Exposure to Social Engagement Metrics Increases Vulnerability to Misinformation

News feeds in virtually all social media platforms include engagement metrics, such as the number of times each post is liked and shared. We find that exposure to these social engagement signals increases the vulnerability of users to misinformation. This finding has important implications for the design of social media interactions in the misinformation age. To reduce the spread of misinformation, we call for technology platforms to rethink the display of social engagement metrics. Further research is needed to investigate whether and how engagement metrics can be presented without amplifying the spread of low-credibility information.

sharing ques onable content and less to fact checking it.
• These findings imply that social media pla orms must rethink how engagement metrics should be displayed such that they do not facilitate the spread of misinforma on or hinder the spread of legi mate informa on.Further research is needed to guard against malicious tampering with engagement metrics at an early stage and to design educa onal interven ons that teach users to priori ze trustworthiness of news sources over engagement metrics.

Implica ons
Online misinforma on is a cri cal societal threat in the digital age, and social media pla orms are a major vehicle used to spread it (Guess et al., 2019;Hameleers et al., 2020;Lazer et al., 2018).As an illustra on, the Interna onal Fact-Checking Network found more than 3,500 false claims related to the coronavirus in less than 3.5 months. 2 Misinforma on can cause serious societal harm in mul ple ways: affec ng public health (Sharma et al., 2020), influencing public policy (Lazer et al., 2018), ins ga ng violence (Arif et al., 2018;Starbird et al., 2014), spreading conspiracies (Samory & Mitra, 2018), reducing trust in authori es (Gupta et al., 2014;Shin & Thorson, 2017;Vosoughi et al., 2018), and increasing polariza on and conflict (Stewart et al., 2018).
Studies have explored the role of environmental, emo onal, and individual factors that impact online misinforma on spread (Coviello et al., 2014;Ferrara & Yang, 2015;Grinberg et al., 2019;Kramer et al., 2014;Yaqub et al., 2020).However, there has been li le empirical research on the effects of interface ele-ments of social media feeds on the spread of misinforma on (Hameleers et al., 2020;Shen et al., 2019).To address this gap, we empirically inves gated how the spread of low-credibility content is affected by exposure to typical social engagement metrics, i.e., the numbers of Likes and Shares shown for a news ar cle.Player behavior in the Fakey game shows near-perfect correla ons between displayed social engagement metrics and player ac ons related to informa on from low-credibility sources.We interpret these results as sugges ng that social engagement metrics amplify people's vulnerability to low-credibility content by making it less likely that people will scru nize poten al misinforma on while making it more likely that they like or share it.For example, consider the recent disinforma on campaign video "Plandemic" related to the COVID-19 pandemic. 3Our results suggest that people may be more likely to endorse the video without verifying the content simply because they see that many other people liked or shared it.
To interpret these findings, consider that the probability of sharing a piece of informa on grows with the number of mes one is exposed to it, a phenomenon called complex contagion (Mønsted et al., 2017;Romero et al., 2011).Social engagement metrics are proxies for mul ple exposures; therefore, they are intended to provide signals about the importance, relevance, and reliability of informa on -all of which contribute to people's decisions to consume and share the informa on.In other words, users are likely to interpret high numbers for engagement metrics for an ar cle as sugges ng that it must be worthy of a en on because many independent sources have validated it by liking or sharing it.
A key weakness in the cogni ve processing of social engagement metrics is the assump on of independence; a malicious en ty can trick people by falsely boos ng engagement metrics to create the percep on that many users endorsed an ar cle.In fact, most disinforma on campaigns rely on inauthen c social media accounts to tamper with engagement metrics, crea ng an ini al appearance of virality that becomes reality once enough humans are deceived (Shao, Ciampaglia, et al., 2018).To prevent misinforma on amplified by fake accounts from going viral, we need sophis cated algorithms capable of early-stage detec on of coordinated behaviors that tamper with social engagement metrics (Hui et al., 2019;Pacheco et al., 2020;K.-C. Yang et al., 2020).
Our findings hold important implica ons for the design of social media pla orms.Further research is needed to inves gate how alterna ve designs of social engagement metrics could reduce their impact on misinforma on sharing (e.g., by hiding engagement metrics or making them less visible for certain posts), without nega vely impac ng the sharing of legi mate and reliable content.A good trade-off between these two conflic ng needs requires a systema c inves ga on of news proper es that can help determine differen al display of social engagement metrics.Such proper es may include types of sources (e.g., unknown/distrusted accounts) and topics (e.g., highly sensi ve or polarizing ma ers with poten al for significant impact on society).
Further research is also needed to design media literacy campaigns, such as Fakey, that teach users to priori ze trustworthiness of sources over engagement metrics when consuming content on social media.Studies could explore the possibility of introducing deliberate pauses when consuming news through a social media feed (Fazio, 2020) and limi ng automated or high-speed sharing.A comprehensive digital literacy approach to reduce the vulnerability of social media users to misinforma on may require a combina on of these interven ons with addi onal ones, such as inocula on (Basol et al., 2020;Roozenbeek & van der Linden, 2019a, 2019b;Roozenbeek et al., 2020), civic online reasoning (McGrew, 2020), cri cal thinking (Lutzke et al., 2019), and examina on of news feeds (Nygren et al., 2019).

