000

0.0
0 20 40 60 80 0 20 40 60 80

x X

(a) (b)

1
0.15

0.9 Nonpar
RNMW
0.8 NMW
0.10

0.7
Scaled TTT−Transform

0.6
h(x)

0.5
0.05

0.4

0.3

0.2
0.00

Empirical
0.1 RNMW
NMW 0 20 40 60 80
0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

(c) (d)

Figure 4: For Aarst data: (a) Histogram and fitted PDFs; (b) Empirical and fitted survival
functions; (c) Empirical and fitted TTT-transforms; (c) Nonparametric and fitted hazard
rate functions.

13
 Table 3: MLEs of parameters, standard errors, AIC, BIC and CAIC for Kumar data.

Model b
α βb b
γ θb b
λ AIC BIC CAIC

NMW 0.034 4.588 × 10−4 0.423 0.657 0.071 410.0 418.9 411.6
(0.022) (3.794 × 10−3 ) (2.351) (0.18) (0.034)
1 1
RNMW 0.055 5.852 × 10−4 2 2 0.065 406.9 412.2 407.5
(0.017) (6.523 × 10−4 ) − − (0.012)

Table 4: K-S statistics for models fitted to Kumar data.

Model K-S

NMW 0.068
RNMW 0.061

The variance-covariance matrix for the fitted RNMW distribution is

 
−4 −6 −5
 2.839 × 10 −3.968 × 10 6.731 × 10 
 
−1 
I =  −3.968 × 10 −6 4.254 × 10 −7 −7.836 × 10−6 .

 
6.731 × 10−5 −7.836 × 10−6 1.498 × 10−4

So, approximate 95 percent confidence intervals for the parameters α, β and λ are [0.022, 0.088],
 
0, 1.864 × 10−3 and [0.041, 0.089], respectively.

6.2 Censored data

In this section, we show how the RNMW distribution can be applied in practice for two
real censored data sets, one of which is presented here for the first time.

6.2.1 Drug data

This data set was collected from a prison in the Middle East in 2011. It represents a sample
of eighty two prisoners imprisoned for using or selling drugs. They were all released as

14
 0.020

1.0
Kaplan−Meier
RNMW
RNMW
NMW

0.8
NMW
0.015

0.6
0.010
f(x)

S(x)

0.4
0.005

0.2
0.000

0.0
0 20 40 60 80 100 0 20 40 60 80 100

x X

(a) (b)
0.25

0.9 Nonpar
RNMW
0.20

0.8 NMW

0.7
Scaled TTT−Transform

0.15

0.6
h(x)

0.5
0.10

0.4
0.05

0.3

0.2
0.00

Empirical
0.1 RNMW
NMW
0 20 40 60 80 100
0
0 0.2 0.4 0.6 0.8 1
X X

(c) (d)

Figure 5: For Kumar data: (a) Histogram and fitted PDFs; (b) Empirical and fitted
survival functions; (c) Empirical and fitted TTT-transforms; (c) Nonparametric and fitted
hazard rate functions.

15
 Table 5: MLEs of parameters, standard errors, AIC, BIC and CAIC for drug data.

Model b
α βb b
γ θb b
λ AIC BIC CAIC

NMW 0.019 2.954 × 10−3 0.484 0.735 0.031 700.9 713.0 701.7
(9.426 × 10−3 ) (3.656 × 10−3 ) (0.504) (0.159) (0.019)
1 1
RNMW 0.038 5.863 × 10−3 2 2 0.026 698.4 705.6 698.7
(0.017) (6.584 × 10−3 ) − − (9.319 × 10−3 )

part of a general amnesty for prisoners. We consider the time from release to reoffending
to be the failure time. Of the eighty two prisoners, sixty six were arrested again for abuse
or sale of drugs. After one hundred and eleven weeks, the others were considered to be
censored.

Both the NMW and RNMW distributions were fitted to the data. Tables 5 and 6 show
the MLEs of the parameters, corresponding standard errors, AIC, BIC, CAIC and the K-S
test statistics. We see that the RNMW distribution has the smaller AIC, BIC and CAIC
values. The K-S statistic values for both distributions are approximately equal to 0.055.

Figures 6a-d show the histogram of the data, PDFs of the fitted NMW and RNMW
distributions, the empirical survival function, the survival functions of the fitted NMW
and RNMW distributions, the empirical TTT-transform, TTT-transforms of the fitted
NMW and RNMW distributions, the nonparametric hazard rate function, and the hazard
rate functions of the fitted NMW and RNMW distributions. We can see that the RNMW
distribution fits the data as well as the NMW distribution.

1
The log-likelihood ratio statistic for testing H0 : θ = γ = 2 versus H1 : H0 is false is
ω = 1.496 with the corresponding p-value of 0.473. Hence, again there is no evidence that
the NMW distribution provides a better fit than the RNMW distribution.

