EUROPEAN JOURNAL OF PURE AND APPLIED MATHEMATICS

Vol. 14, No. 1, 2021, 1-12

ISSN 1307-5543 – ejpam.com
Published by New York Business Global
1

2 A Note on the (Weighted) Bivariate Poisson Distribution

3 R. Bidounga1,∗ , P. C. Batsindila Nganga2 , L. Niéré3 , D. Mizère2
1
4 École Normale Supérieure, Université Marien Ngouabi, BP 69, Brazzaville, Congo
2
5 Faculté des Sciences et Techniques, Université Marien Ngouabi, BP 69, Brazzaville, Congo
3
6 Institut Supérieure de Gestion, Université Marien Ngouabi, BP 69, Brazzaville, Congo

Abstract. In the recent statistical literature, the univariate Poisson distribution has been generalized by
many authors, among them: the univariate weighted Poisson distribution [13], the generalized univariate
Poisson distribution [7], the bivariate Poisson distribution according to Holgate [11], the bivariate Poisson
distribution according to Lakshminarayana, Pandit and Srinivasa Rao [15], the bivariate Poisson distribution
according to Berkhout and Plug [4], the bivariate weighted Poisson distribution according to Elion et al. [8]
and the generalized bivariate Poisson distribution according to Famoye [9]. In this paper, We highlight the
weighted bivariate Poisson distribution and show that it is the synthesis of all the bivariate Poisson distri-
butions which, under certain conditions, converge in distribution towards the bivariate Poisson distribution
according to Berkhout and Plug [4] which can be considered like the standard distribution in N2 as is the
univariate Poisson distribution in N.
8 2020 Mathematics Subject Classifications: 62E10, 62E15, 62H10
9 Key Words and Phrases: Generalized Poisson distribution, convergence in distribution, conditional ditri-
10 bution, bivariate generalized Poisson distribution, punctual duality
11

12 1. Introduction

13 The bivariate Poisson distribution was discussed for the first time by Campbell [6] who consid-
14 ered the limit of the distribution of a two-dimensional contingency table. Practically, at the same
15 period Guldberg [10] obtains the bivariate distribution of independent Poisson distributions as the
16 limit of the distribution of independent binomial distributions. The explicit form of the bivariate
17 Poisson distribution is due a few years later to Aitken [1]. We had to wait Holgate [11] to obtain
18 a bivariate Poisson variable from three independent univariate Poisson variables, i.e. with a non-
19 diagonal variance-covariance matrix. A few years later, Kawamura [12] considered the structure
20 of a bivariate Poisson distribution as the limit of a bivariate Bernoulli distribution and found the
21 results of Holgate. We can refer to Morin [17] for a better edification.
∗
Corresponding author.
DOI: https://doi.org/10.29020/nybg.ejpam.v14i1.3895
Email addresses: rufbid@yahoo.fr (R. Bidounga),
prevot.batsindila@gmail.com (P.C. Batsindila Nganga),leonard.niere@umng.cg (L. Niéré),
domizere@gmail.com (D. Mizère).

http://www.ejpam.com 1 c 2021 EJPAM All rights reserved.

 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 2

22 Several authors have studied bivariate Poisson distributions, in particular Berkhout & Plug [4] and
23 Lakshminarayana et al. [15]. Elion et al. [8] through the crossing of two weighted Poisson
24 distributions revealed the bivariate weighted Poisson distribution. Batsindila Nganga et al. [2]
25 showed that the bivariate Poisson distribution according to Holgate converges in distribution to
26 the bivariate Poisson distribution according to Berkhout & Plug [4].
27 In this paper, we highlight the important role played by the bivariate Poisson distribution according
28 to Berkhout & Plug [4] which allows to generate all the bivariate Poisson distributions. We show
29 that the bivariate weighted Poisson distribution evidenced by Elion et al. [8] is a weighted bivari-
30 ate Poisson distribution. We highlight the weighted bivariate Poisson distribution and show that it
31 is the synthesis of all the bivariate Poisson distributions which, under certain conditions, converge
32 in distribution towards the bivariate Poisson distribution according to Berkhout & Plug [4] which
33 can be considered like the standard distribution in N2 as is the univariate Poisson distribution in N.
34 The rest of this paper is organized as follows. In sections 2 and 3, we respectively recall the notion
35 of univariate weighted Poisson distribution and the notion of generalized Poisson distribution.
36 In section 4, we review the bivariate Poisson distributions according to Berkhout & Plug [4],
37 according to Holgate [11], according to Lakshminarayana et al. [15] and the generalized bivariate
38 Poisson distribution according to Famoye [9] then we show that these distributions converge in
39 distribution towards the bivariate Poisson distribution according to Berkhout & Plug [4]. In section
40 5, we construct the weighted bivariate Poisson distribution and show that under certain conditions,
41 this distribution is equal to the bivariate Poisson distribution according to Berkhout & Plug [4].
42 The section 6 presents the conclusion of this paper.

