What follows is my translation of José Merino’s article for this month’s Nexos magazine. With clearly presented statistical analysis, he shows that the deployment of the Mexican military in the so-called “War on Drugs” has led to a sustained increase in the homicide rate in those places where troops have been deployed. I hope that José publishes a followup analysis with a closer look at municipal-level data. For now we can play around with a handy timeline map that Diego Valle-Jones created with Google Fusion Tables and compare the increase in homicides at the local level with military operations that have taken place in those areas. If you are a data geek interested in public security in Mexico, I also highly recommend Diego’s “Statistical Analysis and Visualization of the Drug War in Mexico.” Finally, on June 15 at noon GMT Sarah Drummond of MyPolice.org and Making a Difference with Data will lead an online discussion about using open data effectively to improve public security policy. Unfortunately, no matter how much evidence is shown to Mexican president, Felipe Calderon, there is no way that he’ll adjust his security policy. Calderon often is not held accountable for his actions because there is no presidential re-election in Mexico, a legacy of its 1910 – 1920 revolution. The following article is meant to be read together with Eduardo Guerrero Gutiérrez’s excellent think piece, “The Root of Violence,” which I won’t have time to translate into English, but hopefully Insight Crime will.


For the first time ever José Merino brings together the three statistical sources of information that measure the number of homicides in the country: INEGI, the National Public Security System, and the database of homicides related to organized crime. Merino finds that the spiral of violence has grown disproportionately in states where the federal government conducted joint operations [between the police and military].

Here’s what we can’t question: in the last three years we see a substantial increase in the number of homicides in Mexico, with monstrous increases in a few, specific states. That’s where the unquestionable ends. But when we ask for the specific increase of homicides, we can only respond with the opportune, but little sexy, “well, it depends.”

There are two sources of information about homicides in Mexico: the National Institute of Statistics and Geography (INEGI), which bases its numbers on death certificates, and the National Public Security System (SNSP), which is based on police investigations. And no, they do not match. Which is better? That depends on who you trust, the forensic experts (INEGI) or the state prosecutors (SNSP). According to INEGI’s data, in 2009 we had the highest homicide rate in Mexico’s recent history. According to the SNSP, the homicide rate was actually higher in 1999 than 2009.

OG Merino1

In January of this year Nexos published an article by Fernando Escalante titled “Death With Permission: Homicides 2008 – 2009.” Escalante concluded that the spike in the homicide rate – following years of its gradual decline – “can be explained by the deployment of the army, marines and federal police throughout a significant portion of the country.”

To support his argument, Escalante offered a series of descriptive graphics that show the evolution of homicide rates in states with joint operations between the military and police versus those without them. The difference was drastic. These graphics can show a pattern in the data, which could or could not implicate a statistical correlation or causal relationship, but nothing more.

I was surprised that the conclusions from the article were quickly echoed by other media as facts, as if they confirmed what everyone already suspected: military presence leads to an increase in the homicide rate.

My intention in this analysis was to use statistical methods that examine causal relationships in order to conclude that Escalante was mistaken, that there was no causation between the joint military operations and the rise in the murder rate. But that’s not how it turned out. Instead, the central conclusion of my analysis is that, using the appropriate statistical methods, Escalante’s conclusions are confirmed: the military operations have caused an increase in the homicide rate in the states where they have taken place.

Finding the Link

We can’t analyze the effect of joint military operations on the level of violence without understanding the trends in homicide statistics in those states with operations compared to states where operations did not take place. In order to do that, we must aggregate the sources of relevant information from INEGI and Secretary of Public Security along with the Associates of Organized Crime (ACO) database from the federal government. I will use all three in my analysis for one simple reason: we’re interested in understanding the relationship between phenomena, rather than defending which source of information is the best or most exact.

According to the statistics in the three databases, in those states with joint military operations, there were also homicide rates much higher than the rest of the states. The difference among states with joint military operations and those without increased dramatically beginning with the implementation of the operations: Michoacán (December 2006), Guerrero and Baja California (January 2007), Nuevo Leon and Tamualipas (January 2008), Chihuahua (April 2008), Sinaloa and Durango (May 2008).

Among those states with joint military operations, Chihuahua’s homicide rate increased 10 points (per every 100,000 citizens) compared to the national average. But even without taking Chihuahua into account, the other seven states with joint military operations have much higher homicide rates than even the most violent states without operatives (I will call these states violent states without operations the G5: Nayarit, Coahuila, Morelos, Quintana Roo, and Sonora). As you can see in the table below, the rest of the country has a much much lower homicide rate than these 13 most violent states.

