Post Nationals Hype Post!!
I wrote about the crossresults vs USAC predictions last week, but now we can look at what actually happened!
The problem is that you really can't judge the quality of a prediction off a single trial. Take, for example, my latest favorite thing to hate: Powerball. Anyone who bought a Powerball ticket I would predict to lose. However, some people won Powerball -- but that doesn't mean the prediction that they were going to lose Powerball was the wrong prediction, right?
Nevertheless, it's fun to look at, anyway. Let's go.
Elite Women
Katie Compton collected her 12th title in a row, but Georgia Gould gave her the toughest run for her money she's gotten in a long time. Guess whose got two thumbs and a website that predicted Georgia on the podium? this guyyyyyy. But guess whose got two thumbs and didn't pick Georgia for a top-5 in his own picks? THIS GUY.
Anyway, big ups to crossresults for calling/getting lucky on Georgia, who skipped all that Europe BS to lay low in Colorado and come out flying at Nationals.
In the last post I highlighted the biggest disagreements between the predictor, let's see how those went:
Crossresults likes:
Elle Anderson (6th in crossresults, 12th in USAC)
Crossresults loved Elle's recent European campaign, and we were right. She crushed it and finished 4th.
Sunny Gilbert (11th in crossresults, 16th in USAC)
Sunny finished 13th, hitting the average of the predictions quite nicely.
Beth Ann Orton (16th crossresults, 23rd USAC)
Beth Ann didn't even start the race because she had a broken hand. Whoops!
USAC likes:
Jena Greaser (9th USAC, 14th crossresults)
Jena finished 11th, right in between the two predictions.
Arley Kemmerer (8th USAC, 15th crossresults)
Arley finished 15th, probably because this happened. How did crossresults know?!
Megan Korol (25th USAC, 31st crossresults)
Megan finished 34th.
So based on this sample of size 1, crossresults was way more accurate on the women's side -- it won disagreements on Gould, Anthony, Anderson, Kemmerer and Korol and didn't really lose any. Good thing I just told you that samples of size one don't mean anything.
Big Rides!
A few women significantly outperformed the average of their two predictions and should be mentioned:Elle Anderson - predicted 9th, finished 4th. (+5)
Rebecca Fahringer - predicted 11.5th, finished 6th (+5.5)
Amanda Nauman - predicted 11.5th, finished 7th (+4.5)
Cassandra Maximenko - predicted 15.5th, finished 10th (+5.5)
There were also some prediction-underperformances, but maaaaan, it's a cross race. Stuff happens. Don't dwell on it.
Elite Men
I gotta admit, the USAC numbers were better here. While both predictions agreed that Hyde/Powers was a virtual tie, we said Logan Owen would be much closer to 4th than the lead. Instead, he crashed the party until the final lap and ended up minutes ahead of 4th. Whoops. I blame Europe.
crossresults likes:
Cody Kaiser (15th crossresults, 24th USAC)
We liked Cody, but we didn't like him ENOUGH. He crushed it with a 10th-place ride.
Tristan Uhl (18th crossresults, 22nd USAC)
Tristan had an amazing ride and WON!!!... the singlespeed race. Then he got 30th in the main event. Crap.
Kevin Bradford-Parish (22nd crossresults, 28th USAC)
Crossresults said 22nd, USAC said 28th, he finished 22nd. TAKE THAT, GHOST OF STEVE JOHNSON
Zach McDonald (25th crossresults, 35th USAC) **
ZMD showed some signs of the form that has put him on Nationals podiums before with a 19th place, fast enough to make crossresults look good, but slow enough that Erin owes me a coffee. YEAH ZACH!
USAC likes:
Dan Timmerman (8th USAC, 12th crossresults)
Dan got the holeshot, went for broke, lost a cleat, finished 28th. Shit happens.
Jeremy Durrin (16th USAC, 21st crossresults)
Jeremy split his predictions perfectly in 18th place.
Jack Kisseberth (21st USAC, 26th crossresults)
JAM Fund Jack had the RIDE OF THE DAY in 11th place, beating crossresults by 15 places and USAC by 10.
Tim Allen (19th USAC, 30th crossresults)
This guy endured more Home Improvement jokes than should be legal and finished a respectable 20th, proving that crossresults disrespects Colorado masters racing too much.
Adam Myerson (25th USAC, 34th crossresults)
Adam split the difference with a 29th place, but he doesn't even care because he has a baby now.
The men's side was a much more event split between the predictions. Crossresults won on Kaiser, ZMD, Bradford-Parrish, and Timmerman, but USAC won on Kisseberth, Allen, and Uhl as well as getting the podium perfectly. We need more samples! Race again!
Big Rides!
Jack Jack Jack Jaaaack - predicted 23.5th, finished 11th (+12.5)Zach McDonald - predicted 30th, finished 19th (+11)
Cody Kaiser - predicted 19.5th, finished 10th (+9.5)
Yannick Eckmann - predicted 16.5th, finished 9th (+7.5)
Tim Allen - predicted 24.5th, finished 19th (+4.5)
Ben Frederick - predicted 16th, finished 12th (+4)
As noted, shit happens, so there were some less-than-big-rides out there too. Notably, both predictors liked Summerhill in 4th, but he landed in 17th. Mechanical, biological, emotional? Who knows.
In summary, you shouldn't draw any conclusions from small sample sizes and we should race Nationals every weekend for a month to figure out which predictor is better.
Comments
FWIW, Arley K was 16th, not 15th. #PDFResults
Because I'm sure you want this to be a bigger time suck than it already is, here's an idea for future race forecasting: instead of, or in addition to, building a model to predict finishing positions, build a model that predicts finishing time relative to the mean or median finishing time for a given race. Having that continuous DV would open the door wider to using Bayesian simulations to estimate things like, say, the probability that the favorite will win, or will finish in the top five, and so on.
Shouldn't be too hard to look at a rider's results from the season (assuming most elites aren't changing massively in ability from September to January) and calculate the standard deviation of their points over that time as well as their current average, and then you're off and running.