43 2. Univariate weighted Poisson distribution

44 Suppose the realization y of the random variable Y of mass function p (y; δ) is recorded with a
45 probability proportional to ω (y); the record y is the realization of a random variable Y ω called
46 weighted version of Y and which has the probability distribution:
ω (y)
pω (y; δ) = P Y ω = y = p (y; δ) , y ∈ N := {0, 1, · · · } , δ ∈ R∗+
 
(1)
Eδ [ω (Y)]
47 called weighted distribution where ω (y) is called weight function, a positive function and Eδ [ω (Y)]
= y∈N ω (y) p (y; δ) the constant of normalization which is the mean relative to the distribution of
P
48

49 Y depending on δ such that 0 < Eδ [ω (Y)] < +∞. The function ω (y) = ω (y; φ) can depend on a
50 parameter φ which represents the mechanism of saving data. Note that ω (y) = ω (y; δ, φ) can also
51 depend on the canonical parameter δ. The data of a weight function makes it possible to generate
52 a weighted probability distribution [13]. In this case, we can say that this distribution is generated
53 by the weight function.
54 In this paper, the distribution p (y; δ) will be called the basic distribution. When the basic distribu-
55 tion is equal to the univariate Poisson distribution of parameter δ, the Expression (1) is called the
56 univariate weighted Poisson distribution.
57 The univariate weighted Poisson distribution has the following characteristics [3]:
!
ω
d
Eδ Y = δ 1+ ln Eδ [ω (Y)]
dδ
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 3

d2
var Y ω = Eδ (Y) + δ2 ln Eδ [ω (Y)] .


dδ2
58 Example 1. The univariate COM-Poisson distribution [5] with a probability mass function:
λy 1
P (Y = y|λ, ν) = ν , y = 0.1, . . . ; λ > 0, ν ≥ 0,
(y!) Z (λ, ν)

59 is a univariate weighted Poisson distribution of weight function ω (y, ν) = (y!)ν−1 and constant of
normalization: E [ω (Y, ν)] = e−λ Z (λ, ν), with Z (λ, ν) = +∞ ν
n=0 λ / (n!) .
n
P
60

61 3. Generalized Poisson distribution

62 The generalized Poisson distribution [7] of a random variable Y has the mass function:
 δy
 (1 + αy)y−1 e−δ(1+αy) , y∈N
 y!


P (Y = y; δ, α) = 

 (2)
for y > m if α < 0,

0


 
63 with max −δ−1 , −m−1 < α < δ−1 , where m (≥ 4) is the largest positive integer such as 1+αm > 0,
64 when α < 0. This distribution has the following characteristics [7]:

Eδ (Y) = δ (1 − αδ)−1
var (Y) = δ (1 − αδ)−3
Eδ e−Y = eδ(s−1) , with ln (s) − αδ (s − 1) + 1 = 0.
 

65 4. Bivariate Poisson distributions

66 4.1. Bivariate Poisson distribution according to Berkhout and Plug [4]

67 Let Yi (i = 1, 2) a random variable which follows the univariate Poisson distribution with pa-
68 rameter δi (i = 1, 2). The vector (Y1 , Y2 ) follows the bivariate Poisson distribution according to
69 Berkhout and Plug [4] if its mass function denoted fBP is equal to
 y1  y
 δ1 −δ1   δ22 −δ2 

fBP (y1 , y2 ; δ1 , δ2 ) =  e   e  , y1 ∈ N, y2 ∈ N, δ1 ∈ R∗+ , δ2 ∈ R∗+ , (3)
y1 ! y2 !

