- Research
- Open Access

# The Kumaraswamy-geometric distribution

- Alfred Akinsete
^{1}Email author, - Felix Famoye
^{2}and - Carl Lee
^{2}

**1**:17

https://doi.org/10.1186/s40488-014-0017-1

© Akinsete et al.; licensee Springer. 2014

**Received:**4 February 2014**Accepted:**15 July 2014**Published:**1 November 2014

## Abstract

In this paper, the Kumaraswamy-geometric distribution, which is a member of the *T*-geometric family of discrete distributions is defined and studied. Some properties of the distribution such as moments, probability generating function, hazard and quantile functions are studied. The method of maximum likelihood estimation is proposed for estimating the model parameters. Two real data sets are used to illustrate the applications of the Kumaraswamy-geometric distribution.

**AMS 2010 Subject Classification:** 60E05; 62E15; 62F10; 62P20

## Keywords

- Transformation
- Moments
- Entropy
- Estimation

## Introduction

*et al*. ([2002]) introduced the beta-generated family of univariate continuous distributions. Suppose

*X*is a random variable with cumulative distribution function (CDF)

*F*(

*x*), the CDF for the beta-generated family is obtained by applying the inverse probability transformation to the beta density function. The CDF for the beta-generated family of distributions is given by

*B*(

*α*,

*β*)=

*Γ*(

*α*)

*Γ*(

*β*)/

*Γ*(

*α*+

*β*). The corresponding probability density function (PDF) is given by

Eugene *et al*. ([2002]) used a normal random variable *X* to define and study the beta-normal distribution. Following the paper by Eugene *et al*. ([2002]), many other authors have defined and studied a number of the beta-generated distributions, using various forms of known *F*(*x*). See for example, beta-Gumbel distribution by Nadarajah and Kotz ([2004]), beta-Weibull distribution by Famoye *et al*. ([2005]), beta-exponential distribution by Nadarajah and Kotz ([2006]), beta-gamma distribution by Kong *et al*. ([2007]), beta-Pareto distribution by Akinsete *et al*. ([2008]), beta-Laplace distribution by Cordeiro and Lemonte ([2011]), beta-generalized Weibull distribution by Singla *et al*. ([2012]), and beta-Cauchy distribution by Alshawarbeh *et al*. ([2013]), amongst others. After the paper by Jones ([2009]), on the tractability properties of the Kumaraswamy's distribution (Kumaraswamy [1980]), Cordeiro and de Castro ([2011]) replaced the classical beta generator distribution with the Kumaraswamy's distribution and introduced the Kumaraswamy generated family. Detailed statistical properties on some Kumaraswamy generated distributions include the Kumaraswamy generalized gamma distribution by de Pascoa *et al*. ([2011]), Kumaraswamy log-logistic distribution by de Santana *et al*. ([2012]) and Kumaraswamy Gumbel distribution by Cordeiro *et al*. ([2012]). Alexander *et al*. ([2012]) replaced the beta generator distribution with the generalized beta type I distribution. The authors referred to this form as the generalized beta-generated distributions (GBGD) and the generator has three shape parameters.

The above technique of generating distributions is possible, only when the generator distributions are continuous and the random variable of the generator lies between 0 and 1. In a recent work by Alzaatreh *et al*. ([2013b]), the authors proposed a new method for generating family of distributions, referred to by the authors as the *T*-*X* family, where a continuous random variable *T* is the *transformed*, and any random variable *X* is the *transformer*. See also Alzaatreh *et al*. ([2012a], [2013a]). These works opened a wide range of techniques for generating distributions of random variables with supports on . The *T*-*X* family enables one to easily generate, not only the continuous distributions, but the discrete distributions as well. As a result, Alzaatreh *et al*. ([2012b]) defined and studied the *T*-geometric family, which are the discrete analogues of the distribution of the random variable *T*.

*F*(

*x*) denotes the CDF of any random variable

*X*and

*r*(

*t*) denotes the PDF of a continuous random variable

*T*with support [

*a*,

*b*]. Alzaatreh

*et al*. ([2013b]) gave the CDF of the

*T*-

*X*family of distributions as

*R*(

*t*) is the CDF of the random variable

*T*,

*W*(

*F*(

*x*))∈ [

*a*,

*b*] is a non-decreasing and absolutely continuous function. Common support [

*a*,

*b*] are [ 0,1], (0,

*∞*), and (-

*∞*,

*∞*). Alzaatreh

*et al*. ([2013b]) studied in some details the case of a non-negative continuous random variable

*T*with support (0,

*∞*). With this technique, it is much easier to generate any discrete distribution. If

*X*is a discrete random variable, the

*T*-

*X*family, is a family of discrete distributions, transformed from the non-negative continuous random variable

*T*. The probability mass function (PMF) of the

*T*-

*X*family of discrete distributions may now be written as

The *T*-geometric family studied in Alzaatreh *et al*. ([2012b]) is a special case of (4) by defining *W*(*F*(*x*))=- ln(1-*F*(*x*)). The rest of the paper is outlined as follows: Section 2 defines the Kumaraswamy geometric distribution (KGD). In Section 3, we discuss some properties of the distribution. In Section 4, the moments of KGD are provided, while Section 5 contains the hazard function and the Shannon entropy. In Section 6, we discuss the maximum likelihood method for estimating the parameters of the distribution. A simulation study is also discussed. Section 7 details the results of applications of the distribution to two real data sets with comparison to other distributions, and Section 8 contains some concluding remarks.

## The Kumaraswamy-geometric distribution

Following the *T*-*X* generalization technique by Alzaatreh *et al*. ([2013b]), we allow the *transformed* random variable *T* to have the Kumaraswamy's distribution, the *transformer* random variable *X* to have the geometric distribution, and *W*(*F*(*x*))=*F*(*x*).

*T*be a random variable with the Kumaraswamy's distribution. The PDF and CDF are defined, respectively, as

where both *α*>0 and *β*>0 are the shape parameters. The beta and Kumaraswamy distributions share similar properties. For example, the Kumaraswamy's distribution, also referred to as the minimax distribution, is unimodal, uniantimodal, increasing, decreasing or constant depending on the values of its parameters. A more detailed description, background and genesis, and properties of Kumaraswamy's distribution are outlined in Jones ([2009]). The author highlighted several advantages of the Kumaraswamy's distribution over the beta distribution, namely; its simple normalizing constant, simple explicit formulas for the distribution and quantile functions, and simple random variate generation procedure.

