In Simple Terms
Two people can hear the same words about a treatment and come away with different ideas about how likely it is to work. A University of Tokyo study found that, for some participants, whether the wording sounded hopeful or discouraging mattered more than the number they thought it meant. Clear numbers and a conversation about what they mean may help people make better-informed medical decisions.
When the Same Word Means Different Odds
In an important medical decision, words such as “likely” and “unlikely” may seem like quick ways to describe the chance of success or failure. But they do not correspond to fixed numbers. A listener may interpret a phrase differently from what a doctor intended, and the emotional tone of the wording may influence a recommendation even when both people are discussing the same probability. This overlap between words and numbers was the focus of a study by researchers at the University of Tokyo, published in Psychonomic Bulletin & Review on October 1, 2026.
When a Word Does Not Mean the Same Percentage to Everyone
People rely on probabilities when making all kinds of decisions, from weighing a medication’s benefits to interpreting a weather forecast or comparing financial options. A probability can be stated as a specific number, such as a 30% chance of success, or described in words, such as “might work” or “is unlikely to work.” A number gives people a value they can compare; a verbal description leaves much more room for personal interpretation.
One listener might take a word like “possible” to mean there is only a small chance, while another might hear it as a sign that success is reasonably likely, or even very likely. That does not necessarily mean either person has misunderstood the language. Everyday words simply do not have precise mathematical boundaries. When the subject is a treatment or medical consent, however, these differences can matter: a patient and a doctor may use the same phrase while imagining very different odds.
The research team examined two connected steps: how a person translates probability words into a number in their mind, and how they then use that estimate, alongside the wording itself, to make a decision. The researchers wanted to know whether autistic and non-autistic people handled this information in the same way, while also recognizing that people may differ considerably within each group.
A Hypothetical Migraine Treatment
The researchers used a scenario based on a familiar kind of decision. Participants were asked to imagine that a close friend was considering an expensive treatment for chronic migraines and to judge how strongly they would recommend it, based on a doctor’s prediction of the outcome. They were not being asked to choose an actual treatment or give real medical advice. The aim was to measure how they responded to information about the chances of success.
The way the prediction was presented varied across trials. In some, participants saw a specific percentage, such as a 30% chance that the treatment would work. In others, the researchers used verbal descriptions with different tones: a more positive phrase suggesting that success was possible, or a more negative one saying that success was unlikely. Participants were then asked to estimate the numerical probability they associated with each verbal description.
This design allowed the researchers to separate two things that can get mixed together in everyday conversation. The first is the number a listener thinks a phrase represents. The second is the effect the phrase’s positive or negative tone has on a recommendation. If two participants assign similar numerical estimates to a phrase but make different recommendations depending on how it is worded, that suggests their decisions are not based on calculation alone.
Who Took Part, and How Was the Sample Put Together?
The research program included two experiments, with 138 adults in total included in the analyses. In the first experiment, the final analyses included 48 people diagnosed with autism spectrum disorder and 41 people who were not autistic. The second experiment used a separate sample of 49 non-autistic participants. Across the two experiments, that made 90 non-autistic participants, alongside 48 autistic participants.
This breakdown explains why different figures may appear in descriptions of the study. The overall total was 138, combining participants from two experiments: one included both comparison groups, while the other added a separate sample of non-autistic participants. The experiment was not an intelligence test or a judgment of any individual’s decision-making ability. It was an effort to understand the patterns that emerge when similar probabilities are presented in different numerical and verbal forms.
The work was led by Hideto Honda, Hirokazu Komazaki, and Professor Kazuhiro Ueda of the Graduate School of Arts and Sciences at the University of Tokyo. The team used hierarchical Bayesian analysis, followed by an analysis that grouped similar responses together. In simpler terms, these methods helped the researchers estimate how much words and numbers influenced participants’ choices, then identify groups of participants who responded in similar ways.
Three Ways of Handling Uncertainty
The findings did not reveal one pattern that separated all autistic participants from all non-autistic participants. Instead, the researchers identified three broad patterns in how people combined a phrase’s meaning with the numerical probability they thought it implied. A pattern strongly influenced by whether the wording sounded positive or negative was more common among autistic participants. A pattern that gave greater weight to the numerical estimate was more common among non-autistic participants. The third pattern included participants from both groups.
In the first pattern, recommendations changed noticeably depending on how the probability was presented: did the wording make the outcome sound encouraging or discouraging? The number participants inferred from the words played a relatively smaller role. As a result, describing the same chance in a positive or negative way did not necessarily produce the same response. The language used could have a strong influence on people’s impressions and choices.
