Worrying about terrorism in the US is essentially pointless

A recent CNN article notes that “The Transportation Security Administration is considering eliminating passenger screening at more than 150 small and medium-sized airports” in the United States. Paul Cruickshank, a terrorism analyst for CNN, intones that this policy change might incite al Qaida and ISIS to attack small airliners to create “a great amount of panic.” Politico Playbook authors Anna Palmer and Jake Sherman, journalists I greatly admire, forwarded the CNN story this morning with the commentary “What could go wrong…”, assumedly implying that this was a poor policy choice relative to the status quo.

What could go wrong? Statistically, probably not much. I am continually surprised that the TSA has any federal, state, local, or popular support at all–save, of course, for its employees and the bureaucracy of the Department of Homeland Security. It has been demonstrated, time and time again, that the TSA does not appear to ever catch terrorists and, to boot, has a nontrivial budget. Here are just a small collections of articles to that effect: Slate, from 2010, Les Abend, writing in CNN in 2015, Vox, from 2016, LA Times, from 2017, and USA Today, from 2018. Of all (publicly-known) unsuccessful terrorist plots in the US post-9/11, none–none!–of the plots were caught by the TSA, and certainly none were foiled by the existence of routine security screenings in small- to medium-sized airports–see the exhaustive list on Wikipedia. Probably the closest incident to fitting these criteria is the infamous “shoe-bomber”, who boarded a plane in Paris and failed to detonate a home-made explosive onboard. No security agency caught him before he attempted to detonate his device; he was restrained by fellow passengers on the plane. This also shouldn’t be surprising by this point, since TSA screeners miss most test weapons.

It isn’t solely that the TSA is useless, although that does appear to be the case. Terrorism just isn’t a big problem in the US, airliner hijackings aren’t common, and when hijackings do occur, they don’t result in many fatalities. Let’s take these points one-by-one.

Here’s a graph of fatalities due to terrorism in the US. The large graph is zoomed in to a level at which we can actually see the datapoints, while the inset graph shows the full range of data. Just the fact that we have to zoom in to see the datapoints is telling; the 9/11 attacks are, by far, the most deadly attacks in this dataset.


Here’s the similar plot of airliner hijackings and deaths directly attributable to those hijackings.


There just isn’t much to see. But, to give the TSA (and DHS more generally) the benefit of the doubt, let’s see if there is information hidden in the data that we can’t see. We can test the hypotheses that:

  1. There exists some year on which the rate of US fatalities due to terrorism changes;
  2. There exists some year on which the number of airliner hijackings and associated fatalities due to hijackings changes.

If the TSA and / or DHS more generally is affecting the rate of US fatalities due to terrorism or airliner hijackings, we might expect to find evidence suggesting that years exists that meet the above criteria. Fitting the most basic of changepoint models (piecewise constant rates, uniform priors), we see that this is not the case. Here’s the changepoint analysis for US fatalities due to terrorism.


Hardly a significant change. The best guess for the switchpoint year was in 1991—and, if any effect does exist, it appears that the later rate is lower than the previous rate. The story for airliner hijackings is similar.


I’m not arguing that the DHS, as a whole, doesn’t do good work. However, the well-known ineffectiveness of the TSA at detecting weapons and would-be terrorists, along with the statistical insignificance of terrorism as a cause of death in the US and miniscule number of fatalities from airline hijackings worldwide make concerns about the TSA cutting funding to small and medium-sized airport screenings seem overblown.

You can analyze the data yourself; everything I used to create the above graphs (and others I didn’t show here!) is here.

Written on August 2, 2018