See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/383040033

Modified-Bivariate-PoissonLindley-Model-Properties-and-Applications-in-
Soccer

Article in International Journal of Computer Science in Sport · August 2024

CITATIONS READS

0 67

3 authors, including:

Halim Zeghdoudi Vinoth Raman

Badji Mokhtar - Annaba University Imam Abdulrahman Bin Faisal University
145 PUBLICATIONS 790 CITATIONS 90 PUBLICATIONS 225 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Vinoth Raman on 12 August 2024.

The user has requested enhancement of the downloaded file.

International Journal of Computer Science in Sport
Volume 23, Issue2, 2024
Journal homepage: http://iacss.org/index.php?id=30

DOI: 10.2478/ijcss-2023-0009

✁✂✄☎✄✆✂ ✝✄✞✟✠✄✟✡✆ ☛✁✄☞☞✁✌✍✎✄✌✂✏✆✑ ✁✂✆✏✒

☛✠✁✓✆✠✡✄✆☞ ✟✌✂ ✔✓✓✏✄✕✟✡✄✁✌☞ ✄✌ ✖✁✕✕✆✠
Allaeddine Haddari1, 2, Halim Zeghdoudi2 and Raman Vinoth3
1
Mathematics Department, Faculty of Mathematics and Computer Science,
Batna, University, Algeria, [email protected]
2
Laboratory of Probability and Statistics (LaPS), BadjiMokhtar-Annaba University, P. O.
Box 12, 23000 Annaba, Algeria, [email protected]
3
Quality Measurement and Evaluation, Department, Deanship of Quality and
Academic Accreditation, Imam Abdulrahman Bin Faisal University, P. O. Box
1982, Dammam 31441, Saudi Arabia, [email protected]

Abstract
This paper presents the bivariate Poisson-new XLindley distribution (BPNXLD),
which may be used to represent dependent and over-dispersed countdata. Among
the characteristics considered are the correlation coefficient, mean, and variance of
the distribution. Acomparison with several Bivariate distributions is included. The
goodness of fit of this novel model iscompared with the bivariate Poisson, bivariate
negative binomial and bivariate Poisson-Lindley distributions using two data sets
from a German Bundesliga season.

KEYWORDS: POISSON-NEW XLINDLEY DISTRIBUTION, BIVARIATE POISSON-

LINDLEY DISTRIBUTION, ESTIMATION, SOCCER DATA SET .
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Introduction
One of the most popular sports in the world is soccer. Both experts and novices love watching
the games and making predictions about their results. Many studies have been conducted on
numerous football modeling topics. Bradley and Terry (1952) used a variety of Bradley-Terry
model extensions to simulate the odds of winning, drawing, or losing. On the other side, Reep
and Benjamin (1968) model the number of passes a team successfully completes up to a change
of possession, either by a shot on goal, being tackled, or an infraction, using a negative binomial
distribution.
Another strategy uses a hierarchical Poisson log-linear model to directly model the number of
goals scored by each team in each game. This method was inspired by the modeling
frameworks of Maher (1982), Dixon and Coles (1997), and Baio and Blangiardo (2010). In
addition, Karlis and Ntzoufras (2003) looked at the bivariate Poisson distribution, which is
rarely used since it is difficult to apply, as a way to model scores in a few sports, namely
football and water polo. Using the help of the bivariate Poisson distribution, a parameter may
be used to model the relationship between two random variables, $X$ and $Y$, which
represent the home and away goals in a football game. In recent years, Tsokos et al. (2019)
conducted acomparison research to assess how well a hierarchical Poisson log-linear model
and several extensions of the Bradley-Terry model performed in predicting soccer match
results (win, draw, or loss).
Many publications analyze data, forecasts, and models related to soccer using probability
distribution and other mathematical techniques. Interesting works concerning soccer data and
modeling can be found in Sadeghkhani and Ahmed (2020), Wheatcroft (2021), Owen (2011),
Marek et al. (2014), Karlis and Ntzoufras (2009), Goddard and Asimakopoulos (2004), oddard
(2005), Constantinou et al.(2012), Boshnakov et al. (2017), Shahin (2023).
By combining the Poisson and Lindley distributions in 1970, Sankaran created the Poisson-
Lindley distribution, of which the probability mass function is
✄ ☛ ✒✁ ✆ ✄ ✆ ✝☎
✞✒✁✂ ✄☎ ✌ ✍ ✁ ✌ ✎✍ ✟✍ ✏ ✏ ✏ ✍ ✄ ✑ ✎✏
✒✟ ✆ ✄☎ ✠✡☞
Many works (see Shanker (2016a, 2016b, 2017), Grine and Zeghdoudi (2017), Zeghdoudi and
Nedjar (2017, 2020)) introduce new discrete distributions by compound Poisson and others
introduce new continuous distributions like Poisson-Amarendra, Poisson-Sujatha, Poisson-
Garima, Poisson Quasi-Lindley, Poisson Pseudo-Lindley, and Poisson-Gamma Lindley
distributions. Two novel distributions, called XLindley (XLD) and new XLindley (NXLD),
were recently introduced by Chouia and Zeghdoudi (2020) and Khodja et al. (2023). The
suggested names for these distributions stem from the fact that they are created as a unique
combination of two distributions: Lindley and Gamma.
Poisson-XLindley (PXLD) and Poisson-new XLindley (PNXLD) are two new discrete
distributions that were recently created by Ahsan-ul Haq et al. (2022) and Seghier et al. (2022,
2023) by the compounding of Poisson and XLindley distributions.
To develop novel families of bivariate distribution, mixing procedures were applied. Several
types of mixed distribution that follow from the univariate instances include the Bivariate
Poisson-Lindley (BPLD), Bivariate Negative Binomial (BNBD), and Bivariate Poisson (BPD).
For additional readings, consider the following: tests for over-dispersion and independence in
BNBD were discussed in (Jung et al., 2009; Cheon et al., 2009); bivariate Poisson-Weighted
Exponential was proposed in Zamani et al. (2014); bivariate Poisson-Lindley was proposed in

