Fitness of 2nd degree polinomials models in fertilizer research

Abstract

The fitness of polinomials models is greatly affected by the model used, the coefficient of variation, localization of the points of maximum response and the number of experiments. Trying to obtain indication of those effects on the fitness of the models in fertilizer research, 2,400 experiments in the factorial design 1/5 (5 x 5 x 5) were simulated. The results showed that the best fitting is obtained when a larger number (ten) of experiments is used, when the experiments have a low coefficient of variation, and when the point of maximum is located at the left of the surface. The quadratic model was the one with the best fitness in this work, independently of the generator model.