An Extension to the Zero-Inflated Generalized Power Series Distributions

International Journal of Mathematics and Statistics Invention (IJMSI)
E-ISSN: 2321 – 4767 P-ISSN: 2321 - 4759
www.ijmsi.org Volume 4 Issue 9 || November. 2016 || PP-45-48
www.ijmsi.org 45 | Page
An Extension to the Zero-Inflated Generalized Power Series
Distributions
Rafid Saeed Abdulrazak Alshkaki
Ahmed Bin Mohammed Military College, Doha, Qatar
ABSTRACT: In many sampling involving non negative integer data models, the number of zeros is observed
to be significantly higher than the expected number in the assumed model. Such models are called zero-inflated
models. These models are recently cited in literature in various fields of science including; engineering, natural,
social and political sciences. The class of zero-inflated power series distributions was recently considered and
studied. Members of the class of generalized power series distributions are some of the well-known discrete
distributions such as; the Poisson, binomial, negative binomials, as well as most of their modified forms. In this
paper an extension to class of zero-inflated power series distributions was introduced, namely, the zero-one
inflated case, and its structural properties were studied. In particular, its moments and some of its moment’s
recurrence relations were obtained along with some of its generating functions. Some of these results were
shown in term of the zero-inflated cases as well.
Keywords: Zero and Zero-One Inflated Models, Power Series Distributions, non-negative integer sampling,
Moments.
I. INTRODUCTION
In many sampling involving non-negative integer data models, the number of zeros is observed to be
significantly higher than the predicated number by the standard assumed distribution. Hence, as a result of mis-
specifying the proper statistical distribution, leads to the so called the zero-inflated distributions. These models
are recently cited in literature in various fields of science including; engineering, natural, social and political
sciences.
The class of zero-inflated power series distributions (ZIPSDs) was recently considered and studied.
Well-known discrete distributions such as, the Poisson, binomial, negative binomial and the logarithmic, are
members of this class.
Gupta et al [1] studied some structural properties and the distribution of the sum of independent the
ZIPSD, as well as estimation of its parameters. Ridout et al [2] considered the problem of modelling count data
with excess zeros, called them zero-modified distributions. Kolev et al [3] proposed the ZIPSDs for modeling of
correlated count data which exhibit overdispersion. Patil and Shirke [4] provided three asymptotic test for
testing the parameters of the ZIPSDs. Murat and Szynal [5] studied some structural properties and estimation of
the parameters of the ZIPSDs. Bhattacharya et al [6] considered a Bayesian test for testing the ZIPSD as a
mixture of a power series distribution (PSD) and a degenerate distribution at zero. Nanjundan and Naika [7]
considered estimation the parameters of ZIPSDs. Edwin [8] studied the construction and estimation of the
ZIPSDs, and considered some its applications.
In this paper, we introduce in Section 2, the definition of the class of discrete PSDs with some of its
members, properties; its mean, variance, probability generating function (PGF). Then, in Section 3, we
introduce the class of ZIPSD, its definition with some of its properties including the mean, variance, and the
PGF. In Section 4 we introduce the definition of the zero-one inflated power series distribution (ZOIPSD), and
it’s mean and variance, PGF, moment generating function(MGF) and the factorial moment generating(FMGF),
and moment recurrences relations, were given in Section 5. Some of these results are shown in term of the
standard and the zero-inflated cases as well.
II. POWERSERIES DISTRIBUTIONS
Let g θ = axx∈T θx
be a power series for θϵΩ = θ; 0 < 𝜃 < 𝜔 , where ω is the radius of convergence of
g θ , and ax ≥ 0 for all x ∈ T ⊆ I = {0,1,2, … }. Then the discrete random variable (rv) X having probability
mass function (pmf);
P X = x =
ax θx
g θ
, x ∈ T
0 otherwise,
(1)
is said to have a PSD with range T and parameter space Ω. The function and g θ is called the series defining
function (sdf). We will denote that by writing X ∼ PSD(θ, g θ ).