70 under the conditions

ln δ1 = x0 β1 (4)

71 and

ln δ2 = x0 β2 + ηy1 , (5)
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 4
 
72 where β1 , β2 and η are parameters and x0 = x1 , x2 , . . . , x p the vector of deterministic variables
 y 
73 or factors. The Expression (4) results in P (Y1 = y1 ; δ1 ) = δ11 /y1 ! e−δ1 is a marginal distribution
 y 
74 of Y1 and the Expression (5) means that P (Y2 = y2 ; δ2 ) = P (Y2 = y2 /Y1 = y1 ) = δ22 /y2 ! e−δ2 is a
75 conditional probability.
76 Thus, we have fBP (y1 , y2 ; δ1 , δ2 ) = P (Y1 = y1 ; δ1 ) P (Y2 = y2 /Y1 = y1 ). When η = 0, then the
77 variables Y1 and Y2 are independent. The generalized linear model of Expression (4) has for
78 response variable Y1 and the model of Expression (5) has for response variable Y2 . The resolution
79 of these models makes it possible to highlight, not only the independence between the variables Y1
80 and Y2 but also the effect of the factor x0 on these same variables. The bivariate Poisson distribution
81 according to [4] has the following characteristics [3]:

Eδ1 (Y1 ) = var (Y1 ) = δ1 (6)

x0 β2 +c2 +δ1 (eη −1)
Eδ2 (Y2 ) = e , (7)

82 where c2 is the intercept of the model (5).

η
var (Y2 ) = Eδ2 [Y2 ] + Eδ2 (Y2 ) 2 eδ1 (e −1) − 1
   
(8)
cov (Y1 , Y2 ) = δ1 Eδ2 [Y2 ] eη − 1 .


(9)

83 The expression (8) shows that the variable Y2 is overdispersed. The Expression (9) confirms the
84 fact that the variables Y1 and Y2 are independent if and only if η = 0. And the covariance is
85 negative, zero or positive depending on whether η is negative, zero or positive.

86 4.2. Bivariate Poisson distribution according to Holgate [11]

87 Let be three univariate random variables V1 , V2 and U independent of Poisson with respective
88 parameters λ1 , λ2 and λ3 . With these three variables, we construct two new dependent variables
89 Y1 and Y2 such as:

Y j = V j + U, where j = 1, 2. (10)

90 Then the joint distribution of the couple (Y1 , Y2 ) is written:

1 ,y2 )
min(y
X λ`3 λy11 −` y −`
λ22
P (Y1 = y1 , Y2 = y2 ) = e−λ1 −λ2 −λ3 ; y1 , y2 = 0, 1, 2, . . . (11)
`=0
`! (y1 − `)! (y2 − `)!

91 By setting δ1 = λ1 + λ3 and δ2 = λ2 + λ3 , we have the following result [2]:

 y1  y
 δ1 −δ1   δ22 −δ2 

P (Y1 = y1 , Y2 = y2 ) =  e   e  × b (y1 , y2 ; δ1 , δ2 , λ3 ) (12)
y1 ! y2 !
92 with
!y2 min(y1 ,y2 )
z`
!y1
λ3 λ3 λ3 X
b (y1 , y2 ; δ1 , δ2 , λ3 ) = e 1− 1− (−y1 )[`] (−y2 )[`] (13)
δ1 δ2 `=0
`!
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 5

93 and z = λ3 / (δ1 − λ3 ) (δ2 − λ3 ) , (−y1 )[`] = (−1)` y1 !/ (y1 − `)!. We denote the distribution given
94 by the Expression (11) by fH (y1 , y2 , λ1 , λ2 , λ3 ). The pair of variables (Y1 , Y2 ) has the following
95 characteristics [11]:

Eδi (Yi ) = var (Yi ) = δi , (i = 1, 2) (14)

cov (Y1 , Y2 ) = λ3 . (15)

