Per What?

Somewhere in Germany a wind turbine is turning, and somewhere on the internet two people are arguing about how many birds it kills. One has a number. The other has a bigger one. Both numbers come from conservationists, both are printed in good faith, and neither settles anything — because neither one is finished. A death toll with no denominator is not a statistic. It is a mood with a citation.

So let us finish them. Germany keeps unusually good books on dead birds, and it also keeps good books on how many turbines, cats, cars, windows and people there are to blame. Divide the first set by the second and something happens that neither side of the wind argument tends to mention: the ranking does not shift, it inverts. Nothing has to be disputed to make it happen. Same figures, same sources, one division.

The tool below runs the division live. Flip between the two views and watch the bars physically re-sort — that re-sorting is the whole essay.

The totals, and who they flatter

Start where every debate starts. Here is what kills birds in Germany each year, as the numbers are normally quoted: an estimated 100 to 115 million die at windows and glass facades; about 70 million on roads and railways; somewhere between 20 and 100 million are taken by cats; 1.5 to 2.8 million collide with power lines, with roughly another million large birds electrocuted on the pylons; about 1.2 million are shot legally by hunters. And wind turbines, the thing the argument is actually about, kill somewhere between 10,000 and 100,000.

Read as totals, this is not close. At the top of its range, wind is about a thousandth of the glass toll and roughly one death in ten thousand across the whole list. You could double every turbine in the country and the national line would not move enough to measure. This is the number the wind industry quotes, and it is true.

It is also incomplete in a way nobody on either side brings up. The largest single human cause of bird death in Germany is not on the list at all, because it does not feel like a cause of death. In 2024 the country slaughtered 694 million birds for the table — about 90% of them broiler chickens — producing 1.6 million tonnes of meat, roughly 13.6 kilograms per person. That is more than twice everything else on the list put together, at the top of every other range. For every bird a turbine kills at the very top of its estimate, Germany eats about seven thousand.

Nobody counts that, and there are decent reasons not to — we will get to them. But notice the shape of the omission. The category birds killed by humans in Germany is drawn to include a cat and a windowpane and exclude a supermarket, and the line was drawn before anyone started counting.

Now divide

A total answers the question how much harm is this doing to birds in aggregate. It is a perfectly good question. It is not the question anyone thinks they are asking when they ask whether wind turbines are bad for birds. That question is about one turbine — about the object, not the fleet. And for that you need the other half of the number.

Germany had 29,226 onshore wind turbines at the end of 2025. It has about 20 million buildings with living space, 49.3 million registered cars, and roughly 17.9 million cats — 15.9 million pets plus an estimated two million feral. Divide:

• One person eating poultry three times a week: about 28 birds a year.
• One German, on average: about 8 birds a year, without deciding anything.
• One building: 5 to 6 birds a year at the windows.
• One cat: 1 to 6 birds a year.
• One car: about 1.4 birds a year.
• One wind turbine: 0.3 to 3.4 birds a year.

The wind turbine is now last. Not by a rounding error either: at the middle of both ranges, a single house with windows in it kills about three times as many birds a year as a single turbine standing in a field. Nothing was fudged. The turbine total was tiny because there are twenty-nine thousand turbines and twenty million houses, and the second number was quietly doing most of the work in the first comparison.

The scariest number in the argument and the most reassuring one are the same number. They differ only in what you divide by.

One Söder

Which brings us to a unit of measurement Bavaria has been supplying free of charge for years. Markus Söder posts photographs of his food — Leberkässemmel, Schweinshaxe, Hendl, the occasional burger disassembled for the camera — with such regularity that the hashtag #Söderisst has become a small genre of its own. Commentators read the meat as positioning: an argument about values conducted entirely through lunch, with almost no words in it.

Three poultry meals a week is 156 meals a year. At a fairly modest 250 grams of meat each, and about 1.4 kilograms of meat off a German broiler, that is roughly 28 birds a year. One wind turbine kills between 0.3 and 3.4. So one Söder is worth somewhere between eight and eighty-two wind turbines, and the width of that answer is not the Söder's fault — it is the turbine estimate, which is uncertain by a factor of ten.

This is a joke and it is also the actual result. Slide the dial in the tool and you can find your own coefficient. The German average, decided by nobody in particular, already sits at eight birds a year — two to twenty-four turbines' worth of bird per resident, per year, before anyone has built anything.

The number nobody audits

Here is the part that should make everyone slightly uncomfortable, whichever side they came in on. The smallest number in this essay is by far the most carefully examined one, and the biggest numbers have barely been examined at all.

The wind figure has been through the wringer. The PROGRESS study searched under roughly 570 turbines at 55 wind farms in the north German lowlands, weekly, for twelve weeks, and found 291 dead birds of 57 species. Wood pigeons and mallards topped the list; buzzards were third; the red kite came twelfth, with five. The central registry has logged 685 buzzards and 637 red kites found at turbines across nearly two decades. Meanwhile figures of 12,000 buzzards and 1,500 red kites a year circulate widely, usually credited to that same study — and Oliver Krüger, who led it, told Correctiv the numbers are "grotesk falsch" and cannot be derived from his data.

