Wow, last week’s Drag Race post made the rounds in the stats and Drag Race circles. It was cross-posted to Jezebel and has been getting some pretty high-profile links. A little birdy told me that Ms. Ru herself has read it. I think I can die a happy man knowing that RuPaul has visited Bad Hessian.

Anyhow, last week I tried to count Coco out. I was reading her like the latest AJS. The library is open. But her response to me was simple — girl, please:

(Also this happened. Wig under a wig.)

(both of these gifs courtesy of f%^@yeahdragrace)

Can that win safeguard Coco from getting eliminated? Let’s look at the numbers after the jump.

I had planned for this week something a little more detailed than what I am actually going to offer. Last week I alluded to the issue of overcounting lipsyncs, such that I include lipsyncs that result in elimination. The problem with this metric is that it seems to overfit the training data and gives an inordinate amount of predictive power to only the lipsync variable. I imagine that this issue comes up in other forms — Adam says this is an issue when modeling vote choice based on party identification. I had mentioned that I wanted to do some sort of out-of-sample validation to see how the models matched up on earlier seasons. But I think I’m going to wait on that, especially since spring break is next week, which will give me a lot more time to do that topic justice. Adam and I have also been going back and forth on using a conditional logit to turn this into a discrete time model. For now, I’m going to provisionally use both models I mentioned last week for fitting predictions and hope you’ll trust me that I’ll try to actually have an answer for why I’m doing so at a later date.

Before getting to my own numbers, here are some insights from fellow Drag Race forecasters. The Dilettwat has some good observations; her predictions for the season are a showdown between Jinkx and Roxxxy. Homoviper ranks them thusly: Jinkx, Alaska, Coco, Ivy, Detox, Alyssa, Roxxxy. Also if you know of other Drag Race forecasters please mention them in the comments!

Here are the predictions for the model I used at the end of last week’s post (the one that uses the uncorrected lipsync variable):

1    Ivy Winters 0.7532603  1.0817218
2  Jinkx Monsoon 0.8151390  0.5921149
3         Alaska 0.9242376  0.9452547
4          Detox 1.1475463  0.5733333
5 Alyssa Edwards 1.2332356  0.3091683
6 Roxxxy Andrews 2.8150008  1.7322992
7  Coco Montrese 4.5556308  0.8606941

Coco’s still on the bottom (there’s a joke here but this is a family blog, people). However, Roxxxy is actually close to her if you factor in confidence intervals. Here’s a graph that plots the change in relative risk between the two episodes, as well as the 95% confidence intervals.


The basic idea is that the greater your relative risk (higher value), the more likely you’re going to get the boot. You can see that before episode 6, Coco was in a class by herself in terms of getting the chop. Everyone else was at about the same footing. Now, Coco and Roxxxy are about wig and wig when factoring confidence intervals. What’s driving that? Well, Roxxxy has the most “highs”, which, in this model, negatively impact the hazard rate but isn’t statistically significant.

The second model is the one without the “elimination” lipsync. The numbers put Detox last and Ivy Winters first:

1    Ivy Winters 0.04116623  0.1485200
2 Roxxxy Andrews 0.09607804  0.1431874
3         Alaska 0.13145644  0.2509301
4  Jinkx Monsoon 0.29485310  0.2550890
5 Alyssa Edwards 0.67310584  0.2306773
6  Coco Montrese 1.41765963  0.5965239
7          Detox 3.13951710  0.7408679

And the graph:


Coco’s still not looking very great this week. There’s an idea that “momentum” can get you out of slumps, and that’s not something that this model accounts for. I could incorporate a lagged variable that considers whether a queen has won a challenge within the last few episodes, serving as a rough “momentum” indicator. Whether this actually bears out, I don’t know. Any guidance here on incorporating that would be more than welcome.

Till next time, the library is closed.