Journal of Applied Probability and Statistics

2018, Vol. 13, No. 2, pp. 1-22

Copyright ISOSS Publications 2018

THE UNIT-WEIBULL DISTRIBUTION AND ASSOCIATED

INFERENCE

J. Mazucheli1 , A. F. B. Menezes1 and M. E. Ghitany2

1
Department of Statistics
Universidade Estadual de Maringá, PR, Brazil
Email: [email protected], Email: [email protected]
2
Department of Statistics and Operations Research
Faculty of Science, Kuwait University, Kuwait
Email: [email protected]

summary
Distributions such as the Simplex, Johnson SB , unit-Gamma and unit-Logistic,
each with support on the unit interval (0,1), are formulated using appropriate
transformation of random variables following inverse-Gaussian mixture, Normal,
Gamma and Logistic distributions, respectively. These distributions can serve
the same purpose of the Beta and Kumaraswamy distributions. In this paper, we
propose a new two-parameter unit-Weibull distribution which is also useful for
modeling data on the unit interval (0,1). Some properties of this new distribution
are studied. Monte Carlo simulations reveal that the maximum likelihood esti-
mators are nearly unbiased and consistent. The potential of this new distribution
is illustrated using two real data sets.

Keywords and phrases: Beta distribution, Maximum likelihood estimation, Monte

Carlo simulation, unit-Gamma distribution, Weibull distribution.
2010 Mathematics Subject Classification: 60E05, 62F10.

1 Introduction
Although the Beta distribution is flexible and has been the most used to modeled data on bounded
domain, in the last years several works have been proposed new distributions on a unit interval.
0
We can mention the following: the Johnson SB distribution [18], the Johnson SB distribution
[19], the Topp-Leone distribution [32], the unit-Gamma distribution [17, 30], the Kumaraswamy
distribution [21], the Arcsine distribution [2], the unit-Logistic distribution [31], the McDonald’s
generalized Beta type I distribution [23], the Simplex distribution [3], the reflected Generalized
Topp-Leone distribution [33], the Beta power distribution [10], the McDonald Arcsine distribu-
tion [11], the Log-Lindley distribution [16], the exponentiated Kumaraswamy distribution [22], the
exponentiated Topp-Leone distribution [25], the Marshall-Olkin extended Kumaraswamy [7], the
reflected generalized Topp-Leone power series distribution [9], the transmuted Kumaraswamy dis-
tribution [29], the size biased Kumaraswamy distribution [28] and the extended Arcsine distribution
[12]. It should be pointed that the majority of these distributions have more than two parameters,
which considering limited amount of data, may produce inaccurate estimates.
Here, following [17] and [30], we propose a new distribution with support on the unit-interval
(0, 1), which arises from a certain transformation on the two-parameter Weibull distribution [35]
with probability density function (p.d.f.)

β
g(y; α, β) = α β y β−1 e−α y , y > 0, α, β > 0, (1.1)

where α and β are the scale and shape parameters, respectively.

Using the transformation X = e−Y , we have a new distribution on (0, 1), which we refer to as
unit-Weibull (UW) distribution. Its cumulative distribution function is given by

F (x; α, β) = exp −α (− log x)β ,

 
0 < x < 1, α, β > 0, (1.2)

and the corresponding p.d.f. is

1
α β (− log x)β−1 exp −α (− log x)β ,
 
f (x; α, β) = 0 < x < 1, α, β > 0. (1.3)
x

Note that α is no longer a scale parameter, since f (αx; α, β) 6= α1 f (x; 1, β). Special cases of the
UW distributions include: the standard uniform distribution over the interval (0,1) (α = β = 1),
the power function distribution (β = 1) and the unit-Rayleigh distribution (β = 2). Therefore,
the new distribution has connection with some well known distributions, and hence, it can be very
useful in many practical situations. Figure 1 shows some possible shapes of the p.d.f. of the UW
distribution for selected values of the parameters α and β.
The purpose of this paper are to introduce and study some properties of the UW distribution.
In Section 2, we present some features of the UW distribution. We discuss the maximum likelihood
estimation and inference of the model parameters in Section 3, where we also derived explicit
expressions for the expected Fisher information matrix. Monte Carlo simulations are conducted in
Section 4 in order to study some properties of the maximum likelihood estimators and to evaluate
the coverage probability of asymptotic confidence intervals. Section 5 shows the comparison between
the new proposed distribution and some other distributions using two real data sets. Finally, some
concluded remarks are given in Section 6.

2
 β = 0.5 β = 1.0
3.00

3.00
α = 0.5 α = 2.0
Probability Density Function

Probability Density Function

α = 1.0 α = 3.0
2.25

2.25
α = 1.5 α = 5.0
1.50

1.50
0.75

0.75
0.00

0.00
0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
x x
β = 1.5 β = 2.0
2.62

2.70
Probability Density Function

Probability Density Function

1.97

2.03
1.31

1.35
0.66

0.68
0.00

0.00

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
x x

Figure 1: Probability density function of the UW distribution for selected values of α and β.

Note that, unlike the Beta distribution, the proposed model has closed form expression for the
quantile function. This fact can be used to introduce a quantile regression model which may be a
more flexible alternative to the classical Beta regression model [8, 15]. As discussed in the statistical
literature [20, 36, 24, 27, 4] the quantile regression analysis has been used in several contexts and its
main advantage when compared with the conditional-mean regressions, such as Beta and Simplex,
is that it provides a complete view of the conditional distribution by studying distinct quantiles.
By employing quantile regression such as conditional-median regressions, practitioners will have a
more robust model for outliers than the usual Beta regression. Another advantage lies on the fact
that if the conditional dependent variable is skewed, the median may be a more appropriate when
compared with the mean. Since the main goal of this paper is to introduce and study some of

3
properties of the UW distribution, the quantile regression issue is not addressed throughout the
text.

2 Statistical Properties
In this section, we explore some statistical properties of the proposed UW distribution.

2.1 Hazard rate function

The hazard rate function of the UW distribution is given by
 
f (x; α, β) αβ (− log x)β−1 exp −α (− log x)β
h(x; α, β) = = , 0 < x < 1. (2.1)
1 − F (x; α, β) x (1 − exp [−α (− log x)β ])
Figure 2 shows some possible shapes of the hazard rate function of the UW distribution for
selected values of the parameters α and β. Figure 2 shows increasing or bathtub shapes of the
hazard rate function of the UW distribution. These shapes are also similar to the shapes of the
Beta distribution (see Ghitany, 2004).

