Ajuste de modelos de platô de resposta via regressão isotônica

Abstract

Isotonic regression is a data transformation technique that provides the incorporation of previous knowledge about the behavior of the dependent variable, widely characterized as a not decreasing, in function of independent variable. For instance, this situation can be occur in response plateau studies, whose the quality fit of the models can be improved by the use of isotonic regression. Among these models, stand out the quadratic polynomial and non linear exponential. Thus, in front of related issues, this work has as main objective to use isotonic regression in context of response plateau models, to propose a new logarithmic model and a nonparametric estimator. Furthermore, other objective is to use a semi parametric estimator for the parameters of segmented linear-increase model. Respectively to these objectives, were used data sets from Zn requirements in bones of Hubbard broiler and children weight-height data measured monthly. The efficiency of the methodologies was evaluated by Monte Carlo simulation, whose results showed a good performance for the proposed logarithmic model and for the used estimation methods.