Benoît Schmutz-Bloch (CREST) joint with Pierre Boyer (CREST), Germain Gauthier (U. Bocconi), Yves Le Yaouanq (CREST) and Vincent Rollet (MIT)
We propose a theory of protest dynamics with heterogeneous protest technology and intensity. The ability to mobilize online reduces the likelihood of coordination failures atboth the extensive (engagement) and intensive (violence) margins. We build a dynamic coordination game with strategic substitutability and endogenous learning, and use it to characterize a crowd-in-then-crowd-out sequence in which social media initially helps launch massive protests, but then encourages radical factions to turn violent, leading moderates to leave the movement. This sequence is illustrated using online and offline data on the 2018 Yellow Vest uprising in France, whose early success and popularity were abruptly undermined by street violence. First, spatial regressions confirm that online and offline mobilizations reinforced each other at the beginning of the movement. Second, our textual analysis reveals that online conversations among protesters progressively radicalized. Using a decomposition with discussant, page, and period fixed effects, we show that (i) half of this trend was due to changes in the composition of online protesters and (ii) more radicalized pages drove out moderate discussants.