Erica Chenoweth initially thought that only violent protests were effective. However after analyzing 323 movements the results were opposite of what Erica thought:
For the next two years, Chenoweth and Stephan collected data on all violent and nonviolent campaigns from 1900 to 2006 that resulted in the overthrow of a government or in territorial liberation. They created a data set of 323 mass actions. Chenoweth analyzed nearly 160 variables related to success criteria, participant categories, state capacity, and more. The results turned her earlier paradigm on its head — in the aggregate, nonviolent civil resistance was far more effective in producing change.
If campaigns allow their repression to throw the movement into total disarray or they use it as a pretext to militarize their campaign, then they’re essentially co-signing what the regime wants — for the resisters to play on its own playing field. And they’re probably going to get totally crushed.
The survivorship bias is highlighted in the summary:
Meaning all unsuccessful campaigns were not considered.
And unless they have very good parameters there have to be countless non-violent and ineffective campaigns.
If you did this as a ratio of failures over success the non-violent numbers would be sky high compared to the violent ones, the rate of failure would indicate violence is the way.
This whole thing seems like a really wordy way of saying “don’t resist”.