By Shankar Vedantam
Monday, December 31, 2007; A03
Two sociologists and a mathematician recently conducted an experiment that provides an intriguing window into the presidential candidate selection that begins this week. Matthew Salganik, Duncan Watts and Peter Sheridan Dodds had a large group of people rate 48 songs. Based on these ratings, the researchers produced a list of the best songs.
They then had eight other large groups of people evaluate the same songs, with one difference: In each of these “parallel universes,” people knew how others in their group were evaluating the music. Did the eight groups come up with the same list of the best songs? No. When people knew how others thought, this changed how they thought.
Since the people in the first “control” group had nothing to go on besides the songs, their ratings were measures of quality. But in the other eight groups, quality played a much smaller role in determining a song’s success. Rather, network dynamics — the mathematical patterns that govern how ideas spread when a large group of people share complex interconnections and simultaneously influence others and are being influenced themselves — explained why some songs became popular.
Did the eight groups exposed to peer pressure agree with one another? Again, no. Each came up with different lists of the best songs.
The experiment, published in Science, suggests that when large networks of people evaluate something together — and it does not matter whether we are talking about songs or “American Idol” contestants or presidential candidates — their conclusions are not only powerfully shaped by the views of others, but by the network that binds them together. The Iowa caucuses, which involve people watching one another and moving from one candidate’s camp to another, have different network properties than a primary where voters don’t have such real-time feedback.
Watts, a sociologist at Columbia University, said his research challenges central beliefs we have about why some musicians become stars and some politicians become presidents. Quality matters, but when voters intensely watch one another, the success of candidates depends at least as much on network dynamics as it does on the quality of the candidates themselves. Because network dynamics are not governed by intuitively simple rules of cause and effect — depending on where they are in a network, people with strong opinions might end up with little influence, while the weak opinions of others get greatly magnified — networks regularly produce outcomes that are partly arbitrary. Each of the eight music “universes” started out the same, but for no good reason, each went off in its own direction.
Once a primary is over, real life does not allow you to go back and rerun the race a second time. But if you could, the music experiment and other research suggests you might arrive at a different result, even though the candidates and voters start out the same. This is disturbing — if Hillary Clinton wins the Democratic primary or Rudy Giuliani wins the Republican nomination, wouldn’t you like to believe that running the race over would give you the same result?
“I am comfortable with the idea that the outcomes we get are often largely arbitrary,” Watts said. “We think there is something we can call quality and it is intrinsic to people and books, and it is timeless and the results we see in the world reflects this quality. If you find it disturbing that that is not how the world works, you should not become a sociologist.”
The real world provides ample evidence for the first part of the music experiment. Voters in Iowa and New Hampshire play an outsize role in determining the winner of a presidential primary for no better reason than that they get to vote before everyone else. What is impossible to see in real life is that the same candidate may not win if you were to run the race over. What causes a Clinton win in one “universe” and a Barack Obama win in another?
In a new paper published in the Journal of Consumer Research, Watts and Dodds debunk the idea that influential people drive races one way or the other. The decisive factor, they show in a series of mathematical models, is not the presence of influential people but people who are easily influenced. Random, insignificant events are vastly magnified by networks of such malleable people influencing one another, and this tilts the race one way or another. Blind chance plays a big role.
Once a winner is declared, however, politicians, voters and the media construct a narrative of how that outcome occurred — they usually point to a set of pivotal characters and crucial turning points. Watts said that these after-the-fact explanations are like explaining a forest fire based on the first spark and a handful of pivotal trees, rather than on the complex relationship between wind, temperature, humidity and fuel.
Most people accept that hurricanes, earthquakes and forest fires can be triggered by random, insignificant events that are greatly magnified by complex networks. We know there is nothing unusual about the pebble that starts an avalanche — most rolling pebbles are insignificant, but a tiny few have giant consequences.
Seeing a presidential election in those terms, however, is troubling: It means that, every four years, we entrust our future to a roll of the dice.