Slop.

I found a counter of the study that seems to justify my thoughts:

Through investigation of revolutions in twentieth century Latin America, Lehoucq (2016) discovered that the sample of resistance movements presented by the NAVCO (Nonviolent and Violent Conflict Outcomes) data set suffers from omitted cases of failed nonviolent movements as well as successful violent movements. Others have pointed out that the NAVCO data feature several observational and coding deficiencies pertaining to the dichotomization of violence/nonviolence (Pressman, 2017; Anisin, 2018; Kadivar and Ketchley, 2018). Many campaigns were labeled as being nonviolent when empirically, these movements actually contained a substantial degree of unarmed violence that manifested in the form of rioting, rock throwing, car and building burning, and even the usage of Molotov cocktails.

Fixing the issues, it found the opposite result:

The analysis reveals surprising findings – nonviolence is less effective than previously assumed, while violent insurgency is a bit more effective than assumed, and campaigns featuring reactive unarmed violence and unarmed violence are the most successful of all. To make sense of these results and their implications, subsequent sections of this study present an overview of the soundness of causal mechanisms that have been associated with nonviolent campaign success. It is demonstrated that causal mechanisms that have been linked to nonviolent strategy and campaign success are not necessarily tied to only nonviolent strategies, but are also complimentary to other forms of resistance.