Path Dependent

Weaker than they feel.

Two papers on the safeguards we trust, a private conscience and a tidy story about a boom, and how little they actually do.

Issue 003Decision Intelligence6 min read

Say a shared spreadsheet lands in the team chat: the office giving drive. Everyone's name is down the side, and a list of amounts runs along the top. You already knew what you wanted to give. You type a higher number anyway.

A September paper found that wanting to look generous inflates giving. How much it inflates depends on whether anyone is watching and on how the question is asked.

1Who is watching, and how you are asked

In a Bonn lab, 697 people chose between taking money for themselves and sending a €350 donation to a charity (Bénabou et al., Games and Economic Behavior, 17 Sep 2026)1. In one version, the personal amount was a single, certain €100: yes or no. In the other, they walked down a list of amounts from €0 to €200, and one line was drawn at random. Half the sessions were double-blind and private. In the other half, people knew a small committee would watch, compare and judge their choice.

The cleaner result is the difference in how much publicity mattered under each way of asking. On the price list, moving from private to public raised giving at €100 by 23.6 percentage points.1 Under the single yes-or-no, the same shift in audience produced only a small, non-significant increase. That difference in sensitivity was significant at the 1-percent level.1

The ranking of the two methods did appear to flip once the choice became public. More people gave under the single yes-or-no when the choice was private, and more on the price list when it was public. Each of those contrasts, though, was only marginally significant. The authors flag this themselves. The firm ground is the interaction. The exact rates are more tentative.

A follow-up with 366 more people, using a university voucher with no moral image at stake, found no meaningful difference between the two ways of asking.1 The pattern showed up when being seen to do good was at stake. It did not show up for a university voucher.

This was a lab experiment with university-recruited adults in Bonn, with a mean age of 24. Only one matched pair per session had their choice paid out for real. It measures how the asking method and the audience inflate recorded generosity, and it says nothing about anyone's character.

Say you're back at that spreadsheet. Amounts run from €0 to €200, and someone will read the names out at the weekly meeting. This is the combination the lab found most sensitive: a watched choice, made on a list. Making a single yes-or-no public barely moved giving in the lab. Making a price list public raised giving at €100 by 23.6 percentage points. Imagine several colleagues who would have said no to a private €100. They now tick a higher number. Same charity, same colleagues, same pay packets. Only the audience and the form moved.

Or imagine a family group chat after a disaster. A relative writes that they have put in €200 and asks everyone to drop their amount below, beside a list of suggested sums. A cousin who had already settled, privately, on a smaller gift types €200 to match the room. Nothing about the appeal changed between the cousin's private decision and the message. An audience arrived, and a list came with it.

If you want to know what you would really give, decide once in private, as a single yes-or-no at a fixed amount, before anyone else's pledge or a public list is in front of you. If you are the one asking, treat a public pledge sheet and a sliding scale as generosity mixed with performance.

2The tidy stories about the bubble mostly didn't hold up

After a boom, everybody has an explanation. People got emotional. They lost their self-control. The wrong kind of buyer was in the room. In a large redo of the classic lab markets, those stories did not hold up.

Researchers collected new data from 166 lab markets and 1,544 traders, and re-tested 17 results from four well-known experimental-finance papers (Huber et al., The Journal of Finance, 29 Oct 2026)2. Of the 14 results originally reported as statistically significant, only three replicated.2 Traders with stronger cognitive reflection, fluid intelligence or Theory of Mind earned more, though the last two of those effects were only about 60% and 40% as large as first reported.2

Findings that excitement, low self-control or the gender mix of the market drive bubbles did not replicate. In several of those tests, the new estimates pointed in the opposite direction and were not significant.

A separate test of a long-standing finding in this literature did hold up. After one pass through the same market, all three bubble measures fell. Experience did not wipe the bubble out. Prices were still too high the second time.2

These were laboratory asset markets with induced fair values, traders in university labs, and experimental points converted into a modest payment. They are a clean test of the original lab findings, and they tell us little directly about housing markets, group chats or company share tips. A failed replication of these specific experiments also leaves open whether real-world investors get emotional.

Say a share tip in a group chat doubles, then collapses. The post-mortem arrives quickly. Someone got emotional and held too long. Someone else just didn't have the willpower to sell. A third theory blames the mix of people in the chat. Those are satisfying stories. They are also the kind of lab finding this study could not repeat. What did repeat is more ordinary. Imagine a friend in the same chat who has already lived through one boom and bust in a similar setting. They are less likely to pile in at the top the second time. Less likely still leaves plenty of room. In the lab, after one pass through the same market, all three bubble measures fell, and prices were still too high the second time.

The same kind of story shows up at Sunday lunch. A price spike in the next suburb gets a confident explanation: everyone got carried away, first-time buyers have no self-control, or we know who to blame. Discount a tidy post-mortem like that, at least until the claim has been tested again. Prefer a second pass with feedback, which is the one explanation here that has now passed twice.

Putting it together

The private pledge and the tidy post-mortem both feel like safeguards. Both do less than they appear to, and both are more conditional than they feel at first.

The same person, with the same money, can look more generous once a committee is watching and the form is a public list of amounts. Privacy takes the audience away, which is when the price list inflated giving most. A single yes-or-no ask moved much less when the audience changed. Image concerns, including self-image, are still there. And some of the neatest stories about a boom, the ones about emotion, willpower or who was in the room, did not survive a careful second test. Experience, which did hold up, shrinks a bubble and leaves it in place.

The habit is a pair of questions, asked before you trust either safeguard too far. Who will see this, and how is the question being asked? Has this story been through a second pass, or is it the satisfying one we reached for first? If the answer changes once the room is watching, or once you've seen the thing play out again, the safeguard was weaker than it felt. A private conscience and a tidy post-mortem will both offer to keep you honest. You can take the offer. You might want a second opinion first.

From Decision Intelligence Who is watching when you decide, and the story you tell afterwards, both get a vote. In our Superior Decisions workshop, your team works through one real, pending decision together. Enquire about a workshop.

Sources

  1. Bénabou, R., Falk, A., Henkel, L., & Tirole, J. (2026). "Eliciting moral preferences under image concerns: Theory and experiment." Games and Economic Behavior 160, 104044, available online 17 Sep 2026. doi.org ↩a ↩b ↩c ↩d
  2. Huber, C., Holzmeister, F., Johannesson, M., König-Kersting, C., Dreber, A., Huber, J., & Kirchler, M. (2026). "Do Experimental Asset Market Results Replicate? High-Powered Preregistered Replications of 17 Key Results." The Journal of Finance, published online 29 Oct 2026. doi.org ↩a ↩b ↩c ↩d