SIAM News Blog
SIAM News
Print

Understanding Social Network Echo Chambers and Polarization

Bottom-up Socio-Cognitive Mechanisms and Top-down Political Campaigns Provide Insight

By Antonio D. Sirianni and Feng Fu

In recent years, social media has become an increasingly important platform for competitive political campaigns. At the same time, partisanship has come to dominate the political sphere and stall political consensus amongst both the political elite and the general population [3, 8]. Researchers are thus seeking to effectively understand the circumstances that can lead to divergence and polarization, and those that might bring about consensus and compromise [2, 4, 6]. 

Figure 1. The structural (1a and 1b) and temporal (1c) evolution of vaccine sentiments of tweets and retweets on Twitter, a common social media platform. Figure courtesy of Feng Fu.
Social media allows users to choose who/what organizations they follow and amplify. Polarization is characterized by both extreme differences of opinion and the tendency of individuals to interact with people whose beliefs are politically similar to their own. These similar parties then further influence the original individual. These interactions can create a highly polarized and segregated network structure in digital environments like Twitter (see Figure 1). At a minimum, the emergence of these social network “echo chambers” weakens the transfer and exchange of information between communities with differing views, thereby impeding healthy public discourse and constructive dialogue. Even worse, homogenous clusters of individuals may be prone to radicalization and extremism. Understanding the causes of political polarization and the emergence of echo chambers is of key importance to the social and information sciences. While polarization does occur organically in social media landscapes, organized political campaigns can also influence, accelerate, and alter this process. 

We recently considered the underlying socio-cognitive mechanisms of political polarization and incorporated the influence of political campaigns on the opinion formation process [7]. Because social interaction and information exchange with people who have different beliefs or behaviors can be stressful or uncomfortable, individuals may respond with a reduced openness to interpersonal influence and search for an environment with self-confirming information. This tendency leads to a dynamic “rewiring” of social networks and forms echo chambers that prohibit the spread of “inconsistent” information, generate social polarization, and impede social consensus. 

We investigate the way in which collective opinion and social network structures co-evolve in the presence of the aforementioned socio-cognitive biases, and explore how two opposing political campaigns can influence this process [7]. Specifically, we characterize and quantify the impacts of confirmation bias (reinforcing opinion resonance via peer influence and campaign influence) and selection bias (avoiding opinion dissonance via network rewiring) on the coevolutionary dynamics of opinion formation and network structure (see Figure 2).

Our work helps us understand the potential causes of and solutions to the problem of partisan polarization and the emergence of echo chambers. In particular, our results show that polarization arises in part from populations that are neither overly open-minded nor overly inflexible in their opinions; the former would lead to uniform homogeneity of opinion and the latter would ultimately form a more diverse set of political clusters. Our model also identifies a potential downside of aggressive political campaigning: individuals who are influenced by candidates might become too radical to persuade their more moderate friends. Integrating real Twitter discourse data with our mechanistic model reveals how the combined effects of absorbing undecided individuals and fluctuating campaign efforts can yield dynamic changes in presidential candidate support levels.

The incorporation of campaign influence into our polarization model offers novel insights about competitive campaign optimization and public messaging. Our work demonstrates that a campaign’s most effective ideological position is not necessarily at the median, contrary to the findings of median voter theorem. That classic result, which is well known in the public choice literature, states that two rational candidates benefit from selecting an ideological position that is identical to the position of the median voter. Median voter theorem is also an example of an “anti-polarization” mechanism — the strategic selection of ideology unites the ideological positions of competitors instead of pushing them apart. However, contemporary models of political influence that focus on voters indicate the ease with which polarization and ideological extremism can emerge. 

Figure 2. Public discourse under the influence of dueling campaigns. Winning majority support in public discourse is determined partly by relative campaign efforts (whose voice is louder) but also more subtly by the divergence of ideological positioning between opposing campaigns. The emergence of a third cluster of undecided individuals along with two major clusters of red and blue supporters is evident. Figure courtesy of Feng Fu.