[My note: It gets confusing, but to review: Mexico has 31 states and one federal district. There are eight states with joint military operations. Of those eight states, Chihuahua is by far the most violent – mostly due to Ciudad Juarez. There are five states that don’t have joint military operations but still have a significant homicide rate. Jose calls these states the “G5.”]

Exp Cuadro1 2

To put it into an international perspective, some of the states with joint military operations have homicide rates along the lines of some of the most violent countries in the world, such as Colombia and Venezuela. Those states that are violent, but without joint military operations (the “G5”) are along the lines of Russia. Meanwhile, the rest of the country is more like Costa Rica or Lithuania.

Now, if we look at the only source of information with complete data for 2010, the homicides of the Associates of Organized Crime database, we see a dramatic increase both in states with joint military operations and the G5. In fact, the increase was so high in the ACO database, that it exceeded the number of 2009 homicides from the INEGI database [which takes into account all types of homicides; not just those related to organized crime].

What happens, then, with the homicide rate after the deployment of joint military operations? (See graphics 2 and 3.)

OG Merino2

OG Merino3

There seems to be a shared pattern (Nuevo León and Tamaulipas are exceptions): an immediate increase in the number of homicides after the start of a joint military operation, then a plateau for the next few months, followed by yet another rise. Once again, Chihuahua is the outlier; even among the most violent states, it has a far higher homicide rate. One could jump to the conclusion that the high correlation between joint military operations and the increase in homicides is fundamentally explained by Chihuahua. But that’s not how it is.

We can start by recognizing that the joint military operations did not occur randomly or by chance, but rather, they took place precisely in those states where violence and organized crime was the most visible. Therefore, if we want to measure the effect of additional violence that was provoked by the operations we have to isolate the operations from other influences, such as the trends of the homicide rates prior to the operations.

Let’s imagine an experiment in which we can compare two states with identical homicide rates for a particular year, but in one state we introduce a joint military operation and in the other we don’t. If there are differences in the homicide rates the following year, they could likely be attributed to the operation. There is a statistical method, propensity score matching, that allows us to do exactly that. It brings together groups of observation (in this case, entities) that are based on a particular variable (the homicide rate from the previous year), and compares the application or not of another variable (the joint military operations). When we use this method based on the three sources of homicide statistics we obtain the following results (see the methodological note at the end).

From this exercise we can reach three conclusions. First: there is a causal effect between the deployment of joint military operations and the rise in the murder rate in all three data sources. Second: the relationship still exists even when we exclude Chihuahua from the data, although the effect is largely reduced using the INEGI data. Third: the effect increases and strengthens when we use municipal data, given that it increases the variance of the data. In order to make comparisons, however, we’ll use the state-level data.

What would have happened if the joint military operations never occurred?

According to the data from INEGI, between 2008 – 2009 we would have had 7,063 fewer homicides (with a confidence interval between 4,170 and 11,328). See graphic 4.

OG Merino4

According to the limited data from the National Security Secretary (SNSP) between 2008 – 2009, the number of murders would have fallen by 5,289 (with a confidence interval between 2,166 and 9,368). See graphic 5.

OG Merino5

In the case of the Associates of Organized Crime database, which covers the entire period between 2008 – 2010, there would have been 11,477 fewer homicides. See graphic 6.

OG Merino6

To say it another way, between 2007 and 2009 INEGI reported 42,064 homicides in total and the SNSP reported 39,563. Without joint military operations those numbers would have been 35,001 and 34,174, respectively. For its part, the Associates of Organized Crime database reported the widely cited and tragic sum of 35,000 related murders between 2007 – 2010. Without joint military operations this total number would have been 22,954. This is the size of the effect of the joint military operations in the states where they occurred: according to all three sources, clear statistical significance. The analysis presented here measures the effect in those states where there are military operations. Of course, it would be naive to think that the impact is merely reduced to these states and doesn’t extend to other areas where drug trafficking cartels are also present.

Behold the numbers of the war on drugs; those which we can attribute to joint military operations with the information that is available to us. What remains to discuss and debate is the violence in Mexico, its causes, and the pertinence of the government’s strategy to combat organized crime.

Methodological note: I used propensity scores and Kernel matching pursuits with error bootstrapping. The database files and the analysis file for Stata is available at http://bit.ly/gOhG0P