What means "1/X weighted multiple regression"

Dragan

Super Moderator
#2
What means 1/X weighted multiple regression?
Is it just WLS with weights = "1/X" or I'm missing smth? :confused:


It appears that what is being discussed is heteroscedasticity in the context of where the variance of the error term (u) is proportional to the square of the independent variable X…i.e. E[u^2] = Sigma^2*X^2.


As such, you transform the model by using 1/X as:
Y/X = b0/X + b1 + u/X


so that E[(u/X)^2] = 1/X^2 * E[u^2] = Sigma^2.