16
 Table 6: K-S statistics for models fitted to the drug data.

Model K-S

NMW 0.0550
RNMW 0.0553

0.05
1.0

Nonpar
RNMW

0.04
NMW
0.8

0.03
0.6
S(x)

h(x)

0.02
0.4

0.01
Kaplan−Meier
RNMW
NMW
0.2

0.00

0 20 40 60 80 100 0 20 40 60 80 100

x X

(a) (b)

Figure 6: For drug data: (a) Histogram and fitted PDFs; (b) Empirical and fitted survival
functions; (c) Empirical and fitted TTT-transforms; (c) Nonparametric and fitted hazard
rate functions.

The variance-covariance matrix for the fitted RNMW distribution is

 
−4 −5 −4
 2.801 × 10 −8.519 × 10 1.138 × 10 
 
I =
−1
 −8.519 × 10
−5 4.335 × 10−5 −6.037 × 10−5 .
 
1.138 × 10−4 −6.037 × 10−5 8.684 × 10−5
 
So, approximate 95 percent confidence intervals for the parameters α, β and λ are 5.553 × 10−3 , 0.071 ,
 
[0, 0.019] and 8.137 × 10−3 , 0.045 , respectively.

17
Table 7: MLEs of parameters, standard errors, AIC, BIC and CAIC for serum-reversal
data.

Model b
α βb b
γ θb b
λ AIC BIC CAIC

NMW 1.74 × 10−3 6.141 × 10−4 0.542 0.438 0.015 783.8 798.8 803.8
(3.115 × 10−3 ) (2.274 × 10−3 ) (0.507) (0.763) (3.764 × 10−3 )
1 1
RNMW 1.799 × 10−3 5.955 × 10−4 2 2 0.014 780.1 789.1 792.1
(1.971 × 10−3 ) (4.174 × 10−4 ) − − (2.061 × 10−3 )

Table 8: K-S statistics for models fitted to serum-reversal data.

Model K-S

NMW 0.115
RNMW 0.107

6.2.2 Serum-reversal data

Serum-reversal data consists of serum-reversal time in days of one hundred and forty eight
children contaminated with HIV from vertical transmission at the university hospital of
the Ribeirão Preto School of Medicine (Hospital das Clı́nicas da Faculdade de Medicina
de Ribeirão Preto) from 1986 to 2001 [24], [17].

Tables 7 and 8 show the MLEs of the parameters, corresponding standard errors, AIC,
BIC, CAIC and the K-S test statistics. We see that the RNMW distribution has smaller
values for AIC, BIC and CAIC. The K-S statistic is only slightly larger for the NMW
distribution.

Figures 7a-d show that both distributions fit the data adequately. However, the log-
1
likelihood ratio statistic for testing H0 : θ = γ = 2 versus H1 : H0 is false is ω = 0.194
with the corresponding p-value of 0.908. Hence, again there is no evidence that the NMW
distribution provides a better fit than the RNMW distribution.

18
 0.025
1.0

Nonpar
RNMW

0.020
0.8

NMW

0.015
0.6
S(x)

h(x)

0.010
0.4

0.005
Kaplan−Meier
0.2

RNMW
NMW

0.000
0.0

0 50 100 150 200 250 300 350 0 100 200 300

x X

(a) (b)

Figure 7: For serum data: (a) Histogram and fitted PDFs; (b) Empirical and fitted survival
functions; (c) Empirical and fitted TTT-transforms; (c) Nonparametric and fitted hazard
rate functions.

The variance-covariance matrix for the fitted RNMW distribution is

 
−6 −7 −6
 3.886 × 10 −4.245 × 10 1.990 × 10 
 
−1 
I =  −4.245 × 10 −7 1.742 × 10 −7 −8.445 × 10−7 .

 
1.990 × 10−6 −8.445 × 10−7 4.246 × 10−6
 
So, approximate 95 percent confidence intervals for the parameters α, β and λ are 0, 5.662 × 10−3 ,
 
0, 1.414 × 10−3 and [0.010, 0.080], respectively.

7 Conclusions and further discussion

The NMW distribution introduced by Almalki and Yuan [2] has been simplified with
its five parameters reduced to three. The simplified distribution has been referred to as
the RNMW distribution. We have studied several analytical properties of the RNMW
distribution and shown it to be a tractable distribution. We have also shown that the
RNMW distribution provides excellent fits to four real data sets: two of them are complete

19
data sets and the other two are censored. By means of the likelihood ratio test, we have
shown that the fit of the NMW distribution is not significantly better than that of the
RNMW distribution. So, the RNMW distribution retains the same flexibility of the NMW
distribution and yet the estimation for the former is much easier.

