Please read below to see what this prediction is based on.
Updated 10:20 AM EST with new Votefair numbers. Alex gained significantly while Jessica and Dexter fell.
Name | Song | WNTS | DI | VF | Not-safe Probability |
---|---|---|---|---|---|
Ben Briley | Bennie and the Jets | 22 | n/a | 0 | 0.562 |
Dexter Roberts | Sweet Home Alabama | 48 | n/a | 2 | 0.468 |
M.K. Nobilette | Make You Feel My Love | 63 | n/a | 4 | 0.387 |
Malaya Watson | I Am Changing | 65 | n/a | 4 | 0.385 |
C.J. Harris | Can’t You See | 73 | n/a | 4 | 0.378 |
Jessica Meuse | Sounds of Silence | 49 | n/a | 11 | 0.201 |
Caleb Johnson | Skyfall | 82 | n/a | 11 | 0.182 |
Sam Woolf | Come Together | 28 | n/a | 15 | 0.135 |
Jena Irene | Decode | 86 | n/a | 15 | 0.112 |
Majesty Rose | Let It Go | 15 | n/a | 17 | 0.111 |
Alex Preston | Falling Slowly | 77 | n/a | 18 | 0.079 |
I’ve addressed the model’s apparent poor accuracy this year in a couple posts, but let me reiterate here. The model uses history to calibrate its expectations. This works well when things that are happening now are like history. But when the rules change a lot, and this year they have, history is not as good a predictor of the future. At this point, I do not yet have enough information to tell how different this year is. Previous weeks have seen events that the model predicted were improbable, but that doesn’t make them impossible. Once we get some better statistics this year (with a few more shows), I will revisit this topic in great detail.
Nobody this week has a crazy high or crazy low chance of being in the bottom 3. Majesty, whom most people (including myself) thought was awful, has great poll numbers for popularity, and has only about a 10% chance of being in the bottom 3 if history is to be trusted. Ben’s numbers are really bad, but he has proven resilient before this year. He’s about 50/50 to be in the bottom 3.
The two WNTS approval ratings that stick out in this list are that of Majesty and Sam. So why are they so much less likely to be in the bottom 3? Because they appear to have a base of popularity—people who will vote for them no matter what they did on the show. This popularity is measured by Votefair. However, I’m dismayed by how few people are voting in that poll this year. Votefair already had a huge sampling bias, because people weren’t randomly selected to take the poll, they were self-selected. If the total number of voters there drops, the error associated with that is hugely amplified.
Dialidol is once again out for the count. That service works by measuring the busy signals on voting phone lines to try to determine whose lines are being called the most. Tonight, no Dialidol lines identified a busy signal on any of the contestants’ phone lines. That variable has been nulled in the projection above, meaning that data that was available in predictions last year is not available now. Internet voting may have just killed Dialidol, but I’ll withhold judgement until a little bit later in the year. The Dialidol forums are mostly quiet about the issue, but it is addressed in one post.