Gossip, grooming, and, well, maybe not really gossip

John Tierney describes some research into the effects of "gossip." In the experiments, subjects repeatedly played a simple money donation game with a series of different people. Well, I'll let Tierney describe:

On each turn, the players would be paired off, and one of them was offered a chance to give 1.25 Euros to the other. If he agreed, the researchers added a bonus of .75 Euro so that the recipient ended up gaining 2 Euros.

So far, so good. If you give 1.25 to the other player, you lose 1.25. If everybody is altruistic, then everyone comes out ahead. But in a single contest, a cheater always wins.

The trick with repeated contests is reputation: cheaters get a bad reputation, and so people stop giving them money. In this research, the experimenters provided two different kinds of information to the players: potential donors could see the record of the other player's previous games, and they could hear a comment from one of that player's previous partners -- something like "nice guy," or "jerk."

Given this information, people didn't make their decisions based only on the records: they rewarded players who had good "buzz."

The donor was told that the source of the gossip didn't have any extra information beyond what the donor could already see for himself. Yet the gossip, whether positive or negative, still had a big influence on the donors' decisions, and it didn't even matter if the source of the gossip had a good reputation himself. On average, cooperation increased by about 20 percent if the gossip was good, and fell by 20 percent if the gossip was negative.
Now, you might think the gossip mattered just in borderline cases -- when the partner had a mixed record of generosity, and the donor welcomed outside guidance in making a tough decision. But the gossip had an impact in other situations, too. Even when a player saw that his partner had a record of consistent meanness, he could be swayed by positive gossip to reward the partner anyway. Or withhold help from a perfectly nice partner just on the basis of malicious buzz.

But wait a minute. I didn't describe the "gossip" in quite the same way that Tierney did -- the German-speaking experiment used different words than "jerk" or "nice guy." And yet, what words in this situation are not going to elicit the same response -- a disproportionate response compared to the numerical record of the games?

The relevant point here is that the positive and negative "gossip" terms carry vastly less information than the statistics. And that's what makes them valuable. The choice before a given player is binary -- should you give your partner 1.25 or not? This choice is made vastly easier by the supply of a single bit of information -- is he a jerk, or not? You may be able to derive this information from the statistics, and if you trust someone else, you may make the wrong decision. But it's not hard to see that this experimental setup primes the players to seek a binary information source, the cost of a "wrong" judgment is pretty low, and there is no benefit in giving "wrong" advice.

To me, this has nothing to do with gossip and the evolution of language. You don't need language to convey binary information about somebody. Particularly in a primate group, where nobody is anonymous.

And if you don't need language, you certainly don't need math. Where in the world did anybody get the idea that people would be innately good at judging the statistical results of repeated trials of a game?