96 The marginal variable Yi (i = 1, 2) is a univariate Poisson variable with parameter δi (i = 1, 2).

97 The variables Y1 and Y2 are dependent because their covariance is strictly positive.
98 By taking
y
δ11 −δ
e 1 = P Y1 = y1 ,
 
y1 !
99 as the marginal distribution of Y1 and
y
δ22 −δ
e 2 = P Y2 = y2 /Y1 = y1 ,
 
y2 !
100 as the conditional distribution of Y2 when we consider Y1 = y1 , under the constraints (4) and (5),
101 we find:

P Y1 = y1 , Y2 = y2 = P Y1 = y1 P Y2 = y2 /Y1 = y1 ,
     
(16)

P Y1 = y1 , Y2 = y2 = P Y1 = y1 P Y2 = y2 /Y1 = y1 = fBP (y1 , y2 ; δ1 , δ2 )

     

102 and

fH (y1 , y2 ; δ1 , δ2 , λ3 ) = fBP (y1 , y2 ; δ1 , δ2 ) × b (y1 , y2 ; δ1 , δ2 , λ3 ) , (17)

103 which are the results found by Batsindila Nganga et al. [2].

104 By setting λ3 = 1/n with n ∈ N∗ , Batsindila Nganga et al. [2] constructed the family of bivariate
Poisson distributions according to Holgate fH,n /n ∈ N∗ , with fH,n (y1 , y2 ; δ1 , δ2 ) = fH (y1 , y2 ; δ1 , δ2 , 1/n).

105

106 By making n tend to infinity, we have the following results [2]:

lim b (y1 , y2 ; δ1 , δ2 , 1/n) = 1 (18)

n−→+∞

107 and

lim fH,n (y1 , y2 ; δ1 , δ2 ) = fBP (y1 , y2 ; δ1 , δ2 ) . (19)

n−→+∞

108 The bivariate Poisson distribution according to Holgate [11] converges in distribution to the bi-
109 variate Poisson distribution according to Berkhout & Plug [4].
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 6

110 4.3. Bivariate Poisson distribution according to Lakshminarayana et al. [15]

111 Lakshminarayana et al. [15] defined the bivariate Poisson distribution, which is the joint distribu-
112 tion of the pair of random variables (Y1 , Y2 ), as the product of Poisson marginal distributions with
113 a multiplicative factor. The probability mass function of this bivariate Poisson distribution that we
114 denote by fLPS is defined by:
 y1  y
 δ1 −δ1   δ22 −δ2  h

  i
fLPS (y1 , y2 ; δ1 , δ2 , λ) =  e   e  1 + λ e−y1 − e−dδ1 e−y2 − e−dδ2 , (20)
y1 ! y2 !

with y1 , y2 ∈ N, (δ1 , δ2 ) ∈ R∗+ 2 , λ ∈ R∗+ et d = 1 − e−1 .

115

116 This distribution has the characteristics (Lakshminarayana et al., 1999):

Eδi (Yi ) = δi , (i = 1, 2)
cov (Y1 , Y2 ) = δ1 δ2 d2 e−c(δ1 +δ2 ) .
 
117 The marginal variables are Poisson with parameters δi (i = 1, 2) et e−dδi = Eδi eYi (i = 1, 2). We
118 have the following result.

119 Proposition 1. Taking into account Expressions (3), (4) and (5), we have the following expression.

fLPS (y1 , y2 ; δ1 , δ2 , λ) = fBP (y1 , y2 ; δ1 , δ2 ) × ψ (y1 , y2 ; δ1 , δ2 , λ) , (21)

  
120 with ψ (y1 , y2 ; δ1 , δ2 , λ) = 1 + λ e−y1 − e−dδ1 e−y2 − e−dδ2 .

121 Proof. The proof is obvious.

122 Corollary 1. By setting λ = λn , n ∈ N, such that limn−→+∞ λn = 0, we build a family of the bivari-
ate Poisson distributions according to Lakshminarayana et al. [15], fLPS ,n (y1 , y2 ; δ1 , δ2 ) /n ∈ N∗

123

124 such that fLPS ,n (y1 , y2 ; δ1 , δ2 ) = fLPS (y1 , y2 ; δ1 , δ2 , λn ).

