In assessing how individuals in a network estimate a quantity, it is often assumed that they take an average or weighted average of the estimates of others they observe, as in the French–DeGroot learning model. However, individuals may actually copy clusters of similar opinions among their peers, a process that we call conformity. For example, if ten individuals estimate a quantity to be 0 and ten individuals guess 100, then although the average estimate is 50, one may conform by choosing either 0 or 100. Here, we a) extend a recent model of conformity to include both personal beliefs and social information, b) evaluate this model and two models of French–DeGroot averaging on human decision-making data, and c) simulate evolutionary dynamics of each model under different kinds of population structure, including static and adaptive networks. Under many of the conditions we analyze, the conformity model provides a significantly better fit to the data than either of the French–DeGroot models. In addition, evolutionary simulations reveal considerable differences among the models; for example, compared to the French–DeGroot models, the conformity model can produce faster shifts in populations toward “extreme” opinions—where individuals’ estimates of a quantity are either very large or very small—and can reduce individuals’ average estimation accuracy. These findings have implications for research on consensus formation, polarization, and wisdom of crowds.
Denton et al. (Thu,) studied this question.