Findings
Finding 1: High levels of social engagement result in lower fact checking and higher liking/sharing, especially for low-credibility content.
For each ar cle shown in the game, the player is presented with a photo, a headline, a descrip on, and a randomly generated social engagement number.Based on this informa on, the player can Share, Like, or Fact Check the ar cle (see Figure 1).The player must Share or Like ar cles from mainstream sources and/or Fact Check ar cles from low-credibility sources to earn points in the game.The Methods sec on provides details on source selec on.
We measured the correla on between the social engagement number η displayed to players and the rates at which the corresponding ar cles from low-credibility sources were liked/shared or fact checked by the players.Given the realis cally skewed distribu on of η values, we sorted the data into logarithmic bins based on the shown social engagement numbers.For each bin ⌊log 10 (η + 1)⌋, we calculated the liking/sharing and fact-checking rates across ar cles and players.We measured correla ons using the nonparametric Spearman test as the data is not normally distributed.For ar cles from low-credibility sources, we found a significant posi ve correla on between social engagement level and liking/sharing (Spearman ρ = 0.97, p < 0.001) and a significant nega ve correla on between social engagement level and fact checking (Spearman ρ = −0.97,p < 0.001).We found similar sta s cally significant rela onships between social engagement level and player behavior for mainstream news ar cle as well, however the correla ons are less strong: ρ = 0.66 (p < 0.001) for liking/sharing and ρ = −0.62 (p < 0.001) for fact checking.
Finding 2: People are more vulnerable to low-credibility content that shows high levels of social engagement.
The previous finding is for the whole player popula on, with measures aggregated across all players.To delve further into the effect of social engagement exposure on individual players, we analyzed whether the displayed social engagement number influenced each player's liking/sharing and fact-checking rates for ar cles from low-credibility sources.We treated each player as an independent en ty and categorized social engagement numbers into three bins: low (0 ≤ η < 10 2 ), medium (10 2 ≤ η < 10 5 ), and high (10 5 ≤ η ≤ 10 6 ).Within each social engagement bin, we counted the number of low-credibility ar cles to which each player was exposed.We then calculated the corresponding propor ons of these ar cles that each player liked/shared or fact checked.Figure 2 plots the mean liking/sharing and fact-checking rates for low-credibility ar cles.Although players were more likely to fact check than like or share low-credibility content, Figure 2 shows that the trends observed at the popula on level held at the individual player level as well.
Since the data is not normally distributed (p < 0.05 using the Shapiro-Wilk test for normality), we used the Kruskal-Wallis test to compare differences among the three bins of social engagement.Liking/sharing (χ 2 (2) = 417.14,p < 0.001) and fact checking (χ 2 (2) = 214.26,p < 0.001) rates for low-credibility arcles differed across bins.To determine which levels of social engagement impacted the rates at which lowcredibility ar cles were liked/shared or fact checked, we conducted post-hoc Mann-Whitney tests across all pairs of social engagement bins, with Bonferroni correc on for mul ple tes ng.We found that liking/sharing as well as fact-checking rates were sta s cally significantly different across all bin pairings (p < 0.001).
We employed the same approach to examine liking/sharing and fact-checking rates for mainstream arcles across the three bins of social engagement.Similar to low-credibility ar cles, Kruskal-Wallis tests revealed a sta s cally significant effect of social engagement level on liking/sharing (χ 2 (2) = 161.80,p < 0.001) and fact checking (χ 2 (2) = 576.37,p < 0.001) rates for mainstream ar cles.

Methods
To inves gate the effect of exposure to social engagement metrics on suscep bility to ques onable content, we developed and deployed Fakey,4 an online news literacy game.

Social media simula on
Fakey simulates fact checking on a social media feed.The user interface of Fakey mimics the appearance of either Facebook or Twi er feeds for players who log into the game through the respec ve pla orms.The game provides players with batches of ten news ar cles in the form of a news feed, as shown in Figure 1.Each ar cle consists of elements typically displayed by popular social media pla orms: photo, headline, descrip on, and social engagement.
For each ar cle from mainstream as well as low-credibility sources, the game displays a single social engagement metric indica ng the combined number of Likes and Shares.Having a single metric decreases the cogni ve workload for players and simplifies the experimental design.The metric uses social engagement values drawn randomly from an approximately log-normal distribu on with a maximum possible value (cutoff) of η = 10 6 .The distribu on is such that roughly 69% of the ar cles display engagement values η > 10 2 and roughly 3% display values η > 10 5 .Although the engagement metric simulated in the game is not drawn from empirical data, the randomly generated metric numbers have a heavy tail similar to those typically observed on social media pla orms (Vosoughi et al., 2018).
Below each ar cle is a set of bu ons to Share, Like, Fact Check, or Skip the ar cle or use a Hint (see Figure 1).Before playing the game, players are instructed that Share is equivalent to endorsing an ar cle and sharing it with the world, Like is equivalent to endorsing the ar cle, and Fact Check is a signal that the ar cle is not trusted.A er playing each round of ten ar cles, players have the op on to play another round or check a leaderboard to compare their skill points with those of other players.

