International Journal of Advanced Research and Publications

ISSN: 2456-9992

The Beta-Halfnormal Distribution And Its

Properties
AKOMOLAFE.A.A, MARADESA.A
Lecturer, Department of Statistics, Federal University of Technology, P.M.B 704, Akure, Nigeria. +2348039137435,
akomolafe01@yahoo.com

Research student, Department of Statistics, Federal University of Technology P.M.B 704, Akure, Nigeria,+2347034361609,
maradprime1@gmail.com

ABSTRACT: In this research, we consider certain results characterizing the generalization of Beta and Half Normal distribution through
their distribution functions and asymptotic properties. The resulting Beta-HalfNormal Distribution [BHND] was defined and some of its
properties like moment generating function, survival rate function, hazard rate function and cumulative distribution function were
investigated. The distribution was found to generalize some known distributions thereby providing a great flexibility in modeling symmetric
heavy tailed, skewed and bimodal distributions.

Keywords: Beta-HalfNormal distribution, Moment generating function, Hazard rate, Survival rate, Assymptotic properties,

1 Introduction √ √
=E(x) = ∫ = ∫ =
√ √
In recent times, researches have used mixed distribution as √ √ √ √
an alternative to some distributions, the half-normal ∫ = ⌊ ⌋ =
√ √ √ √
distribution is a special case of the folded normal and
truncated normal distributions. It was used to model Therefore the mean of half-normal distribution is represented
brownian movement and can also be used in the modeling by (2) below;
measurement data and lifetime data. Advances in distribution
theory indicate that mixed distribution has received √
appreciable consideration in recent years. Many compound E(x) = (2)
√
distributions have been derived based on this approach. Let
X~N(0,σ2), then Y = |X| follows half normal distribution. E(x2)=∫
The Half-normal is a fold at the mean of an ordinary normal √ √
distribution with mean zero, where σ is the scale parameter. ∫ ∫
√ √
∫
The beta distribution is one of the skewed distributions used 2 √ √
in describing uncertainty or random variation in a system. It E(x )= ∫ ∫ =
√ √
can be used to rescale and shift to create mixture of √
∫
distributions with range of shapes. The beta and half-normal √

distribution can be compounded to form the mixture of beta

√
and half-normal distribution which is our proposed ∫ (3)
√
distribution (BHND) .

1.1 Methods :The beta family of distribution became Since y = , then x2=2yσ ; x = √ σ = √ .
popular some years back and several works have been done
regarding beta compounding with other distribution which Since we have established that x = √ , substitute for x =
include beta- √ in (3).
normal(Eugene&Famoye,2002)[2];betaGumbel(Nadarajah&
Kotz,2004)[3],betaWeibull(Famoye,Lee&Olugbenga,2005)[ √ √
4], Beta-exponential (Nadarajah&Kotz, 2006) [5]: bata- = ∫ √ = ∫ √ √
√ √
Rayleigh (Akinsete&Lowe, 2009) [6]:beta-Laplace
(KozubowskiNadarajah, 2008)[7]: beta-pareto (Akinsete, ∫ √
√
Faye & Lee, 200)[8]; beta-Gamma, beta-f,beta-t, beta-beta,
beta-modified weibull, beta-nakagami among others. Let Recall that erf( ) = 1, erf(0) = 0
X~N(0,σ2), then Y = |X| follows half normal distribution, if = √ erf(√ ) - ) = ( √ )
its probability density function is defined as follows: √ √

√
√ )=
√
f(x) = x> 0 (1)
√ Then E(x2) =
Variance of Half-normal distribution can be obtained as (4).
√ √ √
E(x) = ∫ = ∫ ; Let y= ; Var(x) = E(x2) – (E(x))2 =
√ √ √
; dx = (4)

Volume 1 Issue 4, Oct 2017 17

www.ijarp.org
 International Journal of Advanced Research and Publications
ISSN: 2456-9992

