By now you’ve probably heard that TiVo tried to predict who would be removed from American Idol last night. They put out a press release (although I confess that I couldn’t find it on their site, and the closest I could come to an original source on this was here – did they try to pull it?) basically saying that they were able to predict the removal candidate for the last four shows. So they felt confident, and they went live with their next prediction. And, they were wrong.
Then, when proven wrong, they released a statement basically saying that as the field narrowed, the loser became harder to predict (seems to go against logic to me; if you have fewer to choose from, odds are a lot higher that you’ll just get lucky, but also statistically, the numbers should be more distinct).
So here we have it: crowdsourcing failed. One problem, of course, is that the mismatch in demographics between the TiVo viewers and general viewers who vote could be huge. Another problem is that the the original press release doesn’t confirm that the theory was developed prospectively. If it was retrospective, then it might be no more predictive than following stock price histories. Did a statistician figure out the correlation, or did a theorist propose the correlation and then evaluate the data and confirm?
This is all for the better. Exit polls don’t work, and neither does this type of prediction. I, for one, am glad.