Open AccessResidual error analyses are best when independent of model runs
Author(s) -
Schultz Colin
Publication year - 2014
Publication title -
eos, transactions american geophysical union
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.316
H-Index - 86
eISSN - 2324-9250
pISSN - 0096-3941
DOI - 10.1002/2014eo370012
Subject(s) - statistician , figuring , residual , limit (mathematics) , computer science , george (robot) , econometrics , statistics , algorithm , mathematics , artificial intelligence , physics , mathematical analysis , optics
As statistician George E. P. Box famously said, “All models are wrong, but some are useful.” Figuring out when, how, why, and the extent to which models are wrong can unveil new underlying patterns in the system being studied and lead to improved predictive capabilities. At the same time, failing to account for these uncertainties can limit a model's usefulness.