125 We have limn−→+∞ ψ (y1 , y2 ; δ1 , δ2 , λn ) = 1 and therefore

lim fLPS ,n (y1 , y2 ; δ1 , δ2 ) = fBP (y1 , y2 ; δ1 , δ2 ) . (22)

n−→+∞

126 The bivariate Poisson distribution according to Lakshminarayana et al. [15] converges in distri-
127 bution to the bivariate Poisson distribution according to Berkhout and Plug [4].

128 We can therefore notice, through Expression (21), that the bivariate Poisson distribution according
129 to Lakshminarayana et al.[15] is the product of the bivariate Poisson distribution according to
130 Berkhout & Plug with a multiplicative factor. The Expression (22) shows that the bivariate Poisson
131 distribution according to Berkhout & Plug is a limit case of the bivariate Poisson distribution
132 according to Lakshminarayana et al.[15]
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 7

133 4.4. Bivariate generalized Poisson distribution

134 Famoye [9] combines the generalized Poisson distribution of Consul & Jain [7] and the bivariate
135 Poison distribution of Lakshminarayana et al. [15] to construct the distribution whose probability
136 mass function is
2  yi
 δi
Y 
P (Y1 = y1 , Y2 = y2 ) =  (1 + αi yi )
−1 −δ (1+α )
 1 + λ e 1 − c1 e 2 − c2 ,
−y
y y  
−y

i
e i i i
(23)
i=1
yi !
 
137 with ci = E e−Yi , yi ∈ N, δi ∈ R∗+ , αi ∈ R, (i = 1, 2).
138 We will denote the distribution given in Expression (23) by fF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ). This dis-
139 tribution has the following characteristics [9]:

Eδi (Yi ) = δi (1 − αi δi )−1 , i = 1, 2

var (Yi ) = δi (1 − αi δi )−3 , i = 1, 2
cov (Y1 , Y2 ) = λ (c11 − c1 δ1 ) (c22 − c2 δ2 ) ,

with cii = Eδi Yi e−Yi = δi (1 − αi θi si )−1 eδi (1+αi )(si −1)−1 where ln (si )−αi θi (si − 1)+1 = 0 (i = 1, 2) .
 
140

141 We have the following result.

142 Proposition 2. Under the conditions (4) and (5), the Expression (23) becomes

fF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) = fBP (y1 , y2 , δ1 , δ2 ) ψF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) , (24)

143 where
 2 
δ
Y  
ψF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) =   (1 + αi yi ) y i −1
e −αi i yi  1 + λ e−y1 − c1
e−y2 − c2
 .
 (25)
i=1

144 Proof. Note that Expression (23) can still be written

 2 yi
Y δ
 2 
(1 + αi yi )yi −1 e−δi αi yi  1 + λ e−y1 − c1 e−y2 − c2
 Y  
P (Y1 = y1 , Y2 = y2 ) =  i −δi  


e  
i=1 i
y! i=1
 y1   y2
 δ1 −δ1   δ2 −δ2  Y
 2 
α
 
=  e   e   (1 + αi yi ) i e i i i  1 + λ e−y1 − c1 e−y2 − c2 .
y −1 −δ y


y1 ! y2 ! i=1

145 Taking into account Expressions (3), (4) and (5), we get

 2 
(1 + αi yi )yi −1 e−δi αi yi  1 + λ e−y1 − c1 e−y2 − c2 .
Y  
P (Y1 = y1 , Y2 = y2 ) = fBP (y1 , y2 ; δ1 , δ2 ) 



i=1

146 By setting
 2 
(1 + αi yi )yi −1 e−αi δi yi  1 + λ e−y1 − c1 e−y2 − c2 ,
Y  
ψF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) = 



(26)
i=1
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 8

147 it follows that

fF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) = fBP (y1 , y2 , δ1 , δ2 ) ψF (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) .

148 The proof is finished.

149 Corollary 2. In Expression (26), let αi = αin , n ∈ N with limn−→+∞ αin = 0 (i = 1, 2) and λ = λn ,
n ∈ N with limn−→+∞ λn = 0. We can then build a family of Famoye distributions fF,n /n ∈ N such

150

151 that fF,n (y1 , y2 , δ1 , δ2 , α1 , α2 , ) = fF (y1 , y2 , δ1 , δ2 , α1n , α2n , λn ).

