Table 4: Difference-in-differences estimates Conflict dummy (1) (2) CoverageFB·Treat CoverageFB·Post Treat Post Data source Township FE Month-Year FE No. conflict events (3) (4) −0.004 (0.008) 0.004 (0.008) 0.184∗∗∗ (0.018) 0.136∗∗∗ (0.016) −0.002 (0.004) −0.005 (0.005) 0.026∗∗∗ (0.009) 0.023 (0.015) −0.001 (0.008) 0.004 (0.008) 0.034∗ (0.021) −0.017 (0.012) 0.0004 (0.001) −0.003 (0.002) 0.004 (0.005) 0.001 (0.002) GDELT X X ACLED X X GDELT X X ACLED X X ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. Robust standard errors clustered at district level are reported in parentheses. The dependent variable in columns (1)–(2) is an indicator for conflict in a township, and in columns (3)–(4) number of conflict events per 1,000 population. The predictors are standardized. There are 20,790 township-month observations. estimates on violence against civilians (column (6)) and remote violence (column (7)) are marginally statistically significant, and the coefficient on coercion is also relatively large. The average time trend documented in 4 is driven mostly by increases in coercion and fights, and to a lesser degree by assaults and violence against civilians. 6.3. Effect on the Rohingya Crisis The analysis so far has looked at average effects across all regions of Myanmar. Because there are several ongoing conflicts in different parts of the country, it is possible that the estimates are confounded by different regional effects. The previous literature has shown that the effect of mass media and access to communication technology on conflict may be very context specific (see e.g. Adena et al. 2015). There is a lot of anecdotal evidence that in Myanmar Facebook has been used to spread anti-Muslim and anti-Rohingya propaganda, and therefore it could have had a different impact in the Rohingya conflict. As many of the ethnic conflicts in Myanmar are regional, I now focus on the Rakhine state—home for most of the Rohingya population in Myanmar—to gauge the effect of social media on Rohingya-related conflict. I conduct the analysis at village level, to retain enough units of observation, and only consider conflict measured from GDELT. The share of Rakhine villages that experienced at least one conflict event during the treatment period is slightly higher, and the number of conflict events is significantly higher, than on average in Myanmar. Table 6 presents cross-sectional estimates. In contrast to the previous results, in Rakhine villages Face19