Just a quick note to people who follow the site—we had to make some big technical changes last week because the site was getting hammered (the site actually went down on Thursday for about an hour … sorry about that). This included removing the Liveblog feature and the Polls to allow for supercaching. My crack technical team (Jessica) may get either or both working again by Wednesday, but if she doesn’t, that’s why they aren’t there.

Can’t wait for Idol to start next week? Watch my 40 minute retrospective of American Idol season 1!

Transcript after the jump.

Read the rest of this entry »

Hollywood Week is a week away, the new season is almost upon us, and we’re busy tooling up for season 12. I just wanted to preview what’s going on.

My emphasis this year, as opposed to prior years, is going to be to provide a real prediction model. This will be fully (painfully, perhaps) detailed in an article to come. The short version is that a lot of effort went into crafting a much more robust, hopefully much more accurate statistical model for week-to-week predictions of who the bottom 3 will be, with good error estimation to quantify how much uncertainty there is.

I’ve also built a horserace tracking model that is quite good at predicting who will make the Top 3, which is much less noise prone than the week-to-week predicitons. Finally, a similar model to that for the finals has been built for the semi-finals, and I’ll be tracking the tangible variables that go into this model throughout the audition episodes.

I’m switching the blog over to Disqus for commenting. My hope is that this will have some effect on the level of discourse and require less moderation. There may be some hiccups with this, so let us know on our Facebook or Twitter if something is awry.

I’ve also incorporated JQPlot into the site, which will provide not only really spiffy plots, but also ones that are interactive and more informative. Again, if these plots don’t render correctly, please let us know.

Read here: http://www.utsandiego.com/news/2012/may/21/conspiracy-is-always-on-the-idol-ballot/

Point of correction, though. I said “idiots like me”‘ not “people like me”.

Unlike last week, the numbers haven’t moved overnight. Phil is predicted, with 72% chance, to pack it in tonight. It will be interesting to see whether that happens. I would note quickly that Hollie has been underperforming her numbers, so it’s possible the model is getting this totally wrong. Recall, dear reader, that the very song she sang first last night (River Deep – Mountain High) was the song Pia sang when she was eliminated.

There’s no shortage of predictions on the web. Here’s my call along with that of some major sites.

Here are the current values for the contestants on Intrade. They view DeAndre as the least likely to win.

Colton seems to have done himself a world of good in the past few weeks, nearly doubling his chances. Though his stock is not priced as highly as Phil or Jessica’s, it likely will be if he continues to be one of the 3 people who have never been in the bottom 3.

Here are the latest numbers from Intrade. They view Erika Van Pelt as the least likely to win.

Just a note: Dialidol currently has Scotty whooping Lauren, 20.5 to 12.5 on their own scaling. I don’t know if it’s obvious or not, but the way Dialidol works makes it far, far more reliable as the race moves to fewer people. I have always regretted not taking the time to take this into account in the model. The fact that Dialidol has never been wrong in the Finale is almost certainly not a coincidence. If they show Scotty winning by that much, the deal is almost certainly sealed.

Polling results for President don’t move much after the first week of October. This is not surprising: by then, since the campaign has been going on for at least 6 months (usually more like 18 months), people have made up their mind. The individual gaffes, debate performance, and stump speeches, cease making that much of a difference. About the only thing a candidate can do is focus on his Get Out The Vote efforts and hope for the best.

I wonder if the same thing isn’t going on with American Idol in the Top 5 until the finale.

As I noted in the Top 5 pre-game, there was no statistical correlation between how well a person did in the Top 5, and whether or not he or she was eliminated. Some examples:

Season 4: Scott Savol came in with a dismal prior average of 22.7 in the approval ratings, but scored a 41 that week, above Anthony Federov and not far behind Vonzell Solomon. But he was eliminated anyway.

Season 7: Brooke White did much better than her previous score of 27.7, scoring a 39.5, which was better than Jason Castro and David Archuleta. But she was eliminated anyway.

The same can be said about RJ Helton in season 1 and Trenyce in season 2. That’s about half of all years (season 6 had no Top 5). The rest of the people scored the lowest and were eliminated. When half of the contests show no correspondence to the data, you know you have yourself a variable that isn’t correlated at all.

What I believe is happening is that people have more or less picked their favorites by the Top 5. Prior score for the past 3 rounds matters because that’s when people were still picking their favorite. But by now they have one.

Now, the difference to the presidential election is that as we go on, certain contestants are being eliminated. We could then hypothesize about where the votes for the contestant who left are distributed. Likely a fair number are just lost, as the people who liked that person stop voting. But some of them go to someone else. Who thinks Casey fans are going to vote for Lauren Alaina? Not me. I bet they will go to Haley Reinhart.