Like limn−→+∞ ψF (y1 , y2 , δ1 , δ2 , α1n , α2n , λn ) = 1, then

lim fF,n (y1 , y2 , δ1 , δ2 , α1 , α2 , ) = fBP (y1 , y2 , δ1 , δ2 ) ,

n−→+∞

152 the distribution of Famoye [9] converges in distribution towards the bivariate Poisson distribution
153 according to Berkhout and Plug. Expression (24) confirms that the distribution evidenced by
154 Famoye [9] is a bivariate Poisson distribution.

155 5. Weighted bivariate Poisson distribution

156 Definition 1. Consider fBP (y1 , y2 ; δ1 , δ2 ) the basic distribution of the pair of random variables
157 (Y1 , Y2 ). We call the weighted bivariate Poisson distribution, the probability mass function defined
158 by:
ω (y1 , y2 ; δ1 , δ2 , λ)
f ω (y1 , y2 ; δ1 , δ2 , λ) = × fBP (y1 , y2 ; δ1 , δ2 ) , (27)
Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)]

where ω (y1 , y2 ; δ1 , δ2 , λ) is called the weight function, a positive function, and

XX
Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] = ω (y1 , y2 ; δ1 , δ2 , λ) fBP (y1 , y2 ; δ1 , δ2 )
y1 y2

159 the constant of normalization such that 0 < Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] < +∞.
160 Let
ω (y1 , y2 ; δ1 , δ2 , λ)
ψ (y1 , y2 ; δ1 , δ2 , λ) = , (28)
Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)]

161 the normalized weight function ([16], [14]). The expression (28) results in

ω (y1 , y2 ; δ1 , δ2 , λ) = ψ (y1 , y2 ; δ1 , δ2 ) × Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] . (29)

162 From the Expression (29), we can deduce that the constant of normalization Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)]
163 makes it possible to calculate the weight functions and consequently it also generates the weighted
164 bivariate Poisson distribution.
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 9

165 Example 2. suppose that ω (y1 , y2 ; δ1 , δ2 , λ) = ω1 (y1 ) ω2 (y2 ) and Eδ1 δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] =
166 Eδ1 [ω1 (Y1 )] Eδ2 [ω2 (Y2 )] . This last expression does not mean that the random variables Y1 and
167 Y2 are independent. The mass function f ω (y1 , y2 ; δ1 , δ2 ) given in Expression (27) is equal to:
ω1 (y1 ) ω2 (y2 )
f ω (y1 , y2 ; δ1 , δ2 ) = × fBP (y1 , y2 ; δ1 , δ2 ) . (30)
Eδ1 [ω1 (Y1 )] Eδ2 [ω2 (Y2 )]

168 The Expression (30) is the crossing between two univariate weighted Poisson distributions. It is
169 called the bivariate weighted Poisson distribution [8]. Expression (30) shows that the bivariate
170 weighted Poisson distribution is a weighted bivariate Poisson distribution. Its characteristics are
171 ([3]):
Eeη δ1 [ω1 (Y1 )] η −1)
Eδ2 Y2ω2 = e x β2 +c2 +δ1 (e
h i 0

Eδ1 [ω1 (Y1 )]

 
 ω h ω i h  ω i2 
δ η −1) Eδ1 [ω1 (Y1 )] Eδ1 e2η [ω1 (Y1 )]
var Y2 2 = Eδ2 Y2 2 + Eδ2 Y2 2 e 1 (e



2 − 1
Eδ1 eη [ω1 (Y1 )]
 ω ω h ωi d h ω i!
η
cov Y1 , Y2 = Eδ2 Y2 δ1 e + ln Eδ1 e [ω1 (Y1 )] − Eδ1 Y1 1 .
1 2 2


η
dη

172 Proposition 3. If the univariate random variables Y1 and Y2 are punctually dual, then the bi-
173 variate weighted Poisson distribution given by Expression (30) is equal to the bivariate Poisson
174 distribution fBP (y1 , y2 ; δ1 , δ2 ).

