That study appears to be comparing violent protests to non-violent protests.
At 3.5% of the populace taking up arms (not in protest but in war), that would far outnumber armed government officials in most countries. I don't doubt that a government choosing to concede at the point those 3.5% signaled peacefully they are likely to get violence soon, since the government conceding before that happens indicates they are weak enough to not be able to fight it off. Of course, If you have 3.5% of the populace fighting you can defeat even a horribly asymmetric situation, as the Chechens showed when they gained independence in the first Chechen war against Russia where almost everything beyond small arms were obtained via capture from the enemy.
At best your study shows that a government that capitulates before violence is more likely to be defeated, which makes sense since both sides tend to pick violence when they actually think they can win -- and if both sides think they can win then odds are quite good the odds of winning lie somewhere closer to the middle of the odds if the actors are rational. Concession before violence is more likely to indicate the odds lie outside the middle.
> study appears to be comparing violent protests to non-violent protests
No. The 3.5% figure specifically refers to nonviolent resistance [1].
Would note that “new research suggests that one nonviolent movement, Bahrain in 2011-2014, appears to have decisively failed despite achieving over 6% popular participation at its peak” [2]. But the fact remains that it’s harder to identify ineffective mass protests than effective ones.
> which makes sense since both sides tend to pick violence when they actually think they can win
This assumes a lot more rationality than violent resistance (and corrupt governments) tend to have.
Instead, the evidence is that violent resistance fails more often than nonviolent resistance. In part because violent resistance helps the government consolidate power over its own violence apparatus in a way nonviolent protest inhibits.
I listened to an interview with one of the article's authors, and she said the reason non-violent protests defeat a state willing to order violent crackdowns is because the soldiers performing those crackdowns are regular people. They are not the people who most benefit from an authoritarian state. So when they find themselves being told to beat up or shoot a nun sitting in the street, there's a good chance the soldier would defect.
At 3.5% of the populace taking up arms (not in protest but in war), that would far outnumber armed government officials in most countries. I don't doubt that a government choosing to concede at the point those 3.5% signaled peacefully they are likely to get violence soon, since the government conceding before that happens indicates they are weak enough to not be able to fight it off. Of course, If you have 3.5% of the populace fighting you can defeat even a horribly asymmetric situation, as the Chechens showed when they gained independence in the first Chechen war against Russia where almost everything beyond small arms were obtained via capture from the enemy.
At best your study shows that a government that capitulates before violence is more likely to be defeated, which makes sense since both sides tend to pick violence when they actually think they can win -- and if both sides think they can win then odds are quite good the odds of winning lie somewhere closer to the middle of the odds if the actors are rational. Concession before violence is more likely to indicate the odds lie outside the middle.