Top 9 projection and Twitter sentiment

Green = projected safe, Red = projected bottom 3, yellow = too close to call. Probabilities assume a bottom 3. See the methodology post for info on how this works. That post contains all the data and the code.

Name WNTS Rating Popularity Order Safe Probability
Clark 80 38.9 7 98.0%
Nick 59 7.4 6 90.0%
JAX 65 13.7 5 89.2%
Quentin 71 9.5 2 69.5%
Tyanna 52 5.3 4 63.8%
Joey 21 16.8 3 57.1%
Rayvon 24 3.2 9 56.1%
Qaasim 25 3.2 8 51.1%
Daniel 16 2.1 1 25.2%

The last time Daniel Seavey was projected in the bottom 3 (outside the margin of error) was in the Top 12, and he was almost in a dead heat with Sarina-Joi. The model was only barely comfortable projecting him in the bottom 3, and I’m almost sure that he was, although the producers did not release that information.

To say that Daniel is likely to go is a bit of an understatement. His raw safe percentage (before normalization is applied) is 31 percentage points below the cutoff at present. Now, the popularity numbers, as I’ve done in previous weeks, are frozen in their numbers from last week (so that Votefair has time to re-poll people), and thus that might change. Daniel would have to jump from his present percentage of the plurality vote, 2%, to 7%, which would be more popular than Tyanna presently is. I do not see that happening.

Now, of course, that’s not to say that Daniel is definitely going home. But he sang in the first slot, his WNTS scores have been abysmal, and his popularity is now the lowest. With fewer women to serve as cannon fodder, Daniel is not going to have an easy time of it. Worsters better get to it if they want to save him.

I can’t say that I’m too mad at the model’s performance so far. This week had a bottom 3 of Rayvon, Maddie, and Adanna, which were 3 of the 4 lowest rated in the model’s view. (Daniel was the 4th, of course.) No single indicator would have predicted Adanna’s elimination, which is why I bothered with the model in the first place. She had the 4th highest WNTS rating and was tied with Rayvon and Qaasim in popularity. But she sang first, she’s female, and she was a wild card pick. Bing bang boom.

For several weeks there have been a number of people in super-safe territory (> 98%), which is on account of there having been 11 or more contestants and only 3 bottom-3 slots. Now with 9 people, the odds of being that safe are considerably reduced. Clark stays well into that territory, but nobody else does. Joey has fallen significantly, in 6th place and not called safe within the margin of error. At 4th most likely, her fans would be justified being a bit nervous.

To wit, see Twitter’s response to the Top 9:

TwitterSentiment2015-03-26Joey leaps into last place on positive sentiment, as Daniel somehow recovers to above 70%. I have no historical basis to say what that means as far as her chance of being safe, but I present it anyway for your consideration.

The American Idol glass ceiling

In the most recent IdolAnalytics forecast, Adanna Duru is assigned a probability of being safe of only about 25%. Rayvon Owen is assigned a 55% probability of being safe, even though he has an equal percentage of Votefair votes, and Adanna had a significantly higher score for her song according to WhatNotToSing. True, Rayvon sang a bit later in the show, but what accounts for the huge difference?

Let’s look at all finals results without multiple performances by WhatNotToSing rating, and I’ll break it down into men and women:

CompositeThe first thing to notice is that there are significantly more women with high (> 60) ratings who were in the bottom group, 17 women out of 108, versus just 12 men out of 117. Women in this group were in the bottom 3 or eliminated at a rate of 16% versus only 10% for men.

The men in this group (Elliott Yamin, Chris Daughtry, Phil Stacey twice, Matt Giraud, Anoop Desai twice, Adam Lambert, Casey James, Michael Lynche, and Joshua Ledet twice) had an average WNTS of 71.4. The women (Paris Bennett, Carly Smithson  twice, Syesha Mercado twice, Allison Iraheta twice, Pia Toscano, Haley Reinhart, Erika Van Pelt twice, Elise Testone twice, Skylar Laine, Jessica Sanchez, Malaya Watson, and Jessica Meuse) were more numerous and had a slightly higher mean WNTS (72.3). More of them got robbed, and they got robbed even more than the men. Two women with scores above 80 were eliminated (Paris Bennett and Pia Toscano), which has never happened for men.

