People and machines perform tasks differently. Building optimal systems that include people and machines, requires understanding their respective behavioral properties. The task of decision fusion is considered and the performance of people is compared to the optimal fusion rule. Our behavioral experiments demonstrate that people perform decision fusion in a stochastic manner dependent on various factors, whereas optimal rule is deterministic. A Bayesian hierarchical model is developed to characterize the observed human behavior. This model captures the differences observed in people at individual level, crowd level, and population level. The implications of such a model on developing large-scale human-machine systems are presented by developing optimal decision fusion trees with both human and machine agents.