People are bad at predicting whether a direct request will work, and they are wrong in a consistent direction. Across a research program summarized by Vanessa Bohns of Cornell University in a 2016 review in Current Directions in Psychological Science, participants who approached more than 14,000 strangers underestimated how many would say yes by around 48 percent.

This is not a single result. It is a set of related findings accumulated over roughly a decade, with the same asymmetry appearing across different requests, different populations, and different researchers. That consistency is the interesting part, and it is also where the misreading starts.

What the studies actually asked people to do

The founding paper is Francis Flynn and Vanessa Lake’s 2008 study in the Journal of Personality and Social Psychology, titled “If you need help, just ask.” The method was simple. Participants were sent out to ask strangers for a small favor, and before they went, they predicted how many people they would have to approach before someone agreed.

For a questionnaire taking five to ten minutes, participants predicted they would need to ask about 20 people. The actual figure was about 10. For borrowing a cell phone to make a call, the prediction was 10 and the reality was 6. For asking a stranger to walk them to the campus gym, the prediction was 7 and the reality was just over 2.

A third study moved out of the campus setting and followed volunteers fundraising for the Leukemia and Lymphoma Society. Those volunteers expected to approach roughly 210 people and approached about 122. They also underestimated the average donation, predicting about $48 against an actual $64.

The direction never reversed.

The mechanism is not confidence, it is the awkwardness of saying no

The obvious explanation is that requesters are pessimistic or insecure. Bohns argues for something more specific: requesters fail to appreciate how uncomfortable it is to refuse a request delivered in person.

The person asking is focused on their own position. They can feel their own nervousness, they can imagine being turned down, and they have no direct access to what the other person is experiencing. The person being asked is in a different situation entirely. Saying no to someone standing in front of you carries a social cost that the requester, busy managing their own discomfort, does not price in.

That framing does useful work, because it makes a prediction that can be checked: the gap should be widest in person, where refusing is most awkward, and should shrink as the request moves further away from a face.

It also predicts something less comfortable.

The same blind spot applies when the request is a bad one

If the effect were about helpfulness, it would stop at helpful requests. It does not.

In a 2014 paper in Personality and Social Psychology Bulletin, Bohns, Mahdi Roghanizad and Amy Xu ran the same design with unethical requests. In one study, participants enlisted strangers to tell a white lie. In another, they enlisted strangers to commit a small act of vandalism. Two further studies used hypothetical scenarios, including buying alcohol for children and taking office supplies home.

Across all four, the people doing the asking underestimated how much pressure their request carried, and how many people would go along with it. The authors describe instigators failing to recognize the social pressure they place on actors through what feels to them like a casual suggestion.

This is the part that gets left out when the research is repackaged as encouragement to ask for what you want. The finding is not that people are generous. The finding is that a direct request from a person standing in front of you is harder to refuse than the person making it understands, and that holds whether the request is worth agreeing to or not.

A separate line of work on what goes unsaid

A related asymmetry shows up on the other side of the same coin. In 2022, Nicole Abi-Esber, Jennifer Abel, Juliana Schroeder and Francesca Gino published five experiments with 1,984 participants in the Journal of Personality and Social Psychology on constructive feedback.

Their pilot field study is the memorable one. Survey administrators worked with something visible on their own faces, food or marker, and the researchers counted how many of the 155 participants mentioned it. The figure was 2.6 percent.

Across the experiments, people underestimated how much the other person wanted to be told. The authors attribute this to two things pulling together: the giver weighs their own discomfort heavily, and overestimates how badly the receiver will take it. A perspective-taking prompt narrowed the gap more than the other interventions tested. In the final experiment, receiving less feedback was associated with less improvement.

These are different research groups asking different questions, and the overlap is not a coincidence so much as a common cause. Both effects come from the same place: we model our own discomfort in detail and the other person’s only roughly.

What this work does not show

The samples are narrow. Most of the compliance studies were run with undergraduates and passers-by on and around North American university campuses, asking strangers for small, low-cost favors. Whether the same gap holds for large requests, for repeated requests to the same person, or in cultures with different norms around refusal is not something these designs can answer.

The 2016 review is also a review of a research program by one of its central figures, which is normal for the format and worth knowing when reading the effect size. Bohns co-authored much of the work being summarized.

There is a further limit that matters more than either. These studies measure the gap between prediction and compliance. They do not measure whether the person who said yes was glad they did, and they do not follow what happened afterward. Compliance is a behavior, not an endorsement. A study that counted how many people agreed and then asked them a week later how they felt about it would be answering a different and arguably more useful question.

What would settle the open questions

Two things would help. The first is more work outside North American campus populations, since almost everything here rests on a narrow slice of participants approached in a narrow set of situations. The second is a design that separates the requester’s optimism about the outcome from the target’s experience of being asked, which the current paradigm largely collapses into a single yes-or-no measure.

The practical reading people usually take from this literature is that they should ask more often. That is defensible on the evidence for ordinary favors. The 2014 paper is a reminder that the same mechanism does not check whether the request deserved a yes.