Incorporating variability in the process of identification of the global maximum model in Grade of Membership (GoM): methodological considerations

Abstract

The availability of increasingly complex and multidimensional datasets is one of the main causes for the increase in studies employing multivariate analyses based on fuzzy sets. Even though the Grade of Membership method has been widely used in Brazil for empirical studies in health and social sciences, issues regarding identifiability and stability of the final parameters estimated by GoM 3.4 software have not been thoroughly examined. Given the relevance of unique and stable parameters, Guedes et al. (2010) proposed an empirical method to locate a global maximum (GM) with stable parameters. However, the GM locator does not incorporate variability. In the present article, this limitation is circumvented by employing a weighted statistic – weight global maximum (WGM) – similar to the variation coefficient. This indicator does not affect disproportionately situations with very low mean deviations. The WGM locator is shown to decrease the distance of the identified model from the real structure, when compared with the GM locator.