The Data On Nonviolent Campaigns

Start with the question, because the question is better than most people's version of it.

It is not whether nonviolence is morally superior. That is a different argument and it has been going for centuries. The question here is narrow and empirical. When a large movement sets out to remove a government, expel an occupier, or win independence, does it succeed more often when it uses violence or when it does not?

Until about twenty years ago, nobody had checked properly.

What They Did

Erica Chenoweth and Maria Stephan built a dataset of major resistance campaigns from 1900 onward, both armed and unarmed. The original version held 323 of them. To be included, a campaign had to be large, sustained, and aiming at a maximum demand. Toppling a government. Ending an occupation. Breaking away as a new country.

Chenoweth went in expecting the opposite of what she found. That detail matters and it comes back at the end of this lesson.

The headline result is that the nonviolent campaigns succeeded roughly twice as often as the violent ones. Not slightly more often. About double.

Success was defined narrowly and the definition is doing real work, so look at it. A campaign counted as successful only if it fully achieved its stated goal within a year of its peak, and only if the campaign could be shown to have caused that outcome rather than happening to be nearby. That is a hard bar. It excludes a great many campaigns that changed something without getting what they demanded.

Why

The main reason turns out to be arithmetic rather than morality.

Nonviolent campaigns attract far more people, and the gap is enormous. Armed struggle recruits from a narrow band. You need to be physically capable, willing to kill, willing to die, and usually young and male and free of dependents. A strike, a boycott, a mass refusal to show up, or a general slowdown asks for none of that. A sixty year old accountant can participate. So can a mother of three, a nurse, a shopkeeper, a soldier's cousin.

Numbers then convert into something else, which is Lessons 7.2 and 7.3.

The finding is not that nonviolence is nicer. It is that it is easier to join, and joining is what decides these things.

Now The Objections

Here is where the course either keeps its word or does not.

The data may be missing the failures. Campaigns get into a dataset by leaving a record. Large, long running movements leave records. A nonviolent movement crushed in its third week may leave almost nothing, and it is more likely to vanish from history than an armed group that fought for a decade. The people who built the data acknowledge this openly in their own documentation. If the missing cases are mostly failed nonviolent campaigns, the gap is smaller than it looks.

The comparison may not be fair. This is the strongest objection and it deserves your attention. Movements do not choose their methods at random. A campaign may go unarmed because the situation already allowed it, meaning a regime that will not massacre everyone, some press, some allies, some room. A campaign may take up weapons because everything else was already closed off. If so, the nonviolent cases start with better hands, and the data would show a gap even if the method changed nothing at all. Nobody has fully solved this, and honest researchers on both sides say so.

The categories are messier than they sound. Real movements are mixed. Most large unarmed campaigns have some rock throwing, some burned cars, some armed faction operating alongside them. Where you draw the line changes the results, and researchers draw it differently.

Other scholars found cases that were left out. They did, and it produced a real argument in the journals, and the dataset has been revised more than once since.

What Happened To The Finding

And now the part that makes this lesson worth your time.

The finding got worse.

As the data was extended past 2006, the success rate of nonviolent campaigns turned out to be falling, and falling hard. It peaked in the 1990s at around sixty five percent. Since 2010 it has run near a third. Campaigns have also gotten smaller, drawing a much thinner slice of the population than they used to.

Chenoweth has published this herself. It cuts against her own most famous result, and she published it anyway.

Governments appear to have learned. They screen and rotate their security forces to prevent defection, and deploy units with no ties to the people they are facing. They frame movements as foreign operations. They surveil, flood the information space rather than simply censoring it, and trade techniques with each other. Meanwhile a great deal of what now gets called a movement is a large crowd that assembles, is photographed, and goes home.

So the honest current statement is this. Unarmed campaigns have outperformed armed ones by a wide margin over the last century, the size of that margin is contested, and the advantage has shrunk considerably in recent years against opponents who have studied the same research you are reading about now.

That is less satisfying than the version on the poster. It is what the evidence supports.

The Standard

Notice what just happened, because it is the whole reason this finding is usable and you are going to need it in the next module.

The data was assembled so other people could check it, and they did. It was published, downloadable, with the coding rules written out.

Some of the people checking wanted it to be false. Not all of them. Enough.

Some of their objections held up, and changed how the finding is now stated. The dataset was revised. The claim narrowed.

The comparison was to what actually happened, not to what a model calculated should have happened. Real campaigns, real outcomes, counted.

Four items. Any claim that clears all four is worth something. Any claim that clears none of them is worth nothing, no matter how much you like it.