Regression constant *very* negative

[Eleonora Cambone]

Hi,

I'm investigating the relationship between the prices of houses near airports and the distance from airports and number of flights.
I'm conducting a regression on log of prices of properties on different variables, these being dummies of properties characteristics, year trends (these are dummies too), distances, number of flights, a number of interaction terms (such as number of flights * distance), as well as squared terms of distances and flights.

I am also doing the same regression but instead of using prices I am using deflated prices. When I do the latter I get a negative constant which increases in size (becomes "more negative") as I add more variables. How can I solve this?

Thanks

Eleonora

[William Lisowski]

What is there to solve?

Presumably you are concerned that a predicted house price less than 1 (log price < 0) is nonsensical. But that would be a prediction when all your independent variables are equal to zero. Did anything like that ever occur in your data? If not, then that is a prediction that is well outside the range of your data.

Your model fits your data, perhaps, but it doesn't fit where your data aren't without heroic assumptions. Somewhere between where your data are and where they are all zero, the model ceases to hold. But if you're not interested in predicting the prices of houses set in the middle of the runway of an abandoned airport in year 0, it shouldn't be a concern.

[Eleonora Cambone]

That makes perfect sense. Thanks!