β = 0.5 β = 1.0
7.50

7.50
α = 0.5 α = 2.0

α = 1.0 α = 3.0
5.63

5.63
Hazard Rate Function

α = 1.5 α = 5.0
Hazard Rate Function
3.75

3.75
1.88

1.88
0.00

0.00

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
x x
β = 1.5 β = 2.0
7.50

7.50
5.62

5.62
Hazard Rate Function

Hazard Rate Function

3.75

3.75
1.88

1.88
0.00

0.00

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
x x

Figure 2: Hazard rate function of the UW distribution for selected values of α and β.

4
2.2 Moments and associated measures
The rth raw moment of the UW distribution is given by
∞
(−1)n
 
X n
µ0r = E(X r ) = E(e−rY ) = MY (−r) = Γ + 1 .
n=0
n! αn/β β

The skewness and kurtosis measures can be obtained from the expressions

µ03 − 3µ02 µ + µ3
skewness = ,
σ3
µ04 − 4µ03 µ + 6µ02 µ2 − 3µ4
kurtosis = ,
σ4
upon substituting for the raw moments.
However, for the special case β = 1, i.e. the power function distribution, we have
α
µ0r = E(X r ) = , r = 1, 2, . . . .
r+α
In this case, the mean, variance, skewness and kurtosis, respectively, are given by
α α
µ= , σ2 = ,
1+α (1 + α)2 (2 + α)
r
2(1 − α) 2 3(2 + α)(2 − α + 3α2 )
skewness = 1+ and kurtosis = .
(2 + α) α α(3 + α)(4 + α)
Note that, in this case, the skewness can be negative, zero, positive when α < 1, α = 1, α > 1,
respectively.
Setting α = 1 in the last expressions, i.e., the standard uniform distribution, we obtain
1 1 9
µ= , σ2 = , skewness = 0 and kurtosis = .
2 12 5
Similarly, for the special case β = 2, i.e., the unit-Rayleigh distribution, we have
√  
0 r π r 2 /(4α) r
µr = E(X ) = 1 − √ r e erfc √ , r = 1, 2, . . . ,
2 α 2 α

where Z ∞
2 2
erfc(z) = √ e−x dx, z > 0,
π z

is the complementary error function. In this case, the mean and variance of the UW distribution,
respectively, are √  
π 1
µ = 1 − √ e1/(4α) erfc √ ,
2 α 2 α
√    √  2
2 π 1/α 1 π 1/(4α) 1
σ =1− √ e erfc √ − 1− √ e erfc √ .
α α 2 α 2 α

5
 Figure 3 shows the mean, variance, skewness and kurtosis of the UW distribution as a function
of β for various values of α. This figure shows that the skewness can be negative for values β 6= 1
which means that for modeling negatively skewed data, the UW distribution will be a useful model.
1.00

0.10
α = 0.5 α = 4.0

α = 1.0 α = 6.0
0.83

0.08
α = 2.0 α = 10.0

variance
mean
0.67

0.05
0.50

0.03
0.33

0.00
0.0 1.2 2.5 3.8 5.0 0.0 1.2 2.5 3.8 5.0
β β

15.00
1.10

11.58
0.07
skewness

kurtosis
−0.95

8.17
−1.98

4.75
−3.00

1.33

0.0 1.2 2.5 3.8 5.0 0.0 1.2 2.5 3.8 5.0
β β

Figure 3: Mean, variance, skewness and kurtosis of the UW distribution for selected values of α
and β.

2.3 Quantile function and associated measures

The quantile function of the UW distribution is given by
"  1 #
log p β
Q(p) = exp − − , 0 < p < 1. (2.2)
α

The specialpcases α = β = 1, β = 1 and β = 2, respectively, give Q(p) = p, Q(p) = p1/α and

Q(p) = exp(− − log(p)/α). The quartiles of the UW distribution, as well as the special cases, are
obtained by setting p = 14 , 12 , 34 , respectively.

6
3 Maximum Likelihood Estimation
Let x = (x1 , . . . , xn ) be a random sample of size n from the UW distribution with p.d.f. (1.3).
Then, the log-likelihood function of θ = (α, β) is given by
n
X
`(θ; x) = log f (xi ; θ)
i=0
n
X n
X n
X
= n(log α + log β) − log xi + (β − 1) log (− log xi ) − α (− log xi )β . (3.1)
i=1 i=1 i=1

likelihood estimate θ
b of θ is obtained by solving the non-linear equations
n
∂` n X
= − (− log xi )β = 0, (3.2)
∂α α i=1
and
n n
∂` n X X
= + log (− log xi ) − α (− log xi )β log (− log xi ) = 0. (3.3)
∂β β i=1 i=1
n
Equation (3.2) can be solved algebraically for α, giving α b(β) = n .
X β
(− log xi )
i=1
To obtain β,
b we substitute α
b(β) into (3.3) and solve for β. We have
n Pn β
n X n i=1 (− log xi ) log (− log xi )
g(β) = + log (− log xi ) − n . (3.4)
β i=1 X β
(− log xi )
i=1

Equation (3.4) can be solved numerically using, for example, Brent’s method [6] available in
software R [26] through the uniroot function. This method has the advantage that it does not
require computation of the derivative g 0 (β) and initial guess for β can be provided as an interval.
Note that (1.2) satisfies log[− log F (xi ; α, β)] = log α + β log(− log xi ), for i = 1, . . . , n. Thus
a plot of log[− log Fb(x(i) )] versus log(− log x(i) ) would be roughly linear if a UW distribution is
appropriate, where Fb(x(i) ) is the empirical distribution function at the ordered observed value x(i) .
In addition, when the plot is approximately linear, one can obtain empirical estimates of α and β by
fitting a straight line. This empirical estimate of β can be used as initial guess to solve numerically
equation (3.4).
The expected Fisher information matrix of θ = (α, β) based on a single observation is given by
  2 
∂ log f (xi ; θ)
I(θ) = [Iij ] = −E , i, j = 1, 2,
∂θ i ∂θ j
(3.5)
1
 
1
 α αβ (1 − γ − log α) 
=  ,
 
 
1 1
 2 2

αβ (1 − γ − log α) 6β 2 π + 6(1 − γ − log α)

7
where π ' 3.141593 and γ ' 0.577216 is the Euler’s constant.
Under mild regularity conditions (see Lehmann and Casella, 1998, pp. 461–463), the asymptotic
distribution of the MLE θb of θ is such that

√ D
b − θ) −→
n (θ N (0, I(θ)),

D
where −→ denotes convergence in distribution and I−1 (θ) is the inverse of the matrix I(θ), with

 
α2
  6αβ
 π2 π 2 + 6(1 − γ − log α)2 − (1 − γ − log α)
π2 
I−1 (θ) = [σij ] = 
 
.
 
