An analysis of the accuracy of DubStat’s Elo rating system for the 2023/2024 OHSAA wrestling season.
If you’re unfamiliar with the Elo rating system, check out the following articles for an overview on how it works:
What are Elo Ratings and How do they Work?
How does DubStat Calculate Elo Ratings?
The Numbers
For the 2023/24 OHSAA Wrestling season, DubStat’s Elo rating system produced rankings for 9,428 wrestlers and predicted winners for 90,355 matches.
How well did DubStat’s ratings predict winners?
DubStat’s Elo rating system correctly predicted the winner in 73,941 of 90,355 total matches recorded throughout the 2023/2024 OHSAA wrestling season. In other words, the wrestler with the higher Elo rating won 82% of the time.
How does 82% accuracy compare to Vegas?
When compared to Vegas odds, DubStat’s prediction accuracy is higher by 3 to 24 percentage points, depending on the sport.
This doesn’t necessarily mean a similar Elo rating system would do better than Vegas odds at predicting winners in these sports, but it does provide a benchmark for what is possible when predicting winners in competitive games.
How does DubStat compare to NCAA wrestling seeds?
Vegas does not typically produce odds for individual wrestling matches, however, tournament seeds are a type of prediction method we can compare to.
At the NCAA D1 National Wrestling Tournament, the higher seeded wrestler wins 72% of the time. This can be compared to DubStat’s 76% prediction accuracy at the OHSAA State Wrestling Tournament.
Again, this doesn’t necessarily mean a similar Elo rating system would do better than 72% when predicting winners at the NCAA Wrestling Tournament, but it does provide a valuable benchmark.
Read How Accurate were DubStat’s State Tournament Predictions for a full analysis on DubStat’s predictions of state placers, state champions, and team finishes.
Did prediction accuracy change throughout the season?
The system’s prediction accuracy was consistently at 80% or above for every week of the season except for the OHSAA State tournament week where matches were predicted with 76% accuracy.
Was the system better at predicting matches in certain weight classes?
The system’s prediction accuracy was consistent throughout all weight classes with rates ranging from 81% to 83%.
Did the system produce correct “probabilities of winning”?
For every match, DubStat’s Elo rating system calculates a “probability of winning” for both wrestlers based on their Elo ratings (you can use our Matchup Predictor Tool to produce the same probabilities).
The chart below shows the percent of the time wrestlers actually won matches compared to their “probability of winning” the match.
Overall, the “Actual Win %” (blue dots) are within 1-3 percentage points of the “Probability of Winning” (grey line), which means the “probabilities of winning” produced prior to each match are in line with actual results.
For example, there were 1,369 matches this season where the favored wrestler had a 70% chance of winning. The favored wrestler actually won 72% of those matches (blue dot slightly above where the two 70% gridlines meet on the graph).
Similarly out of the 1,263 matches where one wrestler had a 40% chance of winning, that wrestler actually won 39% of the time (blue dot slightly below where the two 40% gridlines meet).
Post Season Prediction Accuracy
For the postseason, DubStat’s Elo rating system correctly predicted:
- 81% of state qualifiers
- 71% of district champions
- 77% of state placers
- 68% of state finalists
- 60% of state champions
Check out the following articles for a full analysis on DubStat’s prediction accuracy at the District and State tournaments:
How Accurate were DubStat’s District Predictions
How Accurate were DubStat’s State Tournament Predictions
Thanks for your work. I found the process and the information very interesting.