23
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Zamani et al. (2015); Marshall and Olkin (1990) studied BNBD and applied it in Karlis and
Ntzoufras (2003).
In this study, the BPNXLD is employed, which was constructed by merging two Poisson-new
XLindley margins with a multiplicative factor parameter. We go over this new class of discrete
distributions' mean, variance, correlation, and joint moment generating function. We looked at
a soccer data set to show how flexible the proposed model is compared to the BPD, BPLD and
BNBD. This is the first time that the classic bivariate Poisson model used by Karlis and
Ntzoufras (2003) has been applied to sport in the literature using the new bivariate model.
The structure of this paper is as follows. We provide a brief overview of the univariate Poisson
new XLindley distribution (PNXLD) in Section 2. The formulation of the new proposed model,
which includes its mean, variance, and correlation coefficient of the BPNXLD, is explained in
Section 3. In section 4, a moment estimation approach was offered for figuring out the proposed
model's unidentified parameters. The BPD, BPLD, BNBD and BPNXLD are compared in
Section 5.

Methods

Univariate Poisson New XLindley Distribution (PNXLD)

The NXLD has the following (p.d.f) (Khodja et al, 2023).
✄
✞☛ ✁✂ ✒✞ ✂ ✄ ☎ ✌ ✒✟ ✆ ✄✞ ☎☎✞✆✒✌✄✞ ☎✍ ✞ ✑ ✎✍ ✄ ✑ ✎✏ ✒✟☎
✝
It can be obtained by mixture of ✞✝ ✒✞ ☎✟✠✞✆✒✄☎ and✞☛ ✒✞ ☎✟✡☞✍✍☞✒✝✍ ✄☎, which
✟ ✟
✞☛ ✁✂ ✒✞✂ ✄ ☎ ✌ ✞✝ ✒✞✂ ✄ ☎ ✆ ✞☛ ✒✞✂ ✄ ☎✏
✝ ✝
Where the rth moment of the NXLD is defined as follows
✟
✎✏✑ ✌ ✒✒✓✔ ☎ ✌ ✏ ✕✖ ✒ ✗ ✆ ✟☎ ✆ ✖ ✒ ✗ ✆ ✝☎✘✏
✝✄
A random variable Y follows Poisson New XLindley distribution if it possesses the following
stochastic representation✒✙✚✛ ☎✟✜✢✣✤✤✢✥✒✛☎ ✦✣✧★ ✆✏ ✍✏ ✞
☎ ✪✫ ✒✛☎✬
✜✒✙ ✌ ✩✚✛☎ ✌ ✍ ✩ ✌ ✎✍ ✟✍ ✝✍ ✮ ✍ ✛ ✑ ✎✍
✩✭
where ✛ represents the average rate (e.g. average number of goals)and ✒✛✚✄☎✟✯✰✱✲✒✄☎.
For ✛ ✑ ✎ and ✄ ✑ ✎. We call the unconditional distribution of Y the Poisson New XLindley
distribution and denote it with ✜✯✰✱✲ ✒✄☎and its probability mass function is obtained as
follows
✵ ✵ ☎ ✪✫ ✒✛ ☎✬ ✄
✜ ✙ ✌ ✩ ✌ ✳ ✜ ✙ ✌ ✩✚✛ ☛ ✁✂ ✛✂ ✄ ✴✛ ✌ ✳
✒ ☎ ✒ ☎✞ ✒ ☎ ✒✟ ✆ ✄✛ ☎☎ ✪✷✫ ✴✛
✶ ✶ ✩✭ ✝
✄ ✵ ✵
✌ ✸✳ ☎ ✪✹✝✡✷✺✫ ✛✬ ✴✛ ✆ ✄ ✳ ☎ ✪✹✝✡✷✺✫ ✛✻✡✝ ✴✛✼
✝✛✭ ✶ ✶
✄ ✖ ✒✞ ✆ ✟☎ ✄✖ ✒✞ ✆ ✝☎
✌ ✸ ✆ ✼✍
✝✛✭ ✒✄ ✆ ✟☎ ✻✡✝ ✒✄ ✆ ✟☎✬✡☛

