The Reason Nothing Is Ever Both

Somewhere along the way you formed a quiet theory of the world: that everything good comes at a price. The most brilliant people you know are a little cruel, or at least careless. The restaurant with the transcendent food has waiters who plainly want you dead. The funniest person at the party turns out, on closer acquaintance, to be a small catastrophe. You can have the looks or the kindness, the talent or the humility, the great plot or the great sentences — but not both. The universe, you’ve concluded, runs on tradeoffs.

I’m here to tell you the universe is mostly innocent. You did the tradeoff yourself, at the door.

The bar that invents a tradeoff

Here is the trick, and it is genuinely a trick of looking, not a fact about the world. Suppose two good qualities — call them Hot and Nice — have nothing whatsoever to do with each other out in the general population. Plenty of people are both, plenty are neither, and knowing one tells you nothing about the other. They are, in the jargon, independent.

But you don’t date the general population. You date, roughly, people who clear some bar — hot enough, or nice enough, or some blend of the two, to make it past your filter and into the set of people you actually consider. And that single, reasonable act quietly poisons the data. Among the people who cleared the bar, Hot and Nice are now negatively correlated, even though they were independent five seconds ago. The ones who got in on looks alone are free to be cruel. The ones who got in on warmth alone are allowed to be plain. The only people your filter reliably throws out are the ones who were neither — and deleting the lower-left corner of the chart tilts everything that’s left into a downward slope. A tradeoff appears out of thin air, and it was never in the people. It was in the doorway.

You can make this exact, which is the part I find almost unreasonably satisfying. Give everyone two independent scores drawn from the same bell curve, and let into your pool only the people whose two scores add up to the top half. Inside that pool, the correlation between the two scores is not zero. It is precisely −1/(π−1), about −0.47 — a solidly negative relationship, conjured out of pure independence by nothing but the cutoff. And it gets worse the choosier you are: admit only the top 10% and the phantom tradeoff stiffens to about −0.71; the top 1%, about −0.83. The pickier you are, the more loudly the world seems to insist you can’t have it all.

It has a name, and it has done damage

This isn’t folk wisdom; it’s a known statistical hazard. Its everyday alias is Berkson’s paradox, and its formal name is collider bias. The one-line version: when you select a group by a standard that two separate traits can each help you meet, you manufacture a tradeoff between those traits inside the group. Joseph Berkson spelled it out in 1946 using hospital patients — two unrelated illnesses can look linked among the admitted, purely because being sick enough to land in the hospital is a bar that either illness can clear.

It would be a cute parlour fact if it stopped at dating and dinner. It doesn’t, and it has cost real understanding. Early in the pandemic, study after study found that smokers seemed strangely underrepresented among hospitalized COVID patients — a result that fed genuine speculation, and at least one clinical trial, about nicotine as protection. A 2020 paper in Nature Communications pointed out the deflating alternative: if you only study people who got tested or admitted, and both smoking and severe COVID nudge you into that group, you’ve rebuilt Berkson’s bar, and the “protection” can be a mirage.

The sharpest case is older and sadder. For decades the data showed that among low-birth-weight babies, the ones with mothers who smoked died less often than those whose mothers didn’t — the “birth-weight paradox,” which for one horrible moment sounds like an argument for smoking while pregnant. It is not. Smoking is only one of the things that can make a baby small; the others — a serious birth defect, say — tend to be far deadlier. So once you look only at the small babies, the smokers’ babies tend to be small for the milder reason, and fare better within that slice. Condition on birth weight and you’ve drawn Berkson’s box around your evidence. (Researchers still argue over how much of the paradox this explains versus other effects — but that slicing by birth weight can invent the protection is not in doubt.)

Notice the door

What I love about all this is that the cure isn’t more data or fancier math. It’s noticing the door. The next time the world seems to be charging you a tax — every smart person smug, every beautiful thing fragile, every easy job dull — ask what bar you used to assemble the evidence. More often than you’d think, the tradeoff lives entirely in your filter, and the instant you widen the door it evaporates.

There are kind geniuses and gorgeous people who are lovely all the way down. You’ve just been quietly declining to let them line up with everyone else, and then blaming the world for the shape of the line.

---

How we know. The exact figure: for two independent standard-normal scores, the correlation among the subset whose sum exceeds its median is −1/(π−1) ≈ −0.4669; tightening the cutoff to the top 10% and top 1% gives ≈ −0.711 and −0.823. These were derived in closed form and cross-checked three independent ways — a 2-D numerical integration and a seeded Monte-Carlo over six million draws — all agreeing; the working lives in this site’s repo (research/berksons-paradox/). Berkson’s paradox: J. Berkson, “Limitations of the application of fourfold table analysis to hospital data,” Biometrics Bulletin 1946;2(3):47–53 — a constructed illustration, not real records; a real-data confirmation came later from autopsy data (Roberts, Spitzer, Delmore & Sackett, “An empirical demonstration of Berkson’s bias,” J. Chronic Dis. 1978;31:119–128). COVID: Griffith et al., “Collider bias undermines our understanding of COVID-19 disease risk and severity,” Nature Communications 2020;11:5749. Birth-weight paradox: the crossover was documented by J. Yerushalmy (Am. J. Epidemiol. 1971;93:443–456) and is discussed at length by A. J. Wilcox (Int. J. Epidemiol. 2001;30:1233–1241); the collider explanation is Hernández-Díaz, Schisterman & Hernán, “The birth weight ‘paradox’ uncovered?”, Am. J. Epidemiol. 2006;164:1115–1120, with the “how much does it explain” debate aired by VanderWeele (Int. J. Epidemiol. 2014;43:1368–1373). The dating, restaurant, and party examples are illustrations of the same proven mechanism, not measurements — the math is real; the anecdotes are mine.