But the effect extends to more average performances as well. The black curves on the plots are the probability of being safe for each of the WNTS ratings (for an average popularity and not having been in the bottom 3 before) controlling for performance order, meaning that the effect of performance order is factored out. The men have a probability of being safe starting at 0.466 for a rating of 0, and 0.965 for a rating of 100. The women have a probability of being safe starting at 0.321 for a rating of 0, and 0.939 for a rating of 100. No matter how you cut it, you are at a significant disadvantage by being a women.

Anyway, this isn’t a surprise to anyone who watches the show. Of the 13 previous seasons, only 4 winners have been women. But here is the mathematical basis for the disadvantage that women face in American Idol, and it’s horrible. I said in last week’s liveblog that the save should only be used on women, and this is the reason I see for that.

Top 11 redux projection and Twitter sentiment (final)

Update 2015-03-25: Final update, 100% Votefair. Adanna edges out of really-sure territory. She’s fairly close with Daniel.

Update 2015-03-23 @ 8 PM: Now with 80% Votefair. Adanna back in the red. Likeliest bottom 3 is her, Daniel, Maddie.

Update 2015-03-20 @ 10 PM: Votefair is registering very few votes. I’m compensating with similar polls on a weighted basis. 60% Votefair, 20% and 20% MJsBigBlog. The numbers will converge to Votefair’s if/when more people vote in that poll. Adanna rises a bit. She is still the most likely, but no longer outside the margin-of-error.

Green = projected safe, Red = projected bottom 3, yellow = too close to call. Please read the methodology post for the details on how the model works. All the code and data are there.

Name WNTS Rating Popularity Order Safe Probability
Clark 81 37.0 9 98.0%
JAX 68 13.0 10 98.0%
Joey 77 16.0 5 98.0%
Quentin 64 9.0 7 98.0%
Nick 42 7.0 4 92.8%
Qaasim 40 3.0 11 92.3%
Tyanna 26 5.0 6 61.3%
Rayvon 25 3.0 3 55.4%
Maddie 15 2.0 8 52.5%
Daniel 6 2.0 2 28.7%
Adanna 48 3.0 1 24.7%

The judges used, some would say squandered, the save on Qaasim tonight. For now, the model is sour on Adanna Duru. She sang at the beginning of the show, she’s female, she had an ok-but-not-great performance, and her popularity is lagging. If her numbers improve, Daniel could be the call. Until that time, history would caution us that the VoteForTheWorst contestants have survived worse.

Joey, Jax, and Clark are frontrunners in the view of the model, and I certainly wouldn’t disagree. These are the people to beat. Any of them would be a shocking elimination.

Nick fell significantly this week in his WhatNotToSing rating, but I have trouble seeing him fall out of the safe zone. Qaasim himself is predicted safe even though he now has been in the bottom group. We’ll see if that holds up.

The model continues to be doubtful about Rayvon. If his popularity among voters doesn’t rise, I’m not sure he can break out. As most likely to be eliminated this week, and now third most likely, things don’t bode well for him.

Twitter predictably ranted about Daniel’s safety:

TwitterSentiment2015-03-20Daniel’s positives are barely over 60%, which is pretty damn bad. Nobody else is below 70. But somebody is voting for this kid.

As for using the save, about 62% of tweets were positive about using it, versus 38% negative.


Top 3 Tracker for the Top 11 (redux)

The Top 3 tracker is a long view of the contest, trying to figure out who will end up in the Top 3 based on the current standings. It works in a similar way to the finals model, but rather than week-to-week, it looks at historical information to decide who most looks like a Top 3 contestant.

Note that these are all calculated before last night’s performances, so these are based on last week’s scores.

All things being equal, everyone’s chances should improve every week, as any time someone is eliminated you have a better chance just by random pick.

Tyanna, Clark, Nick, and Joey look like the most likely by a lot. All of them are better than 50:50. Hit the jump for more info.

Continue reading