Is it fair to compare terrorism and disaster with other causes of death?

Our World in Data presents the empirical evidence on global development in entries dedicated to specific topics.

This blog post draws on data and research discussed in our entries on Causes of Death, Terrorism, and Natural Catastrophes.

Last year the British Royal Statistical Society (RSS) awarded its inaugural International Statistic of 2017 to 'the number 69' because this is the "annual number of Americans killed, on average, by lawnmowers – compared to two Americans killed annually, on average, by immigrant Jihadist terrorists."

This choice sparked controversy between those who consider such figures to be important in understanding relative risks of death, and those who see such comparisons as invalid and misleading.

In this blog we take a look at the scale of deaths from volatile events (here our focus on terrorist attacks and natural disasters), the dangers of relying on statistics from a single year to define risk levels, and whether in light of such evidence, it is still useful to compare their incidence to other causes of death.1

Global deaths from terrorism and natural disasters

Look at the global breakdown of deaths in the chart below for the year 2016. Terrorism- and disaster-related deaths rank very low on the list. In 2016, terrorism and natural disasters combined accounted for less than 0.1 percent of global deaths (0.06 and 0.01 percent, respectively).2

As we discuss in our blog post on 'What does the world die from?', the very low share of terrorism and disaster-related deaths was consistent across most countries in 2016, but not all countries: In several countries, including Iraq, Syria, and Libya, terrorism deaths ranked within the top ten causes.3 It is therefore inaccurate to draw the broad conclusion that the incidence of terrorism deaths is very low everywhere.

It is equally inaccurate to extrapolate or conclude that, based on a single-year snapshot in 2016, the risk of terror-related or natural disaster mortality is very low. Why is this the case?

Terror and disaster-related deaths are different from most other causes of death because of their high volatility. For most causes of death – such as cardiovascular disease, cancers, malaria, homicide, suicide, and even road accidents – the year-to-year changes are small (you can explore the range of trends for Causes of Death in our entry here). Some are increasing or declining over time – for example, deaths from malaria and HIV/AIDS are falling –, but these are slow, gradual changes. The death rates from most causes of death in 2016 is a good predictor of death rates in 2017.

This rule-of-thumb does not apply to deaths related to terrorism or natural disasters. For such volatile cases, deaths can be very low (even zero) for an extended period of time, and then suddenly experience a single-year spike as a result of a large event or even a series of events. Predicting an annual death toll based on the past is therefore less reliable. To have a high degree of confidence in getting it right, I would have to place a large range of error on my estimate: several orders of magnitude in fact.

Let's look at two cases – the 2001 9/11 terror attacks in the United States, and the 2011 Tōhoku earthquake and tsunami (which triggered the Fukushima disaster) in Japan.

Terrorism deaths

In an average year in the United States, the number of deaths from terrorism is very, very low. In 2016, it accounted for an estimated 68 deaths (0.002 percent of total US deaths). Communicating this low level of relative risk was the primary aim of the RSS's choice of International Statistic of 2017. What this statistic fails to capture however is the inherent volatility in this level of frequency.

In the chart below we see the number of fatalities from terrorist attacks in the United States, based on published data and definitions from the Global Terrorism Database (GTD) from 1970 to 2016. For almost the entirety of this 46-year timespan, we see that annual deaths are typically in the range of single digits or tens of deaths.

Then we see a major spike: the 9/11 terror attacks in 2001 which killed approximately 3,000 people – most estimates are close to this figure; reported mortality from the GTD are 3005 deaths. Within the context of this 40 to 50 year trend, this is a dramatic outlier. Within a single year, deaths increased two to three orders of magnitude. Such dramatic spikes are not seen across most causes of death.

This is why we must therefore introduce caution – and perhaps treat such deaths differently – when assessing probability or risk. If we were to take the average annual deaths over this period, or the years spanning 1970 to 2000, we could otherwise make the mistake of concluding that our risk from such events was close to zero. In 2001 we would have been wrong.