 6αβ 6β 2 
− (1 − γ − log α)
π2 π2

The large-sample 100(1 − δ)% confidence intervals for α and β, respectively, are given by

√ √
σ σ
√ 11
b22
b ± zδ/2 β ± zδ/2 √ ,
b
α and b
n n

where σ
b11 and σ b22 , respectively, are the estimated asymptotic variances of the maximum likelihood
estimators α
b, βb and zq is the upper q-th quantile of the standard normal distribution.
Although not considered in this paper, it is important top
note that for a Bayesian analysis we can
use the Jeffreys invariant prior [5] for θ, given by π(θ) ∝ |I(θ)| where |I(θ)| is the determinant
of (10). Two alternative prior joint distributions for α and β can be found, for example, in [1].

4 Simulation Studies

In this section, we carry out Monte Carlo simulations to study the finite-sample behavior of
the MLEs and the asymptotic confidence intervals for the parameters of UW distribution. The
evaluation was performed based on the estimated bias, the estimated root mean-squared error
(RMSE) and the coverage probabilities. We set the samples sizes n = 10, 20, 50, 100, 200 and 500,
α = 0.5, 1.0, 1.5, 2.0, 3.0 and 5.0 and β = 0.5, 1.0, 1.5 and 2.0. For each combination of n, α and
β, we generated random samples from Y ∼ Weibull(α, β) and apply transformation e−Y . The
number of Monte Carlo replications was fixed at M = 10, 000. All simulations were conducted in
Ox Console [13], using the MaxBFGS function to obtain the maximum likelihood estimates of α and
β. The results are reported in Tables 1–4.

8
Table 1: Estimated bias, root mean-squared and coverage probability α and β (β = 0.5).
Bias RMSE CP90% CP95%
α n α β α β α β α β
10 0.0150 0.0833 0.2458 0.1907 0.8465 0.8885 0.9201 0.9426
20 0.0044 0.0366 0.1533 0.1077 0.8784 0.8941 0.9372 0.9456
50 0.0010 0.0143 0.0949 0.0605 0.8930 0.8954 0.9450 0.9463
0.5
100 0.0008 0.0064 0.0671 0.0405 0.8920 0.9023 0.9444 0.9499
200 0.0011 0.0029 0.0477 0.0282 0.8935 0.9020 0.9452 0.9497
500 0.0002 0.0014 0.0296 0.0174 0.9004 0.9029 0.9489 0.9502
10 0.1080 0.0855 0.5083 0.1954 0.8758 0.8843 0.9358 0.9404
20 0.0402 0.0384 0.2765 0.1090 0.8825 0.8932 0.9394 0.9451
50 0.0119 0.0141 0.1568 0.0600 0.8940 0.8987 0.9455 0.9480
1.0
100 0.0059 0.0066 0.1076 0.0408 0.8971 0.8953 0.9472 0.9462
200 0.0039 0.0035 0.0754 0.0284 0.8998 0.8945 0.9486 0.9458
500 0.0017 0.0014 0.0472 0.0176 0.8989 0.8995 0.9481 0.9484
10 0.2618 0.0836 0.8520 0.1909 0.9269 0.8868 0.9628 0.9417
20 0.1092 0.0378 0.4475 0.1086 0.8961 0.8930 0.9466 0.9450
50 0.0392 0.0149 0.2371 0.0609 0.8949 0.8955 0.9460 0.9463
1.5
100 0.0180 0.0072 0.1584 0.0407 0.8984 0.8979 0.9478 0.9476
200 0.0091 0.0035 0.1088 0.0280 0.8954 0.8973 0.9463 0.9473
500 0.0032 0.0015 0.0675 0.0176 0.8996 0.8947 0.9485 0.9459
10 0.4738 0.0823 1.2696 0.1887 0.9436 0.8865 0.9714 0.9415
20 0.1993 0.0376 0.6519 0.1104 0.9129 0.8917 0.9554 0.9443
50 0.0663 0.0142 0.3275 0.0602 0.9051 0.8982 0.9514 0.9477
2.0
100 0.0290 0.0063 0.2180 0.0406 0.9003 0.8989 0.9488 0.9481
200 0.0119 0.0029 0.1491 0.0280 0.8988 0.9012 0.9481 0.9493
500 0.0036 0.0011 0.0922 0.0175 0.9048 0.8988 0.9512 0.9481
10 0.7528 0.0687 1.7562 0.1655 0.9523 0.9113 0.9759 0.9546
20 0.4328 0.0375 1.1766 0.1078 0.9357 0.9008 0.9673 0.9491
50 0.1555 0.0141 0.5872 0.0607 0.9134 0.8923 0.9557 0.9446
3.0
100 0.0708 0.0066 0.3707 0.0407 0.9066 0.8975 0.9522 0.9474
200 0.0358 0.0032 0.2515 0.0283 0.9027 0.8959 0.9501 0.9465
500 0.0154 0.0013 0.1557 0.0176 0.9022 0.8981 0.9498 0.9477
10 0.5698 0.0329 1.9063 0.1254 0.9412 0.9367 0.9702 0.9678
20 0.5397 0.0241 1.6698 0.0916 0.9464 0.9207 0.9728 0.9595
50 0.3418 0.0137 1.1562 0.0592 0.9251 0.9026 0.9618 0.9501
5.0
100 0.1743 0.0070 0.7669 0.0405 0.9049 0.8994 0.9513 0.9484
200 0.0900 0.0036 0.5160 0.0285 0.9024 0.8956 0.9499 0.9464
500 0.0313 0.0014 0.3125 0.0176 0.9053 0.9020 0.9515 0.9497