An Extension to the Zero-Inflated Generalized Power Series Distributions
The discrete distributions; Poisson, binomial, negative binomial, and the logarithmic series distributions, are
well-known member of the PSD’s with sdf’s; eθ
, (1 + θ)n
, (1 − θ)−k
, and − ln 1 − θ , respectively. See
Abdulrazak and Patil [9], [10] and Johnson et al [11] for further details.
The mean of the rv X having PSD(θ, g θ ) can be easily seen to be:
E X = θ
g′(θ)
g(θ)
,
and its variance is given by;
Var X = θ2 g′′(θ)
g(θ)
+ θ
g′(θ)
g(θ)
− θ
g′(θ)
g(θ)
2
Similarly, the PGF of the rv X, GX t , is given by;
GX t = E tX
=
g(θt)
g(θ)
III. ZERO-INFLATED POWER SERIES DISTRIBUTIONS
Let the rv X ∼ PSD(θ, g θ ).as given in (1) with T = 0, 1, 2, … , and let α ∈ 0,1 be an extra proportion
added to the proportion of zero of the rv X, then the rv Y defined by;
P Y = y =
α + 1 − α
a0
g θ
, y = 0
1 − α
ay θy
g θ
y = 1, 2, 3, …
0 otherwise,
(2)
is said to have a ZIPSD, and we will denote that by writing Y ∼ ZIPSD(θ, g θ ; α). The mean of Y is given by;
E Y = 1 − α θ
g′
θ
g θ
= 1 − α E X
and its variance of is given by;
Var Y = 1 − α θ θ
g′′(θ)
g(θ)
+
g′(θ)
g(θ)
− 1 − α θ
g′(θ)
g(θ)
2
= 1 − α Var(X) + α 1 − α θ2 g′(θ)
g(θ)
2
= 1 − α Var X + α (E X )2
The PGF of Y is given by;
GY t = α + 1 − α
g θt
g θ
IV. ZERO-ONE INFLATED POWER SERIES DISTRIBUTIONS
The following table represents the fetal movement data set that was collected, Leroux and Puterman
[12], in a study of breathing and body movements in fetal lambs designed to examine the possible change in the
amount of pattern of fetal activity during the last two thirds of the gestation period.
Number of movements 0 1 2 3 4 5 6 7
Number of intervals 182 41 12 2 2 0 0 1
Edwin [8] considered thatthe number of fetal movement with zero was inflated and studied this case as a ZIPSD
model.
If we look at the above frequencies, we notice that the number of fetal movement with ones is also inflated. This
suggests an extension to the zero-one inflated models.
Let the rv X ∼ PSD(θ, g θ ) as given in (1) with T = 0, 1, 2, … , and let α ∈ 0,1 be an extra proportion
added to the proportion of zero of the rv X, β ∈ 0,1 be an extra proportion added to the proportion of ones of
the rv X, such that 0 < 𝛼 + 𝛽 < 1, then the rv Z defined by;

P Z = z =
α + 1 − α − β
a0
g θ
, z = 0
β + 1 − α − β
a1 θ
g θ
, z = 1
1 − α − β
az θz
g θ
, z = 2, 3, 4, …
0 otherwise,
(3)
and is zero otherwise, is said to have a ZOIPSD with sdf g θ , and inflated parameters α and β. We will denoted
that by writingZ ∼ ZOIPSD(θ, g θ ; α, β).
Note that, if β → 0, then (3) reduces to the form of the ZIPSD. Similarly, the case with α → 0 and β → 0,
reduces to the standard case of PSD.
Although, it does not fit the nature of the supposed model, the inflation parameters α and β may also take
negative values providing that α ∈ max −1, −(1 − β)
p0
1−p0
, 0 and β ∈ max −1, −(1 − α)
p1
1−p1
, 0
without violating that (3) is a pmf. This situation represents the excluding proportion of zero’s and one’s,
respectively, from the standard model given by (1).
V. MOMENTS AND SOME STRUCTURAL PROPERTIES
Let the rvs X ∼ PSD θ, g θ , Y ∼ ZIPSD θ, g θ ; α and Z ∼ ZOIPSD(θ, g θ ; α, β), then the mean of the rv Z is
given by;
E Z = β + 1 − α − β θ
g′ θ
g θ
(4)
Note that;
E Z = E Y + β 1 − θ
g′
θ
g θ
E Z = E Y + β 1 − E(X)
Similarly,
E Z(Z − 1) = 1 − α − β θ2 g′′(θ)
g(θ)
Hence, the variance Z is given by;
Var(Z) = β 1 − β + (1 − α − β)θ θ
g′′(θ)
g(θ)
+ 1 − 2β
g′
θ
g θ
− (1 − α − β)θ
g′(θ)
g(θ)
2
The PGF of Z is given by;
GZ s = E sZ
= α + βs + 1 − α − β
g θs
g θ
= GY s + β(s − 1)
g θs
g θ
The MGF of the rv Z is given by;
MZ t = E etZ
= α + βet
+ 1 − α − β
g θet
g θ
= MY t + β et
−
g θet
g θ
Similarly, the FMGF of Z can be shown to be;
M Z t = E 1 + t Z
= α + β(1 + t) + 1 − α − β
g θ + θt
g θ
= MY t + β 1 + t −
g θ + θt
g θ
Let µr
′
be the rth moment of the rv Z then for r=1, 2, 3, …;
µr
′
= E Zr
= β + 1 − α − β zr
∞
z=1
azθz
g(θ)
Therefore,
µr
′
− β = 1 − α − β zr∞
z=1
az θz
g θ
(5)

