Generating and estimating dependency between binary variables

Abstract

The assumption that binary variables are independent, homogeneously distributed and exchangeable may not reflect reality. Therefore, the present work proposes a method that transforms a multivariate simulation problem dependent of binary variables into a hierarchical dependency model, which allows for easier estimation and simulation. The dependency estimation via Bayesian Inference outperforms the Method of Moments, since it presents a lower error measure and guarantees non-negative estimates bounded by 1. The complexity estimation via Bayesian approach was overcome by important resampling methods via Monte Carlo. The binary simulation proposal was applied to model the probability of death in family groups, considering the broken heart syndrome, when the death of one member affects the probability distribution of another one.