Social Influence BriefingA daily communication practice for people who influence through ideas

Honest Uncertainty

Published · Last reviewed

State uncertainty plainly, and in numbers where you can: a best estimate with its range beside it, in place of words such as “may” or “some say”. In five experiments (total n = 5,780) on facts and numbers in news-style texts, van der Bles and colleagues (2020) found only a small drop in trust, and mostly for verbal uncertainty. A numeric range given with a point estimate did not significantly reduce trust in the number or its source.

Durable practice (not new). This is established communication practice, not a discovery or a trend.

What it fixes

Two breakdowns. One is the confident claim that turns out to rest on a small sample, and the audience learns of the doubt later from someone else. The other is the cloud of hedges (“might”, “possibly”, “some suggest”) that tells the audience you are unsure and tells them nothing about how unsure or why.

How to run it

  1. List each claim in your piece that rests on a measurement, a sample or a forecast.
  2. For each one, look up the range or margin the source itself reports. If the source gives none, write that down instead of inventing one.
  3. Write the number as a best estimate with its range next to it, in the form “about X, with a range from A to B”.
  4. Add one sentence on what the range means for the decision your audience faces.
  5. Where no number exists, name the specific reason for doubt (a small sample, a single study, self-reported data) in place of a vague hedge.

“This figure comes from a survey, so it has a margin. The best estimate is about X, and the range runs from A to B. I am giving you the range so you can see how firm the number is.”

Take the range from the source. If a source reports only a single figure, say so in plain words (“the report gives one figure and no range”) and do not construct an interval of your own. A range you made up conveys precision that does not exist.

Say it out loud

The sentence above is the disclosure: it says what the range is, where it comes from and why you are giving it. It still works with a fully informed listener, because its purpose is to let them judge for themselves. Honest uncertainty also carries an ethical boundary: do not use it to look careful while hiding a weak claim, and do not stack caveats until the claim cannot be found.

Profile translations

How to tell it worked

Action
Put the range next to the number every time the number appears.
Why
A range separated from its number is easy to miss and easy to drop in a retelling.
Verification
A reader who quotes your figure quotes the range with it.
Action
Ask two people to restate how sure you are.
Why
The aim is that the listener leaves with the right degree of confidence.
Verification
They answer with a best figure and a spread, such as “about this, could be between these,” rather than “pretty sure” or “not sure”.
Action
Add a sentence on what the range means for the decision.
Why
A range without a consequence leaves the audience to guess whether it matters.
Verification
A follow-up question concerns what to do at the low or high end, in place of whether the number is real.

Failure signs: stop if you see these

  • You construct a range the source never gave, so the precision is invented.
  • The verbal hedges multiply (“may possibly suggest”) and no number appears.
  • You use the word “estimated” as the only signal, which the study found did not change how uncertain people judged a figure to be.
  • Listeners begin to treat every claim you make as equally shaky, which suggests that the range is not tied to the actual doubt.

Why it works, and how strong the evidence is

Evidence. Moderate to good for its scope. Five experiments (total n = 5,780), including a preregistered replication with a nationally representative UK sample and a field experiment on the BBC News website. The scope is limited to uncertainty about facts and numbers in news-style texts, with UK samples; the field sample was self-selected. Effects on teaching, content or sales communication: Not reported.

van der Bles, van der Linden, Freeman and Spiegelhalter (2020, PNAS) tested the assumption that communicating uncertainty reduces public trust. Participants read news-style texts in which uncertainty about facts (such as unemployment, global warming and immigration figures) was expressed verbally or numerically, and at different magnitudes. People did perceive more uncertainty when it was communicated. The authors observed only a small decrease in trust in the numbers and in the trustworthiness of the source, mostly for verbal uncertainty. A numeric range given with a point estimate did not significantly reduce perceived trust in the number or its source, and across the studies, communicating uncertainty never significantly increased trust either.

The effects differed by how uncertainty was expressed. Words such as “estimated” or “about” did not lead people to perceive uncertainty, and explicit verbal statements of uncertainty were the ones that lowered trust in both numbers and source. The authors recommend, for basic facts and numbers in the media, a numerical range with a point estimate.

Keep to the paper's scope. It addressed uncertainty about measurable facts about the past and present. The authors acknowledge that people may react differently to uncertainty about the future, to emotionally charged topics such as personal health outcomes, and in other cultures. Their prior-attitude results are narrow too: attitudes affected baseline trust but did not change how people responded to uncertainty. Applying the finding to a class, a video or a campaign is a reasonable extension, but an untested one.

Worked example

An illustration, not data. A teacher shares the result of a class survey with parents.

Before

“Most students are keeping up with weekly practice, and results suggest they might be improving.”

After

“In our survey of 120 students, about 62 of every 100 said they completed weekly practice. With a sample this size, the range is roughly 53 to 71 of every 100. That is enough to say most do, and too wide to say whether it rose since last term, which we did not measure.”

The before version hedges with “most”, “suggest” and “might” and gives the listener nothing to repeat accurately. The after version gives a best estimate, a range, the sample size and a statement of what the numbers cannot show. The figures are invented for the illustration, and the margin follows from the sample size under a simple random sample. In practice, take the range from the source or work it out with the person who ran the survey.

Frequently asked questions

Does admitting uncertainty lower the trust people place in you?

In five experiments on facts and numbers in news-style texts, van der Bles and colleagues (2020) found only a small decrease in trust, mostly for verbal uncertainty. A numeric range with a point estimate did not significantly reduce trust in the number or its source.

Is a numeric range better than words like “may” or “probably”?

In that study, verbal expressions of uncertainty were the ones that lowered trust, and words such as “estimated” did not change how uncertain people thought a figure was. The authors recommend a numeric range with a point estimate for basic facts and numbers.

Does the research cover forecasts or personal health?

No. The paper concerned uncertainty about measurable facts in the past and present, with UK participants. The authors note that other kinds of uncertainty, including emotionally charged topics, may provoke different reactions.

What if my source gives no range?

Say so plainly, for example that the report gives one figure and no range, and name the specific reason for doubt. Do not invent an interval.

Will stating uncertainty increase trust?

Not according to this paper: communicating uncertainty never significantly increased perceived trust. The case for stating it rests on accuracy and honesty, with little cost to trust found for numeric ranges.