PHILADELPHIA -- New research from the University of Pennsylvania is challenging some longtime assumptions about why human beings seek and keep their friends, and it reveals a somewhat darker side to the very nature of friendship itself.

Peter DeScioli, who recently earned his Ph.D. at Penn and is currently a fellow at Brandeis University, says his recent research, completed in conjunction with Robert Kurzban, Penn associate professor of psychology, and researchers at two other universities, offers new evidence supporting the so-called "alliance hypothesis" of friendship, which states that individuals' feelings about their friends are based mostly on how those friends feel about them.

This alliance hypothesis, which also states that friendships are valued because of their usefulness during times of conflict, contrasts with the more conventional "reciprocity" theory of friendship, which holds that humans make friends mostly in order to reap the reciprocal benefits of those friendships.

The findings were published on Feb. 3 in the journal Perspectives on Psychological Science.

"Traditionally, one of the main theories about friendship is that friendships are essentially trade relationships," DeScioli said "You do things for your friends, and they do things for you.  The problem with this theory is, if it were true, then we should be good accountants and keep track of what we do for friends and what they do for us.  There is evidence, however, that we don't do that."

To figure out, then, if human beings make friends for reasons other than "trade," the researchers hit the Internet.  Specifically, they went to MySpace, the social networking site that, unlike other sites, allows people to "rank" their friends.  

Leveraging a massive database built from the site — about 11 million data points in all —  the researchers set to out to determine how much of a factor "friend rank" played in an the perceived strength of friendships.

As it turned out, that rank was of massive importance.

Kurzban explained that, if Joe Smith was ranked No. 1 by one of his friends, he was significantly more likely to also rank that friend in the top spot.  For example, the results showed that, comparing first- and second-ranked friends, 69 percent of people chose as their best friend the individual who ranked them better.  And while other factors also played a role in how friends valued their friends — geography, for instance — none of them played as great a role as rank.

"In some ways, this is obvious, right?" Kurzban said.  But “we argue in the paper that, while obvious, no theory predicts this finding, and, further, very little data exist to substantiate the intuition.”

The bottom line?  People greatly value those friends who greatly value them above others.  They are less impressed with those who rank them lower.  We are, it seems, very jealous beings.

"If you think about friendships in terms of alliances, in the context of game theory and international relations, one of the main things you'll find about allies is that they are fundamentally jealous of each other,” DeScioli said.  “If Saudi Arabia is allies with the United States, it's not just concerned about its relationship with the United States.  It's also concerned about the relationship that the United States has with other nations such as Iran.  In reciprocal or exchange relationships, you don't care about how, for example, Wal-Mart feels about other people.  You just care about what you're getting out of the relationship."

This dynamic is so important, the researchers suggest, because humans apparently see their friends as protection — not to mention valuable assets that can be leveraged when conflict arises. 

"You have to build friendships in advance," DeScioli said.  "You have to build a network of people who care about you, before any arguments or scuffles come up.  That's the theory, that humans create these networks so we can potentially use them in the future.  When you have this network of allies, you don't have to worry about direct reciprocity."

The research was completed by Kurzban and DeScioli along with Elizabeth N. Koch from the Department of Computer Science at Carleton College and Department of Computer Science and Engineering, University of Minnesota, Minneapolis, and David Liben-Nowell from the Department of Computer Science at Carleton.  It was supported in part by funding from the National Science Foundation and the International Foundation for Research in Experimental Economics.