Critérios de informação: como eles se comportam em diferentes modelos

Abstract

A phenomenon can be explained by models, which are intended to illustrate certain aspects of the problem, without, however, represent all details and consequently no loss of information. To avoid compromising the understanding of the phenomenon under study, this loss should be minimal. Often more than one model can describe the same phenomenon, given that there is no single methodology to be followed. Thus, when faced with two (or more models) is natural questioning: Among these models which is the most suitable? A good model should be able to balance the quality of fit and complexity, which is generally measured by the number of parameters present here, as more parameters, is more complex, and therefore more difficult to explain. Thus, the selection of the best model becomes then evident. There are several methods used for this purpose, including the Akaike information criterion (AIC), Akaike information criterion corrected (AICc) and Bayesian information criterion (BIC). In this perspective, the aim of this work is to use and evaluate the criteria AIC, AICc and BIC in different areas: selection of standard models, time series and growth models. Simulation studies in each of the three areas were developed, as an application to real data is performed in each of the areas.