Every flash is a death.

Do you want to know how many people die in a second worldwide? Let's ask the machine.

The problem is that that is not true — or not completely true. For the last couple of years we have observed between 60 and 63 million deaths a year, worldwide. A year has 31,536,000 seconds, and dividing one by the other gives about two deaths a second. But that is a yearly average, and nothing about a death is average.

In reality there's a big chance that during any given second nobody dies, because deaths don't happen on a steady rhythm. In fact, if deaths were randomly distributed across the year at the observed annual average, there would be about a 27% chance that exactly two people died in any given second — and roughly the same chance that exactly one person died in that second.

So, how many people die on a given second? Our new (better) answer is: if people die randomly around the year, probably somewhere between 0 and 4 (with a ~95% certainty).

We made a bold assumption when timing the events, that they happened randomly. As we will see this is far from the truth and we'll try improving our timing model. Nevertheless, we'll be equally naive for our first spacial distribution model: deaths will happen at random times and at random places.

I think most people wouldn't get triggered by our timing model, but our spacial model is painful to watch. Most randomly placed dots (~71%) fall in the ocean, as this is the proportion of earth that's covered in water. The dots falling on land fall overwhelmingly on areas without any permanent human settlements (deserts, forests,…).

We can agree that if our goal is an accurate defunction model it should be way better than that. Maybe instead of using the global mortality numbers we could use country level data. The good news is that many countries report their Crude Death Rate —how many people die in a year for every 100000 inhabitants—. The bad news is that many don't provide that data, and virtually all the ones that do overcount or undercount some deaths.

Luckily the WHO has some very smart people working on it and providing good estimates. Even if imperfect the mortality rates by country will be a huge upgrade from rendering random points wherever, with equal chances of it being in Mexico or Lithuania, Togo or the Antartica, on land or in the middle of the ocean.

A country is not an average

Not having deaths pop up in the middle of the ocean is neat. And seeing the popping dots on the countries they should be is also a great improvement. But if you look carefully, you'll notice that the dots appear always exactly in the center of a given country. Sometimes that's in an uninhabited or even uninhabitable part of the country. According to the current simulation, all deaths in Russia happen in Lake Vivi, 400 kilometers away from the closest hospital. But we know that people tend to die where people tend to be, so let's apply a density map to our calculation. This way, the dots will have bigger chances to appear where there's more people and little chances of appearing at unpopulated areas.

Small problem though: the density map is a grid of ~55 km wide cells (0.5°), it's pixelated, so to speak. But country borders aren't pixelated (colonialism got close but not quite), they often follow rivers, mountains or the sea.

The way we go around this is by converting every source, every complexity layer to a grid. And we assign each pixel to whichever country has the most population in that specific cell.

This map is cheating a little bit: it uses a logarithmic scale by default. Otherwise you wouldn't see much. The fact is that the earth, even the land, is mostly empty. Toggle between log and linear scale to see how empty it is.

On a linear scale you barely see anything else than a dozen megacities and some shading around the Indo-Gangetic Plain (the region south of the Himalayas that's home of one-seventh of the world's population).

Now our dots appear mostly in cities and in very populous areas.

CDR by region

A single national rate hides enormous internal spread. In Japan, rural Akita dies at nearly twice the rate of Tokyo; in the US, West Virginia runs well above Utah; in Europe, north-west Bulgaria and eastern Germany far exceed Ireland or the Nordic capitals. Most of the gap is age structure — older regions bury more of their people each year — layered over real differences in health, poverty, and access to care.

An illustrative example of how big the age structure of a place is to its CDR is that of Mexico and Lithuania. In 2000, both countries had life expectancy around 72 years, but Mexico's crude death rate was about 4 per 1,000, while Lithuania's was about 11 per 1,000. Same life expectancy, almost 3× crude death rate difference.

Borders are the wrong unit

Death rate by time of year

From a timing perspective, we're still at the same place that we were at the beginning. Dots still appear at random moments. This looks definitely better than having regular intervals like a metronome, but it's not closer to reality. Randomness feels more organic, but we're striving for something more than feeling.

Do we have something better than random? Yes, at least for some countries, that provide monthly or even weekly CDR numbers. Mortality rates on a given month in a given country vary from year to year, but not by much. Actually, it is quite easy to draw a curve, showing at which time of the year people die less or more. And the data is not only consistent from year to year but also between countries.

