Thursday, April 3, 2014

Social Media and Protests

The Cairo protests that ultimately led to the ouster of Hosni Mubarak received a great deal of attention on Twitter—the most used hashtag in 2011 was #egypt—leading much discussion over whether we were seeing  "a Twitter revolution." But the mere fact that protests occurred at the same time as an increase in calls for regime change on social media does not establish that the latter in any way fueled the former. The same factors that lead people to take to the streets might drive behavior online. Absent a credible identification mechanism, there's no way to settle this matter empirically. But one question we might reasonably ask is whether we can at least identify a clear mechanism by which they might do so.

At this point, you're probably saying to yourself, "Um, yeah. Obviously." Because you're probably thinking that social media can help people learn that they're not alone. That Twitter can help break the fear wall. But there are problems with that argument,\(^1\) as Andrew Little discusses in a fascinating new paper, "Communication Technology and Protest."

Brief Overview

Little argues that two related but distinct coordination problems must be solved.\(^2\) Most obviously, people must choose to coordinate on action as opposed to inaction. But it is no less important that they coordinate on tactics—showing up at the same place, at the same time, issuing the same demands, collectively refraining from violence (or not!)—at least if they are to succeed at anything more than generating a few headlines. Neither of these problems is insurmountable, but neither are they trivial. The average person faces uncertainty both over whether others are willing to protest and how they'd do it if they did. The information that spreads rapidly across social media can address both sources of uncertainty, but the effects are less straightforward than you might think.

Simply put, giving people access to better information about how everyone else feels about the regime can either lead them to take to the streets or stay home, because such information can either reveal that anti-regime sentiment is more widely spread than people realized or that it is less so. You probably didn't consider the second possibility, because the many cases where this has happened are non-events. That is, you may not recall the infamous non-protests in Norway in 2011. Or, less flippantly, those in Saudi Arabia.\(^3\) But this is a real concern, and it exemplifies a more general point—new information can either induce caution in previously optimistic actors or risk-taking in previously pessimistic ones.\(^4\) So while there may indeed be specific cases where social media helped people coordinate on action, we must remain mindful that the very ease with which information about opposition to the regime would spread in the era of social media ensures that the absence thereof can quash would-be rebellions before they even begin.

On the other hand, social media's ability to facilitate coordination on tactics unambiguously promotes greater protest activity. It is perhaps therefore unsurprising that China's government doesn't bother censoring criticism of the regime, but does its best to keep people from overcoming the second coordination problem.

So what is the net effect of social media on protest? Little argues, rather persuasively, that when it comes to the first coordination problem, the causal effect of the technology itself is ambiguous but anti-regime content tends to promote protests. However, he rightly notes that empirical estimates of this effect will be biased upwards, particularly those that focus on specific cases where we already know that protests occurred. The confounding problem alluded to above—that protests\(\leftarrow\)anti-regime sentiment\(\rightarrow\)tweets, in addition to tweets\(\rightarrow\)protests—is a serious one. There may well be a truly causal effect here, but we've probably got an inflated sense of its importance. And so the West might want to reconsider the wisdom of policies such as this one.

Details of the Argument

Little analyzes a series of models, adding layers of nuance with each. But most of the important points are illustrated well by the first and simplest model. A representative citizen decides whether to protest or not, and if so, chooses a set of tactics. Though Little allows other factors to enter into consideration later on, the first model assigns payoffs based simply on the true state of societal dissatisfaction with the regime, denoted \(\omega\), and the difference between their choice of tactics, \(t\), and the average tactics of other protesters, \(\theta\). Specifically, \(u_i(\mbox{protest}) = \omega - k(t-\theta)^2\), where \(k\) determines how costly any given level of deviation from the tactics of others is. For simplicity, the utility of staying home is normalized to 0.

Prior to making their decision, \(i\) receives two signals: one about \(\omega\), denoted \(s_\omega\), and one about \(\theta\), denoted \(s_t\). The latter is simply equal to \(\theta + \epsilon_\theta\), where \(\epsilon_\theta\) is a noise term. The former is just a touch more complicated, being equal to \(\rho\omega + \epsilon_\omega\), where \(\epsilon_\omega\) is another noise term and \(\rho\) indexes the proportion of attitudes towards the regime that are aired publicly. In other words, \(\rho\) tells \(i\) how much they should discount any given criticism of the regime. If there is little censoring, than no one critical tweet conveys much information. If, on the other hand, \(i\) knows that people's attitudes are largely kept out of public view, than a single message can be pretty powerful.

The following figure summarizes the results from the first model.

As we can see, increases in the precision of the tactical signal generally increase the probability that \(i\) protests. (The only time they fail to do so is when that probability is already equal to \(1\)). However, if the proportion of attitudes towards the regime that are aired publicly increases, that may either increase or decrease the probability of protests. When \(i\) is initially unlikely to protest, as is likely to be the case when \(i\) is initially pessimistic about the level of anti-regime sentiment, new information can only increase that probability, and the more information \(i\) has access to, the better. But when \(i\) is initially likely to protest, because \(i\) starts from the assumption that the regime is deeply unpopular, new information can only have one effect—to induce caution. It might take a lot to persuade \(i\) to stay home, especially if it is easy to coordinate tactics with other protesters, but when \(i\) would have been certain to take to the streets absent any new information, there's really only one thing greater access to information about how others feel can have.

There's much more detail in the paper, which I encourage you all to read. But that should give you a good sense of why Little concludes that we shouldn't get too worked up about the ability of social media to solve the first coordination problem, though we have cause to be optimistic about the second, and why it therefore shouldn't surprise us that China tolerates criticism of the regime but does everything it can to keep people from coordinating on tactics.

1. One of which I discussed here. Here however, I focus on an even more profound issue.

2. Not familiar with coordination problems? Check out this lecture (slides).

3. Where, according to Bloomberg, one social activist said, "People here feel that the government is a cash cow that should be preserved."

4. On the relevance of this principle to interstate war, see Arena and Wolford (2012).


  1. Thanks for posting this summary of Little's argument, which for many of us is bit too abstruse. The conclusion about finding the necessary balance between criticism, information and tactics in social activism is important.

    1. A tough balance to strike, but yes, and important one.