2 The proposed Beta-halfnormal Distribution 2.1 Cumulative Distribution Function (BHND)

(BHND) FBHND(x)=
Now by using the logit of beta defined by Jones, the mixture √
∫ [ ( )] [ ( )]
of Beta-Halfnormal distribution can be obtained. Now let X √ √ √

be a random variable from the distribution with parameters = ∫ ; by following the approach of
and defined by (1) using the logit of Beta defined by Jones incomplete beta function, which states B(x;α,β) =
as:
∫ α β
. from this expression, we can conclude

g(x)= [ ] [ ] (5) that ∫ = IZ(a, b) , where

IZ(a, b) is the regularized incomplete beta function.
since x~ HN(0, ), the we need to obtain the cumulative
density of Halfnormal distribution, and it can be obtained as FBHND(z) = IZ(a, b) = (11)
follows:
√ √ The incomplete beta function B(z; a, b) can be defined in
FHND(x)= ∫ ∫
√ √ term of hypergeometric function below:
√ √
= [ ]= [ ]
√ √ B(z; a, b) = (12)

The (12) can also be given by the power series below:

√ √
= [ ] [ ] (6)
√ √ B(z; a, b) = ∑ (13)

The cumulative density can also be obtained in term of error The power series represented by the (13) can be expanded as
function as follow: follow
B(z;a, b) = * +,
√
F(x)=∫ (7) then
√
FBHND(z)=
* + (14)
By using the change of variables and let y= ; ;
√
dx = ; which can be dx = √ by cross FBHND(z) represent the cdf of Beta-Halfnormal
√
multiplication. Now substitute for y and dx in (7) Distribution(BHND) as defined in (14).

√ 2.2 The Assymtotic properties

F(y; ) = ∫ √ ∫
√ √
[ ( )] [
( )– . Therefore, the F(y; ) in √
√ √
term of error function is displayed below √
( )]
√ √
FHND(y; )= ( ) or F(x; )= ( ) (8) = [ ( )] [ ( )]
√
√ √
√ √ √
√
The pdf of Beta-halfnormal can thereby be obtained by = [ ( )] [ ( )]
√ √ √
applying the logit of beta as defined by Jones as expressed in √
(5). = [ ( )] [ ( )] ;
√ √ √
erf( ) = 1, erf(0) = 0
fBHND(x;a ,b,σ) = [ ] [ ] = [ ] [ ]
fBHND(x;a,b,σ)= Limit as x-> 0
√
[ ( )] [ ( )] (9) [ ( )] [
√ √ √
√
The pdf of Beta-halfnormal distribution is represented by (9) √
( )]
above √ √
√ √
Since , then the pdf of = [ ( )] [ ( )]
√ √ √ √
BHND can be written in it short form when we √
= [ ( )] [ ( )]
√ √ √
√
fBHND(x) = (10) = [ ( )] [ ( )] ;
√ √ √
erf( ) = 1, erf(0) = 0
√
= [ ] [ ]
√

Volume 1 Issue 4, Oct 2017 18

www.ijarp.org
 International Journal of Advanced Research and Publications
ISSN: 2456-9992
= 2.4 Moment Generating Function
[ Hosking (1990) describe that when a random variable
following a generalized beta generated distribution, such that
√ √
]
√
= [ ]
√
X~GBG(f,a,b,c) , then = E(F-1U1/c)r , where U~B(a,b) , c=
-1
√ constant and F (x) is the inverse of CDF of the Halfnormal
= distribution, since BHND (a.b,σ) is a special form when c =
√
1. Then the mgf of the proposed distribution m(t) and the
If as x → 0, the pdf (fBHND(x)) also tend to zero and as x→∞ , general rth moment of the beta generated distribution is
the pdf (fBHND(x)) also tend to zero, then it is an indication defined by
that there exist at least on model in pdf of Beta half-normal
distribution
∫ [ ] [ ]
2.3 Hazard Rate Function
The hazard rate is of Beta halfnormal distribution is Cordeiro and de Castro discussed moment generating
expressed as function of for generated beta distribution as :

