An Empirical and Simulation Investigation of Bounded Confidence and Negative Influence in Opinion Dynamics using Stochastic Actor-Oriented Model
Department of Sociology, University of Groningen
14 Sep 2023, 11:45–12:10
Classic opinion dynamics models of assimilation fail to shed light on opinion clustering and polarization. Two micro-processes of social influence, bounded confidence and negative influence, have been proposed as solutions. Empirical evidence for bounded confidence and negative influence is debatable. Two common drawbacks in existing empirical studies are (1) lab experiments that lack external validity; and (2) model designs that do not allow disentangling negative influence from bounded confidence, as well as other social influence mechanisms. In this study, we focus on assessing empirically in a field-setting the evidence for opinion changes being driven by bounded confidence and negative influence, while controlling for additional simultaneous mechanisms identified by earlier field research.
We employ the Stochastic Actor-Oriented Model (SAOM) to analyse The Arnhem School Study (TASS) data, a longitudinal dataset that tracks adolescents’ social networks and opinions’ evolution. Two new SAOM effects were developed: the p-near similarity and q-far similarity effects, capturing the influence of all others in the group (irrespective of friendship ties) whose opinions are sufficiently similar or dissimilar, given the thresholds p and q respectively, allowing us to test the assumptions of bounded confidence and negative influence.
Results show that for the preferences of rap and hip-hop clothing style (ranging from 1 – dislike, to 5 – like), a model including both bounded confidence (threshold = 1) and negative influence from relatively dissimilar others (threshold = 3) provides a good fit to the data and gives a good empirical representation of the development over time of the shape of the opinion distribution as well as opinion polarization. The q-far similarity effect (with q=3) is statistically significant, providing strong empirical evidence for negative influence while no strong evidence for bounded confidence is found.