The RNMW distribution has an exclusive bathtub shaped hazard rate function. Other
hazard rates can be obtained from the NMW distribution. For example, setting γ = θ = 2
we obtain

h(x) = 2αx + β(2 + λx)xeλx

for x > 0, which is an increasing function of x, see Figure 8.

15

α=0.01;β=0.02;λ=0.08
α=0.25;β=0.15;λ=0.025
α=0.0005; β=0.05;λ=0.15
α=0.25;β=0.004;λ=0.007

10
The failure rate function

0
0 2 4 6 8 10 12 14 16 18 20
X

Figure 8: Hazard rate functions of the reduced distribution.

20
A Appendix

The moment generating function of X is

MX (t) = E etx
Z ∞
= etx f (x)dx
0
Z ∞
= etx dF (x)
0
Z ∞ √ √ λx
= 1+ tetx e−α x−β xe dx
0
X
∞
(−β)n (nλ)m tk+1
= 1+ I,
i!j!k!
n,m,k=0

R∞ n √
where I = 0 x 2 +m+k e−α x dx. Using gamma-integral formula, we obtain

X
∞  
(−β)n (nλ)m tk+1 Γ (n + 2(m + k) + 2)
MX (t) = 1 + 2 .
n,m,k=0
i!j!k! αn+2(m+k)+2

B Appendix

The elements of the observed information matrix I(ϑ) for the RN M W (α, β, λ) for com-
plete data are
d 
X 
hα (xi ; ϑ) 2
ℓαα = − ,
h (xi ; ϑ)
i=1
Xd
hα (xi ; ϑ) hβ (xi ; ϑ)
ℓαβ = − ,
i=1
h (xi ; ϑ)2
Xd
hα (xi ; ϑ) hλ (xi ; ϑ)
ℓαλ = − ,
i=1
h (xi ; ϑ)2
Xd  
hβ (xi ; ϑ) 2
ℓββ =− ,
h (xi ; ϑ)
i=1
d

X h (xi ; ϑ) hβλ (xi ; ϑ) − hβ (xi ; ϑ) hλ xi ; φ
ℓβλ = − ,
i=1
h (xi ; ϑ)2
" #
Xd q
h (xi ; ϑ) hλλ (xi ; ϑ) − hλ (xi ; ϑ)2
ℓλλ = 2 − β x5i eλxi ,
i=1
h (x i ; ϑ)

21
where

1
hα (xi ; ϑ) = √ ,
2 xi
(0.5 + λxi ) eλxi
hβ (xi ; ϑ) = √ ,
xi
 
√ 3
hλ (xi ; ϑ) = β xi + λxi eλxi ,
2
 
√ 3
hβλ (xi ; ϑ) = xi + λxi eλxi ,
2
 
√ 5
hλλ (xi ; ϑ) = β xi + λxi eλxi .
2

C Appendix

The elements of the observed information matrix I(ϑ) for the RN M W (α, β, λ) for censored
data are
d 
X 
hα (xi ; ϑ) 2
Iαα = − ,
h (xi ; ϑ)
i=1
Xd
hα (xi ; ϑ) hβ (xi ; ϑ)
Iαβ = − ,
i=1
h (xi ; ϑ)2
Xd
hα (xi ; ϑ) hλ (xi ; ϑ)
Iαλ = − 2 ,
i=1
h (x i ; ϑ)
Xd  
hβ (xi ; ϑ) 2
Iββ =− ,
h (xi ; ϑ)
i=1


Xd
h (xi ; ϑ) hβλ (xi ; ϑ) − hβ (xi ; ϑ) hλ xi ; φ Xq
Iβλ = − 2 − x3i eλxi ,
i=1
h (xi ; ϑ) i∈C
" #
d
X 2 q
h (xi ; ϑ) hλλ (xi ; ϑ) − hλ (xi ; ϑ)
Iλλ = 2 −β x5i eλxi
i=1
h (x i ; ϑ)
Xq
−β x5i eλxi ,
i∈C

22
where

1
hα (xi ; ϑ) = √ ,
2 xi
(0.5 + λxi ) eλxi
hβ (xi ; ϑ) = √ ,
xi
 
√ 3
hλ (xi ; ϑ) = β xi + λxi eλxi ,
2
 
√ 3
hβλ (xi ; ϑ) = xi + λxi eλxi ,
2
q  
5
hλλ (xi ; ϑ) = β xi 3 + λxi eλxi .
2

References

[1] Aarset, M. V. (1987). How to identify bathtub hazard rate. IEEE Transactions on
Reliability, R-36, 106-108.

[2] Almalki, S. J. and Yuan, J. (2013). The new modified Weibull distribution. Reliability
Engineering and System Safety, 111, 164-170.

[3] Bain, L. J. (1974). Analysis for the linear failure-rate life-testing distribution. Tech-
nometrics, 16, 551-559.