h(x) = , M(t)=
∑
where is the survival rate, and is defined as ∫ [ ]
S(x), then the survival rate is expressed as:
( )= Theerefore,M(t)=
h(x)= ∑ ( )∫ [ ( )]
√
√ √
=
The moment generating function for Beta Halfnormal
Distribution is obtained as shown below
where Z = erf ( ), then M(t)=
√
√
∑ ( )∫ [ ( )]
√ √
√ (15)
( ( )) ( ( )) Let a=b=i=1, M(t) become the mgf of parent distribution
√ √ √
( ( ) ) 2.5 Estimation of Parameter
√
√
[ ( )] [ ( )] 2.5.1 Log-Likelihood Function
√ √ √ Cordeiroet. al [2011] gave the log-likelihood function for
( ( ) ) as
√
[ ] [ ] L( ) = [ ] ∑
( ( ) ) ∑ ∑
√
The Generalized Beta distribution reduces to the class of
( ( ) ) Beta generated distribution when c = 1. We hereby have
√
And,
L( ) = [ ] ∑
∑ ∑
√
( ( )) ( ( )) √
√ √ √ = [ ] + ∑ [ ]
√
( ( ) ) ∑ * ( )+ ∑ [ ( )]
√ √ √
√ = ∑ * ( )+ (16)
√
√
= ∑ [ ( )] (17)
√
( ( ) )
√
√
√
√ ∑ [ ] ∑ * ( )+
√ √
( ( ) ) ∑ [ ( )] (18)
√ √
Since * ( )+ can be obtained as follows
√
( ( ) )
√

Volume 1 Issue 4, Oct 2017 19

www.ijarp.org
 International Journal of Advanced Research and Publications
ISSN: 2456-9992
When erf (x) is expressed in term of confluence Substitute (21 and 22) in (18)
hypergeometric function of the first kind √
∑ [ ]
√
= ( )=
√ √
∑ [ ( )]
Therefore √
√
∑ [ ( )]
√
erf(x)= ( ) ( )
, it √
can therefore be expressed by the maclaurin series in (19). =∑ * + ∑
√ ( )
√

= ∑ (19) √
√ ∑ (23)
( )
√

At x = 0,
√ The equation 16, 17 and 23 can be solved using Newton
At x= + Raphson Algorithm to obtain the estimate of ̂ ̂ and ̂σ̂ , the
√
MLE of (a,b, ) .To test the necessary hypothesis about the
significance of the model parameters and the construction of
confidence interval, we need to obtain the second derivative
By following (19) , * ( )+ can be view as of equation (16,17 and 23) with respect to the parameters of
√
σϵ R ; (x < 0 and σ <0 and ( ) ) or (x > 0 and σ >0 interest with a view to constructing Fisher’s information
√ matrix.
and ( ) )
√
2.5.2 Maximum A Posteriori Estimation (MAPE)
By using (19), the series expansion as at σ = 0 represented by When we consider as a random variable, the bayes convert
(20). our belief about the parameter of Halfnormal distribution
( ) (before seeing data) into posterior probability, P( | ) , by
⌊ ⌋√
=[ ( √ ) ] (20) using the likelihood function P(X| ) .The maximum a-
posteriori (MAP) estimate is defined as:

By using (19), the series expansion as at σ = ∞ represented = = argmaxσ

by (21).
The MAP gives the researcher the flexibility of injecting his
√ prior belief about the parameter estimate into the new
= [ ] (21)
estimate.
= ∏
=argmax log∏
√ =argmax∑
Therefore * ( )+ (22)
√ ( )
√
√
Pr(σ|X=x) = and Pr(σ) =
√
Also, * ( )+ can also be obtained as follows L = log [∏ ]
√
Real root =∑
We can solve ;
Domain ( ( ) ) . the root for the ∑
√ logL = √
variable x;
Series at
( ) ∑
√ ( ) ⌊ ⌋√ = + = 0 (24)
[ ]

= = 0 (25)
Series expansion at
√
= , = = 0 (26)
√

√ By solving equation (24,25 and 26) the estimate of (a, b and

therefore * ( )+ (22)
√ ( ) ) can be obtained , and significant for the parameter of the
√
model can be obtained via log likelihood , AIC or BIC.

Volume 1 Issue 4, Oct 2017 20

www.ijarp.org
 International Journal of Advanced Research and Publications
ISSN: 2456-9992

∑ [8] G.M. Cordeiro, C. Alexandra, J.M. Ortega, M.M.

Let
Edwin, Sarabia (2011) Generated distribution ICMA
∑ centre. Discus Disc ion paper in Finance 1-29.

From the Half-normal [9] N. Eugene, C. Lee, and F. Famoye, (2002): Beta-normal
∑ dist and its applications. Communications in Statistic:
distribution as ̂ √ Theory Theory and methods, 31(4), 497-512.

