User:Kfbd/navco

Sometimes people say that nonviolent campaigns only succeed when there are violent campaigns with the same target. Presumably the idea is that the target of the campaign only listens to the demands of the nonviolent campaign because they are scared that if they don't the violent campaign will become stronger.

The NAVCO data includes, for each campaign a year, a target, whether or not the campaign was nonviolent, and whether or not the campaign succeeded. For each campaign, in each year, I defined "concurrent violent campaign" to mean a violent campaign with the same target and within 10 years of the nonviolent campaign under consideration.

Unsuccessful nonviolent campaigns


The above plot shows the how the number of unsuccessful nonviolent campaigns changes as the number of concurrent violent campaigns increases. For instance, the point furthest to the left shows that there were 925 unsuccessful nonviolent campaigns with 5 or less concurrent violent campaigns. Or the point second from the right shows that there were 72 unsuccessful nonviolent campaigns with 31-35 concurrent violent campaigns.

Successful nonviolent campaigns


Similarly, the above plot shows how the number of successful nonviolent campaigns changes as the number of concurrent violent campaigns increases.

Proportion of successful nonviolent campaigns


The above plot shows how the proportion of successful nonviolent campaigns changes as the number of concurrent violent campaigns increases. Don't know what's up with 30.

Lawks knows what any of this means
Even I can see that my analysis is very clumsily done. An actual social scientist would presumably be horrified. And that's not considering the fact I haven't carefully checked my code to see if it did what I think it did. I want to think more carefully about my definition of concurrent. Then I think I want to see what the probability is that that the NAVCO data was generated by a model in which probability of success of a campaign is proportional to number of concurrent violent campaigns. I think this is a test that statisticians can do, but perhaps I misunderstand something important about statistics to think that. I should also read more of Chenoweth and others' work to see if this is something that has already been addressed.