El Farol Bar attendance experiment

The El Farol Bar attendance experiment, devised by Ann M. Bell and William A. Sethares, offered a striking alternative to a coordination problem: people trying to decide whether to visit a bar whose appeal depended on how crowded it would be. By following their own recent experience rather than trying to predict everyone else, the simulated patrons brought attendance close to the bar’s ideal capacity. Yet their stable solution divided the group into regulars and rare visitors, making the experiment a demonstration not only of collective intelligence but also of its uneven benefits.

A different approach to coordination

The experiment addressed the El Farol problem, in which every person’s best choice depends on what everyone else chooses. If too many people go to the bar, it becomes overcrowded; if too few go, potential patrons miss an enjoyable evening. Unlike Brian Arthur’s computer agents, who used diverse strategies to anticipate the behavior of the crowd, Bell and Sethares’s bargoers all followed the same simple rule: they went if their recent experiences had been good and stayed away if those experiences had been bad.

This rule asked little of the participants. They did not try to work out what other bargoers were thinking, and they did not know how crowded the bar had been on nights they had skipped. Their information was local and personal: whether they themselves had recently enjoyed going. The experiment therefore tested whether individual experience, without sophisticated forecasting or explicit coordination, could nonetheless produce a workable group outcome.

Attendance finds a stable level

After enough time had passed for each person to build experience of the bar, weekly attendance settled just below 60 percent of capacity. That result was only a little worse than what an ideal central planner could have achieved. Looking no further than their own experience, the bargoers arrived at an attendance level that kept the bar from being too crowded while still allowing a substantial share of the group to enjoy it.

The apparent success depended on a division within the group. It quickly sorted into regulars, who went nearly every week, and people who went only rarely. The arrangement was defensible in utilitarian terms if everyone got equal pleasure from a visit: more than half the people could attend nearly every week and usually enjoy an uncrowded bar. But a sizable portion of the group rarely went and generally had a bad time when they did. A stable average, in other words, did not mean that everyone shared the benefits equally.

Why no solution is obviously best

Bell and Sethares’s outcome was not clearly superior to Arthur’s, and both approaches seemed surprisingly effective. The experiment brings out why coordination problems are difficult: when each person’s choice depends on everyone else’s, decisions affect one another, and there is no independent reference point that settles what the group should do. The bar’s attendance was not simply a fact waiting to be measured; it was partly produced by the very decisions the participants were trying to make.

That makes this problem different from estimating the weight of an ox, where a crowd can pool judgments about an external reality. At El Farol, the object of prediction changes as people act on their predictions. Against that self-referential difficulty, even getting the average attendance approximately right looks like an achievement. The experiment’s significance lies less in proving that one strategy always wins than in showing how independent, experience-based decisions can generate a workable pattern without a central planner—and how that pattern can leave some participants worse served than others.