Resolving the Abilene Paradox
On a hot afternoon visiting in Coleman, Texas, the family is comfortably playing dominoes on a porch, until the father-in-law suggests that they take a 50-mile (80-km) trip to Abilene for dinner. The wife says, "Sounds like a great idea." The husband, despite having reservations because the drive is long and hot, thinks that his preferences must be out-of-step with the group and says, "Sounds good to me. I just hope your mother wants to go." The mother-in-law then says, "Of course I want to go. I haven't been to Abilene in a long time."
The drive is hot, dusty, and long. When they arrive at the cafeteria, the food is as bad as the drive. They arrive back home four hours later, exhausted.
One of them dishonestly says, "It was a great trip, wasn't it?" The mother-in-law says that, actually, she would rather have stayed home, but went along since the other three were so enthusiastic. The husband says, "I wasn't delighted to be doing what we were doing. I only went to satisfy the rest of you." The wife says, "I just went along to keep you happy. I would have had to be crazy to want to go out in the heat like that." The father-in-law then says that he only suggested it because he thought the others might be bored.
The group sits back, perplexed that they together decided to take a trip which none of them wanted. They each would have preferred to sit comfortably, but did not admit to it when they still had time to enjoy the afternoon.
“In the Abilene paradox, a group of people collectively decide on a course of action that is counter to the preferences of many or all of the individuals in the group. It involves a common breakdown of group communication in which each member mistakenly believes that their own preferences are counter to the group's and, therefore, does not raise objections, or even states support for an outcome they do not want.”
Alexander Kiam’s Masters Thesis in Economics provides a gamified solution, that I played out at a New York City rationality meetup. I remember it going ok, but not great, but I suspect it could be pretty fun. It’s only proven to work in the limit, but that shouldn’t limit you! (I mean, maybe it should, up to you).
Why do this
Things get a bit complicated in step 3, to a degree that I suspect it’s not worth doing in the vast majority of cases. The really interesting bit (according to me) is doing step 1 and repeating until you as a group are better calibrated than you were before, and then testing that calibration in step 2, and noticing how your views on how your preferences relate to the group’s change and whether you’re right about that.
Doing it
The mechanism works for an odd number of people, voting on a binary choice (A vs. B) where they don’t want to reveal their true preferences. (The appendix, which I will post if given permission, has a way to adapt it for even numbers).
Step 1: Everyone figures out whether they believe they support option A more or less than the median in the room, and reveals their answer to this simultaneously.
So in a room of 11 people, let’s say 8 people believe they support A more than the median and 3 people believe they support A less than the median. This can’t be true!
Step 1.5: So if that happens, you run that vote again, and repeat until the difference between the votes is 1, meaning people are about calibrated. Let’s say it gets to 6 people believing they support A more than the median and 5 people believe they support A less than the median.
Step 2: Vote (simultaneous, via raised hands or some other mechanism). (Swap A and B below if B gets more votes in step 1).
If the 6 people above who believed they supported A more than the median vote for A, A wins.
If the 5 people above who believed they supported A less than the median plus any other people vote for B, go to step 3.
If neither of those happen, something went wrong with step 1, so go back and repeat step 1.
Step 3: Take the people referenced in the underlined any other people two lines above (that is the voters who both claimed in the most recent iteration of Step 1 to be more in support of A than the median voter and voted for B in the most recent iteration of Step 2)
Have these people simultaneously reveal if they think it is more likely than not that they supports B the most among this smaller group. Repeat until you narrow it down to one person.
If that person votes for B, B wins, it has more votes. (Why didn’t B win in step 2? Because maybe this person was lying about their preferences to be part of the majority (since it was done simultaneously)).
If she votes for A, the rest of the small group vote. Persumably, they should vote for A, since the B-est member voted for A. If that happens, A wins.
If not, then the narrowing down to one person didn’t work, so you repeat step 3 without the person who already voted.
Continue until a) someone votes for B in Step 3 or b) each Step 3 voter has voted for A in Step 3. In the former case, B wins; in the latter, A does.
