Here are the rules they used. You are the estimator. You will see a jar of coins for ten seconds, from across the room, and then guess what is in it. Before you guess, you get a suggestion from an adviser who was allowed to hold the jar and examine it closely.
The adviser knows more than you do. The adviser is also being paid. Whether that is a problem, and whether you are told about it, is the entire experiment.
Ten seconds. Then guess.
The jar will be visible for ten seconds only, exactly as in the original study.
No further notes on this report.
You are paid on accuracy. Closer to the true value, more money.
Same game. New jar. One addition.
A different jar, a different adviser. Ten seconds again.
Note: The advisor is paid based on how high the estimator is in estimating the worth of the jar of coins.
You have now been told exactly how the adviser gets paid. You are still paid on accuracy.
Your result
What you did
| Round | Adviser said | You said | Actually was | You were off by |
|---|---|---|---|---|
| 1 · undisclosed | $17.30 | — | $15.58 | — |
| 2 · disclosed | $26.38 | — | $19.83 | — |
What happened to the 147 people who did this for real
Carnegie Mellon ran three groups. In the first, the adviser was paid for your accuracy, so there was no conflict at all. In the second, the adviser was paid for how high you guessed, and nobody told you. In the third, same conflict, but printed on the form in bold, exactly like the note you just read.
The part that should bother you
People did discount the advice when they were warned. That is the reassuring half. Here is the arithmetic of the other half, in percentage points measured against the true value of the jar:
The warning worked. It just did not work enough. The adviser moved further than the listener pulled back, and the gap landed on the listener. Which shows up exactly where you would expect, in what people took home:
Being told the truth about the conflict cost people 21% of their earnings versus being kept in the dark. The disclosure did not protect them. It licensed the adviser to push harder, and it handed the listener a warning they could not price.
So what do you actually do with this
Not "distrust everyone who discloses." Disclosure is still better than concealment, and the fix is not to stop disclosing. The fix is to stop treating a disclosure as if it settles anything. When someone tells you how they are paid, that is the beginning of your work, not the end of it.
Two questions survive this experiment intact. Can I see the work, meaning the assumptions and what happens to the answer when one input changes? And: what happens to you if I say no?
Where these numbers come from
Every figure on this page is from Daylian M. Cain, George Loewenstein and Don A. Moore, The Dirt on Coming Clean: Perverse Effects of Disclosing Conflicts of Interest, Journal of Legal Studies vol. 34, no. 1 (2005), pp. 1–25, and from Cain's doctoral dissertation of the same title at Carnegie Mellon's Tepper School of Business, where Loewenstein and Moore were his co-chairs.
Study 1 used 147 Carnegie Mellon undergraduates across six jars worth $10.01, $19.83, $15.58, $27.06, $24.00 and $12.50, a true average of $18.16. Estimators saw each jar for about ten seconds from roughly three feet away. Mean adviser suggestions were $16.48 (no conflict), $20.16 (conflict hidden) and $24.16 (conflict disclosed). Mean estimator guesses were $14.21, $16.81 and $18.14. Estimator earnings per round were $1.64, $1.59 and $1.25. The disclosure note reproduced above is the wording used on the original paper form.
One thing here is not from the study: the two adviser suggestions you were shown. The paper reports condition averages, not a suggestion for each individual jar, so each one was scaled from its jar's true value using that condition's published average. Everything else is as published.
This page is general educational information, not individualized investment advice.