Andrew Markus Guess

We present comprehensive evidence quantifying the magnitude and scope of false polarization (i.e. the impression that opinions on political issues are more polarized than they actually are) on social media, driven by users' selective posting behavior. Using more than a decade of social media behavior connected to individual ideological attitudes in 1.7 million posts, a series of controlled experiments, and a large global survey covering 40 countries (total N = 49,222), we find that (1) strong ideologues post their opinions significantly more often than moderate individuals, and (2) extreme attitudes are posted more often than moderate attitudes. This general pattern replicates robustly across 40 countries, spanning diverse political cultures. Indeed, across studies we find extreme opinions are overrepresented on social media by at least 6% and up to 22% compared to their true prevalence in our samples. Furthermore, the opinions that users post are more hostile and toxic than the beliefs they keep to themselves. These results were not explained by political interest or consistent left-right differences, suggesting that political extremity has an independent effect on social media behavior. Thus, most of the observed polarization of online discourse in recent years may be attributable to the systematic over-representation of pre-existing extreme political views, rather than the causal effects of platforms.

Despite concern about exposure to content from untrustworthy sources on social media, little is known about the frequency or effects of exposure to their content. We examine 2020 data from all active US adults on Facebook and Instagram to measure exposure to content from Pages, groups, and web domains on Facebook and public accounts on Instagram that repeatedly publish misinformation. We find that average users saw relatively little content in their feeds from these untrustworthy sources; exposure was highly concentrated. A multimonth field experiment during the 2020 election among consenting users reduced feed-based exposure to content from untrustworthy sources by approximately 70% on both platforms but had no measurable effects on numerous preregistered outcomes, even among participants with high pretreatment exposure. Our results demonstrate that a feasible platform intervention can successfully reduce exposure to content from untrustworthy sources but suggest that these changes are unlikely to have immediate effects on attitudes and beliefs.

Deceptive online networks are coordinated efforts that use identity deception to pursue strategic political or financial goals. During the US 2020 elections, these networks reached at least 37 million Facebook and 3 million Instagram users, representing 15% and 2% of the platforms' active US adult users, respectively. Only 3 networks out of 49-1 network with explicitly political aims and 2 that appeared to use politics as a lure for profit-were responsible for over 70% of users reached. Notably, accounts unaffiliated with the networks played an important role in facilitating this reach by resharing content the three networks produced. Deceptive networks, regardless of whether their goals were political or financial, reached users who were older, more conservative, more frequently exposed to content from untrustworthy sources, and spent more time on Facebook.

We study the effects of social media political advertising by randomizing subsets of 36,906 Facebook users and 25,925 Instagram users to have political ads removed from their news feeds for 6 weeks before the 2020 US presidential election. We show that most presidential ads were targeted towards parties’ own supporters and that fundraising ads were the most common. On both Facebook and Instagram, we found no detectable effects of removing political ads on political knowledge, polarization, perceived legitimacy of the election, political participation (including campaign contributions), candidate favourability and turnout. This was true overall and for both Democrats and Republicans separately.

The practices of peer review and the broader landscape of academic publishing are under strain. Submission rates are rising, placing greater demands on editorial and reviewer capacity. The use of LLMs in the production of academic papers threatens to accelerate these trends beyond the breaking point. This article is the product of a meeting of editors of academic journals across political science, sociology, and communication science to discuss the issue of LLMs in academic publishing. We argue that peer review is an essential and irreplaceably human component of social science: if elements of the research process traditionally done by humans are substituted by AI, humans should increase their involvement with the evaluation of research. We present a scenario-casting exercise illustrating four possible equilibria for the incorporation of LLMs into the publication ecosystem, and we discuss the various levers that academic journals have at their disposal to navigate the changing landscape. We emphasize that adaptive policies with built-in evaluation mechanisms, feedback loops, and a capacity for revision are required, along with new streams of metascientific data to remain up to date as AI and its adoption continue to evolve.

An enormous body of literature argues that recommendation algorithms drive political polarization by creating “filter bubbles” and “rabbit holes.” Using four experiments with nearly 9,000 participants, we show that manipulating algorithmic recommendations to create these conditions has limited effects on opinions. Our experiments employ a custom-built video platform with a naturalistic, YouTube-like interface presenting real YouTube videos and recommendations. We experimentally manipulate YouTube’s actual recommendation algorithm to simulate filter bubbles and rabbit holes by presenting ideologically balanced and slanted choices. Our design allows us to intervene in a feedback loop that has confounded the study of algorithmic polarization—the complex interplay between supply of recommendations and user demand for content—to examine downstream effects on policy attitudes. We use over 130,000 experimentally manipulated recommendations and 31,000 platform interactions to estimate how recommendation algorithms alter users’ media consumption decisions and, indirectly, their political attitudes. Our results cast doubt on widely circulating theories of algorithmic polarization by showing that even heavy-handed (although short-term) perturbations of real-world recommendations have limited causal effects on policy attitudes. Given our inability to detect consistent evidence for algorithmic effects, we argue the burden of proof for claims about algorithm-induced polarization has shifted. Our methodology, which captures and modifies the output of real-world recommendation algorithms, offers a path forward for future investigations of black-box artificial intelligence systems. Our findings reveal practical limits to effect sizes that are feasibly detectable in academic experiments.

The shift of public discourse to online platforms has intensified the debate over content moderation by platforms and the regulation of online speech. Designing rules that are met with wide acceptance requires learning about public preferences. We present a visual vignette study using a sample (n=2,622) of German and US citizens that were exposed to 20,976 synthetic social media vignettes mimicking actual cases of hateful speech. We find people’s evaluations to be primarily shaped by message type and severity, and less by contextual factors. While focused measures like deleting hateful content are popular, more extreme sanctions like job loss find little support even in cases of extreme hate. Further evidence suggests in-group favoritism among political partisans. Experimental evidence shows that exposure to hateful speech reduces tolerance of unpopular opinions.

Researchers have tested a variety of interventions to combat misinformation on social media (e.g., accuracy nudges, digital literacy tips, inoculation, debunking). These interventions work via different psychological mechanisms, but all share the goals of increasing recipients’ ability to distinguish between true and false information and/or increasing the veracity of news shared on social media. The current megastudy with 33,233 US-based participants tests nine prominent misinformation interventions in an identical setting using true, false, and misleading health and political news headlines. We find that a wide variety of interventions can improve discernment between true versus false or misleading information during accuracy and sharing judgments. Reducing misinformation belief and sharing is a goal that is accomplishable through multiple strategies targeting different psychological mechanisms.