*et al*. [2005]). Many characterizations of the geometric distribution are analogous to the characterization of the exponential distribution. The geometric distribution has been used extensively in the literature in modeling the distribution of the lengths of waiting times. If

*X*is a random variable having the geometric distribution with parameter

*p*, the PMF of

*X*may be written as

*p*is the probability of success in a single Bernoulli trial. The CDF of the geometric distribution is given by

*a*=0, where the random variable

*T*has the Kumaraswamy's distribution with the CDF (6) and the random variable

*X*has the geometric distribution with the CDF (8). Since the random variable

*T*is defined on (0,1), we use the function

*W*(

*F*(

*x*))=

*F*(

*x*) in (3) to obtain the CDF of KGD as

by using Equation (4). Thus, a random variable *X* having the PMF expressed in Equation (10) is said to follow the Kumaraswamy-geometric distribution with parameters *α*,*β* and *q*, or simply *X*~ KGD (*α*,*β*,*q*). One can show that the PMF in Equation (10) satisfies $\sum _{0}^{\infty}g\left(x\right)=1$ by telescopic cancellation.

*T*and a different

*W*(

*F*(

*x*)) function. Suppose a random variable

*Y*follows the Kumaraswamy's distribution in (5), then its PDF is

*T*=- ln(1-

*Y*). By using the transformation technique, the PDF of

*T*is given by

*T*with the CDF in (12) will be called the log-Kumaraswamy's distribution (LKD). We are unable to find any reference to this distribution in the literature. However, it is a special case of the log-exponentiated Kumaraswamy distribution studied by Lemonte

*et al*. ([2013]). By using the LKD and the

*T*-

*X*distribution by Alzaatreh

*et al*. ([2013b]), we can define the log-Kumaraswamy-geometric distribution (LKGD) by using Equation (3), where

*T*follows the LKD,

*X*follows the geometric distribution and

*W*(

*F*(

*x*))=- ln(1-

*F*(

*x*)). By using 1-

*F*(

*x*)=

*q*

^{x+1}and - ln(1-

*F*(

*x*))=- ln

*q*

^{x+1}, the probability mass function of LKGD can be obtained as

which is the same as the KGD in (10) defined by using Kumaraswamy's and geometric distributions. The LKGD, and hence the KGD, is the discrete analogue of log-Kumaraswamy's distribution.

**Special cases of KGD**

- (a)
- (b)
When

*α*=1, the KGD with parameters*α*,*β*and*q*reduces to the geometric distribution with parameter*p*_{*}, where*p*_{*}=1-*q*^{ β }. - (c)
When

*β*=1, the KGD reduces to the exponentiated-exponential-geometric distribution (EEGD) discussed in Alzaatreh*et al*. ([2012b]).

*α*,

*β*and

*q*are given in Figure 1.

## Some properties of Kumaraswamy-geometric distribution

*X*follows the KGD with CDF

*G*(

*x*) in (9). The quantile function

*X*

_{*}(=

*Q*(

*U*),0<

*U*<1) of KGD is the inverse of the cumulative distribution. That is,

where *U* has a uniform distribution with support on (0,1). Equation (14) can be used to simulate the Kumaraswamy-geometric random variable. First, simulate a random variable *U* and compute the value of *X*_{*} in (14), which is not necessarily an integer. The Kumaraswamy-geometric random variate *X* is the largest integer ≤*X*_{*}, which can be denoted by [ *X*_{*}].

**3.0.0.0 Transformation:** The relationship between the KGD and the Kumaraswamy's, exponential, exponentiated-exponential, Pareto, Weibull, Rayleigh, and the logistic distributions are given in the following lemma.

### Lemma1.

*v*] denotes the largest integer less than or equal to the quantity

*v*.

- (a)
If

*Y*has Kumaraswamy's distribution with parameters*α*and*β*, then the distribution of*X*=[log*q*(1-*Y*)] is KGD. - (b)
If

*Y*is standard exponential, then*X*=[log*q*{1-(1-*e*^{-Y/β})^{1/α}}] has KGD. - (c)
If

*Y*follows an exponentiated-exponential distribution with scale parameter*λ*and index parameter*c*, then*X*=[log*q*{1-[1-(1-*e*^{-λ Y})^{c/β}]^{1/α}}] has KGD. - (d)
If the random variable

*Y*has a Pareto distribution with parameters*θ*,*k*and CDF $F\left(y\right)=1-{\left(\frac{\theta}{\theta +y}\right)}^{k}$, then $X=\left[\underset{q}{log}\left\{1-{\left(1-{\left(\frac{\theta}{\theta +Y}\right)}^{k/\beta}\right)}^{1/\alpha}\right\}\right]$ has KGD. - (e)
If the random variable

*Y*has a Weibull distribution with*F*(*y*)=1- exp{-(*y*/*γ*)^{ c }} as CDF, then*X*=[log*q*{1-(1- exp[-(*Y*/*γ*)^{ c }/*β*])^{1/α}}] has KGD. - (f)
If the random variable

*Y*has a Rayleigh distribution with $F\left(y\right)=1-exp\left[\frac{-{y}^{2}}{2{b}^{2}}\right]$ as CDF, then $X=\left[\underset{q}{log}\left\{1-{\left(1-exp\left[\frac{-{Y}^{2}}{2{b}^{2}\beta}\right]\right)}^{1/\alpha}\right\}\right]$ has KGD. - (g)
If

*Y*is a logistic random variable with*F*(*y*)=[1+ exp({-(*y*-*a*)/*b*})]^{-1}as CDF, then*X*=[log*q*{1-(1-[1+ exp{-(*Y*-*a*)/*b*}]^{-1/β})^{1/α}}] has KGD.

*Proof*.

*X*has KGD as given in (10). We will show the result for part (a). Let

*R*be the CDF of the Kumaraswamy's distribution.

which is the PMF of the KGD in (10).

In general, if we have a continuous random variable *Y* and its CDF is *F*(*y*), then *X*=[log*q*{1-(1-*F*^{1/β}(*Y*))^{1/α}}] has KGD.

**3.0.0.0 Limiting behavior:** As *x*→*∞*, ${lim}_{\mathit{\text{x}}\to \infty}g\left(x\right)=0$. Also, as *x*→0, ${lim}_{x\to 0}g\left(x\right)=1-{[\phantom{\rule{0.3em}{0ex}}1-{(1-q)}^{\alpha}]}^{\beta}$. This limit becomes 0 if *q*→1 and/or *α*→*∞*. Thus, the distribution starts with probability zero or a constant probability as evident from Figure 1.

**3.0.0.0 Mode of the KGD:** Since the KGD is also LKGD, a *T*-geometric distribution, we use Lemma 2 in Alzaatreh *et al*. ([2012b]), which states that a *T*-geometric distribution has a reversed J-shape if the distribution of the random variable *T* has a reversed J-shape. We only need to show when the distribution of log-Kumaraswamy distribution has a reversed J-shape.

