boundaries, locations of towns, and railway and road networks. I measure distances from the township centroid to the nearest major city (capital, state/region capital or district town), railway, major road, cell phone tower by MPT, and cell phone tower by another mobile network provider. I use the SRTM elevation data to complement the data with topographic characteristics. 5. Empirical Strategy My empirical strategy is to use mobile phone coverage by a mobile network provider offering the zero-rated plan as an exogenous determinant of social media use. The aim is to compare otherwise similar locations that were differently exposed to Facebook access. To identify the causal effect of mobile phone coverage on probability of conflict, variation in mobile phone coverage must be uncorrelated with all other determinants of the outcome. The endogeneity concern is that cell towers are located strategically in areas that are more prone to conflict. I exploit plausibly exogenous local variation in cell phone signal strength, which is due to topographic variation between cell phone towers and receiver locations. First, I use the Irregular Terrain Model to predict where cell phone signal is strong enough for reception. I then compute the share of each township with reception, and use that as the main independent variable. I conduct both cross-sectional and panel data analysis. 5.1. Linear Probability Model In the cross-sectional analysis I estimate the following linear probability model Yi = βCoverageF Bi + δCoveragei + Xi0 γ + λd + εi (1) where Yi is the outcome in township i, CoverageFB i is the predicted cell phone coverage by MPT and β is the key parameter of interest. To distinguish the effect of Facebook access from cell phone coverage in general, I control for cell phone coverage from other providers, denoted by Coverage i . Xi is a set of township level controls, λd is a district fixed effect, and εi is the error term. The source of exogenous variation that I exploit comes only from terrain differences between locations and cell towers. I control for the factors that might be correlated both with incidence of conflict and having cell phone coverage. Because it is likely that the cell phone towers are located so as to maximize covered population, to control for the demand factors, I include controls for log population, log population density, dummy for below median urban rate, share of 15-64-year-olds, share of population with no ID, share of households with electricity, mobile phone, landline phone, and internet at home. As geographic and topographic characteristics of a town can influence the cost of providing 13