||[Apr. 8th, 2009|11:21 am]
I'm looking at a linear multiple regression investigating demand functions relative to the price of other goods, and have decided to change the functional form of the linear model.
In doing this I have found that a Log-Log model is far better than a Lin-Lin or Lin-Log model as it passes both normality and hetroskedasity.
Why is this is the case?
Also, I have read that it helps make the variable independant or something (basically i have no clue!)
So does anyone know why you'd change the functional form, other than to try and pass tests of normality etc?
If you do know and happen to know a reference could you add that?