Conversition: Your Source for American Idol Predictions

Friday, May 22nd, 2009

It was big news last week when various people tried to use Google trends to predict who the winner of American Idol 2009 would be. Using volume of searches, accounting for known demographic and geographic information, and considering the trends from prior seasons, they felt it was pretty much a sure thing that Adam Lambert would win. Perhaps you heard the news, though, that Kris Allen won the talent contest.


Those of you who have been following MatterMeter may have noticed that MatterMeter correctly predicted the outcome, giving Adam a rating of 67% and Kris a rating of 86%. Let’s just say that to remain focused on the issue at hand here, we won’t talk about sample sizes or confidence intervals associated with these numbers. That’s our little secret.

How is it that so many people predicted incorrectly? Surely, the sample sizes on which the Google trends analyses were based ran into the thousands and possibly millions. Surely, search terms are a reasonable proxy for interest in a topic. Statistically, it seemed to be a reasonable bet.


The missing link, not the Lemur link also discovered this week, is that Google trends does not evaluate feelings or sentiments. Were people searching so fervently for Adam Lambert because he could be counted on to deliver a stunning vocal performance or because he could be counted on to deliver a strange song interpretation that would certainly go viral. Were people searching for Adam Lambert not because they were excited to purchase his upcoming CD but rather because they couldn’t believe that his songs might actually end up on a CD? (For the record, our team thinks Adam and Kris are fabulously talented!)


I’d advise you to keep an eye on MatterMeter and EvoListen next year. We just might get it right again.