9
Table 2: Estimated bias, root mean-squared and coverage probability α and β (β = 1.0).
Bias RMSE CP90% CP95%
α n α β α β α β α β
10 0.0098 0.1691 0.2337 0.3891 0.8501 0.8842 0.9220 0.9403
20 0.0004 0.0783 0.1523 0.2208 0.8765 0.8969 0.9362 0.9470
50 -0.0004 0.0290 0.0939 0.1220 0.8960 0.8995 0.9466 0.9484
0.5
100 0.0004 0.0140 0.0659 0.0821 0.8989 0.9006 0.9481 0.9490
200 0.0003 0.0064 0.0468 0.0567 0.8966 0.9008 0.9469 0.9491
500 0.0004 0.0025 0.0295 0.0352 0.9030 0.9000 0.9503 0.9487
10 0.1060 0.1751 0.5173 0.3890 0.8740 0.8853 0.9349 0.9409
20 0.0380 0.0777 0.2751 0.2187 0.8849 0.8954 0.9407 0.9463
50 0.0125 0.0301 0.1570 0.1212 0.8988 0.8990 0.9481 0.9482
1.0
100 0.0050 0.0149 0.1071 0.0819 0.9012 0.8966 0.9493 0.9469
200 0.0033 0.0068 0.0746 0.0563 0.9027 0.9001 0.9501 0.9487
500 0.0007 0.0029 0.0466 0.0353 0.9052 0.8985 0.9514 0.9479
10 0.2610 0.1754 0.8426 0.3959 0.9261 0.8829 0.9623 0.9396
20 0.1039 0.0776 0.4367 0.2227 0.9012 0.8891 0.9493 0.9429
50 0.0368 0.0288 0.2344 0.1212 0.9005 0.8948 0.9489 0.9459
1.5
100 0.0180 0.0132 0.1560 0.0813 0.9015 0.9024 0.9495 0.9499
200 0.0080 0.0063 0.1068 0.0568 0.9052 0.8933 0.9514 0.9451
500 0.0031 0.0024 0.0670 0.0355 0.9023 0.8951 0.9499 0.9461
10 0.4580 0.1584 1.2515 0.3679 0.9378 0.8901 0.9684 0.9434
20 0.2024 0.0743 0.6800 0.2176 0.9075 0.8936 0.9526 0.9453
50 0.0704 0.0296 0.3330 0.1222 0.9039 0.8935 0.9507 0.9453
2.0
100 0.0324 0.0149 0.2144 0.0813 0.9055 0.9035 0.9516 0.9505
200 0.0156 0.0086 0.1478 0.0559 0.9014 0.9044 0.9494 0.9510
500 0.0068 0.0034 0.0923 0.0347 0.9012 0.9070 0.9493 0.9524
10 0.7296 0.1363 1.7602 0.3359 0.9475 0.9092 0.9734 0.9535
20 0.4091 0.0712 1.1710 0.2143 0.9337 0.9015 0.9663 0.9495
50 0.1440 0.0264 0.5767 0.1188 0.9045 0.8979 0.9511 0.9476
3.0
100 0.0686 0.0133 0.3705 0.0820 0.9020 0.8943 0.9497 0.9457
200 0.0343 0.0063 0.2510 0.0563 0.9021 0.8986 0.9498 0.9479
500 0.0121 0.0026 0.1544 0.0349 0.9014 0.9022 0.9494 0.9498
10 0.5237 0.0614 1.9039 0.2484 0.9379 0.9396 0.9685 0.9693
20 0.5298 0.0472 1.6565 0.1820 0.9399 0.9253 0.9695 0.9619
50 0.3347 0.0272 1.1703 0.1193 0.9184 0.8989 0.9583 0.9481
5.0
100 0.1618 0.0138 0.7705 0.0820 0.9040 0.8988 0.9508 0.9480
200 0.0818 0.0072 0.5189 0.0567 0.9000 0.8959 0.9487 0.9465
500 0.0277 0.0027 0.3129 0.0354 0.8992 0.8966 0.9483 0.9469

10
Table 3: Estimated bias, root mean-squared and coverage probability α and β (β = 1.5).
Bias RMSE CP90% CP95%
α n α β α β α β α β
10 0.0098 0.2541 0.2450 0.5860 0.8376 0.8851 0.9152 0.9408
20 0.0002 0.1145 0.1566 0.3309 0.8651 0.8885 0.9301 0.9426
50 0.0003 0.0399 0.0959 0.1828 0.8869 0.8910 0.9418 0.9439
0.5
100 -0.0001 0.0198 0.0667 0.1214 0.8977 0.9020 0.9475 0.9497
200 -0.0002 0.0098 0.0470 0.0846 0.8988 0.9008 0.9481 0.9491
500 0.0003 0.0040 0.0299 0.0533 0.8948 0.8969 0.9459 0.9470
10 0.1027 0.2559 0.4828 0.5822 0.8807 0.8875 0.9385 0.9421
20 0.0396 0.1162 0.2759 0.3282 0.8868 0.8929 0.9417 0.9449
50 0.0147 0.0398 0.1556 0.1806 0.8971 0.8966 0.9472 0.9469
1.0
100 0.0084 0.0184 0.1074 0.1213 0.9008 0.9022 0.9491 0.9498
200 0.0035 0.0082 0.0741 0.0829 0.9080 0.9091 0.9529 0.9535
500 0.0012 0.0036 0.0467 0.0525 0.9030 0.9010 0.9503 0.9492
10 0.2851 0.2527 0.8907 0.5729 0.9267 0.8892 0.9627 0.9430
20 0.1094 0.1138 0.4514 0.3319 0.8922 0.8909 0.9446 0.9439
50 0.0364 0.0437 0.2345 0.1830 0.8990 0.8963 0.9482 0.9467
1.5
100 0.0164 0.0220 0.1595 0.1225 0.8934 0.8956 0.9452 0.9464
200 0.0076 0.0119 0.1093 0.0852 0.8973 0.8985 0.9473 0.9479
500 0.0030 0.0047 0.0680 0.0530 0.8981 0.8988 0.9477 0.9481
10 0.4765 0.2452 1.2648 0.5596 0.9399 0.8888 0.9695 0.9428
20 0.1984 0.1108 0.6559 0.3220 0.9112 0.8953 0.9546 0.9462
50 0.0722 0.0429 0.3308 0.1806 0.9005 0.9001 0.9489 0.9487
2.0
100 0.0341 0.0207 0.2169 0.1227 0.8992 0.9011 0.9483 0.9493
200 0.0187 0.0102 0.1502 0.0850 0.8992 0.8984 0.9483 0.9478
500 0.0070 0.0040 0.0918 0.0525 0.9034 0.8989 0.9505 0.9481
10 0.7498 0.2087 1.7994 0.5119 0.9459 0.9074 0.9726 0.9526
20 0.4066 0.1074 1.1549 0.3228 0.9312 0.8945 0.9650 0.9458
50 0.1563 0.0444 0.5828 0.1823 0.9104 0.8939 0.9541 0.9455
3.0
100 0.0778 0.0224 0.3743 0.1239 0.9029 0.8999 0.9502 0.9486
200 0.0416 0.0115 0.2496 0.0850 0.9050 0.8982 0.9513 0.9477
500 0.0174 0.0048 0.1542 0.0530 0.8990 0.8990 0.9482 0.9482
10 0.5604 0.0970 1.9143 0.3714 0.9421 0.9440 0.9706 0.9716
20 0.5651 0.0757 1.6948 0.2778 0.9479 0.9183 0.9736 0.9583
50 0.3365 0.0415 1.1552 0.1783 0.9240 0.9018 0.9613 0.9497
5.0
100 0.1676 0.0202 0.7654 0.1225 0.9111 0.8968 0.9545 0.9470
200 0.0792 0.0101 0.5124 0.0847 0.9011 0.9009 0.9493 0.9492
500 0.0311 0.0037 0.3113 0.0529 0.9022 0.8974 0.9498 0.9473

