Remember that all models are wrong; the practical question is how wrong do they have to be to not be useful.
George E. P. Box, Norman R. Draper (2007). “Response Surfaces, Mixtures, and Ridge Analyses”, p.63, John Wiley & Sons