To fill this theoretical gap, our ongoing preliminary work [5]—which accounts for situations like a voting population with a bimodal ideological distribution—identifies and tests a set of realistic extensions to the rational choice voter model that cause strategic candidates to deviate from the median. We then use evolutionary game theory to show that different model parameters can lead to one of the three outcomes: candidates converge to the center (the classic “median voter” result), candidates pick opposing positions that are less polarized than the electorate, or candidates pick opposing positions that are more polarized than the electorate. If we assume that voters follow candidates and parties just as candidates and parties follow voters, then a distributional tipping point may exist wherein voters and candidates chase each other to ideological extremes. 

Recent analysis of a massive voter dataset demonstrates the concerningly high level of partisan sorting across the U.S. [1]. Such geographic political polarization and partisan isolation—together with social network polarization and echo chambers—poses a serious threat to our democracy. Extreme polarization can undermine efforts to build a more inclusive society and even potentially lead to open conflict and violence. Therefore, mitigating processes that accelerate polarization before they reach a point of no return is of utmost importance.

Our work can inform broader public outreach efforts that promote behaviors for the common interest. For example, the public holds a range of opinions on the importance of vaccines, the threats presented by climate change, and the necessity of nonpharmaceutical measures like social distancing and face covering. Convincing large numbers of interacting citizens to behave in ways that protect their health and promote the common interest has become even more important during the ongoing COVID-19 pandemic. Understanding the interplay of competing centralized influence processes that are orchestrated by organized disinformation campaigns, decentralized peer influence, and individual attitude change is essential for the coordination of pro-social behavior. Our research [7] provides insights into the evolution of public discourse in the current media landscape; leveraging these insights can help solve urgent collective action problems. 


Feng Fu presented this research during a minisymposium at the 2021 SIAM Conference on Applications of Dynamical Systems, which took place virtually in May 2021. 

References
[1] Brown, J.R., & Enos, R.D. (2021). The measurement of partisan sorting for 180 million voters. Nat. Hum. Behav., 5, 998-1008. 
[2] Castellano, C., Fortunato, S., & Loreto, V. (2009). Statistical physics of social dynamics. Rev. Mod. Phys., 81(2), 591.
[3] Evans, T., & Fu, F. (2018). Opinion formation on dynamic networks: Identifying conditions for the emergence of partisan echo chambers. R. Soc. Open Sci., 5(10), 181122.
[4] Fu, F., & Wang, L. (2008). Coevolutionary dynamics of opinions and networks: From diversity to uniformity. Phys. Rev. E, 78(1), 016104.
[5] Jones, M.I., Sirianni, A.D., & Fu, F. (2021). Polarization, abstention, and the median voter theorem. Preprint, arXiv:2103.12847.
[6] Kawakatsu, M., Lelkes, Y., Levin, S.A., & Tarnita, C.E. (2021). Interindividual cooperation mediated by partisanship complicates Madison’s cure for ‘mischiefs of faction’. PNAS. To be published.
[7] Wang, X., Sirianni, A.D., Tang, S., Zheng, Z., & Fu, F. (2020). Public discourse and social network echo chambers driven by socio-cognitive biases. Phys. Rev. X, 10(4), 041042.
[8] Yang, V.C., Abrams, D.M., Kernell, G., & Motter, A.E. (2020). Why are US parties so polarized? A “satisficing” dynamical model. SIAM Rev., 62(3), 646-657.

Antonio Sirianni is a postdoctoral fellow with the Program in Quantitative Social Science and the Department of Sociology at Dartmouth College. He uses computational methods to study social networks, organizational behavior, and polarization. His recent work has been featured in Social Networks and Physical Review X
  Feng Fu is an associate professor of mathematics and biomedical data science at Dartmouth College. His research integrates applied mathematics, social science, computer science, evolutionary biology, and statistical physics for a multidisciplinary approach to understanding human behavior and public health in various contexts that range from political polarization to cancer immunotherapy. 

 

blog comments powered by Disqus