11
Table 4: Estimated bias, root mean-squared and coverage probability α and β (β = 2.0).
Bias RMSE CP90% CP95%
α n α β α β α β α β
10 0.0085 0.3453 0.2402 0.7956 0.8399 0.8807 0.9165 0.9385
20 0.0022 0.1565 0.1575 0.4514 0.8685 0.8855 0.9319 0.9410
50 0.0004 0.0569 0.0948 0.2427 0.8908 0.8963 0.9438 0.9467
0.5
100 0.0004 0.0280 0.0667 0.1632 0.8930 0.8983 0.9450 0.9478
200 0.0004 0.0145 0.0474 0.1129 0.8968 0.8990 0.9470 0.9482
500 -0.0001 0.0065 0.0297 0.0703 0.8967 0.9045 0.9469 0.9511
10 0.1012 0.3399 0.4895 0.7702 0.8788 0.8865 0.9374 0.9415
20 0.0401 0.1505 0.2721 0.4336 0.8879 0.8936 0.9423 0.9453
50 0.0151 0.0583 0.1575 0.2437 0.8938 0.8965 0.9454 0.9468
1.0
100 0.0075 0.0289 0.1074 0.1630 0.9021 0.9031 0.9498 0.9503
200 0.0034 0.0145 0.0753 0.1127 0.9001 0.9015 0.9487 0.9495
500 0.0013 0.0053 0.0475 0.0696 0.9013 0.9017 0.9494 0.9496
10 0.2876 0.3408 0.8786 0.7782 0.9323 0.8838 0.9656 0.9401
20 0.1169 0.1585 0.4559 0.4460 0.8911 0.8903 0.9440 0.9436
50 0.0396 0.0604 0.2367 0.2452 0.8938 0.8946 0.9454 0.9458
1.5
100 0.0199 0.0312 0.1600 0.1651 0.8960 0.8967 0.9466 0.9469
200 0.0100 0.0140 0.1098 0.1122 0.8946 0.9042 0.9458 0.9509
500 0.0039 0.0056 0.0677 0.0700 0.9016 0.9009 0.9495 0.9492
10 0.4538 0.3280 1.2459 0.7562 0.9383 0.8909 0.9687 0.9439
20 0.2002 0.1549 0.6651 0.4445 0.9123 0.8876 0.9551 0.9421
50 0.0689 0.0589 0.3319 0.2446 0.9051 0.8939 0.9514 0.9455
2.0
100 0.0356 0.0285 0.2204 0.1640 0.8983 0.9003 0.9478 0.9488
200 0.0179 0.0127 0.1506 0.1126 0.9002 0.9003 0.9488 0.9488
500 0.0065 0.0050 0.0922 0.0703 0.8989 0.8992 0.9481 0.9483
10 0.7609 0.2751 1.8108 0.6710 0.9456 0.9083 0.9724 0.9531
20 0.4247 0.1507 1.1785 0.4312 0.9321 0.8929 0.9654 0.9450
50 0.1562 0.0603 0.5821 0.2448 0.9067 0.8949 0.9522 0.9460
3.0
100 0.0699 0.0277 0.3708 0.1627 0.9058 0.8992 0.9517 0.9483
200 0.0368 0.0132 0.2496 0.1123 0.9034 0.9016 0.9505 0.9495
500 0.0148 0.0056 0.1536 0.0704 0.9032 0.9020 0.9504 0.9497
10 0.5371 0.1235 1.9087 0.5012 0.9422 0.9314 0.9707 0.9651
20 0.5459 0.0960 1.6681 0.3631 0.9468 0.9242 0.9731 0.9614
50 0.3516 0.0569 1.1766 0.2354 0.9233 0.9019 0.9609 0.9497
5.0
100 0.1844 0.0317 0.7775 0.1635 0.9071 0.8987 0.9524 0.9480
200 0.0921 0.0160 0.5167 0.1127 0.9035 0.9004 0.9505 0.9489
500 0.0352 0.0058 0.3129 0.0702 0.9021 0.9036 0.9498 0.9506

12
 Some of the points are very clear from the numerical experiments. Although the biases of α b
and βb goes to zero as sample size increase, both parameters are positively biased. It is also seen
that the RMSE of both parameters decrease as sample size increase. Interestingly, as the value of
α increase their corresponding bias is bigger, while the bias of β decrease. Therefore, estimation
of α becomes better for lower values of α whereas the estimation of β are more accurate for large
values of α.
Also Tables 1–4 show that, as the sample size increases, the coverage probabilities for α and β
are quite close to the nominal levels. Curiously, for large values of α their coverage probability is
greater than the nominal levels, while for β the coverage probability is very close.

5 Applications
In this section we present two applications using two published data sets which demonstrate the
suitability of the proposed UW distribution. The first data set is from [14] and refer to 20 obser-
vations of the maximum flood level (in millions of cubic feet per second) for Susquehanna River
at Harrisburg, Pennsylvania. The second data set refer to 48 observations obtained from 12 core
samples from petroleum reservoirs that were sampled by 4 cross-sections. It should be noted that
this data can be found in [26] on a data.frame named as rock. These data sets are reported in
Table 5.
The proposed two-parameter UW distribution is compared with the following two-parameter
distributions on the unit interval (0,1)

(i) Beta distribution:

Γ(α + β) α−1
f (x; α, β) = x (1 − x)β−1 , α, β > 0.
Γ(α) Γ(β)

(ii) Kumaraswamy distribution:

f (x; α, β) = α β xα−1 (1 − xα )β−1 , α, β > 0.

