1. Introduction Social media provides a platform for sharing content with an unprecedented freedom and ease. They help individuals to find friends and job opportunities and firms to advertise and build business networks. The other side of the coin, however, is that social media can facilitate sharing of simplistic, false and inflammatory messages, and increasing anecdotal evidence suggests that online outrage may lead to violence offline. For example, in 2016, Human Rights Watch (2016) warned Facebook that the platform could be used for propaganda, censorship and surveillance. A UN investigation in Myanmar found that Facebook had been instrumental in spreading anti-Muslim hate speech (Human Rights Council 2018), and a number of NGO’s active in Myanmar also criticized Facebook for not preventing the spread of hate speech on the platform (Mozur 2018). As a result, the CEO of Facebook was called to testify before the U.S. Congress about Facebook’s role in ethnic violence in Myanmar, as well as about data privacy and misinformation campaigns during the U.S. Presidential election (The Washington Post 2018). This paper investigates whether social media affected the intensity of ethnic conflict. Due to the endogenous nature of social media use, causal estimates of its effects are still scarce.1 Furthermore, it is not obvious how to distinguishing the potential effects of social media from those of the broader internet. To this end, I focus on the ongoing Myanmar conflict since in such context internet is mainly accessed via Facebook. Indeed, an important driver of increased internet penetration in many developing countries has been the emergence of zero rated apps. Zero rating means that a mobile network provider agrees to waive data charges associated with a particular app. As the cost of internet remains prohibitive for many people, zero rated content may often be the only justifiable way to access internet (Eisenach 2015). The rapid spread of Facebook in the developing world has led to situations in which Facebook effectively is the internet. A number of reports describe Facebook being so popular that it is considered synonymous to the wider internet. Anecdotal evidence suggests that many users do not know how to access other websites, or do not know they are accessing internet (see e.g. PRI 2017; Regan 2019). This paper focuses on Facebook’s zero rating campaign which was offered in Myanmar between June, 2016 and September, 2017. As the service could only be accessed with a SIM card from one provider, Myanma Post and Telecom (MPT), my empirical strategy is to use mobile phone coverage by MPT as an exogenous determinant of social media use. With information on the locations of cell phone towers, I use a model of electromagnetic signal propagation to calculate the predicted cell phone coverage (Olken 2009). The signal strength in a given location is primarily determined by distance to a cell tower and the terrain between a location and a cell tower. Cell phone coverage provided by other 1. Few recent exceptions I will discuss in Section 2 are Enikolopov et al. (2020), Bursztyn et al. (2019), Campante, Durante, and Sobbrio (2018), Falck, Gold, and Heblich (2014), and Czernich (2012). 2