statistical mirages · metaphor 23 of 100
Everyone you date seems to prove that kindness and good looks trade off. Every actor you've heard of suggests talent and connections are enemies. The world never made these tradeoffs; your selection did — and it manufactures them out of independence, reliably, mathematically.
Suppose that in the whole population two traits are utterly unrelated — two dice, rolled separately, for every person alive. You will never find this out, because you never meet the whole population. You meet the people who cleared a bar: attractive enough overall to end up across a table from you, famous enough to reach your ears, sick enough to occupy a hospital bed.
Conditioning on a shared consequence carves a wedge from the population, and within that wedge the two traits are negatively correlated even though nothing connects them. The correlation is real in your sample and fictional in the world. Below are two thousand people whose traits were drawn independently. Raise the bar, and watch a law of nature appear.
the wedge
The door doesn't care how you cleared the bar, only that you did. So inside the wedge, anyone short on one trait must be long on the other — or they wouldn't be inside. The plainer people in your dating pool really are kinder, not because kindness and looks are connected, but because the plain-and-unkind never made it in. "They must be compensating for something" is the geometry of the cut, felt from within. The door has a slope, and the sample inherits it.
Causal diagrams give the trap its name. Draw the arrows: kindness → dateability ← looks. Two independent causes collide at a shared effect — a collider. Leave the collider alone and its parents stay strangers; condition on it — look only at people for whom it came out one way — and the parents become entangled in your data. Joseph Berkson caught the trap in 1946, in hospital records: two diseases that never interact in the population feud on the ward, because having either badly enough is what gets you admitted. The bed is the collider. So is the shortlist. So is your attention.
what to try
where your samples come from
Once you can see the wedge, you find you have been living in one. The dating pool admits on overall appeal. The shortlist admits on some total impression of credentials and charm — so among the people you interview, the charming hire looks like a gamble on paper. Hospital beds admit on severity, so diseases feud in the charts. News admits on newsworthiness, so importance and sensationalism seem opposed. Restaurants that survive bad locations must be excellent, so location and quality look like enemies to any diner.
Even memory selects: "people I notice" cleared a salience bar built from many traits at once, and inside your recollection those traits quietly trade off. The uncomfortable generalization: you almost never observe a population; you observe the survivors of a door — and every door you don't account for is busy manufacturing structure that isn't there.
the defense
The defense is a single question, asked of every tradeoff you think you've observed: what did these cases have to clear to enter my sample? If the answer is a bar fed by both traits — a sum, an either/or, a vague total impression — the tradeoff is a suspect, not a finding. The cure, where you can get it, is the unselected denominator: everyone who applied and wasn't shortlisted, everyone sick at home. Berkson's fix was exactly this — compare against the neighborhood, not the ward.
Mostly you cannot get that sample, and the defense degrades gracefully into humility: not "the tradeoff is false," but "I have no right to believe it from where I'm standing." That is still worth a great deal — the difference between a rule of thumb and a law of nature, and between judging the people in your world and judging the world.
the mapping
| Mathematics | Life |
|---|---|
| independent X, Y | Two qualities with no real connection — kindness and looks, talent and connections, one illness and another. |
| the collider X → C ← Y | The shared consequence both traits feed: dateability, fame, admission — the door's verdict. |
| conditioning on C | Only ever meeting the people who cleared the bar; sampling by outcome without noticing. |
| the wedge above the cut | Your lived sample — the pool, the shortlist, the ward, the feed, the people you remember. |
| induced negative r | The tradeoff you'd swear is a law of nature: "the good-looking ones are never kind." |
| the missing denominator | Everyone you never met — the only witnesses who could testify that the traits are strangers. |
where the metaphor tears
Budgets are real; time is finite; an hour in the gym is an hour not spent being kind to anyone. Plenty of tradeoffs are made by the world, not by the door. Collider bias is a suspect to interrogate, not a universal acid that dissolves every pattern you dislike. If the lens starts explaining away everything, you've stopped using it and started hiding behind it.
The clean fix — measure the unselected population — is mostly unavailable; nobody will hand you the people who never entered your life. In practice the defense buys discounted confidence, not recovered truth: you learn to trust the tradeoff less, not to know the real correlation. Beware the second-order mirage of "correcting" for a selection process you've merely guessed at.
For decisions confined to the wedge, the induced correlation genuinely holds: among people you can actually date, kindness does predict plainer looks, and the arithmetic doesn't care that the cause is the door. Selection is sometimes the point — a scout choosing among the drafted may lean on wedge correlations legitimately. The sin is exporting them: promoting a fact about your pool into a fact about people.