Linear vs. Log?

[Kevin]

How do you decide to use which when it's not specifically stated?



Kevin

[Gabriel Bowers]

If it's a question for economy of scale or learning economies, then you will need to use a log-linear fit.

 

Otherwise, you may be asked to do both.

[Adi Aloni]

Hi Kevin
My understanding is that you evaluate all the statistical indicators to
decide which one is better: R^2, standard errors, F tests, t tests, DW and
the different visual tests. If one is clearly better than the other, you
choose it. It's also acceptable to say that there is no significant
difference, like we had in one of the HW problems.
Adi.

[Carla Nunes]

Also, if he doesn't ask you to choose one or the other, he might ask which one provided a better forecast.  You can check that based on the RMSE/MAPE.  The one w/ lowest forecast error is the best.

As Adi said, the Rsquare, standard errors, etc will also hint to which model is better.

[David Song]

also note that in Log-linear form, the coefficients are interpreted to
be percentage changes, in linear they are unit changes

[Gabriel Bowers]

Am I correct in saying that the long-linear forecast needs to be converted back to linear before checking RMSE or MAPE?

Gabriel

On Tue, Mar 15, 2011 at 12:20 PM, Carla Nunes <ccrn...@gmail.com> wrote:

[Carla Nunes]

Yes, that's correct.