One thing I’m working on in my doctoral research is understanding why crowdsourcing works in conflict management and resolution…or should at least logically work based on the various theories of conflict management and resolution developed and refined over the last 40 or so years.  In this post, I’m going to use Kenyan election violence as the tangible example, and propose using the term “crowdsharing” instead of “crowdsourcing” since we’re talking about a process through which a local population shares and responds to information laterally.  This is going to be a math-y one, but bear with me.

First, let’s unpack what I mean by “crowdsharing”, as opposed to ‘crowdsourcing.’  When we talk about crowdsourcing we’re talking about a process where a third party gathers data contributed by the population, and reprocesses it for analysis or redistribution.  For the purposes of conflict and violence prevention though, the discussion has been moving away from this kind of top-down process and is focusing more on how communities share information via social media and mobile phone between each other, facilitating a rapid community level response to potential violence before it starts.  I would argue that this is more a process of sharing as opposed to sourcing, hence using the term “crowdsharing” in this post.

But regardless of “sourcing” versus “sharing”, why would we expect to see a decreased risk of violence at the community level when people are using social media and text messages to share information?  First, let’s narrow our scope on what kind of violence we’re talking about.  I don’t think crowdsharing or crowdsourcing matters much once a region is fully engulfed in warfare.  Instead I’ll be focusing on discrete events of violence, such as election violence or riots.

There are a couple of key things we need to know about violence and cooperation though, before we dig into the meat of the “crowdsharing” argument.

1. Violence is not the preferred outcome among the general population, and violence between ethnic or political groups is relatively rare because they have mechanisms for defusing violence before it starts.  Violence is the outcome of a process of perceived risk and a lack of symmetric information about intent across ethnic or political lines: the outbreak of violence is usually due to an information failure or purposeful misrepresentation of risk by leaders.

2. Events of violence are discrete events that occur during particular periods in the calendar, such as election periods.  Thus, violence starts and stops relative to external events, as opposed to being a state of sustained warfare.

3. When there are more mobile phones and the population knows that they can be used to report violence, reach authorities, and gather information about the environment, citizens will make use of their phones in this pursuit since they do not prefer violent outcomes. Thanks to Axelrod’s work on cooperation, we assume that large-N samples of people want to cooperate, and will find ways to cooperate over time if they are able to communicate and modify their behavior based on emerging knowledge of other actors’ behavior.

From these caveats we assume that a violent event occurs at a particular time, and that by sharing information laterally the population can internally prevent or lessen the impact of violence if they are able to communicate.  In effect successfully organizing violence is conspiratorial in nature; violence is likely when fewer actors are knowledgeable or in control of the information space leading up to an event, and there’s a lack of information sharing between ethnic or political groups.  To talk about this in a formal, very simple mathematical way we’ll borrow a couple of variables from Joseph Wilson, who uses them to make the argument that conspiracy theories are basically statistically bunk.  We start with Wilson’s base variables for the likelihood of a conspiracy succeeding:

P(n) & A(n)

Where P(n) is the probability of an event based on the number of actors (n), and A(n) is a proportionality constant that defines the impact of an event relative to number of actors involved (e.g. as more actors are involved the likelihood of a large-scale conspiracy being successful decreases).  From this we can create a probability function to describe the probability of violence breaking out, which noted earlier is similar in nature to a conspiracy (e.g. the active suppression or misrepresentation of information to create information asymmetries that could lead over-estimates of risk and increased likelihood of violence).  Here’s our function:

P = (An)^(1/n)-1, A < n

In this equation (P) equals the probability of an event as a function of impact (A) and the number of actors involved in the process (n).  As (A) and (n) increase the probability of an event decreases proportionally.  In this model (A) must always be smaller than (n); when A = n in the function p = 1 (100% likelihood of success), and when A > n we get (p) > 1 which is logically false.  With that in mind, we will explore this dynamic mathematically.  In the first test we will assume that no one is aware of political manipulation, or has no means for countering false or inflammatory information.  We will start with five actors organizing violence with a success rate of 90% (p=.9).  This will give us (A), a proportionality constant for impact, to later test probabilities of event likelihood:

.9 = (A5)^(1/5)-1

A = .22

What we see is that to achieve a 90% likelihood of an event with an impact of .22 requires five or fewer conspirators.  To keep this discussion framed around a tangible event, let us assume that .22 is an event equal in magnitude to the 2007/8 election violence in Kenya.  Since most Kenyans do not support violence, the Ushahidi mapping platform was invented so that people could report violence using SMS text messages and these can be seen publicly on a digital map by local citizens as well as the international community as a response to an event of violence.

At the next round of elections, international actors have supported the development of the Ushahidi platform and there has been a push to make sure people know that they can use text messages to report violence and communicate among each other at the local level.  The same political actors are involved in trying to perpetuate violence, but with the increased use of mobile phones the total number of actors involved in the election information sharing process has grown.  What is the likelihood of an event occurring that has the same impact (.22) when 25 actors are involved in the communication process (5 original ‘conspiratorial’ actors, plus 20 local leaders “crowdsharing” via mobile phone):

P = (.22(25))^(1/25)-1

P = .19

Increasing the number of actors who may or may not share the original five political actors desire for political violence by twenty has taken the probability of a .22-level event from 90% to 19%.  Another round of hypothetical elections comes up, and more people are involved in the reporting process, now with buy-in from stakeholders including the international community and government.  Now there are 130 actors in the communication system:

P = (.22(130))^(1/130)-1

P = .03

We now have an event probability of 3%, effectively zero if we are talking about a complex event that only happens once every few years, such as election violence.  This result is logically in line with both Axelrod’s and Fearon and Laitin’s assumptions that given enough time, large enough N, and the ability to communicate, groups will find ways to cooperate and resolve conflict in asymmetric information environments.

With this in mind, “crowdsharing”-style information sharing over mobile phone networks and social media should be generating excitement among conflict prevention and peacebuilding professionals.  While the above mathematical model is a very simple, over-stylized way of thinking about information sharing via digital medium, it at least helps us tie our assumptions about digitally supported violence prevention to some well-found theories of conflict and conflict prevention, and provides a formal frame to use for further empirical research.

I owe my father, Walt Martin, sincere thanks for his help thinking through the math in this piece.  He’s a computer scientist, and for any R Mathematical types he’s got some good stuff on using R to determine sink rates for gliders (he’s also a pilot).