Which we obtained p.m.f of PNXLD as follows

IJCSS Volume 23/2024/Issue 2 www.iacss.org

✄✒✄✩ ✆ ✝✄ ✆ ✟☎
✜☛☛ ✁✂ ✒✩✂ ✄☎ ✌ ✍ ✩ ✌ ✎✍ ✟✍ ✝✍ ✄✍ ✮ ✍ ✄ ✑ ✎✏ ✒✝☎
✝✒✄ ✆ ✟☎✬✡☛
With the central first four moments of the PNXLD are obtained as
✄ ✄✆✁ ✄ ☛ ✆ ✝✟✄ ✆ ✟☎ ✝✂✄ ☞ ✆ ✄✎✂✄ ☛ ✆ ✟✝✄ ✆ ✄✄✄
✎✝ ✌ ✍ ✎☛ ✌ ✍ ✎☞ ✌ ✍ ✎✆ ✌ ✏
✝✄ ✂✄ ☛ ✂✄ ☞ ✟ ✄✆
The probability-generating function ✞✝ ✒✧☎ and themoment-generating function ✟✝ ✒✁☎ of the
PNXLD can beobtained as
✵
✡ ✄✧ ✄✒✟ ✆ ✄☎ ✒✝✄ ✆ ✟☎
✞✝ ✒✧☎ ✌ ✒✒✠ ☎ ✌ ✌ ✧ ✬ ✜✒✞ ✂ ✄☎ ✌ ☞ ✆ ✍✍
✝✒✟ ✆ ✄ ☎☛ ✒✄ ✆ ✟ ✌ ✧ ☎☛ ✒✄ ✆ ✟ ✌ ✧ ☎
✬✎✝
and
✄☎ ✠ ✄✒✟ ✆ ✄☎ ✄✒✟ ✆ ✄☎
✟✝ ✒✁☎ ✌ ✒✒✏✑✡ ☎ ✌ ☞ ✆ ✍✏
✝✒✟ ✆ ✄ ☎☛ ✒✄ ✆ ✟ ✌ ☎ ✠ ☎☛ ✒✄ ✆ ✟ ✌ ☎ ✠ ☎

For more details on this section see Seghier et al. 2023.

Bivariate Poisson New XLindley Distribution (BPNXLD)

The joint p.m.f. of ✒✜✒✛✝✓ ✛☛✓✔☎distribution, which was derived from the product of two Poisson
margins witha multiplicative factor parameter, is defined as (Lakshminarayana et al.,1999):
✬ ✬
✛✝ ✕ ✛☛ ✖
✜✒✙✝ ✌ ✩✝ ✍ ✙☛ ✌ ✩☛ ☎ ✌ ☎ ✪✫✕ ✪✫✖ ✒✟ ✆ ✔✒✒✡✝ ✒✩✝ ☎ ✌ ✗✗✗
✡✝ ☎✒✡☛ ✒✩☛ ☎ ✌ ✗✗✗
✡☛ ☎☎☎✍ ✒✂☎
✩✝ ✭ ✩☛ ✭
✩✝ ✍ ✩☛ ✌ ✎✍ ✟✍ ✝✍ ✮ ✂ ✛✝ ✍ ✛☛ ✑ ✎✏
where, ✡✝ ✒✩✝ ☎ and ✡☛ ✒✩☛ ☎ arebounded functions in ✩✝ and ✩☛ respectively. The value of
✒✟ ✆ ✔✘✒✡✝ ✒✩✝ ☎ ✌ ✗✗✗ ✗✗✗☛ ☎✙☎ in (4) is non-negative when✡✚ ✒✩✚ ☎ ✌ ☎ ✪✬✛ and ✗✗✗
✡✝ ☎✒✡☛ ✒✩☛ ☎ ✌ ✡ ✡✚ ✌
✠ ✘✡ ✒✙ ☎✙ ✌ ✠ ✒☎ ✪✝✛ ☎✍ ✧ ✌ ✟✍✝✏
✚ ✚
In a similar manner, the joint p.m.f. of ✜✢✣✤✥✦✒✄✝✓ ✄☛✓ ✔ ☎ is defined as:

✜✒✙✝ ✌ ✩✝ ✍ ✙☛ ✌ ✩☛ ☎
✄✝ ✒✄✝ ✩✝ ✆ ✝✄✝ ✆ ✟☎ ✄☛ ✒✄☛ ✩☛ ✆ ✝✄☛ ✆ ✟☎
✌ ✕✟
✝✒✄✝ ✆ ✟☎✬✕ ✡☛ ✝✒✄☛ ✆ ✟☎✬✖✡☛

✆ ✔✒☎ ✪✬✕ ✌ ✧✝ ☎✒☎ ✪✬✖ ✌ ✧☛ ☎✘ ✒ ☎☎

where ✩✝ ✍ ✩☛ ✌ ✎✍ ✟✍ ✝✍ ✮ ✂ ✄✝ ✍ ✄☛ ✑ ✎✍and
✄✚ ☎ ✪✝ ✄ ✒✟ ✆ ✄✚ ☎ ✒✝✄✚ ✆ ✟☎
✪ ✝
✧✚ ✌ ✠ ✒☎ ☎ ✌✛ ☞ ✚ ✪✝ ✆ ✍ ✍ ✧ ✌ ✟✍✝✏ ✒ ☎
☛
✝✒✟ ✆ ✄✚ ☎ ✒✄✚ ✆ ✟ ✌ ☎ ☎ ☛ ✒✄✚ ✆ ✟ ✌ ☎ ✪✝ ☎

We obtain ✠ ✒☎ ✪✝✛ ☎ in (6) by letting✁ ✌ ✌✟in m.g.f. (3). When✔ ✌ ✎, random variables ✙✝ and
✙☛ areindependent, each is distributed as a marginal PNXLD. Therefore, ✔ is the parameter of
independence.
The mean and variance of ✜✢✣✤✥✦✒✄✝✓ ✄☛✓ ✔☎ are:

