Reexamining the Relative Agreement Model of Opinion Dynamics

Abstract

We present a brief history of models of opinion dynamics in groups of agents, and summarise work from the creation of the Bounded Confidence model (Krause 2000; Hegselmann and Krause 2002) through to the more recent development of the Relative Agreement (RA) model (Deffuant et al. 2002; Deffuant 2006). In the RA model, randomly-selected pairs of agents interact, expressing their opinions and their confidence in those opinions; and each agent then updates their own opinion on the basis of the new information. The two seminal RA papers (Deffuant et al. 2002, Deffuant 2006), both published in JASSS, each present simulation results from the RA model that we have attempted to independently replicate. We have surveyed over 150 papers that cite Deffuant et al. 2002, yet have found no prior independent replications of the key empirical results for the RA model presented in the 2002 paper. We have each written a separate implementation of the RA model (one in Java, one in Python, both published in full as appendices to this paper) which we therefore believe to be the first independent replications of the RA model as published in the 2002 JASSS paper. We find that both our implementations of the RA model generate results that are in good agreement with each other, but both of which differ very significantly from those presented by Deffuant et al.. Our results are presented along with an analysis and discussion where we argue from first principles that our results are more plausible than those published in the 2002 JASSS paper. We close with discussion of the relevance of this model, along with future applicability.