Métodos de estimação baseados em modelos na presença de dados faltantes

Abstract

The missing data are observations that should have been made, but were not for some reason, thus reducing the ability to understand the nature of the phenomenon, in addition to making it difficult to extract information from the analyzed data, since the impact on the results of the studies is not always known. As a considerable part of the statistical techniques were developed to analyze complete data, the missing data usually need to be treated in such a way that the resulting dataset can be analyzed by such established methods. The most used methods to deal with missing data are divided, mainly, between methods of data removal and imputation, being both configurations, in most cases, disadvantageous in terms of the analysis of the final result, either by making the results biased or because we have to work with the uncertainty associated with the imputation of unknown values. In this work, then, we propose some model-based methods for solving the problem of missing data for regression analysis, without having to resort to imputation or removal of information. We verified the performance of the proposed methodologies on simulated data under different scenarios and compared it with the performance of other traditional techniques of imputation and data removal.