[4] Bebbington, M., Lai, C. D. and Zitikis, R. (2007). A flexible Weibull extension.
Reliability Engineering and System Safety, 92, 719-726.

[5] Carrasco, M., Ortega, E. M. and Cordeiro, G. M. (2008). A generalized modified

Weibull distribution for lifetime modeling. Computational Statistics and Data Anal-
ysis, 53, 450-462.

[6] Cordeiro, G. M., Ortega, E. M. and Nadarajah, S. (2010). The Kumaraswamy

Weibull distribution with application to failure data. Journal of the Franklin In-
stitute, 347, 1399-1429.

[7] Famoye, F., Lee, C. and Olumolade, O. (2005). The beta-Weibull distribution. Jour-
nal of Statistical Theory and Applications, 4, 121-136.

23
 [8] Kies, J. A. (1958). The strength of glass. Naval Research Laboratory Report No.
5093, Washington D.C.

[9] Kumar, U., Klefsjö, B. and Granholm, S. (1989). Reliability investigation for a fleet
of load haul dump machines in a Swedish mine. Reliability Engineering and System
Safety, 26, 341-361.

[10] Kuo, W. and Zuo, M. J. (2001). Optimal Reliability Modeling: Principles and Ap-
plications. John Wiley and Sons, New York.

[11] Lai, C. D., Murthy, D. N. P. and Xie, M. (2011). Weibull distributions. Computa-
tional Statistics, 33, 282-287.

[12] Lai, C. D., Xie, M., and Murthy, D. N. P. (2003). A modified Weibull distribution.
IEEE Transactions on Reliability, 52, 33-37.

[13] Meeker, W. Q. and Escobar, L. A. (1998). Statistical Methods for Reliability Data.
John Wiley and Sons, New York.

[14] Miller, R. G., Gong, G. and Muñoz, A. (1981). Survival Analysis. John Wiley and
Sons, New York.

[15] Murthy, D. N. P., Xie, M. and Jiang, R. (2003). Weibull Models. John Wiley and
Sons, New York.

[16] Nelson, W. (1990). Accelerated Testing: Statistical Models, Test Plans, and Data
Analysis. John Wiley and Sons, New York.

[17] Perdoná, G. S. C. (2006). Modelos de Riscos Aplicados à Análise de Sobrevivência

(in Portuguese). Doctoral Thesis, Institute of Computer Science and Mathematics,
University of São Paulo, Brasil.

[18] Pham, H. and Lai, C. D. (2007). On recent generalizations of the Weibull distribu-
tion. IEEE Transactions on Reliability, 56, 454-458.

[19] Phani, K. K. (1987). A new modified Weibull distribution function. Communications

of the American Ceramic Society, 70, 182-184.

24
[20] Rajarshi, S. and Rajarshi, M. B. (1988). Bathtub distributions: A review. Commu-
nications in Statistics-Theory and Methods, 17, 2597-2621.

[21] Sarhan, A. M., Abd El-Baset, A. A. and Alasbahi, I. A. (2013). Exponentiated gen-
eralized linear exponential distribution. Applied Mathematical Modeling, 37, 2838-
2849.

[22] Sarhan, A. M. and Apaloo, J. (2013). Exponentiated modified Weibull extension

distribution. Reliability Engineering and System Safety, 112, 137-144.

[23] Sarhan, A. M. and Zaindin, M. (2009). Modified Weibull distribution. Applied Sci-
ences, 11, 123-136.

[24] Silva, A. N. F. (2004). Estudo evolutivo das crianças expostas ao HIV e notificadas
pelo núcleo de vigilância epidemiológica do HCFMRP-USP (in Portuguese). M.Sc.
Thesis, University of São Paulo, Brasil.

[25] Silva, G. O., Ortega, E. M. and Cordeiro, G. M. (2010). The beta modified Weibull
distribution. Lifetime Data Analysis, 16, 409-430.

[26] Singla, N., Jain, K. and Kumar Sharma, S. (2012). The beta generalized Weibull
distribution: Properties and applications. Reliability Engineering and System Safety,
102, 5-15.

[27] Weibull, W. A. (1951). Statistical distribution function of wide applicability. Journal

of Applied Mechanics, 18, 293-296.

[28] Xie, M. and Lai, C. D. (1995). Reliability analysis using an additive Weibull model
with bathtub-shaped failure rate function. Reliability Engineering and System Safety,
52, 87-93.

[29] Xie, M., Tang, Y. and Goh, T. N. (2002). A modified Weibull extension with bathtub-
shaped failure rate function. Reliability Engineering and System Safety, 76, 279-285.

25
This figure "saad2.jpeg" is available in "jpeg" format from:

http://arxiv.org/ps/1307.3925v1
This figure "saad3.jpeg" is available in "jpeg" format from:

http://arxiv.org/ps/1307.3925v1