It follows that;
d
dθ
μr
′
= 1 − α − β zr
az
d
dθ
θz
g θ −1
∞
z=1
=
1
θ
zr+1
(1 − α − β)
∞
z=1
azθz
g(θ)
−
g′(θ)
g(θ)
zr
(1 − α − β)
∞
z=1
azθz
g(θ)
And using (5), we have that;
d
dθ
μr
′
=
1
θ
μr+1
′
− β −
g′ θ
g θ
μr
′
− β (6)
It follows, with the use of (4), that (6) can be written as;
μr+1
′
= β + θ
d
dθ
μr
′
+
μ1
′
− β
1 − α − β
μr
′
− β
VI. CONCLUSIONS
We introduced the zero-one inflated generalized power series distributions as an extension to class of
zero-inflated generalized power series distributions. Its structural properties were studied, namely, its mean,
variance, probability generating function, moment generating function, and the factorial moment generating
functions. These results were shown in term of the zero-inflated cases as well. Its moment’s recurrence relations
were obtained also. Further studies for this class of distributions need to done, namely, estimation and testing of
its parameters.
REFERENCES
[1]. Gupta, P. L., Gupta, R. C. and Tripathi, R. C. (1995). Inflated modified power series distributions with applications.
Communications in Statistics - Theory and Methods, 24(9), 2355-2374.
[2]. Ridout, M., Dem´etrio, C. G. B. and Hinde, J. (1998). Models for count data with many zeros. International Biometric Conference,
Cape Town, December 1998.
[3]. Kolev, N., Minkova, L.D. and Neytchev, P. (2000). Inflated-Parameter Family of Generalized Power Series Distributions and Their
Application in Analysis of Overdispersed Insurance Data. ARCH Research Clearing House, 2, 295-320.
[4]. Patil, M. K. and Shirke, D. T. (2007). Testing parameter of the power series distribution of a zero inflated power series model.
Statistical Methodology, 4(4), 393-406.
[5]. Murat, M. and Szynal, D. (2007). Non–zero inflated modified power series distributions. Communications in Statistics - Theory and
Methods, 27(12), 3047-3064.
[6]. Bhattacharya, A., Clarke, B. S., and Datta, G. S. (2008). A Bayesian Test For Excess Zeros in a Zero-Inflated Power Series
Distribution. Beyond Parametrics in Interdisciplinary Research: Festschrift in Honor of Professor Pranab K. Sen,, Institute of
Mathematical Statistics, Beachwood, Ohio, USA, 89-104
[7]. Nanjundan, G. and Naika, T. R. (2013). Estimation of Parameters in a Zero-inflated Power Series Model. International Journal of
Statistika and Mathematika, 6(3), 109-114.
[8]. Edwin, T. K. (2014) Power series distribution sand zero-inflated models. Ph.D. Thesis, University of Nairobi.
[9]. Abdul-Razak, R. S. and Patil, G. P. ( 1986). Power series distributions and their conjugates in stochastic modeling and Bayesian
inference. Communication in Statistics, Part l - Theory and Methods, 15(3), 623-641.
[10]. Abdul-Razak, R. S. and Patil, G. P. (1994). Some stochastic characteristics of the power series distributions. Pakistan Journal of
Statistics, 10(1), 189-203.
[11]. Johnson, N. L., Kemp, A. W. and Kotz, S. (2005). Univariate Discrete Distributions, Third Edition, John Wiley and Sons; New
Jersey.
[12]. Leroux, B. G., and Puterman,M. L. (1992). Maximum-Penalized-Likelihood Estimation For Independent And Markov-Dependent
Mixture Models. Biometrics, 545-558.

An Extension to the Zero-Inflated Generalized Power Series Distributions

More Related Content

What's hot (20)

Viewers also liked (20)

Similar to An Extension to the Zero-Inflated Generalized Power Series Distributions (20)

Recently uploaded (20)

An Extension to the Zero-Inflated Generalized Power Series Distributions