25
IJCSS Volume 23/2024/Issue 2 www.iacss.org

✄
✠✒✙✚ ☎ ✌ ✎✚ ✌ ✍ ✧ ✌ ✟✍✝✏
✝✄✚
✄✚ ✆ ✁
✞☞✗✒✙✚ ☎ ✌ ✍ ✧ ✌ ✟✍✝✍
✂✄✚☛
where we can check that
✞☞✗✒✙✚ ☎ ✑ ✠ ✒✙✚ ☎✏
And
✢✁✒✙✝ ✍ ✙☛ ☎ ✌ ✔✒✧✝✝ ✌ ✎✝ ✧✝ ☎✒✧☛☛ ✌ ✎☛ ✧☛ ☎✍ ✒✁☎

where, ✧✚✚ ✌ ✠ ✒✙✚ ☎ ✪✝✛ ☎✍ ✧ ✌ ✟✍✝✍differentiating m.g.f. in (3) with respect to ✁and letting✁ ✌
✌✟, we
have
✂
✟ ✒✁☎✚ ✌ ✠ ✒✙☎ ✪✝ ☎✏
✂✁ ✝ ✠✎✪✝
Thus
✄✚ ☎ ✪✝ ✄ ✒✟ ✆ ✄✚ ☎ ✒✝✄✚ ✆ ✟☎
✧✚✚ ✌ ☞ ✚ ✪✝ ✆ ✍ ✍ ✧ ✌ ✟✍✝✏
☛
✝✒✟ ✆ ✄✚ ☎ ✒✄✚ ✆ ✟ ✌ ☎ ☎ ☛ ✒✄✚ ✆ ✟ ✌ ☎ ✪✝ ☎
Using the variance and covariance in (7), the correlation coefficient is:
✄ ✔✒✧ ✌ ✎ ✧ ☎✒✧ ✌ ✎ ✧ ☎
✄✝☛ ✌ ✝☛ ✌ ✝✝ ✝ ✝ ☛☛ ☛ ☛ ✒ ☎☎
✄✝ ✄☛ ✄✝ ✄☛
From (8), ✙✝ and ✙☛ are independent when ✔ ✌ ✎ and havepositive and negative correlations
when ✔ ✑ ✎ and ✔ ✌ ✎respectively.

Parameter Estimation
By equating the mean and covariance in (7) with the sample moments, onemay derive the
moment estimates of ✜✢✣✤✥✦✒✄✝✓ ✄☛✓ ✔ ☎ . The unique moment estimate of ✄✚ is
✄
✄✆✚ ✌ ✝
✝✚ ✍ ✙✚ ✑ ✎✍ ✧ ✌ ✟✍✝✏
✝✙
✩✝ ☎✪✝ ✒✧☛✠ ☛ ✌ ✧☛✠ ✩✗✗✗☛ ☎✪✝ ,
The moment estimate for ✔ can then be computed using✔✟ ✌ ✤✝☛ ✒✧✝✠ ✝ ✌ ✧✝✠ ✗✗✗
with
✡☞☛✎✝ ✩☛✚ ✡☞☛✎✝✒✩☛✝ ✌ ✩✗✗✗✝ ☎✒✩☛☛ ✌ ✩✗✗✗☛ ☎
✝
✙✚ ✌ ✍ ✧ ✌ ✟✍✝ ☞✥✴ ✍✝☛ ✌ ✍
✥ ✥✌✟
where ✧✝✠ ✍ ✧☛✠ ✍ ✧✝✠ ✝ and ✧☛✠ ☛ areestimated values of ✧✝ ✍ ✧☛ ✍ ✧✝✝ and ✧☛☛ .
The log likelihood function for ✜✢✣✤✥✦✒✄✝✓ ✄☛✓ ✔ ☎ is:
☞
✱✢✡✱ ✌ ✌✒✎✢✡✄✝ ✆ ✏✑✓ ✒✄✝ ✩☛✝ ✆ ✝✄✝ ✆ ✟☎ ✌ ✒✩☛✝ ✆ ✝☎ ✏✑✓✕✝✒✄✝ ✆ ✟☎✘
☛✎✝
✆✎✢✡✄☛ ✆ ✏✑✓✒✄☛ ✩☛☛ ✆ ✝✄☛ ✆ ✟☎ ✌ ✒✩☛☛ ✆ ✝☎ ✏✑✓✕✝✒✄☛ ✆ ✟☎✘ ✒✔☎

✆✏✑✓ ✘✟ ✆ ✔ ✒☎ ✪✬✕✕ ✌ ✧✝ ☎✒☎ ✪✬✕✖ ✌ ✧☛ ☎✙✖

IJCSS Volume 23/2024/Issue 2 www.iacss.org

One can derive the log likelihood estimates of ✒✄✝ ✄☛ ☎ by maximizing the log
✜✢✣✤✥✦
✓ ✓
✔

likelihood inequation (9). The negative expectation of the second derivatives of thelog
likelihood can be used to obtain the Fisher Information matrix.

Results

Application and Comparison

Presentation of data
To illustrate the bivariate Poisson new XLindley model, we shall examinemany football data
sets (see [ 1,2]).The GermanBundesliga football championship comprised 18 teams which
competed in both home and away matches, resulting in a total of 306 matches. The file
underinvestigation originates from the bivpois package developed by Karlis and Ntzoufras
(2003) and has four variables, namely:
the amount of goals scored by the home team (g1);
the amount of goals scored by the visiting team (g2);
team1: name of the home team;
team2: name of the visiting team;
For the investigation's aims, let's set:
the home team's goal total is represented by X,
while the away team's goal total is represented by Y.
The distribution of matches by score is shown in Table 1
Table 1. Distribution of matches according to scores with the fit of BPNXLD of data set I.

