MTH-416, REGRESSION ANALYSIS Lecture Notes - Lecture 15: Indian Institute Of Technology Kanpur, Generalized Linear Model, Poisson Regression

• School: Indian Institute of Technology - IIT Kanpur
• Department: Mathematics
• Course: MTH-416, REGRESSION ANALYSIS
• Professor: All

Document Summary

The usual regression model is based on the assumption that the random errors are normally distributed and hence the study variable is normally distributed. In case, the study variable is a dichotomous variable taking only binary values, viz. , 0 and 1, then logistic regression is used where the study variable follows a bernoulli distribution. Similarly, we consider the situations where the study variable is a count variable that represents the count of some relatively rare event. For example, the study variable can be a count of patients with some rare type of disease with one or more explanatory variables like age of variables, haemoglobin level, blood sugar etc. In another example, the study variable can be the number of defects in the car engine of a reputed car maker which again depends on one or more explanatory variables. Assumption of normal or bernoulli distribution for the study variable will not be appropriate in such situations.