175 Proof. If Y1 and Y2 are punctually dual [13], then ω1 (y1 ) ω2 (y2 ) = 1, ∀ (y1 , y2 ) ∈ N2 . So
176 Eδ1 [ω1 (Y1 )] Eδ2 [ω2 (Y2 )] = 1, therefore f ω (y1 , y2 ; δ1 , δ2 ) = fBP (y1 , y2 ; δ1 , δ2 ).

177 Example 3. In Expression (13), let

min(y 1 ,y2 )
λ3 y1 λ3 y2
X z`
eλ3
   
178 ψ (y1 , y2 ; δ1 , δ2 ) = b (y1 , y2 ; δ1 , δ2 , λ3 ) = 1− δ1 1− δ2 (−y1 )[`] (−y2 )[`] .
`=0
`!
From the Expression (28), if we take

Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] = e−λ3 ,

as the constant of normalization, then the weight function is equal to (Cf. Expression (29)):
!y2 min(y 1 ,y2 )
z`
!y1
λ3 λ3 X
ω (y1 , y2 ; µ1 , µ2 , λ) = 1 − 1− (−y1 )[`] (−y2 )[`] .
δ1 δ2 `=0
`!

179 We deduce, from Definition 1, that the bivariate Poisson law according to Holgate [11] is a
180 weighted bivariate Poisson distribution.
 R. Bidounga et al. / Eur. J. Pure Appl. Math, 14 (1) (2021), 1-12 10

Example 4. From Expression (21), we have

  
ψ (y1 , y2 ; µ1 , µ2 , λ) = 1 + λ e−y1 − e−dµ1 e−y2 − e−dµ2 , with d = 1 − e−1 .

If we take
Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] = e−(δ1 −δ2 )λ ,
as the constant of normalization, the weight function is equal to
h   i
ω (y1 , y2 ; µ1 , µ2 , λ) = 1 + λ e−y1 − e−dµ1 e−y2 − e−dµ2 e(δ1 −δ2 )λ .

181 We deduce, from Definition 1, that the bivariate Poisson distribution according to Lakshminarayana
182 and al. [15] is a weighted bivariate Poisson distribution.

183 Example 5. Let Yi (i = 1, 2) be random variables of COM-Poison [5] with parameters (δi , νi ) ,
184 (i = 1, 2). The COM-Poisson distribution is a weighted univariate Poisson distribution (Cf. Ex-
185 pression (1)) with a weight function

ωi (yi , δi ) = (yi )1−νi , (i = 1, 2) , (31)

186 and the constant of normalization

Eδi [ω (Yi )] = e−δi Z (δi , νi ) , (32)

with Z (δi , νi ) = +∞ νi
n=0 δi / (n!) . Taking into account the Expressions (31) and (32), the distribu-
n
P
tion given by the Expression (30), called bivariate COM-Poisson distribution [5], is a weighted
bivariate Poisson distribution of weight function

ω (y1 , y2 , δ1 , δ2 , ν1 , ν2 ) = (y1 )1−ν1 (y2 )1−ν2

and the constant of normalization

Eδ1 ,δ2 ω (y1 , y2 , δ1 , δ2 , ν1 , ν2 ) = Z (δ1 , ν1 ) Z (δ2 , ν2 ) e−δ1 −δ2 .

 

187 Example 6. From Expression (25), we have

 2 
(1 + αi yi )yi −1 e−αi δi yi  1 + λ e−y1 − c1 e−y2 − c2 .
Y  
ψ (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) = 



i=1

If we take
Eδ1 ,δ2 [ω (Y1 , Y2 ; δ1 , δ2 , λ)] = 1,
as the normalization constant and the weight function is equal to

ω (y1 , y2 ; δ1 , δ2 , λ) = ψ (y1 , y2 , δ1 , δ2 , α1 , α2 , λ) .

