Nonlinear models are appropriate to describe nutrient release, since they estimate quantities of practical interest and they have goodness of fit. Although its inferential process is based on asymptotic arguments, there are ways to know the nonlinearity intensity. In this work, we evaluate the nonlinearity of two nonlinear regression models through the curvatures of Bates e Watts, bias of Box and the least squares estimator sampling properties by simulation study. The data are from the study, over time, of the K release from 4 animal manure in combination with 2 soils. The exponential model was more appropriate in terms of inferential and practical aspects, since by all measures showed lower nonlinearity.
curvature of Bates e Watts; bias of Box; bootstrap; half-life
