Tag: drug traffic

Criminal Activities and Migration

When something unpleasant is happening around you, the most natural reaction you might have is to try and get away from it. For instance, if you live in Mexico and there is a lot of criminal activity in the neighborhood, you might want to move, if you can afford to. This seems intuitive, but is it true? Science is often the process of systematically testing whether our intuitions about the world are true; you never know when the easy answer is wrong if you never investigate it!

This is what I did with Roxana Gutiérrez-Romero: I went back to my old love – the investigation of Mexican drug trafficking – and this resulted in the paper “Displacement and disconnection: the impact of violence on migration networks and highway traffic in Mexico” which was recently published in the Spatial Economic Analysis journal.

The question is simple: do we see a disproportionate increase in emigration (whether local within Mexico or international) from municipalities that experience spikes of criminal violence? Answering this question is quite hard, and it involves controlling for many other potential explanations that might drive emigration. Roxana did a remarkable job in figuring out how to control for those other factors, leaving me to worry about a simpler, networky question. Let’s focus on local internal migration for now.

Nodes are municipalities, connected by migration links. I’ll get to the color meaning in the text later.

We can take a snapshot of internal migration by creating a network of municipalities. Two municipalities are connected by a directed edge weighted by the number of people who change their residence from one to the other. The picture above depicts just that.

A single network doesn’t tell us much, we need two different points in time. For each pair of municipalities, we have data about the migration links for several five-year intervals (2005-2010, 2010-2015, 2015-2020). We cannot simply compare the edge weights in two subsequent snapshots, as the change we observe might be just a random fluctuation. Moreover, how do we know whether a change is significant when there are many migration links? Immigration from a municipality might have increased at the same time as all other incoming links have decreased. For this reason, the question morphs into a network science one: if we have an edge observed across two different five-year intervals with two different weights, how do we know the edge weight changed in a statistically significant way?

We got help from an unexpected ally: network backboning. Normally, network backboning is the process of determining whether an edge measured with a noisy process exists. However, by using the Noise-Corrected approach I developed with Frank Neffke a while ago, we can do more than that.

See, noise-corrected backboning achieves the task of verifying an edge’s existence by modeling it, estimating its expected weight and variance. The same edge at different times will have different weights and different variances. By using bootstrapping, a fancy word that means “draw many random numbers from a distribution characterized by the edge’s weight and standard deviation,” we can create an edge weight distribution and figure out whether the edge’s weight truly increased, decreased, or stayed the same.

This is what you see in the picture above: green edges showed an increase in migration, red edges a decrease, and yellow ones stayed about the same. We can aggregate a municipality’s net migration change, which we use for the node’s color. As a robustness check, we create the same network, but using highway traffic instead of migration:

We can use this estimation of increased/decreased migration/traffic as the variable we want to predict in a big and complicated regression that takes into account many possible alternative explanations – ask Roxana for the painfully precise details, she worked literal years on it.

What Roxana found was that our intuition is accurate: violence indeed is associated with increased emigration. We also checked international emigration to the US (which accounts for 90% of Mexican emigration) and found a similar effect: violence is associated with a 5% rise in emigration to the US and a 3% drop in return migration from the US.

So, for once, we don’t have a puzzling counter-intuitive result: we indeed see violence and criminal activity discouraging people to stay around. It would be interesting to see whether this holds in different contexts and scenarios.

Destroying Drug Traffic, One Query at a Time

in·tel·li·gence NOUN: a. The capacity to acquire and apply knowledge.

The intelligence process, like in Central Intelligence Agency, is the process any person or organization should go through when making important operative decisions. But this is a description of a perfect world. In reality, organizations have to face phenomena that are very complex. When the organization itself is significantly smaller than the complexity it has to face, its members have to rely on intuition, art or not solidly grounded decisions.

This is usually the case for the local police when facing organized crime. Large crime organizations like the Italian Camorra or the drug cartels in Mexico are usually international. If you read Roberto Saviano’s Gomorrah, you’ll realize that Camorra operates as far as Germany or Scotland, while drug cartels usually span from Colombia to the US passing through Mexico. On the other hand, most of their activities happen at the local level: kidnappings, killings, drug traffic. Their main adversary is not usually a broadly operating institution like the FBI, but the local police. But for the local police, to gather a satisfying amount of information to face them is usually hopeless.

With this problem in mind, I teamed up with a Mexican colleague of mine at Harvard, Viridiana Rios. Our aim was to develop a system to enable a cheap and cost-effective way to gather intelligence operations about criminal activities.

The problem with criminal activities is that they are not only part of the complex organism of organized crime. They are also usually hidden from the public. Of course, no head of a mafia family wants to conduct his business en plein air. However, whether he likes it or not, some of these activities reach the general public anyway. This happens because, “luckily”, bad news sells a lot of newspapers. Criminal activities usually leave a clear footprint in the news. Mexican drug traffic in this is also particular, for the tradition of leaving the so called narcomensajes. These messages are writings painted on walls or on highway billboards. They are used by the criminal organizations to threaten each other or the government and the police. A narcomensaje looks like this:

When we design a system for tracking the activities of crime organizations, we want this system to be as automatic as possible. Therefore, we use some computer science tricks and we rely on the information present on the websites of newspapers. Web knowledge has a lot of problems: it’s big, it’s about many different things and it’s subject to reliability concerns. However, Google News deals with most of these problems by carefully selecting topics and reliable sources. What was left for us to do, was to systematically query the system with its APIs and clean the results. The details of this process are in a paper presented by me this week at the Conference for Information and Knowledge Management (CIKM 2012).

We did not have any way to understand if our queries connecting drug traffickers to Mexican municipalities were capturing real connections. For this reason, we performed the very same task using Mexican state governors. With our great surprise, we were able to detect with high accuracy their real patterns of activities. (Not that we are drawing a parallel between organized crime and politics, just to be clear!) This indicates that our method of tracking people’s activities by using Google News data is valid. Here are some maps of some state governors. In red the municipalities where they are detected and with a large black border their state:

What did we find?

Mexican drug traffic follows a fat-tail distribution. The meaning? There is an incredible amount of municipalities with a weak drug traffic presence and some others are an explosive factory where the employees have to carry flamethrowers. Moreover, it really looks like a hydra: destroying one hub is likely just to generate another hub, or ten smaller hubs.

And the system is growing fast, jumping from one order of magnitude to a larger one in less than a decade.

We are also able to classify cartels with several features: how much they like to compete or to explore the territory. In the future, this may be used to predict where and when we will see a spike of activity for a particular drug cartel in a particular municipality (in the picture, the migration patter of the Los Zetas cartel).

Apart from the insights, our methodology really aids the intelligence problem, whenever there are no sufficient resources to perform an actual intelligence task. We used the case of criminal activities, but the system is fairly general: by using a list of something other than a drug cartel and something other than a Mexican municipality, you can bend the system to give you information about your favorite events.