Internet search engines: A crystal ball into the future?
Can Internet searches help industry watchers predict the success of a movie or the spread of an epidemic?
Plenty of current studies show that when a trend in politics, consumer goods or even disease rises or falls, the Internet search volume for related terms rises and falls along with it.
Thus, said researchers from a division of Yahoo! Research, the Internet might appear at a given point in time to be “a snapshot of the collective consciousness, reflecting the instantaneous interests, concerns, and intentions of the global population.”
But does that mean the Internet provides a better predictor of what people are going to buy, or how many people are unemployed, or whether they’re getting sick, than the standard predictors used for such trends? The researchers decided to find out.
In a study published online Monday in the journal Proceedings of the National Academy of Sciences, they looked at movie box-office revenue, video game sales and billboard rankings for music. They analyzed whether the related search volume reflected each product’s real-life success. For movies and video games, the volume of searches matched up pretty well with real outcomes -- though not quite as well for music.
But in almost all cases, standard forecasting methods based on publicly available information such as film production budgets, or critics’ video game reviews were more accurate than search-based predictions. (The exception was non-sequel video games: Because such new games lack a built-in audience, predictions have to rely on critics’ reviews, a rather meager data source.)
But when the researchers combined the search results with the standard predictors, they found that taking search volume into account enhanced the accuracy of their predictions for music and for non-sequel video games. For movies and sequel video games -- think Final Fantasy XIII -- that tactic didn’t make much difference.
The study also looked at searches related to influenza, and it found a high correlation with flu data from the Centers for Disease Control and Prevention. But the search volume wasn’t necessarily any more accurate a predictor than the existing mathematical models.
Their conclusion? “These results illustrate two points: first, although search data are indeed predictive of future outcomes, alternative information sources often perform equally well or even better; second, search appears to be most useful when key indicators (e.g., past sales performance, production budgets, etc.) do not exist or are unavailable,” the authors wrote.
Of course, they don’t mention one other possibility: Maybe the search results would have been more accurate if they were using Google instead of Yahoo. Just throwing it out there.
-- Amina Khan / Los Angeles Times
Back to the Booster Shots blog.