(iii) Johnson SB distribution:

(   2 )
β 1 1 x
f (x; α, β) = √ exp − α + β log , α ∈ R, β > 0.
2 π x (1 − x) 2 1−x

(iv) Unit-Logistic distribution:

β eα xβ−1 (1 − x)β−1
f (x; α, β) = 2 , α ∈ R, β > 0.
[xβ eα + (1 − x)β ]

(v) Simplex distribution:

(x − α)2
  
 2
 1
3 −2 1
f (x; α, β) = 2 π β {x (1 − x)} exp − 2 , α ∈ (0, 1), β > 0.
2β x (1 − x) α2 (1 − α)2

13
 (vi) Unit-Gamma distribution:

β α β−1
f (x; α, β) = x (− log x)α−1 , α, β > 0.
Γ(α)

(vii) Extended Arcsine distribution:

√
α−1 √
α β−1
     
αβ 2 2
f (x; α, β) = 1 − arcsin x 1 − 1 − arcsin x , α, β > 0.
π(x − x2 )1/2 π π

(viii) Exponentiated Topp-Leone distribution:

α−1 β−1
f (x; α, β) = 2 α β (1 − x) [x (2 − x)] [1 − xα (2 − x)α ] , α, β > 0.

Table 5: Flood level data and Petroleum reservoirs data.

Data Set I
0.26, 0.27, 0.30, 0.32, 0.32, 0.34, 0.38, 0.38, 0.39, 0.40, 0.41, 0.42, 0.42, 0.42, 0.45,
0.48, 0.49, 0.61, 0.65, 0.74
Data Set II
0.09, 0.11, 0.12, 0.12, 0.13, 0.14, 0.15, 0.15, 0.15, 0.15, 0.15, 0.16, 0.16, 0.16, 0.16,
0.17, 0.17, 0.18, 0.18, 0.18, 0.18, 0.19, 0.19, 0.20, 0.20, 0.20, 0.20, 0.20, 0.20, 0.23,
0.23, 0.23, 0.23, 0.24, 0.25, 0.26, 0.26, 0.28, 0.28, 0.28, 0.29, 0.31, 0.33, 0.33, 0.34,
0.42, 0.44, 0.46

The maximum likelihood estimates and their corresponding standard errors for both data sets
are given in Table 6. To check the suitability of the suitability of the UW distribution and the above
eight competing distributions, we consider three goodness-of-fit tests (Kolmogorov-Smirnov statistic
(KS), Anderson-Darling statistic (AD) and Cramér-von Mises statistic (CvM)). In order to compare
the UW distributions to the above eight competing distributions, we consider the likelihood-based
statistics (Akaike’s Information Criterion (AIC) and the Bayesian information criterion (BIC)). The
results for both data sets are presented in Table 7.

14
 Table 6: Maximum likelihood estimate (standard-error) for α and β.
Data Set I Data Set II
Distribution α
b βb α
b βb
1.0248 3.9036 0.0602 5.1130
UW
(0.2399) (0.6806) (0.0236) (0.5754)
6.7569 9.1117 5.9422 21.2070
i
(2.0946) (2.8518) (1.1815) (4.3472)
3.3634 11.7906 2.7186 44.6540
ii
(0.6034) (5.3604) (0.2935) (17.5720)
0.6143 1.9262 2.8736 2.1525
iii
(0.2438) (0.3045) (0.3269) (0.2197)
1.3599 3.5915 5.2285 3.8274
iv
(0.4793) (0.6886) (0.6913) (0.4611)
0.4309 1.0923 0.2197 1.1637
v
(0.0269) (0.1727) (0.0113) (0.1188)
8.7310 9.7251 17.9510 11.3100
vi
(2.7099) (3.1068) (3.6307) (2.3197)
9.1631 141.9528 14.1764 101.1463
vii
(1.7151) (126.7324) (1.6670) (52.7009)
4.6064 4.0442 3.1358 13.6413
viii
(0.9496) (1.4752) (0.3642) (4.1988)
UW: unit-Weibull, i: Beta, ii: Kumaraswamy, iii: Johnson SB , iv: Unit-Logistic,
v: Simplex, vi: Unit-Gamma, vii: Extended Arcsine and viii: Exponentiated
Topp-Leone.

A close inspection of Table 7 reveals that the UW distribution outperforms the competing
distributions for both data sets, since it has the smallest AIC and BIC values. This conclusion is
also support by the probability-probability plots in Figures 4 and 5, where we can see again that
the UW provides the best fit among the considered models.

15
 Table 7: Goodness-of-fit measures (p-values) and likelihood-based statistics.
Data Set I
Distribution KS CvM AD AIC BIC
UW 0.1448 (0.7958) 0.0512 (0.8742) 0.3434 (0.9013) -28.3430 -26.3515
i 0.1988 (0.4082) 0.1236 (0.4847) 0.7327 (0.5303) -24.1245 -22.1330
ii 0.2109 (0.3359) 0.1636 (0.3528) 0.9322 (0.3936) -21.7324 -19.7409
iii 0.1935 (0.4424) 0.1153 (0.5187) 0.6930 (0.5627) -24.5257 -22.5342
iv 0.1391 (0.8339) 0.0547 (0.8529) 0.4804 (0.7648) -25.4724 -23.4809
v 0.2098 (0.3424) 0.1447 (0.4087) 0.7970 (0.4815) -24.3065 -22.3150
vi 0.1955 (0.4293) 0.1178 (0.5084) 0.7046 (0.5530) -24.3769 -22.3854
vii 0.1543 (0.7275) 0.0564 (0.8419) 0.3890 (0.8581) -27.8320 -25.8405
viii 0.2063 (0.3625) 0.1432 (0.4136) 0.8142 (0.4692) -23.1852 -21.1937
Data Set II
Distribution KS CvM AD AIC BIC
UW 0.1007 (0.7143) 0.0383 (0.9434) 0.2338 (0.9782) -112.2416 -108.4992
i 0.1428 (0.2819) 0.1301 (0.4577) 0.7771 (0.4971) -107.2004 -103.4580
ii 0.1533 (0.2092) 0.2060 (0.2566) 1.2892 (0.2358) -100.9831 -97.2407
iii 0.1252 (0.4390) 0.0862 (0.6587) 0.5190 (0.7267) -109.9699 -106.2275
iv 0.0979 (0.7467) 0.0557 (0.8435) 0.4054 (0.8427) -109.9063 -106.1639
v 0.1297 (0.3945) 0.0965 (0.6041) 0.5569 (0.6888) -110.1133 -106.3709
vi 0.1365 (0.3325) 0.1130 (0.5263) 0.6793 (0.5756) -108.2175 -104.4751
vii 0.1138 (0.5628) 0.0492 (0.8829) 0.2957 (0.9411) -111.9385 -108.1961
viii 0.1525 (0.2145) 0.1866 (0.2957) 1.1477 (0.2880) -102.7118 -98.9694
UW: Unit-Weibull, i: Beta, ii: Kumaraswamy, iii: Johnson SB , iv: Unit-Logistic, v: Simplex,
vi: Unit-Gamma, vii: Extended Arcsine and viii: Exponentiated Topp-Leone.

