Table 5: Difference-in-differences estimates on different conflict types CoverageFB·Treat CoverageFB·Post Treat Post Observations Data Township FE Month-Year FE Coerce (1) Assault (2) Fight (3) Mass violence (4) Battle (5) Civilians (6) Explosion (7) −0.008 (0.007) −0.012 (0.007) 0.091∗∗∗ (0.016) 0.052∗∗∗ (0.012) −0.002 (0.004) 0.0003 (0.004) 0.031∗∗∗ (0.008) 0.015∗ (0.009) −0.003 (0.006) 0.005 (0.007) 0.077∗∗∗ (0.016) 0.041∗∗ (0.016) −0.001 (0.002) −0.003 (0.005) 0.009∗∗ (0.004) −0.003 (0.006) −0.002 (0.004) −0.001 (0.005) 0.012 (0.010) 0.003 (0.012) 0.001 (0.002) −0.004∗ (0.002) 0.023∗∗ (0.009) 0.032∗∗ (0.013) −0.006 (0.004) −0.007∗∗ (0.003) 0.006 (0.007) 0.006 (0.009) 20790 GDELT X X 20790 GDELT X X 20790 GDELT X X 20790 GDELT X X 20790 ACLED X X 20790 ACLED X X 20790 ACLED X X ∗ p<0.1; ∗∗ p<0.05; ∗∗∗ p<0.01. Robust standard errors clustered at district level are reported in parentheses. All regressions include township and month-year fixed effects. The predictors are standardized. There are 20,790 township-month observations. book availability is associated with a small increase in probability of conflict. Although the estimates are small and only marginally significant, they demonstrate that there is important regional heterogeneity. Moreover, the estimates are likely biased down as they do not account for the large number of Rohingya fleeing from Myanmar during the conflict. There are reports of completely burned down Rohingya villages, and an influx of people filling into refugee camps in Bangladesh, which could mechanically reduce subsequent violence.11 The estimated effects on number of conflict events are very imprecise, but suggest a negative effect on the intensive margin. Examining different conflict types reveals that increasing probability of conflict is driven particularly by increased fighting (see Table C.3 in the Appendix). Although the Rohingya have been subjected to discrimination for decades, the antiMuslim hate campaign and Buddhist nationalism has intensified during the past decade. Violence in the Rakhine state flared up in 2012 and since then there have been increasing reports of attacks, particularly against the Rohingya (Human Rights Council 2018). Therefore, these results are consistent with Adena et al. (2015) who show that the effectiveness of propaganda varies with the receivers’ predisposition towards the message. Similarly, Bursztyn et al. (2019) suggest that social media use may aggravate xenophobic attitudes and lead to more hate crimes when intolerant views are already prevalent. The results presented in this section support the view that pre-existing ethnic tensions and animosity may be an important determinant for the impact of social media in conflict. 11. For information on the refugee crisis, see https://www.unocha.org/rohingya-refugee-crisis. 20