Simulation of the accuracy and coverage in experiments including mixtures with binomial and poisson responses studied using generalized linear model and ordinary least squares aproaches

Abstract

This article provides an analysis of experimental designs simplex centroid

mixtures with binomial and Poisson responses using a Generalized

Linear Model. Throughout a simulation process, in which a model was

assumed known, it was compared accuracy and coverage of the confidence

intervals around the expected average obtained under a model

based on normal theory with that provided by a Generalized Linear

Model. Simulations were carried out using specialized software, and

random perturbation to the true average expected in each experimental

point was introduced repeatedly. The new response vectors were analyzed

in each of the two models. Accuracy and coverage were reported

in each case depending on the magnitude of the induced disturbance

and were used as a measure of comparison of estimation uncertainty

and to evaluate the performance of both models. In an experimental

design with mixtures including no normal reponse, a Generalized Linear

Model produced better results in terms of accuracy and coverage than

an Ordinary Least Squares based on a transformation.