*t*, we obtain

where *V*(*t*)=*α* *β* *e*^{-t} (1-*e*^{-t})^{α-2} [1- (1-*e*^{-t})^{
α
}]^{β-2} is positive. For *β*≥1 and *α*≥ 1, it is straight forward to show that *Q*(*t*)≤ 0. For *β*<1 and *α*≤1, the function *Q*(*t*) is an increasing function of *t*. It is not difficult to show that ${lim}_{t\to 0}Q\left(t\right)=\alpha -1\le 0$ and ${lim}_{t\to \infty}Q\left(t\right)=0$. Thus, for *α*≤ 1 and any value of *β* and *q*, *Q*(*t*)≤0 and so the PDF of the log-Kumaraswamy distribution is monotonically decreasing or has a reversed J-shape. Hence, the KGD has a reversed J-shape and a unique mode at *x*=0 when *α*≤1.

When *α*>1, it is not easy to show that the KGD is unimodal. However, through numerical analysis of the behavior of the PMF, and its plots in Figure 1 for various values of *β* and *q*, we observe, that for values of *α*>1, the KGD is concave down or has a reversed J-shape with a unique mode.

## Moments

*r*

^{ t h }raw moment is given by

*α*, if

*α*is a positive integer. When

*β*=1 in the above, we have,

*r*=1 and

*α*=1, the expression for the first moment, or the mean of the KGD may be written as,

which is the mean of the geometric distribution, a special case of KGD.

*g*(

*x*) is given by (10). Observe that the series is absolutely convergent by using the ratio test and hence the series in (17) is absolutely convergent. Thus, interchanging the order of summation has no effect. Using Equation (17), the

*r*

^{ t h }moment may be written as

*β*

^{(i)}=

*β*(

*β*-1)(

*β*-2)…(

*β*-

*i*+1), and similarly for (

*α*

*i*)

^{(j)}. Also,

is the polylogarithm function, (http://mathworld.wolfram.com/Polylogarithm.html).

*α*=1=

*β*, the expressions for the central moments of the geometric distribution are as follows:

The results for this special case may be found in standard textbooks on probability. See for example, Zwillinger and Kokoska ([2000]).

*M*(

*t*)) and the probability generating function (

*φ*(

*t*)) can be simplified further. In the case of

*φ*(

*t*), we have

which reduces to the result in Alzaatreh *et al*. ([2012b]) when *β*=1.

*α*and

*β*for the numerical computation are from 0.2 to 10 at an increment of 0.1, while the values of

*q*are from 0.2 to 0.9 at an increment of 0.1. For brevity, we report the mode, the mean and the standard deviation in Table 1 and the skewness and kurtosis in Table 2 for some values of

*q*,

*β*and

*α*. From the numerical computation, the mean, mode and standard deviation are increasing functions of

*q*. From Table 1, the mean, mode and standard deviation are decreasing functions of

*β*but increasing functions of

*α*. For

*α*≤ 1, the skewness and kurtosis are decreasing functions of

*q*but increasing functions of

*β*. For

*α*>1, the skewness and kurtosis first decrease and then increase as both

*q*and

*β*increase. The skewness and kurtosis are decreasing functions of

*α*. Some of these observations can be seen in Table 2 while others are from the numerical computation. Instead of Tables 1 and 2, contour plots may be used to present the results in the tables. However, it becomes difficult to see the patterns described above.