A cluster of similar curves

Add or remove any country with a directly-measured curve to compare.

From observations to a continuous multiplier

Weekly, monthly and quarterly describe how often observations arrive. They are not smoothing methods. Choose any quality-gated country with weekly observations below to see what each cadence preserves, compare the former circular three-month average, and toggle orders 1–4 of the continuous harmonic curve. Order 4 is the model the globe now uses.

Loading the smoothing comparison…

In these charts a factor above 1 means deaths fire faster than the annual average, and a factor below 1 means they fire slower. The production curve is a pooled order-4 Fourier regression over every complete non-COVID year: weekly observations stay weekly, counts are converted to daily intensity, and the result can be evaluated continuously on any day of the year.

Many temperate countries follow a similar pattern: winter is significantly more deadly than summer. Closer to the equator the annual signal is usually flatter or follows rainy and dry seasons rather than a winter/summer contrast, so the harmonic model does not assume that every country has a winter peak.

Southern-hemisphere curves are shifted by half a year only when the chart compares their seasonal shape with northern countries. Their production curves retain their true calendar phase.

These choices on my wording reveal an assumption that, even if quite obvious, could be flawed: that change in rates over a year are mostly based on climatic seasons. In theory they could reflect something different: think social mechanisms like Ramadan or data artifacts like late reporting of defunctions, default defunction dates, and so on.

Trying to understand what causes seasonality is very interesting, but it isn't critical for the countries from which we have observed seasonal data. What do we do with the other ~100 countries that only provide yearly CDR?

When I'm missing the data I need I always ask the question: are there any proxies to the data I'm looking for, that are easier to get? Are they causes of the phenomenon we study, consequences of it, or do they share a common cause with our subject? In this case I can think of a couple, each with its own strengths and weaknesses.

Before committing to one, it's worth seeing what actually tracks the strength of a country's seasonal swing among the countries that already report a curve. Many are roughly unimodal, while some tropical climates can have broader or multiple rainy-season features. Amplitude is still a useful first summary — how far the curve moves away from its annual mean — without claiming that every shape has one winter peak.

In the charts that follow, every dot is a country with observed seasonal data. The y-axis shows amplitude: the higher a dot is, the bigger the difference in mortality between summer and winter. The x-axis places countries on each of the proposed proxies.

The bottom axis shows absolute latitude — how far a country is from the equator. Points at the left are tropical, points near the right edge are near the poles.

We see an expected pattern: amplitude is lowest between the tropics, as countries there don't experience astronomical seasons. And an — at least for me — unexpected one: above 35° seasonality decreases, instead of climbing with colder and longer winters. A possible explanation is that above a certain threshold people understand that they should adapt to winter and implement social, behavioural and housing adaptations. That Spain, with a more temperate climate, shows higher winter mortality than Sweden would point in this direction.

Here we put all countries and regions in their respective climatic bucket. If this is to be a good proxy we should see that the dots in each climate are packed together.

Well, this doesn't change the picture much. We confirm that there's a correlation, that seasonality is lower between the tropics, and that climate — just as latitude — best predicts seasonality where seasonality is low.

How many older people live in a country seems like a very poor proxy. I just see a cloud of dots here. Possibly because richer countries tend to have older populations and one thing offsets the other. Shading the countries by GDP per capita confirms this: countries on the left are lighter (poorer) than countries on the right.

More of the same. There's no visible correlation between how rich a country is and how strongly its mortality is affected by the seasons. The cause is probably the same as above: richer countries offset their seasonality with stronger health systems and better adaptations.

How about neighbouring countries? That's better. There's a strong correlation here. Bordering-neighbour adjacency is the strongest proxy — the lowest median error, the highest correlation, and the only one with positive skill against both the mean-curve floor and latitude. Latitude and climate class land close together just behind, with climate edging latitude on the typical country.

So how did your ranking do?

You put the five in an order before any of this was on screen. Here is the order the charts above actually produce, and where your guess landed against it.

The seasonality tables have not loaded, so there is nothing to score the ranking against yet.

Bordering regions, not just bordering countries

Adjacent regions inside the same country track each other even more tightly than bordering countries do. That agreement is why the nearest-region reconstruction beats every other region-level proxy. Two Russian regions, Ingushetia and Chukotka, had unusable raw weekly data — zero-rate weeks or excessive spike noise — and are imputed from the average of their nearest good neighbours rather than shown as-is or dropped.

