There are some references like this that says in linear regression, this is not necessary for variables to have normal distribution. but I read some books that say otherwise:
Joseph Hilbe in a book called "Modeling Count Data" on page 3 states that
The traditional linear regression model is based on the normal or Gaussian probability distribution.
A book entitled "Geographically Weighted Regression" authored by Fotheringham, Brunsdon and Charlton on page 188 says:
The basic linear regression model assumes a Gaussian distribution for the dependent variable, which has a number of properties: for example, it is symmetrical about some mean value and it admits values anywhere in the interval $(-\infty,\infty). $
Now I am just really confused!