**Mode, mean and standard deviation (SD) of KGD for some values of**
α
**,**
β
**and**
q

q=0.4 | q=0.6 | q=0.8 | ||||||||
---|---|---|---|---|---|---|---|---|---|---|

α | β | Mode | Mean | SD | Mode | Mean | SD | Mode | Mean | SD |

0.4 | 0.4 | 0 | 1.60 | 2.48 | 0 | 3.16 | 4.51 | 0 | 7.74 | 10.40 |

0.6 | 0 | 0.83 | 1.52 | 0 | 1.72 | 2.81 | 0 | 4.39 | 6.54 | |

0.8 | 0 | 0.48 | 1.04 | 0 | 1.06 | 1.96 | 0 | 2.82 | 4.60 | |

1.5 | 0 | 0.10 | 0.39 | 0 | 0.29 | 0.79 | 0 | 0.92 | 1.96 | |

2.0 | 0 | 0.04 | 0.22 | 0 | 0.13 | 0.49 | 0 | 0.49 | 1.26 | |

4.0 | 0 | 0.001 | 0.04 | 0 | 0.01 | 0.11 | 0 | 0.07 | 0.35 | |

0.6 | 0.4 | 0 | 1.87 | 2.59 | 0 | 3.67 | 4.69 | 0 | 8.97 | 10.80 |

0.6 | 0 | 1.04 | 1.65 | 0 | 2.14 | 3.02 | 0 | 5.43 | 6.99 | |

0.8 | 0 | 0.64 | 1.17 | 0 | 1.41 | 2.18 | 0 | 3.71 | 5.08 | |

1.5 | 0 | 0.18 | 0.51 | 0 | 0.48 | 1.00 | 0 | 1.48 | 2.42 | |

2.0 | 0 | 0.08 | 0.32 | 0 | 0.26 | 0.67 | 0 | 0.91 | 1.67 | |

4.0 | 0 | 0.005 | 0.07 | 0 | 0.04 | 0.20 | 0 | 0.21 | 0.62 | |

0.8 | 0.4 | 0 | 2.08 | 2.66 | 0 | 4.08 | 4.81 | 0 | 9.93 | 11.04 |

0.6 | 0 | 1.21 | 1.74 | 0 | 2.49 | 3.16 | 0 | 6.27 | 7.27 | |

0.8 | 0 | 0.79 | 1.27 | 0 | 1.71 | 2.33 | 0 | 4.47 | 5.38 | |

1.5 | 0 | 0.26 | 0.60 | 0 | 0.68 | 1.16 | 0 | 2.01 | 2.74 | |

2.0 | 0 | 0.13 | 0.40 | 0 | 0.41 | 0.82 | 0 | 1.35 | 1.99 | |

4.0 | 0 | 0.01 | 0.12 | 0 | 0.08 | 0.31 | 0 | 0.43 | 0.87 | |

1.5 | 0.4 | 1 | 2.61 | 2.79 | 1 | 5.06 | 5.00 | 3 | 12.22 | 11.44 |

0.6 | 0 | 1.68 | 1.89 | 1 | 3.39 | 3.39 | 2 | 8.38 | 7.76 | |

0.8 | 0 | 1.21 | 1.44 | 1 | 2.54 | 2.59 | 2 | 6.43 | 5.92 | |

1.5 | 0 | 0.54 | 0.81 | 0 | 1.31 | 1.47 | 1 | 3.61 | 3.35 | |

2.0 | 0 | 0.35 | 0.62 | 0 | 0.95 | 1.14 | 1 | 2.77 | 2.61 | |

4.0 | 0 | 0.08 | 0.29 | 0 | 0.38 | 0.63 | 0 | 1.42 | 1.46 | |

2.0 | 0.4 | 1 | 2.87 | 2.83 | 2 | 5.55 | 5.06 | 4 | 13.35 | 11.57 |

0.6 | 1 | 1.93 | 1.94 | 1 | 3.85 | 3.47 | 4 | 9.45 | 7.92 | |

0.8 | 1 | 1.44 | 1.50 | 1 | 2.97 | 2.68 | 3 | 7.45 | 6.10 | |

1.5 | 0 | 0.73 | 0.89 | 1 | 1.69 | 1.58 | 2 | 4.51 | 3.57 | |

2.0 | 0 | 0.51 | 0.71 | 1 | 1.30 | 1.26 | 2 | 3.61 | 2.83 | |

4.0 | 0 | 0.18 | 0.40 | 0 | 0.65 | 0.77 | 1 | 2.11 | 1.69 | |

4.0 | 0.4 | 2 | 3.56 | 2.89 | 3 | 6.79 | 5.16 | 8 | 16.19 | 11.79 |

0.6 | 1 | 2.59 | 2.02 | 3 | 5.05 | 3.59 | 7 | 12.21 | 8.19 | |

0.8 | 1 | 2.09 | 1.59 | 2 | 4.14 | 2.81 | 6 | 10.13 | 6.41 | |

1.5 | 1 | 1.32 | 1.01 | 2 | 2.77 | 1.74 | 5 | 7.00 | 3.94 | |

2.0 | 1 | 1.08 | 0.84 | 2 | 2.34 | 1.43 | 4 | 5.99 | 3.23 | |

4.0 | 1 | 0.64 | 0.61 | 1 | 1.57 | 0.96 | 3 | 4.24 | 2.10 | |

6.0 | 0.4 | 2 | 3.99 | 2.90 | 4 | 7.55 | 5.19 | 10 | 17.92 | 11.87 |

0.6 | 2 | 3.01 | 2.04 | 4 | 5.79 | 3.63 | 9 | 13.90 | 8.29 | |

0.8 | 2 | 2.50 | 1.61 | 3 | 4.87 | 2.86 | 8 | 11.80 | 6.53 | |

1.5 | 1 | 1.72 | 1.03 | 3 | 3.47 | 1.80 | 7 | 8.59 | 4.08 | |

2.0 | 1 | 1.46 | 0.87 | 2 | 3.02 | 1.50 | 6 | 7.55 | 3.38 | |

4.0 | 1 | 1.02 | 0.62 | 2 | 2.21 | 1.02 | 5 | 5.71 | 2.27 |

**Skewness and kurtosis of KGD for some values of**
α
**,**
β
**and**
q

q=0.4 | q=0.6 | q=0.8 | |||||
---|---|---|---|---|---|---|---|

α | β | Skewness | Kurtosis | Skewness | Kurtosis | Skewness | Kurtosis |

0.4 | 0.4 | 2.4529 | 8.4883 | 2.3789 | 8.0612 | 2.3360 | 7.8215 |

0.6 | 2.8022 | 10.920 | 2.6534 | 9.9332 | 2.5656 | 9.3787 | |

0.8 | 3.1918 | 14.035 | 2.9430 | 12.172 | 2.7954 | 11.133 | |

1.5 | 4.9662 | 32.736 | 4.0792 | 23.245 | 3.5894 | 18.553 | |

2.0 | 6.8722 | 60.013 | 5.0708 | 35.519 | 4.1724 | 25.214 | |

4.0 | 30.524 | 990.63 | 12.671 | 196.82 | 7.1320 | 72.567 | |

0.6 | 0.4 | 2.2480 | 7.2663 | 2.1956 | 7.0004 | 2.1684 | 6.8666 |

0.6 | 2.4465 | 8.4649 | 2.3387 | 7.8659 | 2.2823 | 7.5649 | |

0.8 | 2.6694 | 9.9363 | 2.4883 | 8.8459 | 2.3932 | 8.3020 | |

1.5 | 3.6723 | 17.839 | 3.0582 | 13.119 | 2.7529 | 10.999 | |

2.0 | 4.6998 | 27.856 | 3.5337 | 17.209 | 2.9979 | 13.043 | |

4.0 | 14.964 | 238.38 | 6.6621 | 54.450 | 4.0916 | 23.691 | |

0.8 | 0.4 | 2.1208 | 6.5743 | 2.0847 | 6.4108 | 2.0685 | 6.3383 |

0.6 | 2.2269 | 7.1361 | 2.1498 | 6.7675 | 2.1151 | 6.6063 | |

0.8 | 2.3522 | 7.8283 | 2.2198 | 7.1642 | 2.1603 | 6.8772 | |

1.5 | 2.9553 | 11.535 | 2.5010 | 8.8497 | 2.3055 | 7.8006 | |

2.0 | 3.5861 | 16.072 | 2.7466 | 10.413 | 2.4048 | 8.4499 | |

4.0 | 9.2202 | 90.141 | 4.3332 | 22.881 | 2.8543 | 11.487 | |

1.5 | 0.4 | 1.9024 | 5.5182 | 1.9015 | 5.5273 | 1.9038 | 5.5384 |

0.6 | 1.8485 | 5.2051 | 1.8400 | 5.2063 | 1.8433 | 5.2269 | |

0.8 | 1.8105 | 4.9322 | 1.7862 | 4.9087 | 1.7894 | 4.9374 | |

1.5 | 1.8191 | 4.4671 | 1.6645 | 4.1552 | 1.6529 | 4.1894 | |

2.0 | 1.9494 | 4.6197 | 1.6281 | 3.8307 | 1.5890 | 3.8375 | |

4.0 | 3.3608 | 11.0492 | 1.7629 | 3.5961 | 1.4574 | 3.0767 | |

2.0 | 0.4 | 1.8322 | 5.2176 | 1.8440 | 5.2733 | 1.8503 | 5.2983 |

0.6 | 1.7260 | 4.6855 | 1.7442 | 4.7835 | 1.7563 | 4.8321 | |

0.8 | 1.6352 | 4.2020 | 1.6551 | 4.3386 | 1.6733 | 4.4121 | |

1.5 | 1.4620 | 3.0364 | 1.4309 | 3.2239 | 1.4664 | 3.3814 | |

2.0 | 1.4564 | 2.6068 | 1.3301 | 2.7152 | 1.3700 | 2.9254 | |

4.0 | 2.0958 | 3.6392 | 1.1838 | 1.6496 | 1.1546 | 1.9922 | |

4.0 | 0.4 | 1.7344 | 4.8228 | 1.7559 | 4.9001 | 1.7627 | 4.9251 |

0.6 | 1.5605 | 4.0589 | 1.6019 | 4.1951 | 1.6151 | 4.2403 | |

0.8 | 1.4043 | 3.3984 | 1.4675 | 3.5897 | 1.4874 | 3.6530 | |

1.5 | 1.0049 | 1.9031 | 1.1406 | 2.2381 | 1.1833 | 2.3432 | |

2.0 | 0.8155 | 1.2870 | 0.9925 | 1.7112 | 1.0500 | 1.8310 | |

4.0 | 0.4669 | -0.1633 | 0.6666 | 0.8077 | 0.7728 | 0.9349 | |

6.0 | 0.4 | 1.7069 | 4.7081 | 1.7246 | 4.7731 | 1.7310 | 4.7964 |

