Top 11 redux Elimination Projections

The model is relatively sure of the bottom 3. With a binary value model it thinks

Contestant Elim. Probability
Naima 0.216
Paul 0.278
Thia 0.260

A continuous variable analysis is way more sure of Naima

Contestant Elim. Probability
Naima 0.448
Paul 0.102
Thia 0.079

My personal feeling is that the second predictions are better, and the significance is better anyway. Naima is toast, then Paul and Thia in a dead heat. This could change if the major scoring services change a lot. I don’t think that’ll happen. Bye Naima!

Bookmark the permalink.
  • trashy09

    Where did you get this projection?

    • Reuben

      I explained the model in an earlier post. It’s a simplistic guess, really. A logistical regression for past contests is used, and the model is scored with tonight’s preliminary dialidol and WNTS scores. The response values are then calculated from the logit in R.

      • trashy09

        DialIdol score shows Stefano lands in the bottom, while the raw numbers shows Paul. This is just weird, because whoever sits in the bottom at DialIdol score happens to be at bottom on raw score as well. What’s your take?

        • Reuben

          Generally speaking WNTS and dialidol scores are not quite dispositive. Dialidol has an ok but not great, rate of picking the lowest vote getter so early in the contest (the statistics are in another post). However, it is good at predicting the bottom 3, with an average of about 60%. WNTS scores are somewhat more variable, though they’ve been dead-on this year. I’m guessing Stefano is on the bubble, so to speak, but the model assigns him an advantage by being male and due to a slightly higher approval on WNTS. I would also say that while Stefano is last in dialidol, I’d be wary of claiming too much statistical significance based on the raw score there. On the scale of the numbers, the bottom 4 contestants could be dead even due to dialidol sampling error.

          I’m a bit more worried about Lauren Alaina. Her WNTS scores and dialidol ranking are wildly different. Not sure what’s going on there at all.

          Note that this is all still a guess. Americans are routinely unpredictable.

          • trashy09

            Should we believe on dialidol score or the raw numbers?

  • Jessica

    The thing about a model is that it’s just a model; you look at the information, assess it, check out the methodology and see if it sounds reasonable, and then decide what you think is going to happen and whether you think it’s a good model. You shouldn’t believe the dial idol score OR the raw numbers; you need to think for yourself 🙂 American Idol is in a lot of respects a chaotic system, so it’s difficult to predict anything accurately. One of the goals of this site is to come up with a model that gets things mostly correct, since most models out there currently do little better, if at all, than a coin flip.