METHODOLOGY
Every number needs context.
A probability is useful when you can understand what it measures, where it comes from, and what might change it.
This page sets out the standards that guide our explanations and the questions to ask when reading an odds claim. For evidence about a particular story, look at that article’s sources, assumptions, and review information.
ON THIS PAGE
01
Start with a clear question
Before an odds claim can mean much, it needs a definition. What counts as the event? Who or what is included? Over what period?
A chance per trip and a chance over a lifetime answer different questions. A figure for one country, age group, or set of conditions may not describe another.
Our standard is to make the relevant population, time period, and conditions clear. If a question is too vague to support a useful number, the explanation should say so.
02
Trace the evidence
Our preferred starting points are original research, official datasets, and the organizations responsible for collecting the information. Reporting and other secondary sources can add context or help locate the original evidence.
A source’s date matters, as do its methods and coverage. A recent article may quote an older study. A large dataset may still leave out the people or situations a reader cares about.
The goal is to give readers enough source information to follow the claim back to its basis. When an original source is unavailable or evidence is limited, that limitation belongs in the story.
03
Show how the answer is reached
Some probabilities come from a mathematical model. Others are estimates based on observed events. A forecast introduces further assumptions about what may happen next. An explanation should identify which kind of number it uses.
For calculations, the important ingredients are the inputs, the method, and the assumptions. Rounding should make a result easier to read without suggesting more precision than the evidence supports.
A simple example: two heads in a row
Assume a fair coin and two independent tosses.
Each toss has a 1/2 chance of heads. Multiply those probabilities:
1/2 × 1/2 = 1/4 = 25%
Under this model, HH, HT, TH, and TT are equally likely. One of those four outcomes is two heads.
“Independent” means the first toss does not change the probability of the second. The answer depends on that assumption and on the coin being fair. A 25% chance does not guarantee one success in every four attempts.
04
Leave room for uncertainty
A neat-looking number can hide a wide range of possible answers. Sample size, missing information, changing conditions, and different definitions can all affect an estimate.
When the evidence supports a range, our standard is to preserve that range. Comparisons need compatible populations and timeframes, and should explain meaningful differences rather than imply they are interchangeable.
Sometimes the most accurate answer is: there isn’t enough evidence to calculate reliable odds.
05
Make the visual explain the number
A chart should help you understand the relationship being described. Its labels, units, scale, and timeframe need to agree with the accompanying explanation.
Illustrations serve a different purpose: they help tell the story. A dramatic image is not evidence, and a conceptual illustration should not be read as a photograph of a real event.
06
Make review and corrections visible
A publication date tells you when a story appeared. A last-checked date, when shown, should refer to an actual review of the relevant claims or sources. A named reviewer should reflect a review that person performed.
Articles without review details should not be assumed to have received an independent fact-check. Likewise, a recent page update does not necessarily mean its underlying data is recent.
Our correction standard is to acknowledge substantive changes to a number, calculation, source, or conclusion with a note explaining what changed. Minor spelling and formatting fixes do not need the same treatment.
A note on AI
AI can assist with research organization, drafting, code, and illustrations. It is not a source or a fact-checker. Any claim developed with AI still needs supporting evidence, and a generated calculation still needs to be checked. Responsibility for published material remains with the publisher.
Something doesn’t add up?
Send us the article URL, the claim you’re questioning, and any source or calculation that may help. Specific feedback makes it easier to investigate.