0.6 | 1.5197 | 3.8941 | 1.5522 | 4.0017 | 1.5641 | 4.0428 | |

0.8 | 1.3558 | 3.2107 | 1.4034 | 3.3521 | 1.4207 | 3.4078 | |

1.5 | 0.9572 | 1.7788 | 1.0513 | 1.9565 | 1.0843 | 2.0406 | |

2.0 | 0.7743 | 1.2870 | 0.8984 | 1.4383 | 0.9394 | 1.5299 | |

4.0 | 0.3102 | 0.6548 | 0.5863 | 0.5921 | 0.6444 | 0.6837 |

## Hazard rate and Shannon entropy

*et al*. ([2012b]). Observe that

*L*

_{1}>

*L*

_{2}when

*α*<1. For

*α*<1, we check the behavior of

*h*(

*x*). The function

*h*(

*x*) is monotonically decreasing when

*α*<1 if

*h*(

*x*)≥

*h*(

*x*+1) for all

*x*. When

*α*=1, observe that

*h*(

*x*) is a constant. For values of

*α*<1, we numerically evaluate

*d*(

*x*)=

*h*(

*x*)-

*h*(

*x*+1) for

*α*and

*q*from 0.1 to 0.9 at an increment of 0.1. All the values are positive, which indicates that the function

*h*(

*x*) is monotonically decreasing. Similarly, we analytically evaluate

*d*(

*x*) for small values of

*α*>1 and the difference

*d*(

*x*) is always negative. Numerically, we use the values of

*q*from 0.1 to 0.9 at an increment of 0.1 with values of

*α*from 1.5 to 10.0 at an increment of 0.5. All the

*d*(

*x*) values are negative which indicates that the function

*h*(

*x*) is monotonically increasing. Thus, we have a decreasing hazard rate when

*α*<1 and an increasing hazard rate when

*α*>1. For

*α*=1,

*L*

_{1}=

*L*

_{2}and we have a constant hazard rate. The graphs of the hazard rate function defined in Equation (24) are shown in Figure 2 for various values of the parameters. We see in Figure 2 that the hazard rate decreases for values of

*α*<1 and increases for

*α*>1.

*X*with probability mass function

*g*(

*x*), the Shannon entropy is defined as,

*x*) is a measure of the information carried by

*g*(

*x*), with higher entropy corresponding to less information. Substituting Equation (10) in Equation (25), we have

*α*=1 for simplicity. We may now write the entropy as

*β*=1 in Equation (27), we have, for the geometric distribution,

Note that when *β*=1, *p*=*q*=1/2, (*x*)=2. It is not difficult to show that (*x*) is an increasing function of *q* for any given *β*. This is consistent with the pattern of the standard deviation. We also note that ${lim}_{\beta \to \infty}\mathcal{S}\left(x\right)=0$, with the proviso that 0 log0=0. This indicates that smaller values of *β* increase the uncertainty in the distribution, while higher values of *β* increase the amount of information measured in terms of the probability. Actually, a zero entropy indicates that all information needed is measured solely in terms of the probability. In a way, the KGD has smaller entropy (more probabilistic information) than the geometric distribution for values of *β*>1.

## Maximum likelihood estimation

We discuss the maximum likelihood estimation of the parameters of the KGD in subsection 6.1. Subsection 6.2 contains the results of a simulation that is conducted to evaluate the performance of the maximum likelihood estimation method.

### Estimation

*n*be taken from KGD, with observed frequencies

*n*

_{ x },

*x*=0,1,2,…,

*k*, where $\sum _{x=0}^{k}{n}_{x}=n$. From Equation (10), the likelihood function for a random sample of size

*n*may be expressed as

Setting the non-linear Equations (30), (31) and (32) to zero and solving them iteratively, we get the estimates $\widehat{\mathbf{\theta}}={(\widehat{\alpha},\widehat{\beta},\widehat{q})}^{T}$ for the parameter vector **θ**=(*α*,*β*,*q*)^{
T
}. The initial values of parameters *α* and *β* can be set to 1 and that of parameter *q* can be set to 0.5.

*∂*

^{2}

*l*/(

*∂*

*i*

*∂*

*j*)=-

*l*

_{ ij }, where,

*i*,

*j*∈{

*α*,

*β*,

*q*}. Under conditions that are fulfilled for parameters in the interior of the parameter space but not on the boundary, the asymptotic distribution of $\sqrt{n}(\widehat{\mathbf{\theta}}-\mathbf{\theta})\phantom{\rule{1em}{0ex}}\phantom{\rule{1em}{0ex}}\text{is}\phantom{\rule{1em}{0ex}}\phantom{\rule{1em}{0ex}}{N}_{3}(0,{\mathcal{I}}^{-1}(\mathbf{\theta}\left)\right)$. The asymptotic multivariate normal distribution ${N}_{3}(0,{\mathcal{I}}^{-1}(\widehat{\mathbf{\theta}}\left)\right)$ of $\widehat{\mathbf{\theta}}$ can be used to construct approximate confidence intervals for the parameters. For example, the 100(1-

*ξ*)

*%*asymptotic confidence interval for the

*i*

^{ t h }parameter

*θ*

_{ i }is given by

where _{i,i} is the *i*^{
t
h
} diagonal element of ${\mathcal{I}}^{-1}\left(\mathbf{\theta}\right)$ for *i*=1,2,3, and *z*_{ξ/2} is the upper *ξ*/2 point of standard normal distribution. See for example, Mahmoudi ([2011]).

The expressions for the elements *l*_{
ij
} are given in the Appendix.

### Simulation

*α*,

*β*and

*q*. The different sample sizes considered in the simulation are

*n*=250, 500 and 750. The parameter combinations for the simulation study are shown in Table 3. The combinations were chosen to reflect the following cases of the distribution: under-dispersion (

*α*=6,

*α*=4,

*q*=0.8), over-dispersion (all other cases), monotonically decreasing (

*α*=1.6,

*β*=2.0,

*q*=0.6), and unimodal with mode greater than 0 (all other cases). For each parameter combination and each sample size, the simulation process is repeated 100 times. The average bias (actual - estimate) and standard deviation of the parameter estimates are reported in Table 3. The biases are relatively small when compared to the standard deviations. In most cases, as the sample size increases, the standard deviations of the estimators decrease.

