The noisy voter model under the influence of contrarians

TL;DR

This paper investigates how contrarians influence the phase transition behavior of the noisy voter model, revealing conditions under which bimodal or unimodal opinion distributions emerge in the presence of anti-herding agents.

Contribution

It introduces the effect of contrarians into the noisy voter model and analyzes the resulting phase transition dynamics at the mean-field level, highlighting the conditions for different steady states.

Findings

Contrarians eliminate the bimodal phase when their number is four or more.

The presence of contrarians shifts the system towards a unimodal steady state.

Steady state distributions are well approximated by Gaussian functions under certain conditions.

Abstract

The influence of contrarians on the noisy voter model is studied at the mean-field level. The noisy voter model is a variant of the voter model where agents can adopt two opinions, optimistic or pessimistic, and can change them by means of an imitation (herding) and an intrinsic (noise) mechanisms. An ensemble of noisy voters undergoes a finite-size phase transition, upon increasing the relative importance of the noise to the herding, form a bimodal phase where most of the agents shear the same opinion to a unimodal phase where almost the same fraction of agent are in opposite states. By the inclusion of contrarians we allow for some voters to adopt the opposite opinion of other agents (anti-herding). We first consider the case of only contrarians and show that the only possible steady state is the unimodal one. More generally, when voters and contrarians are present, we show that the bimodal-unimodal transition of the noisy voter model prevails only if the number of contrarians in the system is smaller than four, and their characteristic rates are small enough. For the number of contrarians bigger or equal to four, the voters and the contrarians can be seen only in the unimodal phase. Moreover, if the number of voters and contrarians, as well as the noise and herding rates, are of the same order, then the probability functions of the steady state are very well approximated by the Gaussian distribution.