Content selec on
We followed the prac ce of analyzing content credibility at the domain (website) level rather than the ar cle level (Bovet & Makse, 2019;Grinberg et al., 2019;Lazer et al., 2018;Pennycook & Rand, 2019;Shao, Ciampaglia, et al., 2018;Shao, Hui, et al., 2018).Each ar cle in the game is selected from one of two types of news sources: mainstream and low-credibility.
For mainstream news, we manually selected 32 sources with a balance of moderate liberal, centrist, and moderate conserva ve views: ABC News Australia, Al Jazeera English, Ars Technica, Associated Press, BBC News, Bloomberg, Business Insider, Buzzfeed, CNBC, CNN, Engadget, Financial Times, Fortune, Mashable, Na onal Geographic, New Scien st, Newsweek, New York Magazine, Recode, Reuters, Techcrunch, The Economist, The Guardian, The Independent, The New York Times, The Next Web, The Telegraph, The Verge, The Wall Street Journal, The Washington Post, Time, and USA Today.The game obtains mainstream news ar cles via the News API. 5e selected the set of low-credibility sources based on flagging by various reputable news and factchecking organiza ons (Shao, Ciampaglia, et al., 2018;Shao, Hui, et al., 2018).The selected low-credibility sources tend to publish fake news, conspiracy theories, clickbait, rumors, junk science, and other types of misinforma on.The game uses the Hoaxy API6 to retrieve ar cles from these low-credibility sources.
For each round, the game randomly selects five ar cles each from mainstream and low-credibility sources.For a given source, any ar cle returned by the News or Hoaxy API is shown to the player regardless of topic, without further selec on or filtering except for ensuring that the same story is not shown to the same player mul ple mes across rounds.

Data collec on
The game is available online through a standard web interface and as a mobile app via the Google Play Store and the Apple App Store.The mobile app is available in the following English-speaking countries: Australia, Canada, United Kingdom, and United States.People from other countries can play the game on the web.
Our analysis uses data from a 19-month deployment of the game, between May 2018 and November 2019.During this period, we adver sed the game through several channels, including social media (Twi er and Facebook), press releases, conferences, keynote presenta ons, and word of mouth.We recorded game sessions involving approximately 8,606 unique players7 and 120,000 news ar cles, approximately half of which were from low-credibility sources.We did not collect demographic informa on, but we collected anonymous data from Google Analy cs embedded by the game's hos ng service.Players originated from the United States (78%), Australia (8%), UK (4%), Canada (3%), Germany (3%), and Bulgaria (2%).

Limita ons
Our news literacy game emulates relevant interface elements of popular social media pla orms, such as Facebook and Twi er, without raising ethical concerns of real-world content manipula on (Kramer et al., 2014).Yet, conduc ng the study in a simulated game environment rather than an actual pla orm presents clear limita ons as the experience and context are not iden cal.For example, we limited cogni ve burden on players by capturing only Like and Share ac ons; these were the earliest ones deployed on social media pla orms and, as such, are the most common across pla orms and the most familiar to users.
The even mix of ar cles from mainstream and low-credibility sources is not necessarily representa ve of the propor on of misinforma on to which social media users are exposed in the wild.Further, the factchecking framing of the game primes players to expect misinforma on, poten ally making it more likely to be spo ed.These factors might make players more suspicious within the game compared to the real world, correspondingly increasing fact-checking rates.However, there is no reason to believe that these factors impact our results regarding the influence of social engagement metrics.
While this study is focused on user interac on elements, other factors related to users and content can affect the spread of ques onable content.To respect privacy, we chose not to collect any player informa on apart from game analy cs.However, knowledge about the background of the players (e.g., educa on, demographics, poli cal affilia on) might provide further insight into vulnerability to misinforma on.Similar refinements in insight would be provided by examining whether certain types of content are more likely to be influenced by social engagement metrics.These are important avenues for future research.

Figure 1 :
Figure 1: A news post in the social media feed simulated by the Fakey game.

Figure 2 :
Figure 2: Mean rates of liking/sharing and fact checking low-credibility ar cles for low, medium, and high social engagement levels.Error bars indicate the standard error.