Probability words
What "likely" and "very likely" mean to two expert systems, why they disagree, and this manual rule: say the number.
When you say "likely," a listener hears a number, and different listeners hear different ones. Two expert scales show how far apart the same word can sit. Communicating uncertainty uses the table to show why a word alone is not enough.
The two scales are written for their own authors: IPCC lead authors and US intelligence analysts. Neither is a standard for an interview, and neither tells you what number to say. This manual rule is a convention: pair any probability word with the number or range, and say where it came from. Whether a number is sound is for the author of the estimate to say.
The two scales side by side
| Term | IPCC AR5 (probability) | ICD 203 (probability) | Note |
|---|---|---|---|
| Virtually certain / almost certain | 99 to 100 percent | 95 to 99 percent ("almost certain" or "nearly certain") | ICD ranges never reach 0 or 100 percent. |
| Very likely | 90 to 100 percent | 80 to 95 percent ("very likely" or "highly probable") | The same words, different lower bounds. |
| Likely | 66 to 100 percent | 55 to 80 percent ("likely" or "probable") | IPCC ranges nest, so a "very likely" event is also "likely"; ICD bins do not overlap. |
| About as likely as not / roughly even chance | 33 to 66 percent | 45 to 55 percent | Width differs by a factor of more than three. |
| Unlikely | 0 to 33 percent | 20 to 45 percent ("unlikely" or "improbable") | An event the IPCC calls unlikely can be a 30 percent chance. |
| Very unlikely | 0 to 10 percent | 5 to 20 percent ("very unlikely" or "highly improbable") | |
| Exceptionally unlikely / almost no chance | 0 to 1 percent | 1 to 5 percent ("almost no chance" or "remote") |
Sources: IPCC, Guidance Note for Lead Authors of the Fifth Assessment Report (2010), Table 1; Office of the Director of National Intelligence, Intelligence Community Directive 203 (2015, technical amendments 2022), probability table. Rows are matched by the plain meaning of the term; the IPCC scale also has optional terms (extremely likely, 95 to 100 percent; more likely than not, above 50 percent) that ICD 203 does not.
What the research says
- People read verbal probability terms in very different ways, and attempts to fix a translation, such as the IPCC's, have not succeeded. Giving the number beside the word produced more accurate interpretation than either alone, in studies the National Academies reviewed. Research finding Source: National Academies of Sciences, Engineering, and Medicine, Communicating Science Effectively: A Research Agenda (2017), Chapter 2.
- The FDA's guide recommends numeric likelihoods over words alone, which raise risk perception and are read differently by different people. Research finding Source: FDA guide (2011), Chapter 7.
- The IPCC and ICD 203 both keep a second scale for confidence, separate from likelihood. ICD 203 says not to put a confidence level and a likelihood in the same sentence.
This manual rule
- Use the word only with the number. "Likely, about 2 in 3" tells a listener more than either. If you have a count, say the count: "2 of the 14 blooms in the past five summers."
- Say where the number came from. Your own data, a study, an agency.
- If there is no number, say that. "We do not have a number yet; the state lab results are due Thursday" is an honest statement. Do not borrow a word from a scale you did not use.
Dr. Lena Ostrowski, a fictional hydrologist, is asked, "Is the water going to be toxic?" She does not answer "likely" or "unlikely." She says that in her own records, 2 of the 14 blooms on this lake in the past five summers tested above the state's recreational advisory level, that this bloom's results are due from the state lab, and that the state health department decides about an advisory. Known: her five-year record. Open: this bloom's result; the state lab settles it.