In a lot of papers you can find the AAD% (Absolute average deviation) to estimate the goodness of a new equation, or to estimate a new method in comparison with literature methods. Do you think it is the best estimator?
I mean, sometimes you have some very small denominators that could affect your results. Although, if you're minimizing with some software, the minimization method probably doesn't use the AAD% (there are a lot of problem with absolute value!).
So I'm wondering if you want to compare your results with other results, is it better to compare with more estimators, such as RMSE (Root mean square error)?
Do you have some other ideas?
Thanks.
As you suggest I feel more comfortable with the RMSE method
as a general approach you may consider any construction of a merit function
e.g. to qualify the goodness of a model to determine the angle of error of an astigmatic component, instead of using the RMSE of the angle of error, I prefer using the RMS of the sine of the angle of error, since this is the productive metric of interest, although the measurement value is the angle itself
In my researches, I usually report AAD%, MSE, R2 and F-value. The latest is essential to justify deviation and degree of freedom (number of parameters vs. number of data points) simultaneously.
Dear Paul,
Thanks a lot! very interesting papers.
Happy new year!!!
Hi Mariano,
Currently I use other estimators! I think that is really important to compare your results with many kind of estimators as possible, because each type of function or data presents different results about the tests. We can't forget this!
- Badges
- Science topic
- Similar topics
- Mathematics
- Applied Mathematics