« Back to Results

Experiments on Polarization

Paper Session

Friday, Jan. 7, 2022 10:00 AM - 12:00 PM (EST)

Hosted By: Economic Science Association
  • Chair: Andrea Robbett, Middlebury College

Social Identity and Belief Polarization

Kevin Bauer
Goethe University-Frankfurt
Yan Chen
University of Michigan
Florian Hett
Johannes Gutenberg University Mainz
Michael Kosfeld
Goethe University-Frankfurt


We study the role of social identity and `groupiness' for the polarization of
political beliefs using an online experiment with a representative sample of the US population,
deployed the week before the 2020 US presidential election. In our experiment, approximately
1,000 participants were incentivized to predict policy-sensitive statistics a year after the election
conditional on its outcome. To update their initial predictions, individuals can select or
exogenously receive articles on the respective topics curated from different news sources. We find
that groupiness, i.e. the behavioral sensitivity to group contexts, is systematically associated with
belief polarization. While participants generally display a partisan bias in both the demand for and
processing of information, this bias is significantly amplified for groupy subjects. Hence, the
susceptibility to adopt polarized political opinions seems related to the instinct to understand the
environment through intergroup distinctions. Finally, we test the effectiveness of reducing the
salience of intergroup distinctions to alleviate biases in opinion formation by de-labeling
information sources. We find that de-labeling is effective in reducing bias for information demand
but not for information processing.

Strategic Behavior with Tight and Loose Norms

Eugen Dimant
University of Pennsylvania
Michele Gelfand
Stanford University
Anna Hochleitner
University of Nottingham
Silvia Sonderegger
University of Nottingham


A large body of literature has shown that social norms can be a powerful driver of human behavior, and norm interventions often focus on shifting beliefs about average or majoritarian behavior. This work addresses a less studied aspect of norms, namely their strength or tightness. We investigate how individuals react to different distributions of co-player behavior within the context of a strategic game -- the public good game (PGG) -- where subjects strongly condition their actions on those of their opponent. We focus on the difference between tight (i.e., characterized by low behavioral variance) and loose (i.e., characterized by high behavioral variance) norm environments. Our data show that individual responses vary considerably in these two cases, with loose norms generating more variance in individual responses compared to tight norms. In other words, “tight breeds tight” and “loose breeds loose”. We also find that, in loose environments, personal values have a stronger predictive power for individual behavior than in the case of tight norms. This shows that it is crucial to consider differences in the strength of norms in order to understand their impact on behavior.

Polarization and Group Cooperation

Andrea Robbett
Middlebury College
Peter Hans Matthews
Middlebury College


Does increased partisanship undermine the ability of politically heterogeneous groups to function and cooperate in apolitical settings? On the eve of the 2020 U.S. elections, we conducted an online experiment in which Democrats and Republicans played repeated public goods games, both with and without punishment. Absent punishment, mixed party groups are less cooperative and efficient than homogeneous groups. However, polarized groups fare no worse than those in which political affiliations are unknown. We find no differences in cooperation across groups that are able to punish free-riding behavior. Thus, knowing that one is in a group with likeminded individuals can serve as a substitute for an enforcement mechanism, but polarized groups can, at some efficiency cost, achieve similar contributions when sanctions are possible.

The Supply of Motivated Beliefs

Michael Thaler
Princeton University


There has been a growing interest in better understanding the supply of and demand for
“fake news”. Fake news is especially prevalent in politics (e.g. Lazer et al. 2018, Allcott and
Gentzkow 2017), an environment in which many individuals receiving information prefer
to hold particular beliefs. In such environments, people often distort how they process
information in particular ways that favor their preferred beliefs: motivated reasoning (Kunda
1990; Benabou and Tirole 2011; Kahan 2016). This paper argues that, when an individual
believes that others engage in motivated reasoning, reputation can have a perverse effect:
when the individual is incentivized to appear more truthful, they may choose less truthful
messages. Using a series of online experiments, I first show that people believe that others engage in
politically-driven motivated reasoning about topics such as the U.S. economy, immigration,
and COVID-19, and then show that people often choose less truthful news in order to take
advantage of others’ biases. In particular, the results highlight three complementary features
that lead people who care about their reputation to choose a message that is false: (1) when
the topics evoke motivated beliefs; (2) when they know the receiver’s (political) preferences
over beliefs; and (3) when the truth and the receiver’s preferred message are at odds. I also
show that many people are willing to pay to learn the politics of a receiver, and when they
do, they are more likely to choose false information.
Ro'ee Levy
Massachusetts Institute of Technology
Laura Gee
Tufts University
Catherine Eckel
Texas A&M University
Cesar Martinelli
George Mason University
JEL Classifications
  • C9 - Design of Experiments