Modelos multiníveis

Abstract

Model a dataset is to find a reliable mathematical representation for the phenomenon under study, the ideal case would be to specify a parsimonious model, ie, one that involves a minimum number of parameters to be estimated without compromising the explanation of the chosen model. The classical linear regression models are often applied in various areas of knowledge, such as agronomy, social sciences, management, engineering, biology, education, and so on. However, in situations where the data have investigated a structure hierarchy or grouping, or which are characterized by the presence of aggregated units in other experimental larger units, these data need special treatment. Thus, a statistical technique that has been most used are those of the multilevel models that have been developed for the analysis of data that have a hierarchical structure, or group structure, because they take into account the dependency within each hierarchical level and between levels. This paper aims to present the structure of multilevel models, and compare them with the classical linear regression models. An application from microdata Saeb with the 2011 edition will be used. All procedures and implementations were done using the software free R code.