X/Y 0 1 2 3 4 5 6 Total
0 16(17.2) 12(12.3) 16(15) 2(2.1) 3(2.8) 1(1.1) 1(0.5) 51
1 23(24.4) 41(40.3) 17(16.1) 10(11) 10(8.5) 1(1.6) 2(2.1) 104
2 19(19.4) 29(28.5) 15(15.3) 8(7) 1(2.1) 2(0.9) 0(0.4) 74
3 11(9.9) 14(14.3) 12(11.4) 1(0.9) 3(3.5) 0(0.4) 1(1.4) 42
4 10(9.1) 8(8.7) 3(2.7) 0(0.01) 0(0.01) 0(0.01) 0(0.01) 21
5 4(3.1) 3(3.1) 2(2.1) 0(0.01) 0(0.01) 0(0.01) 0(0.01) 9
6 2(1.5) 1(0.8) 0(0.2) 0(0.01) 0(0.2) 0(0.01) 0(0.01) 3
7 1(1.1) 1(0.6) 0(0.01) 0(0.01) 0(0.01) 0(0.01) 0(0.01) 2
Total 86 109 65 21 17 4 4 306

Table 2. Descriptive statistics of data set I and II.

Globalstatistics Home goals(X) Away goals(Y)
SetI SetII SetI SetII
Minimum 0 0 0 0
Maximum 7 8 6 6
Mean 1.76 1.66 1.35 1.55
Wins 47% 40% 29% 38%
Standard deviation 1.40 1.31 1.30 1.25
Variance 1.11 1.03 1.25 1.008
Correlation 0.11 0.08

27
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Table 3. General statistic of data set I and II.

SetI SetII SetI SetII
Numberofrounds 34 34 Averageofgoalspermatch 3.11 3.21
Homevictories 143 123 Bestroundingoals 14th (41goals) 10th (35goals)
Awayvictories 90 115 Totalnumberofgoals 953 982
Drawnmatches 73 68 Averagegoalsperround 28.03 28.83

Table 4. Various statistics Wins, Drawn matches and Losses of data set I.
Wins Drawnmatches Losses
BayernMunchen 24 ArminiaBielefeld 13 SpVggGreutherFürth 22
BorussiaDortmund 22 EintrachtFrankfurt 12 HerthaBerlin 19
BayerLeverkusen 19 VfBStuttgart 12 ArminiaBielefeld 16
RBLeipzig 17 FCKöln 10 FCAugsburg 16
FCUnionBerlin 16 SCFreiburg 10 VfLBochum 16
SCFreiburg 15 BorussiaMönchengladb 9 VfLWolfsburg 16

Table 5. Various statistics victories,Tie games and Losses (Home case).

Home victories Tie games at home Losses at home
BayernMunchen 13 ArminiaBielefeld 10 SpVggGreutherFürth 8
BorussiaDortmund 13 EintrachtFrankfurt 7 HerthaBerlin 8
RBLeipzig 11 FSVMainz05 5 VfBStuttgart 7
FSVMainz05 10 FCUnionBerlin 5 VfLWolfsburg 7
BayerLeverkusen 10 SCFreiburg 5 EintrachtFrankfurt 6
FCUnionBerlin 10 SpVggGreutherFürth 5 FCAugsburg 6

Table 6. Various statistics victoriess, Tie games and Losses of data set I (Away case).
Away victories Tie games (Away) Losses (Away)
BayernMunchen 11 VfBStuttgart 8 pVggGreutherFürth 13
BorussiaDortmund 9 FCKöln 6 FSVMainz05 12
BayerLeverkusen 9 BayerLeverkusen 5 ArminiaBielefeld 11
RBLeipzig 7 BorussiaMönchengladb 5 HerthaBerlin 11
EintrachtFrankfurt 6 EintrachtFrankfurt 5 VfLBochum 11
FCUnionBerlin 6 RBLeipzig 5 FCAugsburg 10

Tables1-6 represents general statistics and descriptive statistics of data set I and II.

Crockett quick and Index of dispersion tests for bivariate Poisson new XLindley
distribution
Crockett quick test
Crockett (Rayner and Best (1979)) proposed the test statistic
☛ ☛
✝☛
☛
✝ ✌ ✝✍✝☛ ✝ ☛ ✌ ✝✍✝☛ ✆ ✙
✙ ☛ ✝✝ ☛ ☛☛
✞✌✥
✝✒✙✝✝ ☛ ✙✝☛ ☛ ✌ ✍✝✆☛ ☎
where ✙✝✝ and✙✝☛ are the sample means; ✝ ✌ ✍✝☛ ✌ ✙✝✝ , ☛ ✌ ✍☛☛ ✆ ✙✝☛ ; ✍✝☛ , ✍☛☛ the sample variances;
and ✍✝☛ the samplecovariance.
IJCSS Volume 23/2024/Issue 2 www.iacss.org

The statistic T asymptotically follows a ☛ distribution with twodegrees of freedom for n large
and the ✞✶ hypothesis is rejected if✞ ✑ ☛☛ ✝✪☞ for a level test .
✓
✔

