I’ve been using our Social Media Command Center to analyze this season’s American Idol fan reaction to the contestants. In addition to the application’s utility for tracking immediate reaction to the live broadcasts, I was also looking at whether the volume for particularly candidates vis-à-vis their competitors were predictive of voting outcome.

We’ve done voting predictions based on social media volume before, but in the sphere of political campaigns the data wasn’t always predictive. There are undoubtedly a few reasons for this:

1. The “one person, one vote” nature of elections. You can tweet as many times as you want about a candidate, but you can vote only once.
2. Demographic skew. While I know many senior citizens who are on Facebook, by and large they post to their own network and not publicly.
3. Relatively low volume of geo-coded data. When you’re looking to call the results state by state, you have to rely on individual posts to come with either location data from a mobile device or with a location given in the author’s profile. You do get sufficiently high volume in populous states, but not, for example, for states like Alaska.

Idol, of course, is different. For one thing, geographic location, while interesting, doesn’t have any bearing on the voting, and for another, you can vote as many times as you want. For a third, people who take the time to tweet seem more likely to pick up the phone or text in order to vote, especially for contestants towards whom they are positively inclined. So the working assumption I had was that if I looked at the relative volumes of positive comments for contestants, I could predict at least who was going to be in the Bottom 3.

This worked well…for a few weeks, I was 100%. Then came the week that saw Jessica Sanchez get the lowest number of votes and be saved by the judges. The social media numbers certainly did not predict that. Ok, an anomaly, I figured.

Then came the following week, when Colton Dixon got sent home. This was a complete shock – Colton had consistently been the first- or second-most talked about contestant. My confidence was shaken.

After that, though, things returned to normal. And thanks to my friends at FOX, I had the privilege of attending a Wednesday evening live performance when the show was down to five contestants. Being in the live studio audience really brought home how talented all of the contestants really are – I hadn’t had a real favorite up to that point, but I left blown away by Joshua Ledet’s voice. Unfortunately, the numbers tell me he’s going home tonight.

All contestants’ volume spikes during the live broadcasts pretty significantly, and it’s interesting to compare the numbers there, but it’s also useful to track volume when the show isn’t airing – it gives you a picture of what the base level of support looks like. If we look at the numbers, Jessica and Phillip have very similar sorts of volume ranges, both during the show and when it isn’t on, and theirs is roughly four times the volume that Joshua has.

Some other interesting views on the data relate to audience: you can see that all three contestants are more talked about by females vs. by males, but that Phillip’s female bias is much more pronounced.

Also, note Jessica’s volume from the Philippines eclipses volume from the U.S., which is quite unusual for a U.S.-based show.

Next week will be very interesting. Thanks again for FOX for the opportunity to experience the show live.