The equation derived above can also be solved using [10] F. Famoye, C. Lee., and O. Olumolade,(2005). The beta-
iteration method [Newton Rapson] to obtain parameter Beta-Weibull distribution. Journal of Statistical Theory,
estimates using maximum likelihood estimation technique. 4,121–136
Taking the second derivatives of the equations 24,25, 26
with respect to the parameter, it is possible to derive the [11] F. Famoye,C. Lee, and N.Eugene(2004). Betanormal
interval estimate and hypothesis tests on the model normal distribution: bimodality properties and
parameters. This will be shown in our further research. application. Journal of Modern Applied Statistical
Methods, 3, 85-103.
3 Conclusions
By compounding two or more probability distributions, we [12] M.C. Jones,(2004). Families of distribution arinsing of o
get the corresponding hybrid distribution with increased from distribution of order Statistics test 13:1-43
number of parameters which is believed to give the newly
compounded distribution more flexibility, consistency, [13] O.I. Shittu, A.K. Adepoju , (2013). On the Beta-Nakaga
stability, sufficiency uniqueness and wider applicability as Nakagami Distribution: Progress in Applied Mathe-
compare to its parent distribution. Therefore, the hybrid Mathe matics 5:49-5
distribution Beta-halfnormal Mixture distribution is said to
have vast applicability extending beyond modeling simple Author’s Profile
Brownian movement but can be used to model statistical
behavior of stochastic processes such as the growth of First Author
tumors’ cells in an oncology’s study of benign
transmogrifying to malignant tumor, studying the Akomolafe Abayomi Ayodele
consumers’ buying behavior and in complex epidemiological received a Bsc in Statistics
studies because of its increased number of parameters which (University of Ilorin, Nigeria),
give the hybrid distribution more flexibility to model many Msc and Phd in Statistics
stochastic phenomena. (University of Ibadan, Nigeria)
in 2004 and 2011 respectively.
References Since then he has been a
[1] A.Alzaatreh, C.Lee and F. Famoye,(2013).A new university lecturer that
method for generating families of continuous specializes in statistical
distributions. Metron, 71(1), 63–79. inference, Stochastic process and applied sample survey.
Having acquired more than 15years of university teaching
[2] A.Akinsete, F.Famoye, and C. Lee,(2008). The beta- experience, he is presently a senior lecturer at the Federal
pareto distribution, Statistics 42(6), 547-563. University of Technology Akure (FUTA), Nigeria.

[3] N.I. Badmus, Ikegwu,M. Emmanuel,(2013). The Beta-

Weighted Weibull Distribution: Some properties and
application to Bladder Cancer Data, Applied and
Computational Mathematics 2:5.
[4] W. Barreto-Souza, G.M Cordeiro and A.B. Simas
(2011). (2011). Some results for beta Frechet
distribution. Communications in Stat Statistics Theory
&Methods, 40, 798–811.
[5] I.W. Burr,(1942). Cumulative frequency functions.
Annals of Mathematical Statistics, 13, 215-232.
[6] G.M. Cordeiro and M. de Castro,(2011). A new family
of generalized distributions. Journal of Statistical Com
Compu putation and Simulation 81(7), 883-898.
[7] G.M.Cordeiro, A.B. Simas, and B.D. Stosic,(2011).
Closed Closed form expressions for moments of the beta
Weibull d. Distribution,Annals of the Brazilian
Academy of Sciences, 8 83,357-373.
Volume 1 Issue 4, Oct 2017
www.ijarp.org
Second Author
Maradesa, Adeleke received
Bachelor of Technology (Btech)
in Statistics (Federal University
of Technology Akure, Nigeria)
and National diploma in Statistics
(Federal School of Statistics
Ibadan, Nigeria) in 2016 and
2012 respectively. He received
adequate training in General Household Survey (GHS),
Multiple Indicator Cluster Survey, Farm Survey and Retail
price at National Bureau of Statistics, Nigeria (NBS)
between 2010 and 2012. He is currently a research student at
Federal university of technology Akure Nigeria. His areas of
research interest are: Mathematical Statistics, Stochastic
Process, and Probability Distribution Theory (Hybrid
Distribution) with application to epidemiological studies,
genetic modeling and microarray.
21