0

I would like to compare actual with predicted values and now if the predicted values are on average above or below the 1:1 line which represents complete agreement between model and reality.

All of the methods for measuring error such as MAE,RMSE,etc only provide the magnitude of the error but not the direction. How do I know if I am overall over or under-predicting actual values? And since these values span across several order of magnitude, should they be logged before calculating the error?

I would appreciate some advice that can be understood by non-statisticians and easily implemented in R.

In case it might matter, the predicted values do not come from a statistical model but a mechanistic model (biophysical) so there's no variability associated to them. All values, predicted and actual, and their difference can be positive or negative or zero.

kjetil b halvorsen
  • 63,378
  • 26
  • 142
  • 467
Herman Toothrot
  • 249
  • 2
  • 10
  • One common choice is explained at https://stats.stackexchange.com/questions/251600. – whuber Oct 27 '20 at 18:08
  • @whuber thanks for the suggestion but that answer is way over my head, I am not even sure how it relates to my question. ELI5 – Herman Toothrot Oct 27 '20 at 19:34
  • It might help to peruse some of our [threads concerning loss functions](https://stats.stackexchange.com/search?q=loss+function), then. – whuber Oct 27 '20 at 19:35

0 Answers0