A Bayesian Analysis of South American Terrorist Attacks
Terrorism is a significant issue in the current world. The media has undoubtedly heightened our awareness of the issue and according to the graph below, with good reason. In my research, I used the Global Terrorism Database assembled by the National Consortium for the Study of Terrorism and Responses to Terrorism (START).
Clearly there has been a drastic rise in terrorist bombings since about 2004. It’s important to note, however, that drastic changes such as the change in terror since 2004 may not only be due to a rise in terror but also a change in the way terrorism is measured. It’s difficult to disentangle the effect this may have had. With that said, I chose to do my analysis on South American bombings, specifically, the 10 years pre and post Pablo Escobar’s death in 1993 so this potential change in measuring standards should not be an issue.
When doing a Bayesian analysis, we must pick a “prior”. A prior is simply a prior assumption in the form of a probability distribution expressing a prediction before new evidence is provided. This prior influences our final prediction when new evidence is provided. The prior I chose was based on the 198 bombings per year that I knew occurred from 1970–2015 in South America.
I had suspicions that bombings would diminish significantly after Pablo Escobar’s death but had never seen statistics to say one way or the other.
There were 591 bombings a year in the 10 years prior to Escobar’s death. Amazingly, there were only 106 bombings a year in the 10 years after his death. The timeline of bombings in South America each year can be seen below.
The mean bombings per year dropped 485! The charts below illustrate how the mean and standard deviations shifted after Pablo’s death.
The charts below show a 95% highest posterior density (HPD) range represented by the horizontal black bar. If the vertical green bar falls inside the HPD this means the samples can be considered to have been taken from the same population. If the green bar falls outside this 95% HPD, we can say the samples came from statistically significantly different populations. As we are concerned with the change in the mean, I will focus on the graph on the right. As we can see, that green bar is way off the map. It is not remotely close to the 95% HPD. Thus, the populations are deemed statistically significantly different.
Although, the populations changed drastically, we cannot say that Pablo Escobar’s death caused this drop. However, this massive drop in attacks could signal that his death was a large contributing factor.