**Bias and standard deviation for maximum likelihood estimates**

Actual values | Bias with standard deviation in parentheses | ||||||
---|---|---|---|---|---|---|---|

n | α | β | q | $\widehat{\alpha}$ | $\widehat{\beta}$ | $\widehat{q}$ | Mode |

250 | 1.6 | 2.0 | 0.6 | -0.1295(0.2099) | 0.1128(0.4697) | 0.0409(0.0733) | 0 |

4.0 | 2.0 | 0.4 | -0.3136(0.7844) | 0.1047(0.4617) | 0.0237(0.0698) | 1 | |

4.0 | 2.0 | 0.6 | -0.1977(0.5714) | 0.1021(0.4571) | 0.0198(0.0571) | 2 | |

4.0 | 2.0 | 0.8 | -0.2467(0.5521) | 0.0852(0.4020) | 0.0126(0.0293) | 4 | |

6.0 | 4.0 | 0.8 | -0.3931(0.8570) | 0.2084(0.9507) | 0.0121(0.0278) | 5 | |

500 | 1.6 | 2.0 | 0.6 | -0.0437(0.1436) | 0.0121(0.4496) | 0.0153(0.0665) | 0 |

4.0 | 2.0 | 0.4 | -0.3142(0.5876) | 0.1383(0.4421) | 0.0290(0.0639) | 1 | |

4.0 | 2.0 | 0.6 | -0.2015(0.4728) | 0.0808(0.4504) | 0.0194(0.0527) | 2 | |

4.0 | 2.0 | 0.8 | -0.2747(0.4630) | 0.1526(0.3781) | 0.0157(0.0278) | 4 | |

6.0 | 4.0 | 0.8 | -0.3283(0.7771) | 0.2187(0.9375) | 0.0116(0.0267) | 5 | |

750 | 1.6 | 2.0 | 0.6 | -0.0533(0.1305) | 0.1184(0.4250) | 0.0278(0.0669) | 0 |

4.0 | 2.0 | 0.4 | -0.3181(0.5939) | 0.1707(0.4195) | 0.0328(0.0622) | 1 | |

4.0 | 2.0 | 0.6 | -0.1787(0.4775) | 0.0562(0.4469) | 0.0159(0.0522) | 2 | |

4.0 | 2.0 | 0.8 | -0.1717(0.4739) | 0.0813(0.4117) | 0.0101(0.0299) | 4 | |

6.0 | 4.0 | 0.8 | -0.2472(0.5940) | 0.1947(0.8090) | 0.0087(0.0227) | 5 |

## Applications of KGD

We apply the KGD to two data sets. The first data set is the observed frequencies of the distribution of purchases of a brand *X* breakfast cereals purchased by consumers over a period of time (Consul [1989]). The other data set is the number of absences among shift-workers in a steel industry (Gupta and Ong [2004]). Comparisons are made with the generalized negative binomial distribution (GNBD) defined by Jain and Consul ([1971]) and the exponentiated-exponential geometric distribution (EEGD) defined by Alzaatreh *et al*. ([2012b]).

### Purchases by consumers

*X*breakfast cereals. The original data in Table 4 was taken from Chatfield ([1975]).

**The number of units**
r
**purchased by observed number (**
f
_{
r
}
**) of consumers**

Expected | ||||
---|---|---|---|---|

r-value | Observed | EEGD | GNBD | KGD |

0 | 299 | 300.11 | 298.98 | 299.33 |

1 | 69 | 63.10 | 68.03 | 65.59 |

2 | 37 | 41.13 | 41.12 | 42.02 |

3 | 34 | 30.68 | 29.89 | 30.92 |

4 | 23 | 24.36 | 23.51 | 24.28 |

5 | 20 | 20.06 | 19.32 | 19.81 |

6 | 12 | 16.92 | 16.32 | 16.58 |

7 | 18 | 14.52 | 14.04 | 14.14 |

8 | 14 | 12.61 | 12.25 | 12.21 |

9 | 9 | 11.06 | 10.80 | 10.67 |

10-11 | 17 | 18.47 | 18.16 | 17.72 |

12-15 | 27 | 26.69 | 26.63 | 25.50 |

16-35 | 63 | 62.29 | 62.95 | 63.23 |

Total | 642 | 642 | 642 | 642 |

Parameter | $\widehat{\alpha}=0.2910\left(0.0212\right)$ | $\widehat{\theta}=0.9658\left(0.0148\right)$ | $\widehat{\alpha}=0.3373\left(0.0924\right)$ | |

Estimates | $\widehat{\theta}=0.9267\left(0.0076\right)$ | $\widehat{m}=0.2264\left(0.0339\right)$ | $\widehat{\beta}=1.5114\left(1.3648\right)$ | |

$\widehat{\beta}=0.9882\left(0.0126\right)$ | $\widehat{q}=0.9592\left(0.0544\right)$ | |||

| 4.34 | 3.92 | 4.11 | |

| 10 | 9 | 9 | |

| 0.9307 | 0.9166 | 0.9040 | |

AIC | 2471.3 | 2472.8 | 2473.0 | |

LL | -1233.64 | -1233.38 | -1233.50 |

The data contains the frequency of consumers who bought *r* units of brand *X* over a number of weeks. The data is fitted to the KGD, EEGD, and the GNBD. We are not sure how the frequencies for the (10-11), (12-15) and (16-35) were handled in previous applications of the data set. In our analysis, the probabilities for the classes (10-11) and (12-15) were obtained by adding the corresponding individual probabilities in each class. When finding the maximum likelihood estimates, the probability for the last class was obtained by subtracting the sum of all previous probabilities from 1. The results in Table 4 show that all the three distributions provide adequate fit to the data. Since the parameter *β* in KGD is not significantly different from 1, it may be more appropriate to apply the two-parameter EEGD to fit the data. Also, the likelihood ratio test to compare the EEGD with the KGD is not significant at 5% level.

### Number of absences by shift-workers

*x*-value, of 248 shift workers in the years 1947 and 1948. Arbous and Sichel ([1954]) used the negative binomial distribution (NBD) to fit the data. Gupta and Ong ([2004]) proposed a four-parameter generalized negative binomial distribution to fit the data and compared it to the NBD and the GPD. The chi-square value for their distribution was 8.27 with 15 degrees of freedom. The chi-square value obtained by Gupta and Ong for the GPD was 27.79 with 17 degrees of freedom (DF). The DF =

*k*-

*s*-1, where

*k*is the number of classes and

*s*is the number of estimated parameters.