188 The bivariate generalized Poisson distribution according to Famoye [9] is a weighted bivariate
189 Poisson distribution, that is to say a bivariate Poisson distribution.
 REFERENCES 11

190 6. Conclusion

191 We have reviewed the bivariate Poisson distributions and determined the functional relationships
192 that exist between them. We have highlighted the important role played by the bivariate Poisson
193 distribution according to Berkhout and Plug [4] which allows to generate all the bivariate Pois-
194 son distributions. The bivariate weighted Poisson distribution evidenced by Elion et al. [8] is a
195 weighted bivariate Poisson distribution. The weighted bivariate Poisson distribution that we have
196 defined is the synthesis of all the bivariate Poisson distributions which, under certain conditions,
197 converge in distribution towards the bivariate Poisson distribution according to Berkhout and Plug
198 [4]. This last distribution can be considered as the standard distribution in N2 as is the univariate
199 Poisson distribution in N.

200 Acknowledgements

201 The authors recognize the role of insightful comments and observations from anonymous ref-
202 erees and the editor which greatly improved the clarity of the paper.

203 References

204 [1] A.C. Aitken. Statistical Mathematics, volume 134. Oliver and boyd, eds edition, 1944.
205 [2] P. C. Batsindila-Nganga, R. Bidounga, C.C. Kokonendji, and D. Mizère. Comparison be-
206 tween two bivariate Poisson distributions through the phi-divergence. Afrika Statistika,
207 12(3):1481–1494, 2017.
208 [3] P. C. Batsindila-Nganga, R. Bidounga, and D. Mizère. The covariance structure of the bivari-
209 ate weighted Poisson distribution and application to the Aleurodicus data. Afrika Statistika,
210 14(2):1999–2017, 2019.
211 [4] P. Berkhout and E. Plug. A bivariate Poisson count data model using conditional probabili-
212 ties. Statistica Neerlandica, 58(3):349–364, 2004.
213 [5] R. Bidounga, E. G. Mandangui Maloumbi, R. Mizélé Kitoti, and D. Mizère. The New Bivari-
214 ate Conway-Maxwell-Poisson Distribution Obtained by the Crossing Method. International
215 Journal of Statistics and Probability, 9(6), 2020.
216 [6] J. T. Campbell. The Poisson correlation function. Proc. Edinburgh Math. Soc, 4:18–26,
217 1934.
218 [7] P. C. Consul and G. C. Jain. A generalization of the Poisson distribution. Technometrics,
219 15:791–799, 1973a.
220 [8] L.-L. Elion, M. Koukouatikissa Diafouka, R. Bidounga, C.G. Louzayadio, D. Mizère,
221 R. Makany, and G. Kissita. The bivariate weighted Poisson distribution: Statistical study
222 of the arterial hypertension data according to the weight and blood sugar level. Far East
223 Journal of Theorical Statistics, 52:365–393, 2016.
 REFERENCES 12

224 [9] F. Famoye. A new bivariate generalized Poisson distribution. Statistica Neerlandica,
225 64(6):112–124, 2010.

226 [10] A. Guldberg. On discontinuous frequency functions of 2 variables. Skand. Aktuar., 17(6):89–
227 117, 1934.

228 [11] P. Holgate. Estimation for the bivariate Poisson distribution. Biometrika, 51:241–245, 1964.

229 [12] K. Kawamura. The structure of bivariate Poisson distribution. Kodai Math. Sem. J., 25:246–
230 256, 1973.

231 [13] C. C. Kokonendji, D. Mizère, and N. Balakrishnan. Connections of the Poisson weight
232 function to overdispersion and underdispersion. J. Statist. Plann. Inférence., 138:1287–1296,
233 2008.

234 [14] C. C. Kokonendji, A. Y. Touré, and R. Abid. On General Exponential Weight Functions and
235 Variation Phenomenon. Sankhya A, pages 1–17, 2020.

236 [15] J. Lakshminarayana, S.N.N. Pandit, and K. Srinivasa Rao. On a bivariate poisson distribu-
237 tion. Communications in Statistics Theory and Methods, 28(2):267–276, 1999.

238 [16] D. Mizère, R.A. Makany G. Kissita, and N. Yumba. Determination of the mean dual dis-
239 tribution through the Poisson weight function. Far East Journal of Theorical Statistics,
240 29(1):53–63, 2009.

241 [17] A. Morin. Loi de Poisson multivariée. Journal de la société statistique de Paris, 134(2):3–13,
242 1993.