Popula r A rticle

Journal Home: www.bioticainternational.com

[ [
Article: RT912

Biotica H o w t o c i t e t h i s a r t i c l e?
Chittora, 2022. Poisson Regression: A Regression
Research Approach for Count Response Variable. Biotica Research
Today 4(2): 129-131.
Today A bstrac t
129
Vol 4:2
I
n regression analysis when the response variable is not continuous
131 in nature or it is in number or count data in that situation response
2022 variable will not follow the assumption if regression. When the
response variable is in number or count in that situation Poisson
regression will give the valid estimates of the regression parameters.
Poisson Regression: A The characteristics of the Poisson regression mean and variance

Regression Approach for must be the same and our response variable should follow the
assumption of Poisson model. Poisson Regression models are best
Count Response Variable used for modeling events where the outcomes are counts. Poisson
Regression helps us analyze both count data and rate data by allowing
Vaibhav Chittora us to determine which explanatory variables (X values) have an effect
on a given response variable Y.
Dr. Y.S. Parmar University of Horticulture and Forestry, Nauni,
Solan, Himachal Pradesh (173 230), India Introduction

R
egression analysis is a measure of the functional
relationship between the two variables. Or it can be
define as a technique for determining the statistical
Open Access relationship between two or more variables when a change
Corresponding Author in a dependent variable is associated with and depends on a
change in one or more independent variables.
Vaibhav Chittora
e-mail: [email protected] Definition of Linear Regression

Key wo r d s
Count data, Linear regression, Parameters estimation, Poisson
I n statistics, linear regression is a linear approach to
modeling the relationship between a dependent variable
and one or more independent variables. If it involves
only one independent variable it is called as simple linear
regression.
Distribution
Y=β0+β1X+ε
Whereas, Y is dependent variable, X is independent variable,
is error term.
Assumption of Linear Regression
1. Linear relationship
2. Y and X are continuous variables
3. Error are independently identically distributed as mean
zero and constant variance, σ2ε..
• If the assumption of continuity of dependent variable fails
then, a simple linear regression is not appropriate.
• Then depending upon the scales of measurement of
dependent variable we have to modify the idea of simple
linear regression.
Article History Introduction of Poisson Regression

P
Received on: 17th February 2022
Revised on: 21st February 2022 oisson regression is similar to regular multiple regression
Accepted on: 22nd February 2022 except that the dependent (Y) variable is an observed
count that follows the Poisson distribution (Bastos and
E-mail: [email protected] Rosa, 2013). Thus, the possible values of Y are the nonnegative

© 2022 129
Chittora, 2022

integers: 0, 1, 2, 3, and so on. Estimating Model Parameters β0,β1

Assumptions of Poisson Regression To simplify the notation we introduce xi0=1 and can now
1. The dependent variable y is count. write (1) as
2. Counts must be positive integers (i.e., 0,1,2,3… k). E[Yi]=⋋i=exp(∑1j=0βj xji),
3. Count must follow a Poisson distribution. Therefore, the Where YiϵP0(⋋i) for i = 1
property mean=variance holds. The likelihood function is calculated as
4. Explanatory variables must be continuous, dichotomous ⋋y
L(β)= ∏ki=1 P (Yi=yi)=∏ki=1 i e-⋋i
or ordinal. y!
Where ⋋i=⋋i (βp) is a function of βp=(β0,β1)
5. Observations are independent of each other.
The ML-Estimates β p*=(β 0*,β 1*) are the values of β that
Applications maximize the log likelihood function,
It is applicable where positive integer outcome L(β)=-∑ki=1log(yi!)+∑ki=1yi log(⋋i)-∑ki=1⋋i
• The dependent variable is taking positive integer values By setting the first order derivatives of the log-likelihood equal
for example: to zero, we get a system of (P+1) non-linear equations in βj
1. Number of families visiting a restaurant in a given hour.
2. Number of accident claims being made by car insurance
holder in a year.
3. Expected number of credit cards a customer may have.
• Most of the time the dependent variable is smaller values
like 0,1,2, and there are few somewhat larger values. Extension of P-Independent Variable
• The average is likely to be a small value in which case the The count data is assumed to be observations from random
response variable y has a Poisson distribution. variables yiєP0 (⋋i) (called response or dependent variables)
• In such cases the response variable y has a Poisson with mean value ⋋i=⋋i(xi1,……,xip ). The variables xi1,……xip,
distribution. are called explanatory variables and are assumed to measure
factors that influence the count data (Consul and Famoy,
• If the average is large value, then for the response variable
1992).
y, the normal distribution is a good approximation for the
Poisson distribution (Cameron and Trivedi, 2013). We restrict ⋋i to be a log-liner function,
• In Poisson regression-based on input variables, you tend to ⋋i=exp(β0+β1xi1+ ---+βpxip)……….. (2)
predict the count (i.e., Dependent variable) (Deb and Trivedi, And thus the probability that Yi=y is,
1997).
• In all such cases independent variable can be Continuous
or Continuous + dummy variable (categorical) and dependent
variable is count variable.
Poisson Regression Model Estimating Model Parameters β0,……βp

T
he count data is assumed to be observations from
random variables y i ϵP 0 (⋋ i) (called response or To simplify the notation we introduce xi0=1 and can now
dependent variables) with mean value ⋋i=⋋i (x1). The write (1) as
variable x1 is called explanatory variable and is assumed to E[Yi]=⋋i=exp(∑pj=0βj xij),
measure factor that influence the count data. Where YiϵP0 (⋋i) for i = 1…k
We consider ⋋i to be a log-liner function, The likelihood function is calculated as
⋋i=exp(β0+β1x1) ………………... (1) ⋋y
L(β)= ∏ki=1 P (Yi=yi)=∏ki=1 i e-⋋i
And thus the probability that Yi=y is, y!
Where ⋋i=⋋i (βp) is a function of βp=(β0,…..βp)
The ML-Estimates βp*=(β0*,…..,βp*) are the values of β that
maximize the log likelihood function,
L(β)=-∑ki=1log(yi!)+∑ki=1yi log(⋋i)-∑ki=1⋋i

© 2022 130
Biotica Research Today 2022, 4(2):129-131

By setting the first order derivatives of the log-likelihood equal

to zero, we get a system of (P+1) non-linear equations in βj
References
Bastos, L.S., DA, Rosa, J.M.C., 2013. Predicting probabilities for
the 2010 FIFA world cup games using a Poisson-Gamma
model. Journal of Applied Statistics 40, 1533-1544.
=∑ki=1(yi-⋋i)xij=0,…,p Cameron, A.C., Trivedi, P.K., 2013. Regression analysis of count
Usually, the equation system must be solved with some data. Cambridge University Press 53, 560-590.
numerical method, because it is not possible to get the Consul, P.C., Famoye, F., 1992. Generalized Poisson regression
likelihood equations in the closed form. model. Communications in Statistics-Theory and
Methods 21, 89-109.
Conclusion Deb, P., Trivedi, P.K., 1997. Demand for Medical Care by the

I
Elderly: A Finite Mixture Approach. Journal of Applied
n regression analysis if the dependent variable does not
Economics 12, 313-336.
follow the assumption of regression like continuity of
response variable in that situation ordinary least square
will not give us valid estimation of regression parameters and
in this situation when dependent variable is count data and
follows a Poisson distribution then it is always better to go for
Poisson regression for fitting the regression model.

© 2022 131