We encountered an interesting result from the Eagles-Panthers Thursday night game. In a poll, users picked the Eagles to win, but with scores, they picked the Panthers. So in this case, the binary was correct and the precision was not.
A quick reminder: we think the value of the wisdom of the crowd is in precision. There are several ways to determine where the public lands on either side of a given line, but over the course of a season, that number will be close to 50%. That’s how the sportsbooks make their money.
The value we believe that Crowdsourced Scores has is that it allows the contrarian view to have a bit more weight. An Eagles vote and a Panthers vote have the same value in a poll, but when there is a prediction of Eagles 28, Panthers 21 and another of Panthers 24, Eagles 23, the aggregate score leans Eagles. This becomes especially important when we consider the lines since the delta between the crowd prediction and the lines is how we define value.
The Game 1 CPR – 3 for 3
Correct: Straight-up, Against the Spread, Over/Under. Another Trifecta!
The chart above shows the how that crowd average stabilized over the week. Initial submissions favored the Eagles quite strongly, but the scores evened out as the game got closer. The final spread predicted by the crowd was Eagles +0.6, so it was basically a coin flip. We’re certainly happy that the crowd was on the right side, even if it’s just by a hair. Meanwhile, the total stayed fairly steady throughout. The final predicted total of 47 was not within a margin of safety (more than a touchdown), but it never came close to the line of 44.
The Cynthia Frelund Experience
Cynthia Frelund’s crowd (@cfrelund) was 2.5 for 3 tonight, getting the spread and the total correct, and I’m giving them a half point for getting it right with the poll. I cut off prediction collection at 50 total predictions, and it’s interesting to consider whether 50 is a satisfactory cutoff or whether it’s not a big enough sample size.
Additionally, there is a potential that Ms. Frelund’s audience is not quite as decentralized as we’d like. It’s possible that they are influenced by other users, by Cynthia herself, or something else entirely. When you have a prediction this close one way or the other, there are a lot of different ways that you can look at the data to see if you can improve your aggregation, and there is a lesson in here, for sure. It’s just not clear what it is. 🙂
Do you see a way for us to improve our data collection? Let us know in the comments!