Now compare that with the number sitting a thousand times higher on the same chart. The 100-to-115-million glass figure is an extrapolation. It has no confidence interval attached in the places it is quoted, it has never been fact-checked by anybody, and it has never had a study author publicly disown a derivative of it — because nobody is trying to stop a window from being built. The scrutiny a number receives has almost nothing to do with how large it is, or how uncertain, and almost everything to do with whether it is currently load-bearing in a political argument.

Numbers get audited where they are contested, not where they are wrong. The uncontested ones just sit there being enormous.

The honest catch: the third denominator

If the essay stopped here it would be a defence of wind power, and it would be cheating in exactly the way it has spent four sections accusing everyone else of cheating. There is a third denominator, and on that one the ranking moves again.

Per-unit rates treat every dead bird as one dead bird. Conservation does not, and it is right not to. What matters to a species is deaths measured against the population that has to absorb them, and against the speed at which it can replace them — a bird that raises three broods a summer is in a different position from one that raises a single chick and takes years to start. Germany carries a specific liability on exactly that axis: more than half of the world's red kites breed in this country. For a species like that, national totals barely matter. A few hundred adult deaths a year in the wrong districts can weigh more than a very large number spread thinly across abundant species on an entire continent.

That is a real argument, and it sits one step away from a bad one. The bad version says the other killers only take the cheap birds — sparrows, blackbirds, the ones there are plenty of — so a hundred million deaths can be waved through as background noise. It does not survive contact with the data. NABU's own list of frequent glass victims runs kingfisher, thrushes, sparrowhawk, kestrel, woodcock, and the organisation notes that some species — goshawk, sparrowhawk, woodcock, kingfisher — may be suffering disproportionately high mortality at glass. A window does not check the species first. A red kite has no exemption from one.

Now the number that has been missing from this essay, and it is the strongest one in it against wind power. Researchers in Saxony-Anhalt assembled 658 reports of dead red kites from 1951 to 2019, of which 277 had a clear cause of death. In the first half of that span, power lines (nearly 29%) and human persecution (27%) led. Among the finds from 2000 onwards, collisions with wind turbines are 45.6% — against 13% for vehicles, 4.9% for power lines, 1.5% for persecution. On the species Germany is most answerable for, in a wind-heavy state, turbines are the leading identified cause of death by a distance.

And that figure carries the same disease as everything else here, running the other way. It is a statistic about carcasses that were found, and finding is not random. Turbines get searched: operators are required to look, inside a marked radius, on a schedule, by people who are paid to do it. Nobody walks a transect along every road verge, and a poisoned kite on private land turns up by accident or never. Part of that 45.6% is mortality and part of it is search effort, and the study cannot separate the two for you. The share is real. What it is a share of is a question about who was out looking.

Every number in this essay is a picture of where somebody went looking. None of them is a picture of what died.

So the third denominator is not the true one hiding behind the other two. It is the third choice, and it has the worst data of the three. Deaths per population at risk is the right question for conservation and the hardest one to answer honestly, because the species breakdown you need only exists where a legal mandate and a budget put somebody in a field with a clipboard. The well-monitored cause ends up looking specific; the unmonitored one ends up looking like a vague enormous mass. That difference is about funding, not about birds.

Which leaves the poultry comparison, the funniest number here and the one that needs the most care. One distinction holds, and it is narrow: a broiler's death subtracts nothing from a wild population, because that bird was bred into a supply chain and was never in any population's books. That is a real difference in kind, and it is why the tool has a switch to strip those rows out and watch what they were doing. What it is not is a verdict that 694 million animals are of no consequence — that is a different argument on a different denominator again, and this essay is not where it gets settled. Putting them on the chart makes one smaller point, and it is about arithmetic: the boundary of the category was drawn before the counting started, and it decided the answer as surely as the division did.

How to read a scary number

The wind-and-birds fight is a good specimen because both camps are quoting real figures from real conservationists and reaching opposite conclusions without either one lying. That is not what disinformation looks like. It is what an unfinished number looks like when two people finish it differently and neither notices they did.

You will meet the same structure everywhere once you have the shape of it: crime per capita versus crime per neighbourhood, deaths per journey versus deaths per kilometre, emissions per country versus emissions per person. The denominator is where the argument actually lives. It is just quieter than the numerator, and it never makes the headline.

Push the switch above and watch the order fall over; then find something in your own week that only ever gets quoted as a total. How Common Is It, Really? is the same discipline aimed at your intuitions about frequency, and The Base Rate Machine is the version that runs on your own worries. When the number is a risk rather than a body count, Carcinogen to Scale does the equivalent job on hazard labels, and Dread vs. Data measures the gap between how frightening a thing is and how often it actually happens.