Index of Dispersion Test

Similar to the Crockett test, Best and Rayner (1997) also suggest revisingthe statistic and
recommending it.
✙✝☛ ✍✝☛ ✌ ✝✍✝☛☛ ✆ ✙✝✝ ✍☛☛
✁☛ ✌ ✥
✙☛ ✙✝ ✌ ✍✝☛☛
✗✗✗✗✗

The null distribution of this statistic ✁☛ is asymptotically a ☛ with 2n-3 degrees of freedom.
Hypothesis ✞✶ is thereforerejected if ✁☛ ✑ ☛☛☞✪☞ ✝✪☞ for a test of level .
✓
✔

In our case,we test:

Hypothesis ✞✶ : the data can be modeled with BPNXLD.
Hypothesis ✞✝ : the data can not be modeled with BPNXLD.

Table 7. ✂✄☎✆✝✟✠✠ ✡✌✍✆✝ ✎✏✑ ✒✏✑✟✓ ☎✔ ✑✍✕✖✟✄✕✍☎✏ ✠✟✕✠✕ ✗✎✘✌✟✕ ✔☎✄ ✙✚✛✜✢
Crockett quick test Index of dispersion test
Set data I (n = 8) T= 0.66 ✣✤ ✥ ✦✧★★
Set data II (n = 8) T= 0.645 ✣✤ ✥ ✩✧✪✪

Hypothesis ✞✶ is not rejected because T and ✁☛ ✌ ☛✫ ☞ . According table 3 to crockett quick and
✓

Index of dispersion tests,the two data sets can be modeled with BPNXLD.
Comparison between BP, BPLD, BNBD and BPNXLD
This subsection presents a comparative analysis of many bivariatedistributions, including the
BPD, BPLD, BNBD with five parameters✒✄✝ ✍ ✄☛ ✍ ✍ ✬✝ ✍ ✬☛ ☎ (see Famoye (2010)) andBPNXLD.
✔

Figure 1. Estimate values using BPNXLD and real values of matches according to scores of data set I.

29
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Figure 2. The P P plot of BPNXLD according to data set I.

Table 8.
Distributions Parameters -LL AIC AICc BIC ✡

BPD ✌
✞☛ ✥ ✩✧✩✪★✁ ✞✌ ✥ ✄✧✂✪★✁ ☎✆ ✥ ✄✧✄✪✝ 152.76 305.53 305.99 322.69 50.698
BPLD ✌ ✥ ✦✧✄✩✟✁ ✞
✞ ✌ ✥ ✩✧✝✠✩✁ ☎ ✆ ✥ ✄✧✪✟ 118.54 237.09 237.55 254.25 27.734
☛

✞☛ ✥ ✄✧✟✄✁ ✞
✌ ✌ ✥ ✄✧✠✩✁ ☎
✆ ✥ ✄✧☞✟✁
BNBD 110.97 231.94 232.14 250.56 26.692
✍☛ ✥ ✄✧✩✩✁ ✍ ✥ ✄✧✂✄

BPNXLD ✌ ✥ ✄✧✪✄✟✁ ✞
✞ ✌ ✥ ✩✧✩✄★✁ ☎ ✆ ✥ ✄✧✄★ 89.16 179.32 179.78 195.85 6.29
☛

Table 9.
Distributions Parameters -LL AIC AICc BIC ✡

BPD ✥ ✩✧✄✝✁ ✞✌ ✥ ✄✧✂✦✁ ☎✆ ✥ ✄✧✄★✂

✌
✞☛ 188.65 383.30 383.38 394.47 100.8
BPLD ✞☛ ✥ ✩✧☞✪✁ ✞
✌ ✌ ✥ ✩✧★✟✁ ☎ ✆ ✥ ✄✧✝☞ 154.98 315.96 316.04 327.13 77.5
✌ ✥ ✄✧✠★✁ ✞
✞ ✌ ✥ ✄✧✦✪✁ ☎✆ ✥ ✄✧✪✪✁
☛
BNBD 121.20 252.40 252.60 271.02 20.3
✍☛ ✥ ✄✧✩✄✁ ✍ ✥ ✄✧✝✂

BPNXLD ✞☛ ✥ ✄✧✂✪✂✁ ✞
✌ ✌ ✥ ✩✧✄✪✁ ☎ ✆ ✥ ✄✧✄✟★ 122.96 251.92 252.00 263.09 20.2

As shown in Table 8, 9 and Figures 1, 2, the BPNXLD offers the smallest ☛ , BIC, AIC and
AICc values in comparison to the other bivariate distributions, and as a result, best fits the data
of all the distributions taken into consideration.

