Fear Factor: Guns vs Terrorism

I‘ve been pretty quite here recently – some intense projects and my travel schedule haven’t left me much time to write.  I do have a few half-written posts that I’ll try to finish up soon.  In the meantime, here’s a short series that veers a bit from pure technology and into the interconnected realm of data analytics and social sciences…

A few weeks ago I was talking to a family member in the U.S. (I’m a U.S. citizen currently living in Germany) and we were discussing the recent spate of weather and other natural disasters that were hammering the states. When we were done he said, “Well as crazy as it is here I’d take this any day over what you’re dealing with.”

I was a bit confused, and asked what disaster he was referring to. He clarified, “No, I mean all of the terrorists driving trucks into crowds and setting off bombs on trains and stuff.”

Ah, right. I’ve heard similar statements several times since I moved to Europe and never quite understood them – after all, while horrific, the sheer number of terror related deaths in either Europe or the U.S. is in the dozens or low hundreds; I was pretty confident that the probability of being a victim of a terrorist is far lower than many other forms of violent crime or preventable death. I replied, “You know, there are more gun deaths each day in the US than terrorism deaths in Europe every year. What you should be afraid of is walking out your door.”

Not surprisingly, we agreed to disagree and the conversation ended cordially. However, it got me thinking: Was I right that someone in Europe is less at risk from an Islamic (or other radical) terrorist than an American is from another American with a gun?  If not, why is the fear factor from terrorism so much greater than gun violence?

The first question sounded like a straightforward data analytics exercise, so I busted out a Jupyter notebook to explore, grabbed some data and challenged the hypothesis.

To analyze terrorism I chose the Global Terrorism Dataset (GTD), a very comprehensive collection of worldwide terrorism over the last half century. The gun violence datasets were harder to come by, in part due to the successful lobbying efforts by the National Rifle Association (NRA) which blocks government research on gun violence, so I chose to work with the Centers for Disease Control (CDC) Multiple Causes of Death dataset which classifies all deaths in the US, including deaths by firearms. The latest year that the GTD and CDC set fully overlap is 2015, so I chose that as the year to focus on.

Terrorism 

Let’s start by looking at terrorism.  Worldwide, there was a significant spike in terrorism over the most recent decade, with the vast majority of the increase coming from the middle east, Africa, and south Asia.

 

If we zoom into this decade and look only at the US and Western Europe, this is what we see:

Look at the Y axis on both of the above graphs – it’s clear that it’s much safer to be in Europe or the US that many other parts of the world (two orders of magnitude safer). While Europe has seen a relative spike in terrorism related deaths since the end of 2015, it also has roughly double the population of the US so to get a better picture of how this compares to US deaths we need to look at deaths per million residents. Here’s what we get:

2015 terror deaths EU: 171.0 total, or 0.23 per million residents
So in 2015 a European had roughly a 1 in 4,000,000 chance of dying in a terrorist attack. That sounds pretty small.  Just out of curiosity, I wonder how that compares to terrorist attacks on American soil:

2015 terror deaths US: 44.0 total, or 0.14 per million residents
I hate to write this because some knucklehead will quote it out of context, but on the surface Europeans have roughly twice the probability of being terror victims than Americans when adjusted for population (in 2015 at least). But that’s like saying a person is twice as likely to be killed by a bear than by a shark – both numbers are so low that doubling either is still a low number.  (In fact, the odds of dying in a shark or bear attack aren’t too far off than death by a terrorist, but that’s for another article.)
Let’s look at the other side of the problem.

Gun Deaths in the U.S. (round 1)

Ok, how does that compare to the risk of dying from a gun in the US? Here’s a high-level breakdown of US gun deaths in 2015:

suicide     22060
homicide    13018
accident      489
other         284

The rough numbers/ratios above have been quoted quite a bit over recent years – roughly 35K gun deaths per year with ~1/3 homicides and ~2/3 suicides – so no big surprises there. 
 
Since terror attacks are essentially homicides, let’s look at gun homicides per million so we can compare with the terrorist threat:

2015 gun homicides US: 13018 total, or 40.29 per million residents
So, at ~40 gun homicides per million residents, an American is ~175x more likely to die from a gun homicide in the US than a European is from a terrorist in Europe.  Hmm.
 
But… it could be argued that this isn’t a fair comparison.  I’ve heard several arguments that have gone something like this: “Terrorists tend to strike random, killing innocent, unsuspecting victims.  U.S. gun violence mostly happens in places like Chicago, St Louis and Detroit and involves gangs and criminals.  In other words, U.S. gun violence is about ‘them’, and we’re not ‘them’.”
 
So how can we whittle the dataset down to “not them”?

Gun Deaths in the U.S. (round 2) 

Let’s see what we can find as we drill into the CDC data…

On an absolute basis, American men are ~6X more likely to be victims of gun violence, while on a percentage basis, men and women have similar levels of root cause, with suicide being the major contributor.

How about race?

The differences here are striking – blacks and hispanics are far more likely to die from homicide while whites are overwhelmingly likely to take their own life. To get a different perspective, let’s look at this on a percentage basis:

Again, some striking differences in intent between different racial groups.  (My gut tells me the homicide rate roughly correlates with average income level, but that’s an analysis for another day.)
 
Ok, maybe education plays a role, either directly or as a proxy for socio-economic status:

Again, a pretty strong correlation.

And now let’s look at age. Here are two views, one broken down by intent and the other by race:

(Note: the bump around 50 is due to a spike in white male suicide…  Remember, remember the month of Movember…)

While tragic, the suicides, accidents and undetermined cause events aren’t relevant to this analysis so we’ll exclude those to focus exclusively on homicides and revisit the age vs race graph in this light:

So, it appears that gun deaths skew heavily towards young black and hispanic males without college degrees.  It feels wrong removing men from the equation since most of the comments I’ve heard relating to this hypothesis have come from men, so let’s just filter on the other dimensions and look at whites over 30 with college degrees:

2015 gun homicides US (white, over 30, college degree): 392 total, or 1.21 per million residents

So even this limited demographic is still ~5X more likely to die from gun in the US than a European is from a terrorist attack.

Conclusion

Ok, let’s review:

  • In 2015 a person in Europe had less than one in a million chance of being killed by a terrorist.
  • That same year, a person in the U.S. had a probability up to 40 in a million of being killed by another American with a gun.

At this point I think I can be pretty confident that my original hypothesis is correct: an American is at much higher risk of being killed by another American with a gun than a European is of being killed by a terrorist.

In the course of exploring this data I have to admit I was surprised at some of the things I found and want to explore them further – for example:

  • What’s going on with terrorism in the rest of the world?
  • How does the casualty rate from guns and terrorism compare with other preventable deaths?
  • Why is the fear factor orthogonal to the reality of the actual risks? 

Stay tuned.

(If interested, you can look at the code of the analysis on this Kaggle kernel.)