Model validation through the linear regression fit to actual versus predicted values

Abstract Validation of predictive models in agriculture is often carried out with the simultaneous F test of zero intercept and unit slope when regressing predicted outputs on real-system values or vice versa. Such a test is not suitable for this intended use, because it does not take into consideration the behaviour of the non-controlled part (the lack of fit) of such models. Moreover, the test is not informative when several models are available and, consequently, it would not be able to pick up the best one. A correct proof for model validation should be based on the mean square error of the models being evaluated.