Orazani et al. (2021)2 is a meta-analysis of lab experiments. The experiments showed people news articles about (real or hypothetical) violent or nonviolent protests and measured their favorability toward the protesters’ cause. The meta-analysis found that:
Nonviolent advocacy had a positive effect (d = 0.25, p < .00001)
Violence had a non-significant negative effect (d = –0.04, 95% CI [–0.19, 0.12], p = .65)
The methodology here is flawed. “People are less favorable to groups that are called violent” is straightforward and easy to anticipate, but it ignores second-order effects. Namely, successful violent organizations tend to intimidate people into refraining from calling them violent, or into redirecting their outrage against violence and disorder onto their opponents. Most of the other studies you cite have the same issue.
Moreover, a literature review on this topic is subject to publication bias. Nobody wants to write the “violent protests work” paper. No reviewer wants to sign off on it. Even where methodology is perfect, these things will shape how results are phrased.
The methodology here is flawed. “People are less favorable to groups that are called violent” is straightforward and easy to anticipate, but it ignores second-order effects. Namely, successful violent organizations tend to intimidate people into refraining from calling them violent, or into redirecting their outrage against violence and disorder onto their opponents. Most of the other studies you cite have the same issue.
Moreover, a literature review on this topic is subject to publication bias. Nobody wants to write the “violent protests work” paper. No reviewer wants to sign off on it. Even where methodology is perfect, these things will shape how results are phrased.