found on NationalGeographic.com
Attention "American Idol": Hits Are Tough to Predict
Mason Inman
for National Geographic News
February 13, 2006 -- The judges on the hit reality TV show American Idol are supposed to be able to use their experience and shrewd ears to suss out the next great pop star.
But predicting which artists will top the charts and which will flop is a crapshoot, a new study suggests.
Which songs wind up getting attention has something to do with quality—a tone-deaf singer with no rhythm is unlikely to be a hit.
But because fans often influence each other, almost anything can happen, the study suggests.
To test how songs win over listeners, researchers at New York's Columbia University set up a Web site called Music Lab. More than 14,000 volunteers logged on and were able to listen to songs donated by 48 virtually unknown bands.
The researchers solicited most of the participants through a teen-interest Web site called Bolt. The listeners were required to agree to Music Lab's consent form, which made it clear that they would be taking part in an experiment.
After hearing a song the listeners could choose to download it. The researchers—Matt Salganik, Peter Dodds, and Duncan Watts—kept tabs on the number of downloads of each song.
In some experiments the researchers showed this data to listeners, so participants could see which songs were snagged the most.
When listeners could see what other people had chosen, a few songs' popularity snowballed, while other tunes languished.
Also, by separating the listeners into separate groups, or "worlds," the researchers could see if the same bands would consistently come out on top. The scientists assembled each world to ensure that the groups were similar in terms of age, location, and so on.
Though listeners in the different worlds roughly agreed on which songs were best, there was a lot of variation.
The study found that when listeners could see how often a song got downloaded, it was actually harder for researchers to predict before the experiment which songs would eventually end up on top.
Some songs, such as "She Said" by Parker Theory, consistently won. But others, like "Lockdown" by 52metro, were more volatile.
In one world "Lockdown" came in first, with 114 downloads. In another it was 40th, with only 7 downloads.
The study suggests that this unpredictability is inherent in a system where people's tastes interact, Salganik says.
The researchers published their results in Friday's issue of the journal Science.
The Paradox of Success
The researchers were motivated to study music downloading to understand a paradox about how things become popular, Salganik said.
Take the first Harry Potter novel. "It has been tremendously successful and been translated into more than 60 languages," Salganik said.
"That would make you believe Harry Potter is somehow different from all the other books that didn't have nearly this level of success," he said.
"But this difference wasn't at all clear ahead of time," Salganik said. Many publishers, thinking it wouldn't succeed, rejected the book.
The same phenomenon shows up with music and movies.
"A lot of this unpredictability is invisible" to the public, Salganik said, because it never sees the things that fail.
One of the most spectacular bombs, he points out, was Irish singer Carly Hennessy. Record company MCA threw 2.2 million U.S. dollars into producing and promoting her album Ultimate High. It sold only 378 copies in the first three months.
"I went online and bought her CD for a penny," Salganik said. "It actually doesn't sound that much different from a lot of other things you hear. It just seemed like she had an unlucky break."
Testing, Testing, 1, 2 …
The Music Lab experiment was not quite like the real world, Salganik says. Most of the listeners were teens, and there were no celebrity endorsements or advertising to affect the results.
Yet the experiment's results are similar to those from mathematical models of the success of movies, said economist David Walls of Canada's University of Calgary. He and economist Art De Vany of University of California, Irvine, have modeled how people tell each other about movies.
"When [people] see a movie they like, they make a discovery, and they tell their friends about it," Walls and De Vany wrote in a 1996 article in The Economic Journal.
This creates a feedback that unpredictably leads some movies to break out and others to nosedive, Walls said.
Sociologist Michael Macy of Cornell University finds the Music Lab study convincing.
"They're showing that it's very hard to predict which songs will become popular," he said. "That's not just because we don't know what peoples' tastes are," Macy said.
"It's because random fluctuations early on in the reports about what's popular become exaggerated due to the social influence."
However, economist Stan Liebowitz, at the University of Texas in Dallas, is skeptical about how much the study proves.
To improve the data, he suggests tricking some of the listeners by giving them fake data so that the songs that were actually most popular would appear to be the least popular, and vice versa.
