# How do I determine which functional form is correct?

I have an assignment that asks me to explore different series of data with respect to the effects that an incinerator built in 1981 in Massachusetts, has on the price of houses in that area.

I have around 30 different sets of data with regards to the houses ranging from 'number of bathrooms', 'distance from the incinerator', 'square footage of house' to things like 'price before incinerator was built' and the 'age of the house'.

I've started off with which variables I believe explain the changes (or no changes) in the house prices in the area and have around 6 explanatory variables with 'price' as the dependent. I know the Ramsey RESET test can be used to confirm functional form is correct; however, I'm not sure on when I should log a variable, square a variable or just keep it a linear functional form?

Should I adopt the attitude of trying out different functional forms on different variables until I come to a point where my regression passes the RESET test, White test for heteroscedasticity and the Normality tests, or should I try and think about what the data looks like before I start taking the log of variables randomly.

Thanks for any responses in advance, sorry if this question seems strikingly simple, I'm new to all this.

-
I would be wary of applying the test to many different scenarios as you may "unwittingly happen onto a solution by chance". What I mean is that if you repeat the test many times, you are not using the original test anymore (most tests are not designed to handle being used repeatedly on the same data). See this paper by Salzberg. –  Andrew Apr 21 '12 at 19:52
Thanks for the quick reply and link to paper –  Aaron Apr 21 '12 at 20:47
I've added the "model selection" tag. Browsing through the previous Q and A on this topic should get you started on some of the issues raised in your question and @Andrew's answer. stats.stackexchange.com/questions/tagged/… –  Peter Ellis Apr 21 '12 at 21:57
Thanks Peter, much appreciated. –  Aaron Apr 21 '12 at 22:46
Do you have just cross-sectional data (ie at one point in time) or longitudinal (multiple observations per house). If the former, it's going to be difficult to explain the "change in prices", although still not impossible to go some way in that direction. But if you've got longitudinal data that complicates things quite substantially too. –  Peter Ellis Apr 21 '12 at 22:48
show 1 more comment