How did the 9/11 terrorist attacks impact on terrorism's share of total deaths in the United States? In the charts below we see causes of death in the US in 2001 ranked by absolute numbers, and their percentage of total deaths. In 2001, terrorism accounted for 0.13 percent of US deaths. It is true that this is several orders of magnitude higher, and has climbed several ranking places versus 2016. It is also true that it still places relatively low within total deaths – one out of 770 deaths was caused by terrorism.

Is it possible that terrorism could ever fall within some of the top causes of death in the United States? To account for more than one percent of deaths (which would place it in the top ten), mortality would have to increase by a further order of magnitude from the 9/11 attacks, to more than 30,000 deaths. Such a high death toll would be unprecedented; only in the last few years have global terrorism deaths exceeded 30,000, and only in Iraq have deaths from terrorism exceeded 10,000 per year.4 The probability of such an occurrence is likely to be very low, although impossible to rule out completely.

Natural disaster deaths

Natural disaster events show similar volatility. Let's look at the example of the 2011 Tōhoku earthquake and tsunami in Japan. In the chart below we see annual deaths from natural catastrophes (which includes earthquakes, tsunamis, extreme weather, heatwaves, floods, volcanic activity, and wildfire) for Japan in the period since 1990. In a typical year in Japan we see deaths from natural disasters amounting to tens or hundreds of people. This places it at – or very close – to the bottom of the rankings in causes of death in an average year.

However, in 2011 we see the dramatic impact of the 2011 earthquake and tsunami, which killed over 19,000. Like our example of terrorism in the United States, this represented an increase of several orders of magnitude relative to the longer-term average. Natural disaster events – usually those of earthquakes or tsunamis which are highly unpredictable and difficult to react to – produce similar volatile results across the world. The 2010 earthquake in Haiti and the 2004 Sumatra tsunami in the Indian Ocean show dramatic instances of such trends; more than 230,000 died in the Haiti earthquake, and an estimated 230,000 across countries were killed by the 2004 Sumatra tsunami.

Instances of such events are by nature volatile and unpredictable; interpreting annual risk for a given year based on data from previous years is highly speculative, if not impossible.

In the charts below – which presents absolute numbers of death, and each cause as a percentage of total deaths – we see the impact of the 2011 earthquake and tsunami in Japan. Natural disasters climbed significantly up the list of causes of death, falling just under the top ten, and accounting for 1.5 percent of total deaths. The results in the case of Haiti are dramatic: total deaths over the past few decades have averaged around 100,000 per year. In 2010, total deaths more than tripled to 335,000 with natural disaster deaths accounting for almost 70 percent of the total. 

Treating comparisons of volatile and non-volatile events with caution

These examples highlight the volatile nature of low frequency, (potentially) high-impact events and their effects within the overall scale of mortality causes. How does this affect the validity of making such comparisons? Is it useful to do so?

If we reflect on the split reactions such comparisons typically create, I think both sides have a valid case to make. It is true that many tend to overestimate the incidence of terrorism and disaster-related events, especially relative to other causes of death. Single, high-impact events tend to claim breaking news and newspaper headlines; this can give the impression of heightened risk from such causes. Risk from other and dominating causes, such as cardiovascular diseases, cancer, diabetes, liver disease, and road accidents receive significantly less attention because they are  persistent.

It is also true, however, that assessing risk and incidence of volatile events from a single year is misleading. Looking at mortality trends within a wider context (both temporally and geographically) is essential to assess the potential impact of such events. Equally important is to be aware that historical trends cannot be used to interpret the maximal scale of future events. We cannot conclude that 3000 deaths is an upper limit of terrorism deaths in the United States, just as we cannot assume that an earthquake or tsunami will not kill more 19,000 in Japan in the future.

We must therefore treat comparisons between volatile and non-volatile events with caution. The instance of a 'low' year for terrorism or disaster-related deaths should not be interpreted with complacency; death tolls from such events may in fact be low partly as a result of mitigation efforts against them. If we look at the number of deaths from natural disasters over the course of the 20th and 21st centuries, we see a significant decline over time – not as a result of fewer events, but from more emphasis on adaptation and resilience measures against them.5

Relative comparisons of causes of death can nonetheless provide important context for mortality rates across the world, so long as we are aware of the limitations of comparing causes of highly different natures and frequency patterns.