Plain-language answer
There is no single number that settles whether a coincidence “means something.” There is, however, a method: a sequence of questions that separates what actually happened from how it’s being interpreted, forces any probability claim to state its assumptions, and ends with an honest account of what is known, plausible, unsupported, and unresolved — rather than a verdict dressed up as one.
Why it matters
Most viral “impossible odds” claims fail at step three or four below, long before any arithmetic is needed: the facts haven’t been checked, or the matching rule was invented after the result was already known. Working through the method in order catches these failures early, instead of building an elaborate probability calculation on a foundation that was never solid.
The method
- Write the exact claim. Preserve the original wording and cite where it came from. Paraphrasing early is how claims quietly get more dramatic with each retelling.
- Separate event from interpretation. What actually occurred, and what meaning or mechanism is someone assigning to it? Keep these as two separate sentences.
- Verify the facts. Prefer contemporaneous or primary records over retellings, and log any contradictions you find rather than silently picking the version that fits best.
- Ask when the matching rule was chosen. Was the specific pattern specified before the events occurred, or constructed afterward to fit them? A prediction and a retrospective match are different kinds of evidence.
- Define a match, precisely. Fix tolerances: how close a date counts as “the same,” how loose a resemblance counts as “the same name,” whether order or location matters. Vague tolerances let almost anything count.
- Choose the reference class. Compared with which people, places, time period, and set of events? Different reasonable choices can give very different numbers — state which one you used.
- Count the opportunities. Include unnoticed failures and every implicit comparison, not just the one that was noticed. This is the step that “multiple opportunities” reasoning lives in.(Dunn, 1961)
- Model the dependencies. Shared culture, geography, seasonality, social networks, and common causes all break the naive assumption that separate “chances” are independent of each other.
- Estimate the uncertainty. Use a range and a sensitivity check where inputs are unclear, rather than a single confident-sounding figure.
- Compare explanations. Weigh chance under your model against a real causal connection, a reporting or memory error, a selection effect, fraud, and plainly unknown mechanisms — not just “chance” versus “not chance.”
- Update proportionally. Ask how expected the evidence is under each competing explanation; a low probability under one model is not, by itself, proof of another model.(Diaconis & Mosteller, 1989)
- State the residual. Finish with an explicit account of what is established, what is plausible, what is unsupported, and what remains genuinely unresolved.
The warning this method exists to enforce
Multiplying several guessed, non-independent probabilities together — “a 1-in-10 name match, times a 1-in-365 date match, times a 1-in-1000 location match” — produces an impressive-looking number that is usually meaningless. Real events share causes and contexts in ways that make naive independence assumptions wrong, and a guessed input raised to a false assumption of independence does not become more reliable by being multiplied with other guesses. If you can’t defend each input and the independence between them, state that plainly instead of reporting the product.
Worksheets and tools
Each of these develops one step of the method above into a standalone, practical tool:
- Reference-class worksheet — step 6, choosing and stating a comparison group explicitly.
- Source checklist — step 3, tracing a claim back toward a checkable primary source.
- Pre-registration for everyday predictions — step 4, applied as a personal habit for testing your own predictions.
- Common red flags — a fast triage list to run before committing to the full method.
- Worked example — all twelve steps applied to a real, already fact-checked case.
Common misconception
Working through this method is not the same as trying to prove a coincidence was “just chance.” Step 12 can just as easily conclude that a causal connection is plausible, or that the honest answer is “unresolved.” The method’s job is to make sure whichever conclusion you reach is one the evidence actually supports.
Related
- What is a coincidence? separates the components — event, surprise, improbability, meaning, cause — that this method walks through in order.
- Multiple opportunities and selection effects develops step 7 in full.
- Probability is a model, not a verdict underlies steps 6 through 9.
Key takeaways
- Most coincidence claims are best evaluated as a sequence of questions, not a single probability calculation.
- The facts, the matching rule, and the reference class should all be settled before any arithmetic is attempted.
- Never report the product of several guessed, non-independent probabilities as if it were a defensible number.