In order to discriminate between the UW distribution with each competing distribution, we

apply the Vuong likelihood ratio test of non-nested distributions [34]. The Vuong test statistic is
defined as
n
1 X f (xi | mθ)
b
T = 2
√ log
ω
b n i=1
g(xi | mbγ)

16
where !2 " !#2
n n
2 1 X f (xi | mθ)
b 1 X f (xi | mθ)
b
ω
b = log − log
n i=1 g(xi | mbγ) n i=1 g(xi | mbγ)

1 P n f (xi | mθ)
b
is an estimator for the variance of √ log , f (xi | mθ)
b and g(xi | mb
γ ) are the
n i=1 g(xi | mbγ)
corresponding rival densities evaluated at the maximum likelihood estimates. It was demonstrated
D
that, when n → ∞, T → N (0, 1). At a significance level γ%, we reject distribution equivalence if
|T | < zγ/2 , where zq is the upper q-th quantile of the standard normal distribution.
The results of Voung test are given in Table 8. This table shows that UW and Extended Arcsine
distributions are equivalent for data set I. This table also shows that UW and Beta, Johnson SB ,
Unit-Logistic, Simplex, Unit-Gamma and Extended Arcsine distributions are equivalent for data
set II. However, the UW distribution has smallest AIC and BIC as shown in Table 7. Therefore
the UW distribution provides the best fit among the competing distributions.

Table 8: Observed values of Voung statistic (p-values).

Comparisons Data Set I Data Set II
UW vs Beta 2.4722 (0.0067) 1.4354 (0.0756)
UW vs Kumarasawamy 2.7266 (0.0032) 2.2934 (0.0109)
UW vs Johnson SB 2.3338 (0.0098) 0.9908 (0.1609)
UW vs Unit-Logistic 2.3340 (0.0098) 1.0926 (0.1373)
UW vs Simplex 2.0095 (0.0222) 0.9470 (0.1718)
UW vs Unit-Gamma 2.4346 (0.0075) 1.2882 (0.0988)
UW vs Extended Arcsine 1.4977 (0.0671) 0.2632 (0.3962)
UW vs Exponentiated Topp-Leone 2.5305 (0.0057) 2.0813 (0.0187)

17
 1.00

1.00

1.00
unit−Weibull ● Beta ● Kumaraswamy ●
● ● ●
● ● ●
● ● ●
0.75

0.75

0.75
● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities
● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

0.00
● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities
1.00

1.00

1.00
Johnson SB ● unit−Logistic ● Simplex ●
● ● ●
● ● ●
● ● ●
0.75

0.75

0.75
● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities
● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

● ● 0.00 ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities
1.00

1.00

1.00

unit−Gamma ● extended Arcsine ● exponentiated Topp−Leone ●

● ● ●
● ● ●
● ● ●
0.75

0.75

0.75

● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities

● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50

● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25

● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

0.00

● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities

Figure 4: PP-Plots of the fitted distributions — Data Set I.

18
 1.00

1.00

1.00
unit−Weibull ●
●
●
Beta ●
●
●
Kumaraswamy ●
●
●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.75

0.75

0.75
● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

0.00
● ● ●
● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities
1.00

1.00

1.00
Johnson SB ●
●
●
unit−Logistic ●
●
●
Simplex ●
●
●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.75

0.75

0.75
● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

0.00

● ● ●
● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities
1.00

1.00

1.00

unit−Gamma ●
●
●
extended Arcsine ●
●
●
exponentiated Topp−Leone ●
●
●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.75

0.75

0.75

● ● ●
Empirical probabilities

Empirical probabilities

Empirical probabilities

● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.50

0.50

0.50

● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.25

0.25

0.25

● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
● ● ●
0.00

0.00

0.00

● ● ●
● ● ●

0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00 0.00 0.25 0.50 0.75 1.00
Theoretical probabilities Theoretical probabilities Theoretical probabilities

Figure 5: PP-Plots of the fitted distributions — Data Set II.

19
6 Conclusions
In this paper a new two-parameter distribution, called the UW distribution with support on (0, 1),
is introduced and studied in details. Maximum likelihood estimators of the parameters and their
standard errors were derived. We also proposed a starting-point strategy using the fact that UW
cumulative distribution function can be linearized. Random sample for the distribution can be
easily simulated by simple transformation of samples generated from the Weibull distribution. A
simulation study was carried out to examine the bias and root mean-squared error of the maximum
likelihood estimators of the parameters as well as the coverage probability of the confidence inter-
vals. Applications of the proposed distribution to two real data sets showed better fit than many
other well-known two-parameter distributions with support on (0,1), such as Beta, Kumaraswamy,
Johnson SB , unit-Logistic, Simplex, unit-Gamma, extended Arcsine, and exponentiated Topp-
Leone distributions. Finally, since the UW is very flexible and has a closed form expression for
the quantiles, its quantile regression model will be useful in many applications and is now under
investigation.

Acknowledgments
The authors would like to thank the Editor-in-Chief, Associate editor, and two referees for helpful
comments and suggestions which greatly improved the presentation and quality of the paper.