Where the extra deaths fall, month by month

Each square is half a degree of the same grid the globe samples, coloured by the deaths this month's season adds to, or takes from, an ordinary month there. Drag through the year and the hemispheres trade places: Europe's January surplus is southern Africa's January deficit, and by July they have swapped.

The texture inside a country is population, not seasonality. Ten countries have measured regional curves, and inside this frame that comes down to South Africa, western Russia, and climate-modelled India and China; Europe has none at all. Everywhere else every cell of a country shares one curve, and the outlines say which unit that curve belongs to — solid where somebody counted deaths in it, dashed where it was reconstructed from a neighbour, a climate class or a latitude. What changes from cell to cell inside one of those borders is how many people live there, not when they die.

A war is not a Poisson process

The previous layers capture long-term mortality trends, which account for most deaths worldwide. But current mortality also needs finer-grained factors, the largest being conflicts. ACLED's real-time Research feed publishes weekly country and Admin-1 aggregates with fatalities and regional centroid coordinates. Those centroids are an approximation, not individual event locations.

For all other layers we multiplied the base crude death rate by a seasonality or density factor. Here we get 12 complete, reported weeks ending at the oldest publication date shared by all six ACLED regions. So what is our factor for the current week?

We use a recency-weighted average to estimate the current week's mortality. Specifically, a robust exponentially weighted moving average (Robust EWMA): use recent weeks more than older weeks, but damp suspiciously extreme values before averaging.

For example:

Take seven weeks from the 12-week window:

We calculate the 10th and 90th percentiles.

Then we update the numbers above and below these caps:

To calculate the weights we choose a half-life: how many weeks it takes to halve a week's impact on the final estimate. A fitted value would require a longer validation exercise; here the production default is four weeks.

So the weights to halve the impact every four weeks look like this:

Then we do a weighted average using these weights. Which gives us 28.4.

Both choices — how fast a week's influence decays, and how hard the extremes get pulled in — are mine, not the data's. The globe always uses the default four-week/P10–P90 result, distributed by each Admin-1 region's share of the 12-week fatalities and annualised. The sliders below are a counterfactual demonstration; dragging them does not change the globe.

The map reports fatalities over the same 12 complete weeks at ACLED's Admin-1 centroids. For the globe, each centroid is assigned to the nearest populated sampling-grid cell within the same country.

Age, then sex, then one cause

Age and sex come from the UN World Population Prospects table of deaths by age and sex, in its 2023 figures. Cause comes from the World Health Organization's Global Health Estimates — the 2021 release, read at 2019, the last year before the pandemic, so that covid is not drawn as how people ordinarily die — taken at its most specific level — the recognisable causes, not the clinical groupings above them — and reduced to the strongest eight per country, sex and age band, with everything else folded into "other causes".

They are sampled in that order, so a cause is only ever drawn from the age and sex band that plausibly dies of it. Draw the cause first and you get twenty-year-olds with dementia.

Both tables ship as JSON in the repository, so the feed needs no runtime API call and reads the same offline. The cause table used to come from the IHME Global Burden of Disease, hand-exported from a results tool that puts every download behind a sign-in and a hundred-thousand-row cap. The WHO table is one request and no account, which is most of why it is the one here.

Then the month tilts it

The same three tables do not give the same answer in January as in July. Every age band and every cause group carries a twelve-point curve, normalised to average one across the year, so the annual totals stay exactly what the tables say while the mix inside a month moves: a northern winter draws older, and more respiratory. South of the equator those curves run half a year out of phase, which is measured rather than assumed.

The curves come from monthly death records in Brazil and Mexico — the only two countries here whose microdata carries a date and a cause together — and reach everywhere else through the same climate-zone transfer the seasonality chapter uses. Age is covered completely. Cause is not: 56 of the 90 labels borrow the curve of their ICD-10 chapter, and the other 34 are left flat, multiplied by one. The largest of them is "other causes" itself — the residual that the strongest-eight truncation folds everything else into, typically a third of the weight in the adult bands and up to a half. Leaving it flat is a decision rather than an oversight: it is a bag of unrelated deaths with no chapter of its own, and handing it the average death's curve would assert a shape I have not measured.