On the Methods for Explaining Polarization of Private and Unobservable Opinions

Abstract

Polarized opinions are everywhere. From opposite attitudes towards Hawaiian pizza to the partisan divide in theUnited States, we have experienced enough opinion polarization in recent years. Sadly, it is usually a sign of follow-up criticism when people start to talk about "opinion polarization". The term, which should neutrally describe a widespread social phenomenon, has been proven to be associated with different dismaying outcomes, ranging from hostility to civil wars. Given its harmful consequence, few would doubt the urgent need for a solution to this long-lasting issue, and such a solution requires a deep understanding of opinion polarization in real-life situations. The urgent need has motivated remarkable research efforts in the past few decades. Especially in the domain of computational sociology, a considerable amount of opinion dynamics models have been proposed to explain opinion polarization from microscopic mechanisms that govern interactions between individuals. A common feature of these models, which probably results from their roots in statistic physics, is that opinions are observable and can be directly affected by other opinions just like a "spin" in the famous Ising model. However, opinions in real life are of fundamental difference from "spin" in the sense that it is by nature private and unobservable, whose expression, transmission, and inference largely depend on observable behaviors: even if people are allowed to verbally exchange opinions, how these opinions are translated into words and how these words are inferred by both parties still play a critical role in the dynamics of opinions. Thereby, we could put forward a thesis (which we did, literally) that there is a fundamental discrepancy between opinion polarization in the literature and opinion polarization in real-life situations that would deteriorate our trust in these models, let alone the solutions generated accordingly.