References
[1] Achcar, J. A., Coelho-Barros, E. A., Cordeiro, G. M., 2013. Beta generalized distributions
and related exponentiated models: A Bayesian approach. Brazilian Journal of Probability and
Statistics 27 (1), 1–19.

[2] Arnold, B. C., Groeneveld, R. A., 1980. Some properties of the Arcsine distribution. Journal
of the American Statistical Association 75 (369), 173–175.

[3] Barndorff-Nielsen, O., Jørgensen, B., 1991. Some parametric models on the Simplex. Journal
of Multivariate Analysis 39 (1), 106–116.

[4] Bayes, C. L., Bazán, J. L., Castro, M., 2017. A quantile parametric mixed regression model
for bounded response variables. Statistics and Its Interface 10 (3), 483–493.

[5] Box, G. E. P., Tiao, G. C., 1973. Bayesian inference in statistical analysis. Addison-Wesley
Publishing, New York.

[6] Brent, R. P., 1973. Algorithms for Minimization without Derivatives. Prentice-Hall, Englewood
Cliffs, N.J.

[7] Castellares, F., Lemonte, A. J., 2015. On the Marshall-Olkin extended distributions. Commu-
nications in Statistics - Theory and Methods 45 (15), 4537–4555.

20
 [8] Cepeda-Cuervo, E., 2001. Variability modeling in generalized linear models. Ph.D. thesis,
Mathematics Institute, Universidade Federal do Rio de Janeiro.

[9] Condino, F., Domma, F., 2017. A new distribution function with bounded support: The
reflected generalized Topp-Leone power series distribution. METRON 75 (1), 51–68.

[10] Cordeiro, G. M., dos Santos, R. B., 02 2012. The Beta power distribution. Brazilian Journal
of Probability and Statistics 26 (1), 88–112.

[11] Cordeiro, G. M., Lemonte, A. J., 2012. The McDonald Arcsine distribution: A new model to
proportional data. Statistics 48 (1), 182–199.

[12] Cordeiro, G. M., Lemonte, A. J., Campelo, A. K., 2016. Extended Arcsine distribution to pro-
portional data: Properties and applications. Studia Scientiarum Mathematicarum Hungarica
53 (4), 440–466.

[13] Doornik, J. A., 2007. Object-Oriented Matrix Programming Using Ox, 3rd ed. London: Tim-
berlake Consultants Press and Oxford.

[14] Dumonceaux, R., Antle, C. E., 1973. Discrimination between the Log-Normal and the Weibull
distributions. Technometrics 15 (4), 923–926.

[15] Ferrari, S., Cribari-Neto, F., 2004. Beta regression for modelling rates and proportions. Journal
of Applied Statistics 31 (7), 799–815.

[16] Gómez-Déniz, E., Sordo, M. A., Calderı́n-Ojeda, E., 2013. The Log-Lindley distribution as an
alternative to the Beta regression model with applications in insurance. Insurance: Mathemat-
ics and Economics 54, 49–57.

[17] Grassia, A., 1977. On a family of distributions with argument between 0 and 1 obtained by
transformation of the Gamma distribution and derived compound distributions. Australian
Journal of Statistics 19 (2), 108–114.

[18] Johnson, N. L., 1949. Systems of frequency curves generated by methods of translation.
Biometrika 36 (1/2), 149–176.

[19] Johnson, N. L., 1955. Systems of frequency curves derived from the first law of Laplace. Tra-
bajos de Estadistica 5 (3), 283–291.

[20] Koenker, R., Bassett, G., 1978. Regression quantiles. Econometrica 46 (1), 33–50.

[21] Kumaraswamy, P., 1980. A generalized probability density function for double-bounded random
processes. Journal of Hydrology 46 (1), 79–88.

[22] Lemonte, A. J., Barreto-Souza, W., Cordeiro, G. M., 02 2013. The exponentiated Ku-
maraswamy distribution and its log-transform. Brazilian Journal of Probability and Statistics
27 (1), 31–53.

21
[23] McDonald, J. B., 1984. Some generalized functions for the size distribution of income. Econo-
metrica 52 (3), 647–663.

[24] Mitnik, P. A., Baek, S., 2013. The Kumaraswamy distribution: Median-dispersion re-
parameterizations for regression modeling and simulation-based estimation. Statistical Papers
54 (1), 177–192.

[25] Pourdarvish, A., Mirmostafaee, S. M. T. K., Naderi, K., 2015. The exponentiated Topp-Leone
distribution: Properties and application. Journal of Applied Environmental and Biological
Sciences 5 (78), 251–256.

[26] R Core Team, 2017. R: A Language and Environment for Statistical Computing. R Foundation
for Statistical Computing, Vienna, Austria, ISBN 3-900051-07-0.

[27] Santos, B., Bolfarine, H., 2015. Bayesian analysis for zero-or-one inflated proportion data using
quantile regression. Journal of Statistical Computation and Simulation 85 (17), 3579–3593.

[28] Sharma, D., Chakrabarty, T. K., 2016. On size biased Kumaraswamy distribution. Statistics,
Optimization and Information Computing 4, 252–264.

[29] Shuaib, K. M., Robert, K., Lena, H. I., 2016. Transmuted Kumaraswamy distribution. Statis-
tics in Transition new series 17 (2), 183–210.

[30] Tadikamalla, P. R., 1981. On a family of distributions obtained by the transformation of the
Gamma distribution. Journal of Statistical Computation and Simulation 13 (3–4), 209–214.

[31] Tadikamalla, P. R., Johnson, N. L., 1982. Systems of frequency curves generated by transfor-
mations of logistic variables. Biometrika 69 (2), 461–465.

[32] Topp, C. W., Leone, F. C., 1955. A family of J-Shaped frequency functions. Journal of the
American Statistical Association 50 (269), 209–219.

[33] van Drop, J. R., Kotz, S., 2006. Modeling Income Distributions Using Elevated Distributions
on a Bounded Domain. In: Distribution Models Theory. World Scientific Press, Singapore,
Ch. 1, pp. 1–25.

[34] Vuong, Q. H., 1989. Likelihood ratio tests for model selection and non-nested hypotheses.
Econometrica 57 (2), 307–333.

[35] Weibull, W., 1951. A statistical distribution of wide applicability. Journal of Applied Mechanics
18, 293–297.

[36] Yu, K., Lu, Z., Stander, J., 2003. Quantile regression: applications and current research areas.
Journal of the Royal Statistical Society: Series D (The Statistician) 52 (3), 331–350.

22