**The number of absences among shift-workers in a steel industry**

Observed | Expected | |||
---|---|---|---|---|

x-value | Frequency | EEGD | GNBD | KGD |

0 | 7 | 10.18 | 9.86 | 6.52 |

1 | 16 | 16.41 | 16.44 | 18.21 |

2 | 23 | 18.30 | 19.32 | 21.40 |

3 | 20 | 18.57 | 19.84 | 20.92 |

4 | 23 | 18.02 | 19.09 | 19.31 |

5 | 24 | 17.03 | 17.73 | 17.45 |

6 | 12 | 15.82 | 16.13 | 15.64 |

7 | 13 | 14.52 | 14.50 | 13.97 |

8 | 9 | 13.22 | 12.93 | 12.46 |

9 | 9 | 11.95 | 11.49 | 11.11 |

10 | 8 | 10.74 | 10.17 | 9.90 |

11 | 10 | 9.62 | 9.00 | 8.83 |

12 | 8 | 8.59 | 7.95 | 7.87 |

13 | 7 | 7.65 | 7.02 | 7.01 |

14 | 2 | 6.79 | 6.21 | 6.25 |

15 | 12 | 6.02 | 5.49 | 5.57 |

16 | 3 | 5.33 | 4.86 | 4.96 |

17 | 5 | 4.71 | 4.30 | 4.42 |

18 | 4 | 4.16 | 3.81 | 3.94 |

19 | 2 | 3.67 | 3.38 | 3.51 |

20 | 2 | 3.23 | 3.00 | 3.13 |

21 | 5 | 2.85 | 2.67 | 2.79 |

22 | 5 | 2.51 | 2.38 | 2.49 |

23 | 2 | 2.21 | 2.11 | 2.21 |

24 | 1 | 1.94 | 1.88 | 1.97 |

25-48 | 16 | 13.96 | 16.44 | 16.16 |

Total | 248 | 248 | 248 | 248 |

Parameter | $\widehat{\alpha}=1.5255\left(0.1471\right)$ | $\widehat{\theta}=0.0026\left(0.0001\right)$ | $\widehat{\alpha}=3.1859\left(1.6702\right)$ | |

Estimates | $\widehat{\theta}=0.8767\left(0.0093\right)$ | $\widehat{m}=1242.96\left(4.259\right)$ | $\widehat{\beta}=0.1292\left(0.0882\right)$ | |

$\widehat{\beta}=254.77\left(20.785\right)$ | $\widehat{q}=0.4099\left(0.2374\right)$ | |||

| 12.18 | 9.62 | 7.78 | |

| 17 | 16 | 16 | |

| 0.7891 | 0.8857 | 0.9551 | |

AIC | 1517.8 | 1517.3 | 1515.2 | |

Log-likelihood | -756.88 | -755.66 | -754.62 |

We re-analyzed the data for the EEGD, the GNBD, and the GPD. We obtained a chi-square of 9.62 for the GPD, which is much smaller than the 27.79 provided by Gupta and Ong. Thus, our estimates from the GPD (not reported in Table 5) differ significantly from the results in Gupta and Ong ([2004]). When finding the maximum likelihood estimates, the probability for the last class was obtained by subtracting the sum of all previous probabilities from 1. In view of this, the results obtained from the EEGD are slightly different from those of Alzaatreh *et al*. ([2012b]) who applied the EEGD to fit the data. We apply the KGD to model the data in Table 5, and the results from the table indicate that the KGD, EEGD and GNBD provide good fit to the data.

If *m*→*∞*, the GNBD with parameters *θ*, *m* and *β* goes to the GPD with parameters *α* and *λ*, where *α*=*m* *θ* and *λ*=*θ* *β* on page 218 of Consul and Famoye ([2006]). We fitted the GPD to the data and we got the same log-likelihood with $\widehat{\alpha}=3.2250$ and $\widehat{\lambda}=0.6597$. From the GNBD, we obtained $\widehat{m}\widehat{\theta}=1242.96\times 0.002591=3.22\approx 3.2250=\widehat{\alpha}$ and $\widehat{\beta}\widehat{\theta}=254.77\times 0.002591=0.66\approx 0.6597=\widehat{\lambda}$.

We observe that the parameter *β* in the KGD is significantly different from 1. This makes the KGD a more appropriate distribution over the EEGD. The likelihood ratio statistic for testing the EEGD against the KGD is ${\chi}_{1}^{2}=-2(754.62-756.88)=4.52$ with a *p*-value of 0.0335. Thus, we reject the null hypothesis that the data follows the EEGD at the 5% level. The likelihood ratio test supports the claim that the parameter *β* is significantly different from 1, and hence the KGD appears to be superior to the EEGD.

## Conclusion

Discrete distributions are often derived by using the Lagrange expansions framework (see for example Consul and Famoye [2006]) or using difference equations (see for example Johnson *et al*. [2005]). Recently, Alzaatreh *et al*. ([2012b], [2013b]) developed a general method for generating distributions and these distributions are members of the *T*-*X* family. The method can be applied to derive both the discrete and continuous distributions. This article used the *T*-*X* family framework to define a new discrete distribution named the Kumaraswamy-geometric distribution (KGD).

Some special cases, and properties of the KGD are discussed, which include moments, hazard rate and entropy. The method of maximum likelihood estimation is used in estimating the parameters of the KGD. The distribution is applied to model two real life data sets; one consisting of the observed frequencies of the distribution of purchases of a brand *X* breakfast cereals, and the other, the observed frequencies of the number of absences among shift-workers in a steel industry. Two other distributions, the EEGD and the GNBD are compared with KGD. It is found that the KGD performed as well as the EEGD in modeling the observed numbers of consumers. The results also show that the KGD outperformed the EEGD in modeling the number of absences among shift-workers. It is expected that the additional parameter offered by the Kumaraswamy's distribution will enable the use of the KGD in modeling events where the EEGD or the geometric distribution may not provide adequate fits.

## Appendix

### Elements of the information matrix

The values of *A*_{
x
}, *B*_{
x
}, *C*_{
x
}, *D*_{
x
}, *E*_{
x
}, and *F*_{
x
} are given in Section 6.

## Authors' contributions

The authors, viz AA, FF and CL with the consultation of each other carried out this work and drafted the manuscript together. All authors read and approved the final manuscript.

## Declarations

### Acknowledgment

The authors are grateful to the Associate Editor and the anonymous referees for many constructive comments and suggestions that have greatly improved the paper.

