Why 1.95 deaths per second?
The headline number on this site is a division sum with a surprising amount of argument inside it.
The Demise Ledger counts at 1.9489 deaths per second. That figure comes from a single division: 61,500,000 annual deaths ÷ 31,556,926 seconds in a year. The seconds are not controversial. The 61.5 million is where all the interesting problems live.
Where an annual death total comes from
There is no global register of deaths. Nobody is counting. What exists instead is a reconstruction, assembled from three quite different kinds of evidence.
The best of these is civil registration1: a country records deaths as they are certified, codes each to a cause, and publishes totals. Where this works (much of Europe, Japan and parts of the Americas), the data are genuinely good. WHO's current assessment is that nearly 40% of the world's deaths are still not registered. In much of sub-Saharan Africa, that share is far higher, and parts of South Asia are in the same position.
Where registration is incomplete, statisticians fall back on household surveys and censuses, which ask about deaths in the household over a recent period, and on demographic models2 that infer mortality from population structure3, fertility4 and life tables5. Both are legitimate. Neither is an actual count.
So the global death total is a modelled6 estimate resting on a partially observed base. That does not make it bad. It makes the last two digits meaningless.
Why the number moved
Through the 2010s, United Nations figures rose from about 54 million deaths a year to about 58 million, not because the world was becoming more dangerous, but because it held more people and, crucially, more older people. A population that ages will bury more of its members each year even as every individual's risk falls.
Then COVID-19 broke the trend. WHO's estimates for 2021 put total global deaths at 68 million. The pandemic's full toll is best captured by excess mortality7. WHO's assessment is that there were 14.9 million excess deaths across 2020 and 2021, within a range of 13.3 million to 16.6 million, compared with 5.4 million deaths reported as COVID-19. The gap is that wide partly because reported counts depend on testing and certification8 capacity that was overwhelmed in exactly the places worst affected.
Deaths then fell back as the pandemic receded. The UN estimate for 2023 is 61.7 million, so a recent year outside the pandemic peak sits in the low 60 millions. That is why this site uses 61.5 million.
| Annual deaths | Per second | Roughly corresponds to |
|---|---|---|
| 55,000,000 | 1.74 | Mid-2010s estimates |
| 58,000,000 | 1.84 | Late pre-pandemic |
| 61,500,000 | 1.95 | Used by this site |
| 68,000,000 | 2.15 | WHO estimate for 2021, pandemic peak |
The Ledger uses 61.5 million, which is above the pre-pandemic baseline and below the peak. It is a round number in a moving range, and the site is frank about that. A tracker showing 1.74 and one showing 2.15 could both be defensible, depending on which year they were anchored to. This is worth remembering whenever you meet a counter that does not say which year it is anchored to.
What 1.95 per second actually means
It does not mean that two people die every second. It means that, on average over a long enough period, deaths occur at that rate. In any given second, the count is most often one or two. About one second in seven has no deaths, and about three seconds in ten have three or more.
This is why the main counter on the Ledger arrives unevenly. It models deaths as a Poisson process9, drawing each interval from an exponential distribution10, so the gaps cluster and stretch the way independent random events genuinely do. A counter that ticked metronomically would be easier to build and would teach you something false.
The scale problem
Sixty-one and a half million a year, spread across the same 31,556,926 seconds, is about 168,400 a day, or about 7,020 an hour. Those numbers are all equally true and all equally difficult to feel. Psychologists call the failure scope insensitivity11: our emotional response to a number of lives barely changes as the number grows. We respond to one identifiable person far more strongly than to a statistic, no matter how large.
The Ledger's response is to stop presenting the number and start presenting the rate, so that the count can accumulate in front of you at the pace it actually happens. Whether that works is a matter of taste. But it is the argument the site is making.
For the real figures, go to the sources rather than to this site: the WHO Global Health Estimates, the Global Burden of Disease study and the UN World Population Prospects are all freely available and considerably more careful than anything here.
Notes
- Civil registration. A government's official record of births and deaths. ↩
- Demographic models. Calculations that estimate how many people die when the deaths themselves were not all recorded, using what is known about ages, births and survival. ↩
- Population structure. How a population is spread by age and sex: how many children, how many adults, and how many older people. ↩
- Fertility. In this context, the rate at which children are born in a population. It is not a statement about any one person's ability to conceive. ↩
- Life table. A table of the chance of dying at each age, used to work out how long people in a population can expect to live. ↩
- Modelled. Worked out from other information, rather than counted death by death. ↩
- Excess mortality. Deaths above the number expected from earlier years. For the pandemic, that includes deaths from the virus itself and deaths that followed because care and ordinary life were disrupted. ↩
- Certification. A doctor or official recording that a person has died, and of what. ↩
- Poisson process. A model of events that happen independently, at a steady average rate, but not on a fixed timetable. ↩
- Exponential distribution. The pattern of waiting times between those events. Short gaps are common, and long gaps are rarer. ↩
- Scope insensitivity. The tendency to feel much the same about a large number of deaths as about a smaller one. The feeling does not grow in step with the number. ↩