Discussion
It is clear from Tables 2-6 that, on average, the home side scores more goals than the away
team. This is known as the "home (field) advantage,"which is discussed here and isn't exclusive
to soccer. Right now, the Poisson new XLindley distribution is a helpful idea to teach.
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Additionally, Table 1 demonstrates that the variance and mean are not equal, which means that
the Poisson distribution is inappropriate in this circumstance (see Karlis and Ntzoufras (2003)
and Zamani et al. (2014)). However, the Poisson new XLindley distribution is appropriate
because the variance is greater than the mean. Table 2 shows that the average number of goals
scored at homeand away in data set II (2019-2020 season) are nearly the same because
therewere no spectators present at home matches owing to quarantine measuresrelated to the
Covid-19 pandemic.
Hence, using negative log-likelihood -LL, Akaike Information Criteria AIC, Akaike
Information Corrected Criteria AICc and Bayesian Information Criteria BIC, the choice of the
best fitted model is evaluated. Inaddition, we calculate the anticipated frequencies for each
distribution, and the quality of fit of the model is evaluated using Pearson's chi-squaretest.
Tables 8, 9 and Fig 1, 2 show that all distributions have positivevalues for ✂, which suggests a
positive connection. A positive correlation is shown by a positive value for ✂ indistributions.
Onthe other hand, BNBD is the most suitable option based on -LL for the dataset II (2019-2020
season) because there were no spectators present owing tothe quarantine imposed by the Covid-
19 pandemic. The BPNXLD offers the bestfit for the two data sets based on AIC, AICc and
BIC. Also, we can seethat this study demonstrates the adaptability of the BPNXLD for sport
datasets.

Conclusion
In this paper, the bivariate Poisson XLindley distribution was used to fit a sample of bivariate
count data. The results show that, compared to the bivariate Poisson and bivariate Poisson-
Lindley distributions, the bivariate Poisson new XLindley distribution fits dependent and
excessively dispersed count data with a positive correlation. This implies that the distribution
could be a good substitute. We believe it would be interesting to use the bivariate Poisson-new
XLindley model to model actuarial or dependability data. The significance and potentiality of
the proposed bivariate distribution are empirically demonstrated using several sport data sets..

Acknowledgements
The authors are grateful for the comments and suggestions by the referee and the Editor
Professor Arnold Baca.Their comments and suggestions greatly improved the article.

References
Ahsan-ul-Haq, M., Al-Bossly, A., El-Morshedy, M., & Eliwa, M. S. (2022). Poisson XLindley
distribution for count data: statistical and reliability properties with estimation
techniques and inference. Computational Intelligence and neuroscience, 2022(1),
6503670.
Baio, G., & Blangiardo, M. (2010). Bayesian hierarchical model for the prediction of football
results. Journal of Applied Statistics, 37(2), 253-264.
Best D. J., Rayner J. C. W. (1997). Crockett.s test of fit for the Bivariate Poisson. Biometrical
Journal, 39(4):423.430.
Boshnakov, G., Kharrat, T., & McHale, I. G. (2017). A bivariate Weibull count model for
forecasting association football scores. International Journal of Forecasting, 33(2),
458-466.

31
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Bradley, R. A. and Terry, M. E. (1952). Rank analysis of incomplete blockdesigns: I. The

method of paired comparisons. Biometrika, 39, 324-345.
Cheon S., Song S.H., Jung B.C. (2009). Tests for independence in a bivariatenegative binomial
model. J. Korean Statist. Soc., 38: 185-190.
Chouia, S., & Zeghdoudi, H. (2021). The XLindley distribution: Properties and
application. Journal of Statistical Theory and Applications, 20(2), 318-327.
Constantinou, A. C., Fenton, N. E., & Neil, M. (2012). pi-football: A Bayesian network model
for forecasting Association Football match outcomes. Knowledge-Based Systems, 36,
322-339.
Cox, D.R. and D.V. Hinkley, (1979). Theoretical Statistics. 1st Edn., CRCPress, ISBN-10:
0412161605, pp: 528.
McNeil, D. (1979). A QUICK TEST OF FIT OF A BIVARIATE DISTRIBUTION.
In Interactive Statistics: Proceedings of the Applied Statistics Conference, Sydney,
February 8-9, 1979 (p. 185). North-Holland.
Dixon, M. J., Coles, S. G. (1997). Modelling association football scores andinefficiencies in
the football betting market. Applied Statistics, 46(2), 265.
Famoye, F. and P.C. Consul, (1995). Bivariate generalized Poisson distribution with some
applications. Metrika, 42: 127-138.
Famoye, F. (2010). On the bivariate negative binomial regression model. Journal of Applied
Statistics, 37(6), 969-981.
Ghitany M. E., Atieh B., Nadarajah S. (2008). Lindley distribution and itsapplications. Math.
Comput. Simulation, 78, pp. 493-506.
Goddard, J. (2005). Regression models for forecasting goals and matchresults in association
football. International Journal of Forecasting, 21(2), 331-340.
Goddard, J., & Asimakopoulos, I. (2004). Forecasting football results and the efficiency of
.xed-odds betting. Journal of Forecasting, 23(1), 51-66.
Holgate, P., (1964). Estimation for the bivariate Poisson distribution. Biometrika, 51: 241-245.
Johnson, N.L., S. Kotz and N. Balakrishnan, (1997). Discrete Multivariate Distributions. 1st
Edn., Wiley, New York, ISBN-10: 0471128449, pp: 328.
Jung, B.C., M. Jhun and S.M. Han, (2009). Score test for overdispersionin the bivariate
negative binomial models. J. Statist. Comput. Simulat., 79:11-24.
Karlis, D. and I. Ntzoufras, (2003). Analysis of sports data by using Bivariate Poisson models.
Statistician, 52: 381-393.
Karlis, D., &Ntzoufras, I. (2009). Bayesian modelling of football outcomes: using the
Skellam.s distribution for the goal difference. IMA Journal of Management
Mathematics, 20(2), 133-145.
Kocherlakota, S., & Kocherlakota, K. (2017). Bivariate discrete distributions. CRC Press.
Khodja, N., Gemeay, A. M., Zeghdoudi, H., Karakaya, K., Alshangiti, A. M., Bakr, M. E.,
Hussam, E. (2023). Modeling voltage real data set by a new version of Lindley
distribution. IEEE Access, 11, 67220-67229.
Lakshminarayana, J., S.N.N. Pandit and K.S. Rao, (1999). On a bivariatePoisson distribution.
Communicat. Statist. Theory Methods, 28: 267-276.
Lindley, D. V. (1958). Fiducial distributions and Bayes' theorem. Journal of the Royal
Statistical Society. Series B (Methodological), 102-107.
IJCSS Volume 23/2024/Issue 2 www.iacss.org

