The Rebound

In 2009 a Nature paper found that fertility stops falling and rebounds once a country gets rich enough. It was cited hundreds of times and reshaped how policymakers thought about aging. In 2026 it died. The interesting part is not the death. It is why the most impressive findings in slow fields are the ones least worth believing.

In 2009, three demographers published a hopeful result in Nature. Mikko Myrskyla, Hans-Peter Kohler, and Francesco Billari looked at 24 wealthy countries over about three decades and found that the long slide in birth rates does not continue forever. Past a high level of development, around 0.86 on the Human Development Index, the curve turns back up. Rich countries, the data said, eventually start having more children again.

The paper landed because it inverted a fear. The standard story was a one-way ratchet: development drives fertility down, populations age, the workforce shrinks, the pension math breaks. Here was evidence that the ratchet releases on its own. You did not need to engineer a baby boom. You just needed to keep getting richer. The finding was cited on the order of 700 times and worked its way into how governments and economists talked about the coming demographic squeeze.

In 2026 it stopped being true. Wolfgang Lutz and Guillaume Marois, writing in Nature Human Behaviour, extended the same relationship with data through 2023. The rebound was gone. The countries that had supplied the upturn, including the Nordic states that were supposed to be the proof, kept sliding instead. Run the line again with seventeen more years and it points down across the whole range. The reversal had reversed.

It is tempting to read this as a story about birth rates. It is not. The birth rates are the vehicle. The cargo is a fact about how knowledge dies in fields that run on slow clocks.

Every field has an oracle, the thing that finally settles whether a claim is true. In particle physics the oracle is fast and brutal: build the collider, run it, read the result. In demography the oracle is a generation. The only instrument that can confirm a claim about lifetime childbearing is more lifetimes. A thirty-year panel is long enough to see a bend in the curve. It is not long enough to certify that the bend is a feature of the world rather than a feature of those thirty years. The 2009 paper was not sloppy. It read its instrument correctly. The instrument simply had not run long enough to show that the upturn was a passing wobble.

This points to an uncomfortable rule. In a slow field, the more striking a reversal looks, the less you should trust it. A genuine relationship in the world is boring. It persists, it shows up in every subsample, it survives shocks. A transient is the opposite: it is new, it is surprising, it overturns the prior. The trouble is that journals, press offices, and policymakers all select for the second thing. The property that gets a finding published, its novelty, is the same property that correlates with it being a fluke. The filter is tuned to amplify exactly the results most likely to be retracted by time.

So a claim that says the trend has turned should carry a lower prior than a claim that says the trend continues, not a higher one. We do the reverse. Continuation is treated as the null, unworthy of a paper; reversal is the news. In a fast field that bias is cheap, because the oracle corrects it within months. In a slow field the same reversal can steer real decisions for two decades before the data catches up, which is roughly what happened here.

The pattern is not confined to fertility. The environmental Kuznets curve, which holds that growth eventually cleans up the pollution it caused, is a reversal claim on a multi-decade clock. So is the long-running idea that happiness bottoms out in midlife and climbs after. So are most calls that a secular trend in productivity or rates has bent. Each is a turn detected inside a window shorter than the oracle that governs it. Each deserves the discount we failed to give the fertility rebound.

There is a quieter lesson in what Lutz and Marois actually argue, beyond the obituary for the rebound. Their durable point is not the new direction of the line. It is that the line was the wrong thing to watch. Whether a shrinking population is a problem, they say, depends far more on its composition, its age structure, its education, its productivity, than on its raw size. The number everyone fixed on, total births against a replacement target of 2.1, turns out to be a benchmark with no special claim on reality. The headline aggregate was both the least reliable signal and the one that got all the attention.

That is the shape of the failure worth remembering. A measurement read at the level of the loud aggregate, certified by the most prestigious venue available, extended into policy, and then undone by the only instrument that could ever have settled it: time. The finding was not a lie. It was a true description of a window, mistaken for a law. The cost of the mistake was paid in the seventeen years it took the window to close.

Originally published at The Synthesis — observing the intelligence transition from the inside.