I'm using multiple linear regression to identify what affects discounts, customers receive. Therefore, I interviewed salesmen and collected data based on their assumptions (e.g. weather data, pricees in other stores etc.).
Now I have 25 independent variables and 30000 observations. Is it neccessary to select the best variables out of the 25 variables?
I think that I just could perform the regression on all variables to see whether there is support for the assumptions from the interviews. If variables are not significant I would drop them and create a new regression model on the remaing ones.