Lord, D. and S.R. Geedipally, (2011). The negative binomial-Lindley distribution as a tool for
analyzing crash data characterized by a large amount of zeros. Accident Anal. Prevent.,
43: 1738-1742.
Loukas S. and Kemp C. D. (1986), The Index of Dispersion Test for the Bivariate Poisson
Distribution. International Biometric Society, Vol. 42, No. 4,pp. 941-948.
Maher,M. J. (1982).Modelling association footballscores. Statistica Neerlandica,
36(3),109.118.
Mahmoudi E. and Zakerzadeh H. (2010), Generalized Poisson-Lindley distribution,
Communications in Statistics- Theory and Methods, 39, 1785 - 1798.
Marek, P., .edivá, B., & µToupal, T. (2014). Modeling and prediction of icehockey match
results. Journal of quantitative analysis in sports, 10(3), 357-365.
Mitchell, C. R., & Paulson, A. S. (1981). A new bivariate negative binomial distribution. Naval
Research Logistics Quarterly, 28(3), 359-374.
Owen, A. (2011). Dynamic Bayesian forecasting models of football matchoutcomes with
estimation of the evolution variance parameter. IMA Journal of Management
Mathematics, 22(2), 99-113.
Paul, S.R. and N.I. Ho, (1989). Estimation in the bivariate poisson distribution and hypothesis
testingconcerning independence. Communicat. Statist.Theory Methods. 18: 1123-1133.
Reep, C., & Benjamin, B. (1968). Skill and chance in association football. Journal of the Royal
Statistical Society. Series A (General), 131(4), 581-585.
Sadeghkhani, A., & Ahmed, S. E. (2020). The application of predictive distribution estimation
in multiple-inflated poisson models to ice hockey data. Model Assisted Statistics and
Applications, 15(2), 127-137.
Sankaran, M., (1970). The discrete Poisson-Lindley distribution. Biometrics, 26: 145-149.
Seghier, F. Z., Zeghdoudi, H., & Raman, V. (2023). A Novel Discrete Distribution: Properties
and Application Using Nipah Virus Infection Data Set. European Journal of
Statistics, 3, 3-3.
Shanker, R. (2016a). The discrete poisson-amarendra distribution. Int. J. Stat. Distrib.
Appl, 2(2), 14-21.
Shanker R. (2016b), The discrete Poisson-Sujatha distribution. International Journal of
Probability and Statistics. 5(1):1-9.
Shanker R. (2017), The discrete poisson-garima distribution. Biometrics & Biostatistics
International Journal, 5(2):48-53.
Seghier, F. Z., Ahsan-ul-Haq, M., Zeghdoudi, H., & Hashmi, S. (2023). A new generalization
of poisson distribution for over-dispersed, count data: mathematical properties,
regression model and applications. Lobachevskii Journal of Mathematics, 44(9), 3850-
3859.
Tsokos, A., Narayanan, S., Kosmidis, I., Baio, G., Cucuringu, M., Whitaker, G., & Király, F.
(2019). Modeling outcomes of soccer matches. Machine Learning, 108, 77-95.
Wheatcroft, E. (2021). Forecasting football matches by predicting match statistics. Journal of
Sports Analytics, 7(2), 77-97.
Zamani, H., Faroughi, P., & Ismail, N. (2014, June). Bivariate Poisson-weighted exponential
distribution with applications. In AIP Conference Proceedings (Vol. 1602, No. 1, pp.
964-968). American Institute of Physics.

33
 IJCSS Volume 23/2024/Issue 2 www.iacss.org

Zamani, H., P. Faroughi and N. Ismail, (2015). Bivariate Poisson-Lindley Distribution with
Application, Journal of Mathematics and Statistics, 11 (1): 1-6.
Zeghdoudi, H., & Nedjar, S. (2017). On Poisson pseudo Lindley distribution: Properties and
applications. Journal of probability and statistical science, 15(1), 19-28.
Shahin, S. (2023). Sports Data Analysis by using Bivariate Poisson Models in the Bayesian
Framework. Quaid-e-Awam University Research Journal of Engineering Science and
Technology, 21(1), 7-15.
Wheatcroft, E. (2021). Forecasting football matches by predicting match statistics. Journal of
Sports Analytics, 7(2), 77-97.
Singh, A., Scarf, P., & Baker, R. (2023). A unified theory for bivariate scores in possessive
ball-sports: the case of handball. European Journal of Operational Research, 304(3),
1099-1112.

[1] Data Set I :https://www.the-sports.org/football-soccer-2021-2022-german-bundesliga-

epr114545.html. 2023
[2] Data Set II :https://www.the-sports.org/football-soccer-2019-2020-german-bundesliga-
epr98318.html. 2024

View publication stats