An experimental study of persuasion bias and social influence in networks

Abstract

In many areas of social life, individuals receive information about a particular issue of interest from multiple sources. When these sources are connected through a network, then proper aggregation of this information by an individual involves taking into account the structure of this network. The inability to aggregate properly may lead to various types of distortions. In our experiment, four agents all want to find out the value of a particular parameter unknown to all. Agents receive private signals about the parameter and can communicate their estimates of the parameter repeatedly through a network, the structure of which is known by all players. We present results from experiments with three different networks. We find that the information of agents who have more outgoing links in a network gets more weight in the information aggregation of the other agents than under optimal updating. Our results are consistent with the model of "persuasion bias" of DeMarzo et al. (2003).

Dispersed Information and the Origins of Aggregate Fluctuations

“…DeMarzo et al (2003) argue that agents may have difficulties with the weighting process and propose a model in which agents are subject to persuasion bias, i.e., they fail to properly account for possible repetitions of the information they receive. Brandts et al (2015) present experimental evidence consistent with the model of persuasion bias of DeMarzo et al (2003).…”

“…DeMarzo et al (2003) argue that persuasion bias, defined as the individual failure to properly adjust for possible repetitions when processing information, is consistent with psychological evidence, while Brandts et al (2015) present experimental evidence showing that hat agents have difficulties in assessing correctly the information they receive. It is highly unlikely that agents in a complex network are able to objectively assess the precision of information embedded in the actions of other agents and to understand how to manage the information in order to avoid the persuasion bias.…”

Gender Bias in Opinion Aggregation

“…However, there are also results showing that certain forms of structured communication can lead to information cascades where people fail to learn the truth, but instead herd on wrong information (Anderson and Holt, 1997). The difficulty of reaching a consensus and learning the truth has also been pointed out in recent experimental literature on opinion dynamics in social networks (Corrazini et al., 2012; Brandts et al., 2015; Chandrasekhar et al., 2015; Grimm and Mengel, 2020). Other reasons for failing to learn an objective truth are social pressure, as in the classic experiment by Asch (1995), or a desire to be perceived favorably by other group members (Isenberg, 1986).…”

Treasure Hunt: Social Learning in the Field

“…These papers do not focus on information aggregation. The foundational model of naive information aggregation in 4 Also closely related are the lab experiments on observational learning by Choi, Gale and Kariv (2012) who test for the implications of a Bayesian learning model in three-player networks; by Mueller-Frank and Neri (2014) who test for general properties of the rules of thumb people use in updating; and by Corazzini, Pavesi, Petrovich and Stanca (2012) and Brandts, Giritligil and Weber (2014) who test between different variants of boundedly rational models. These studies too feature restricted communication and perfect diffusion.…”