People Say Yes About Twice as Often as You Expect

Across 14,000 strangers, people underestimated their odds of a yes by 48 percent. Why the estimate fails, and what it means for a request sent to a screen.

In 2008 Francis Flynn and Vanessa Lake sent people out to ask strangers for things. Small things, mostly. Directions, the loan of a phone, a few minutes with a questionnaire, money for a charity.

Before each attempt the asker wrote down a prediction: how many strangers would I have to approach before someone agrees?

Most guessed too high, by about half.

The finding has held up better than most things in social psychology from that era. By the time Vanessa Bohns — the same researcher, publishing under her earlier name in the 2008 paper — reviewed the accumulated work in 2016, the experiments covered more than fourteen thousand strangers, and the average underestimation came out at 48 percent. Whatever number you have in your head for your odds of getting a yes, doubling it will usually get you closer than your first guess did.

Why the estimate is wrong

Bohns's explanation is not flattering to anyone, and I think that is the point of it.

When you ask someone for something to their face, you hand them a problem. Saying no costs them a small, immediate thing: a second of appearing unkind, in front of a person who is standing right there. You, the one asking, never feel that cost, because you are not the one who would have to say the word.

What you feel instead is the price of yes — ten minutes and a detour you didn't plan.

So both sides do the arithmetic on different ledgers, and the asker's ledger is missing the entry that actually decides it.

Read that way, the yes is not generosity but friction avoidance, which is a bleak thing to build anything on.

A second reading, from 2022

Xuan Zhao and Nicholas Epley ran a series of studies — scenarios, recalled experiences and live interactions, 2,118 participants in total — and found the askers wrong in three directions at once. They underestimated how willing people were to help and how good the helper would feel afterwards, and they overestimated how much of an imposition the whole thing was.

The specific error Zhao and Epley identified: askers assume a yes is compliance when it is mostly something else. People help because they want to, feel better for having done it, and the person who asked never sees that part.

One account says people say yes because refusing is awkward. The other says people say yes because helping is genuinely pleasant and we are bad at believing it. The behaviour looks the same either way; the implication does not.

Which one holds when nobody is standing in front of you

There is a test buried in the same literature, and it is unkind to the optimistic reading.

In 2017 Mahdi Roghanizad and Bohns had 45 people each ask ten strangers to fill in a survey. Half asked in person, half sent an email, and both groups predicted roughly the same result for themselves — about half would say yes.

The authors report the face-to-face requests succeeding 34 times more often. Thirty-four times, not thirty-four percent — though a ratio that large is partly a function of how close to zero the email condition landed.

Take away the physical presence and something collapses. Some of that is trust, since an email from a stranger reads as spam in a way a human being standing there does not. The rest is the mechanism Bohns described. Nobody has ever felt awkward closing a tab.

What this means for a map of requests

Good People Map is, structurally, a wall of asks and offers that nobody is standing in front of. Every request on it is the email condition.

I would rather not conclude that, but the evidence points there and pretending otherwise would only produce a worse product.

Two things follow. The first is that low response rates on a public request board are not evidence that people are unkind; they are what happens when you strip a request of the one thing that makes requests work. We asked in the weakest available format and then read the result as a verdict on human nature.

The second is what follows from that, though it is a hypothesis and not something we have tested: put the face back. A name and a photograph instead of an anonymous pin, a specific person with a specific problem instead of a category, wording that sounds like someone talking rather than a listing. None of it reproduces a person standing in front of you, but it is the nearest a screen gets to the condition where the 48 percent gap was measured in the first place.

And if Zhao and Epley have the second half of it right, there is a gap in how the map works today. Their finding was that helpers end up happier than the asker predicted. There is a thanks button on deeds already done, and it quietly adds karma to the person who did them. What does not exist is the other direction: someone who answers a request is never told what their help led to, so the one piece of evidence that would correct everybody's estimate stays invisible.

That seems like a cheap thing to add, and we have not added it.

Sources: Flynn, F. J., & Lake, V. K. B. (2008). «If you need help, just ask: Underestimating compliance with direct requests for help». Journal of Personality and Social Psychology, 95(1), 128–143. — Bohns, V. K. (2016). «(Mis)Understanding Our Influence Over Others: A Review of the Underestimation-of-Compliance Effect». Current Directions in Psychological Science, 25(2), 119–123. — Zhao, X., & Epley, N. (2022). «Surprisingly Happy to Have Helped: Underestimating Prosociality Creates a Misplaced Barrier to Asking for Help». Psychological Science. — Roghanizad, M. M., & Bohns, V. K. (2017). «Ask in person: You're less persuasive than you think over email». Journal of Experimental Social Psychology, 69, 223–226.