Log RMSE to linear RMSE

[#1]

Hi all,

I am trying to compare the RMSE values of a linear fit and log fit model.

However, to my understanding the RMSE for the linear fit is in the same units as the response (for instance MPG), but the log fit has log MPG as unit. Is there a possibility within JMP that allows the calculation of RMSE in the original scale?

Kind regards, Janneman

[#2 · Accepted solution]

Hi @Janneman and welcome: One interpretation of the RMSE in log-scale is this; it is an estimate of the Coefficient of Variation (CV = sigma/mean) in the original scale. And CV is unitless.

So, as an approximation: RMSE(ln-scale) = RMSE(orig)/Mean(orig) gives:

RMSE(orig) = RMSE(ln scale)*Mean(orig).

[#3]

Hi @MRB3855,

Thank you for the warm welcome and thank you for the reply.

I was wondering if you have any suggestions on how to evaluate the performance of a linear model and log-log model?

I am trying to evaluate which model is better in terms of predictability and accuracy.

My first guess was to compare R2 values and RMSE values. 

Kind regards, Janneman