Analyzing patterns in Global Terrorism
A group of us independently analyzed data on the timing, type, and location of terrorist attacks around the world. The data was from the Global Terrorism Database (http://start.umd.edu/gtd/).
I first did some exploratory data analysis on the number and types of attacks vs. time. Most categories of attacks follow a similar pattern of increases and decreases over time (figure 1.) However, the location of the attacks (figure 2) over time has varied, with attacks in Central and South America decreasing, and attacks in the Middle East and South Asia decreasing over the past decades.
I then looked at the time series properties of the data shown in figures 1 and 2. The best predictor of the number of attacks in a given year, from what is contained in the data, is the # of attacks in the previous year. Partial autocorrelations beyond 1 year don’t contain significant information for the prediction.
We then each chose an event of interest to see if it significantly impacted the number of attacks. I performed two Bayesian tests to determine whether the 1996 Guatemala peace accords, signed on December 28th, 1996, affected the # of attacks in that area. The results showed that there is greater than a 95% chance that the proportion of attacks in Central America vs the rest of the world experienced a real decline after 1996. I also found that there is about an 85% chance that the number of attacks in Guatemala significantly declined after the peace accords.
Others in my group similarly looked at major events such as the Northern Ireland peace accord and the start of the Iraq war. These also corresponded to a significant change in the number of attacks in those regions. However, the impact in one region isn’t necessarily seen in other regions.