Social Simulation Conference 2022

Our work on “Disagreement on the Diagnostic Value of Evidence in Scientific Communities: an Opinion Dynamics Model” (Matteo Michelini, Javier Osorio, Wybo Houkes, Dunja Šešelja and Christian Straßer) has been accepted for the Socia Simulation Conference 2022.

Scientific disagreements may persist even if evidence is shared in a community, as shown for instance by the limited success of so-called consensus conferences (Stegenga, 2016). One possible explanation is that scientists attribute different degrees of significance to the same experimental result due to different background assumptions, methodologies or theoretical commitments. In this paper, we study the opinion dynamics in a community of scientists who disagree on the diagnostic value of evidence. We examine the conditions for scientists to converge on the right hypothesis despite initial disagreement on the significance of experimental results. Moreover, we explain how polarization can emerge in a community of Bayesian agents who are exposed to the same evidence (and treat evidence as certain). We develop an agent-based model (ABM) based on bandit models, common in formal studies of scientific communities (Zollman, 2007), in which agents may assign different diagnostic values to the same piece of evidence. Consequently, although agents share their evidence with each other, they may draw different conclusions from it. While agents share evidence within their social network, they exchange information about their diagnostic values only with those who have a similar opinion to theirs, following the rules of a bounded confidence model (Hegselmann and Krause, 2002). In this way, our ABM naturally extends network epistemology models, such as (O’Connor and Weatherall, 2018) and (Zollman, 2010). Our results show that an initial disagreement on the diagnostic value of evidence can, but does not necessarily, lead to polarization in a scientific community, depending on both the network structure and the confidence interval within which scientists share their opinions. These findings shed light on how different ways of interpreting evidence affect polarization in scenarios in which no detrimental epistemic factors are present, such as biases, deceptive information, uneven access to evidence, uncertainty about evidence etc. Moreover, our analysis suggests additional conditions and interventions for convergence, for example, at consensus conferences.

References.