Plain-language answer
A conditional probability is the chance of one thing given that another thing is already known to be true — written P(A given B). It answers a different question from the plain probability of A on its own, P(A), and the two are easy to mix up. Coincidence reasoning depends on conditional probability constantly, because the interesting question is almost always “how likely was this pattern, given everything else we already know?” — not the probability of the pattern in a vacuum.
Why it matters
A classic confusion: mixing up P(evidence given innocent) with P(innocent given evidence). Knowing that a rare event is unlikely to occur by chance — P(match by chance) is tiny — is not the same as knowing that chance is an unlikely explanation for that match, once you also account for how many other explanations were available and how likely the match was under each of them. Confusing these two conditional probabilities (sometimes called the prosecutor’s fallacy in legal contexts) makes small chance-probabilities look like much stronger evidence against coincidence than they really are.
Worked example: frequencies make the condition visible
People are far more accurate at conditional-probability problems when the same numbers are presented as natural frequencies instead of percentages. Compare:
- “A condition has a 1% base rate. A test for it is 90% accurate for people who have the condition and gives a 9% false-positive rate for people who don’t. If someone tests positive, what’s the probability they have the condition?”
- “Out of 1,000 people, 10 have the condition. Of those 10, 9 test positive. Of the 990 who don’t have the condition, about 89 also test positive. Out of the 98 people who test positive, how many actually have the condition?”
The second version, phrased in natural frequencies (about 9 out of 98, roughly 9%), reliably produces far more correct answers than the equivalent single-event-probability version, even though the underlying arithmetic is identical.(Gigerenzer & Hoffrage, 1995) The lesson generalizes directly to coincidences: “a match this close happens by chance only 1% of the time” sounds like strong evidence against chance, until it is reframed as “out of everyone who ever looks for this kind of match, how many will find one by chance alone, and how does that compare with how many will find one for any other reason?”
Common misconception
“The chance of this exact match happening randomly is tiny, so something else must be going on” treats P(match given random chance) as if it were P(chance given this match) — the very confusion the frequency-format research was designed to expose. The second quantity depends not only on how rare random matches are, but on how many opportunities there were for a match to be sought out and noticed, and how plausible the competing explanations are.
Limits and open questions
Reframing a problem in natural frequencies helps enormously but does not supply numbers that were never estimated in the first place — it only makes existing numbers easier to reason about correctly. For most everyday coincidence stories, the relevant conditional probabilities (how many opportunities existed, how flexible the matching criteria were) are not known precisely and have to be estimated, with the uncertainty stated explicitly rather than hidden behind a single misleadingly precise figure.
Related
- Base rates covers the prior information that a conditional probability updates.
- Independence and dependence looks at when P(A given B) is simply equal to P(A) — meaning B tells you nothing about A at all.
Key takeaways
- “How likely is this, given what we already know?” is usually the right question — but it is easy to accidentally answer a different conditional question instead.
- The probability of a rare match occurring by chance is not the same as the probability that chance is the explanation, once other possibilities and opportunities are counted.
- Reframing a probability problem as natural frequencies (out of 1,000 people…) rather than percentages makes conditional reasoning much less error-prone, for readers and analysts alike.