Britain has had enough of experts? Social networks and the Brexit referendum

We investigate the impact of social media on the 2016 referendum on the United Kingdom membership of the European Union. We leverage 18 million geo-located Twitter messages originating from the UK in the weeks before the referendum. Using electoral wards as unit of observation, we explore how exogenous variation in Twitter exposure affected the vote share in favor of leaving the EU. Our estimates suggest that in electoral wards less exposed to Twitter the percentage who voted to leave the EU was greater. This is confirmed across several specifications and approaches, including two very different IV identification strategies to address the non-randomness of Twitter usage. To interpret our findings, we propose a model of how bounded rational voters learn in social media networks vulnerable to fake news, and we validate the theoretical framework by estimating how Remain and Leave tweets propagated differently on Twitter in the two months leading to the EU referendum.