For people in the second pattern, the numerical probability they inferred appeared to be the main factor in their decisions. If they understood a phrase to mean that the chance of success was low, that estimate was more closely tied to their recommendation than whether the words sounded positive or negative. This does not mean they were completely unaffected by language. Rather, within the task, their inferred number had a stronger presence in their decisions.
In the third pattern, the two influences appeared more evenly balanced: the numerical probability suggested by the phrase and the positive or negative tone of its wording. This mixed pattern shows that people do not simply fall into two fixed categories, and that an autism diagnosis alone does not predict how every individual will respond. The cluster included autistic and non-autistic participants, highlighting the importance of individual differences.
The Numbers Participants Read Into Words
The study also found that autistic participants assigned verbal probability descriptions numerical estimates that were less distinct from one another than those assigned by non-autistic participants. In general, the gaps between the percentages they associated with different expressions were smaller. This does not mean every autistic participant interpreted the words in the same way, nor does it reveal how any one person was thinking. It describes an overall difference found in the group data.
Notably, the similarity of participants’ numerical estimates did not stop wording from influencing their decisions. A person might associate several phrases with fairly similar numbers but still change their view when the wording shifts from encouraging to discouraging. This suggests that asking “What number does this word mean to you?” is not enough to understand a choice. It also matters to ask, “How did the way it was put affect how you viewed the outcome?”
It is important not to treat this finding as a complete description of autistic people or as a rule that applies in every situation. Multiple response patterns appeared among the participants, and one group included both autistic and non-autistic people. These findings are a reason to be cautious about applying group averages to an individual—and to ask that person what they understood and what influenced their decision.
What the Findings Could Mean in a Clinic
Imagine a doctor saying that a treatment has a 30% chance of success. If the same percentage is described as a “possible” outcome one time and as “unlikely” another time, a patient may come away with different impressions even though the underlying number has not changed. The study suggests that some people whose decisions are sensitive to wording may give that difference more weight than the numerical estimate they inferred.
The findings do not mean that doctors should stop using words or assume that numbers alone will solve every communication problem. A percentage can be hard to understand if it is given without context: compared with what is the chance of success? Which group of patients does it apply to? And what counts as success? The practical lesson is to pair the number with a clear explanation, rather than replace one way of communicating with another.
For example, a doctor could give the percentage, explain what it means for the patient’s situation, describe the limits of the prediction, and discuss other benefits and risks to consider. The doctor could then ask how the patient understood the probability, or invite them to explain in their own words what they took from it. That kind of conversation can uncover misunderstandings instead of relying on a quick question about whether the information was clear.
The same idea applies beyond health care. Teachers and others explaining uncertain information may benefit from making their meaning explicit, not relying on flexible words alone, and checking that the other person has understood the point as intended. The study does not prescribe a single script for everyone to use. Instead, it encourages people to remember that a phrase that seems clear to the speaker may not carry the same meaning for the listener.
What the Experiment Does Not Settle
The findings describe response patterns in a hypothetical task, but they do not identify the precise psychological reason why someone falls into one pattern rather than another. The researchers raised possibilities for future study, including how people infer intended meaning from everyday language and how much mental effort it takes to process information in a given situation. The study does not establish that any one factor explains the results.
Making an imagined decision about treatment for a friend is not the same as making a real decision about one’s own care. In a real situation, other factors may come into play, including past experience, health, concerns, and how the doctor explains the options. The team therefore called for further research, including studies that follow participants over time and cognitive tests that could help explain the mechanisms behind the differences. It remains important to distinguish what this task actually showed from what still needs to be investigated.
The paper, whose title concerns understanding verbal and numerical probability information and making decisions among autistic and non-autistic people, was published in Psychonomic Bulletin & Review. Its digital object identifier is 10.3758/s13423-026-03032-0. The research was supported by Japanese Grants-in-Aid for Scientific Research, numbers JP22H03911 and JP25H01131.
The Bottom Line
The University of Tokyo study suggests that describing probabilities in words is not a neutral substitute for giving people numbers. Positive or negative wording can affect a decision even when someone has formed a numerical estimate of the chance. The researchers found three response patterns: one more influenced by how a phrase was framed, one that gave greater weight to numbers, and one that balanced the two. Each pattern included people with different diagnostic backgrounds, and the results should not be applied to every individual. Still, they offer a practical reason to explain probabilities and their context clearly, then use conversation to check that the message has come across as intended.