Amid growing belief that social media exacerbates polarization, little is known about the causal effects of social media on ethnic outgroup attitudes. Through an experiment in Bosnia and Herzegovina where users refrained from Facebook usage during 1 wk of heightened identity salience, we find that—counter expectations—people who deactivated their accounts reported lower outgroup regard than the group that remained active, but this effect was likely conditional on the level of ethnic heterogeneity of respondents’ residence. Additionally, we replicate findings from a study on US users: Deactivation led to a decrease in news knowledge and suggestive improvements in subjective wellbeing. Our findings bring nuance to popular beliefs, frequently dichotomous and simplistic, of social media’s impact on societal dynamics.
Despite the belief that social media is altering intergroup dynamics—bringing people closer or further alienating them from one another—the impact of social media on interethnic attitudes has yet to be rigorously evaluated, especially within areas with tenuous interethnic relations. We report results from a randomized controlled trial in Bosnia and Herzegovina (BiH), exploring the effects of exposure to social media during 1 wk around genocide remembrance in July 2019 on a set of interethnic attitudes of Facebook users. We find evidence that, counter to preregistered expectations, people who deactivated their Facebook profiles report lower regard for ethnic outgroups than those who remained active. Moreover, we present additional evidence suggesting that this effect is likely conditional on the level of ethnic heterogeneity of respondents’ residence. We also extend the analysis to include measures of subjective well-being and knowledge of news. Here, we find that Facebook deactivation leads to suggestive improvements in subjective wellbeing and a decrease in knowledge of current events, replicating results from recent research in the United States in a very different context, thus increasing our confidence in the generalizability of these effects.