Beyond Salah: Does Exposure to Muslim Celebrities Reduce Islamophobia?

Can exposure to celebrities from stigmatized groups reduce prejudice?

In our 2021 paper, we answer this question using the case of Mohamed Salah — a visibly Muslim, elite soccer player (Alrababah 2021). Using data on hate crime reports throughout England and 15 million tweets from British soccer fans, we find that after Salah joined Liverpool F.C., hate crimes in the Liverpool area dropped by 16% compared with a synthetic control, and Liverpool F.C. fans halved their rates of posting anti-Muslim tweets relative to fans of other top-flight clubs. An original survey experiment suggests that the salience of Salah’s Muslim identity enabled positive feelings toward Salah to generalize to Muslims more broadly. Our findings provide support for the parasocial contact hypothesis — indicating that positive exposure to out-group celebrities via the mass media can spark real-world behavioral changes in prejudice.

This result begs the question, however –- is Salah unique?

Under what conditions should we expect to see “Salah effects” among other Muslim players? Is there something special in Liverpool confounding the effect? Is there heterogeneity in fan responses that is relevant for policy? Perhaps most importantly — does Islamophobia take hold once more if Muslim players stop scoring? During our study period, Salah had an outstanding season at the individual, club, and national team levels. The external validity and mechanisms underpinning “Salah effects” thus remain unclear, limiting our understanding of how and when Muslim celebrities can reduce prejudice at scale.

Research Design, Data, and Outcomes

To answer these questions, we move away from the focus on one player in one country, and instead expand our data to include thousands of players in a 7-year period across 7 European countries: England, Germany, France, Italy, Spain, Portugal and the Netherlands. Replicating the event study style approach in Alrababah et al. 2021, we first collect data on identifiably Muslim players, and add variables on other player characteristics like nationality, ethnicity, and performance. We plan to scrape player data on ethnicity and nationality from games like FIFA and Fantasy apps, which we have successfully done for each of the 500+ players in the English Premier League as a proof of concept. Information on player religion comes from triangulating information from public sources, including players’ social media accounts, that indicate an affiliation with Islam or Muslim communities. Second, we use Twitter data made available to us by coauthor Julian Hinz – a sample of 1% sample of all tweets — to measure our main outcome of interest; the portion of tweets about Islam or Muslims that are Islamophobic. To identify relevant tweets, we begin by identifying all tweets about Muslims using a generic keyword search and using word2vec, a neural network that processes text, to find other relevant terms in the data. Successfully using this method in our prior study resulted in scraping 15 million fan tweets, filtered down to 44,000 tweets that reference Muslims specifically. We then take a sample of relevant tweets and train three native English speakers to code each tweet at anti-Muslim or not. Filtering on the Twitter users’ geolocation, and extracting their history of team follows, we know where fans are located and which team they support (and by extension, which teams they consider rivals).

Third, we study heterogeneity in Islamophobic fan behavior based on context-level variables like the ethnic composition of the fans’ home city, team-level variables like the earlier presence of Muslim players on the squad, and player-level variables like nationality, ethnicity, performance, and the salience of the players’ Muslim identity. To do so, we propose collecting original data on players’ behaviors (e.g. praying) when starting games and celebrating goals, as well as calculating an index for the “Muslimness” of the players’ name (based on the F-Index used to calculate name “Foreignness” among immigrants; see Abramitzky, Boustan and Eriksson [2020]). Finally, we aim to pin down causal mechanisms through an original survey experiment of English Premier League fans. To complement our observational analyses, we will run social media ads targeting soccer fans living in the U.K. We will randomize treatments that stress the salience of the players’ Muslim identity (high vs. low salience), their performance (good vs. bad), and their team affiliation (team vs. national). In doing so, we combine descriptive trends on how these factors correlate with fan behavior with causally-identified evidence.