Um estudo do desempenho do método de entropia máxima generalizada na identificação de efeitos ativos em experimentos fatoriais sem réplicas

Abstract

In the study of the influence of factors in a process, one of the most used approaches is based on unreplicated factorial experiments. It is common to use the normal probability plot to analyze data from this type of experimental design, but there are controversies about its use due to subjectivity and difficulty of interpretation, especially when the magnitude of the effect is moderate. The Generalized maximum entropy and lenth’s method are alternatives proposals to avoid this subjectivity. The objective of this work is to evaluate the performance of the generalized maximum entropy method to identify active effects in unreplicated factorial designs. A simulation study is done considering an experiment with 16 treatments and different scenarios for quantity of active effects and their respective magnitudes. For the evaluation of the performance of the generalized maximum entropy method the type I and type II error percentages are used in comparison with the Lenth method. The simulation results indicate that the generalized maximum entropy approach facilitates the identification of active effects of moderate magnitude. In addition, the performance of the method remains virtually unchanged when the amount of active effects increases.