## Authors’ Affiliations

## References

- Akinsete A, Famoye F, Lee C: The beta-Pareto distributions.
*Statistics*2008, 42(6):547–563. 10.1080/02331880801983876MathSciNetView ArticleMATHGoogle Scholar - Alexander C, Cordeiro GM, Ortega EMM, Sarabia JM: Generalized beta-generated distributions.
*Comput. Stat. Data Anal*2012, 56: 1880–1897. 10.1016/j.csda.2011.11.015MathSciNetView ArticleMATHGoogle Scholar - Alshawarbeh E, Famoye F, Lee C: Beta-Cauchy distribution: some properties and its applications.
*J. Stat. Theory Appl*2013, 12: 378–391. 10.2991/jsta.2013.12.4.5MathSciNetView ArticleGoogle Scholar - Alzaatreh A, Famoye F, Lee C: Gamma-Pareto distribution and its applications.
*J. Mod. Appl. Stat. Meth*2012a, 11(1):78–94.Google Scholar - Alzaatreh A, Famoye F, Lee C: Weibull-Pareto distribution and its applications.
*Comm. Stat. Theor. Meth*2013a, 42(7):1673–1691. 10.1080/03610926.2011.599002MathSciNetView ArticleGoogle Scholar - Alzaatreh A, Lee C, Famoye F: On the discrete analogues of continuous distributions.
*Stat. Meth*2012b, 9: 589–603. 10.1016/j.stamet.2012.03.003MathSciNetView ArticleGoogle Scholar - Alzaatreh A, Lee C, Famoye F: A new method for generating families of continuous distributions.
*Metron*2013b, 71: 63–79. 10.1007/s40300-013-0007-yMathSciNetView ArticleGoogle Scholar - Arbous AG, Sichel HS: New techniques for the analysis of asenteeism data.
*Biometrika*1954, 41: 77–90. 10.1093/biomet/41.1-2.77View ArticleMATHGoogle Scholar - Chatfield, C: A marketing application of characterization theorem. In: Patil, GP, Kotz, S, Ord, JK (eds.)Statistical Distributions in Scientific Work, volume 2, pp. 175-185. D. Reidel Publishing Company, Boston (1975). Chatfield, C: A marketing application of characterization theorem. In: Patil, GP, Kotz, S, Ord, JK (eds.)Statistical Distributions in Scientific Work, volume 2, pp. 175-185. D. Reidel Publishing Company, Boston (1975).Google Scholar
- Consul PC:
*Generalized Poisson Distributions: Properties and Applications*. Marcel Dekker, Inc., New York; 1989.MATHGoogle Scholar - Consul PC, Famoye F:
*Lagrangian Probability Distributions*. Birkhäuser, Boston; 2006.MATHGoogle Scholar - Cordeiro GM, de Castro M: A new family of generalized distributions.
*J. Stat. Comput. Simulat*2011, 81(7):883–898. 10.1080/00949650903530745MathSciNetView ArticleMATHGoogle Scholar - Cordeiro GM, Lemonte AJ: The beta Laplace distribution.
*Stat. Probability Lett*2011, 81: 973–982. 10.1016/j.spl.2011.01.017MathSciNetView ArticleMATHGoogle Scholar - Cordeiro GM, Nadarajah S, Ortega EMM: The Kumaraswamy Gumbel distribution.
*Stat. Methods Appl*2012, 21: 139–168. 10.1007/s10260-011-0183-yMathSciNetView ArticleGoogle Scholar - de Pascoa MAR, Ortega EMM, Cordeiro GM: The Kumaraswamy generalized gamma distribution with application in survival analysis.
*Stat. Meth*2011, 8: 411–433. 10.1016/j.stamet.2011.04.001MathSciNetView ArticleMATHGoogle Scholar - de Santana TV, Ortega EMM, Cordeiro GM, Silva GO: The Kumaraswamy-log-logistic distribution.
*Stat. Theory Appl*2012, 3: 265–291.Google Scholar - Eugene N, Lee C, Famoye F: The beta-normal distribution and its applications.
*Comm. Stat. Theor. Meth*2002, 31(4):497–512. 10.1081/STA-120003130MathSciNetView ArticleMATHGoogle Scholar - Famoye F, Lee C, Olumolade O: The beta-Weibull distribution.
*J. Stat. Theory Appl*2005, 4(2):121–136.MathSciNetGoogle Scholar - Gupta RC, Ong SD: A new generalization of the negative binomial distribution.
*Comput. Stat. Data Anal*2004, 45: 287–300. 10.1016/S0167-9473(02)00301-8MathSciNetView ArticleMATHGoogle Scholar - Jain GC, Consul PC: A generalized negative binomial distribution.
*SIAM J. Appl. Math*1971, 21: 501–513. 10.1137/0121056MathSciNetView ArticleMATHGoogle Scholar - Johnson NL, Kemp AW, Kotz S:
*Univariate Discrete Distributiuons*. John Wiley & Sons, New York; 2005.View ArticleGoogle Scholar - Jones MC: Kumaraswamy's distribution: a beta-type distribution with some tractability advantages.
*Stat. Methodologies*2009, 6: 70–81. 10.1016/j.stamet.2008.04.001View ArticleMATHGoogle Scholar - Kong L, Lee C, Sepanski JH: On the properties of of beta-gamma distribution.
*J. Mod. Appl. Stat. Meth*2007, 6(1):187–211.Google Scholar - Kumaraswamy P: A generalized probability density function for double-bounded random processes.
*Hydrology*1980, 46: 79–88. 10.1016/0022-1694(80)90036-0View ArticleGoogle Scholar - Lemonte AJ, Barreto-Souza W, Cordeiro GM: The exponentiated Kumaraswamy distribution and its log-transform.
*Braz. J. Probability Stat*2013, 27: 31–53. 10.1214/11-BJPS149MathSciNetView ArticleMATHGoogle Scholar - Mahmoudi E: The beta generalized Pareto distribution with application to lifetime data.
*Math. Comput. Simulations*2011, 81: 2414–2430. 10.1016/j.matcom.2011.03.006MathSciNetView ArticleMATHGoogle Scholar - Nadarajah S, Kotz S: The beta Gumbel distribution.
*Math. Probl. Eng*2004, 2004(4):323–332. 10.1155/S1024123X04403068MathSciNetView ArticleMATHGoogle Scholar - Kotz S: The beta exponential distribution.
*Reliability Eng. Syst. Saf*2006, 91: 689–697. 10.1016/j.ress.2005.05.008View ArticleGoogle Scholar - Singla N, Jain K, Sharma SK: The beta generalized Weibull distribution: properties and applications.
*Reliability Eng. Syst. Saf*2012, 102: 5–15. 10.1016/j.ress.2012.02.003View ArticleGoogle Scholar - Zwillinger D, Kokoska S:
*Standard Probability and Statistics Tables and Formulae*. Chapman and Hall/CRC, Boca Raton; 2000.MATHGoogle Scholar

## Copyright

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/4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly credited. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/), applies to the data made available in this article, unless otherwise stated.