Predicting Margin of Victory
Let's talk about predicting the margin of victory for a little bit. This has been a major challenge since I started this project. The overall goal was to use a formula to calculate and assign a rating score for each team and then use that score to compare and estimate both the predicted winner and the predicted margin of victory. I'm pleased with the accuracy for most sports with regards to predicting the winner. Accurately predicting the margin of victory, however, has proved to be more evasive.
Sport-Specific Challenges
Some of this difficulty is inherent given the differences between sports. Football and basketball can have extremely large margins or be decided by just one or two points. Soccer, on the other hand, does not have such dramatic differences, especially with mercy rules in place at the high school level. Volleyball scoring limits outcomes even further, as matches are capped at three sets, making the margin of victory either one or two. Lacrosse falls somewhere in between, with games sometimes decided by a single goal and other times by large margins. It is generally understood that extremely large differentials may raise concerns about sportsmanship.
Approach to Estimation
I still don't know the best way to estimate the margin of victory. Ideally, the ratio of predicted margin to actual margin would be as close to one as possible. To account for differences across sports, I have used a multiplier. This multiplier tends to be higher for sports like football and basketball and lower for sports like volleyball.
I have also experimented with making the multiplier a dependent variable, using factors such as maximum and average margins of victory. However, this approach has only been minimally effective. As a result, I continue to adjust the multiplier manually in an effort to improve accuracy.
Ongoing Challenges
As with most predictive models, accuracy tends to improve later in the season when more data is available. Early-season predictions remain less reliable. Despite ongoing adjustments, creating a consistent and standardized method for predicting margin of victory across multiple sports remains a significant challenge.
Share Your Ideas
If you have any new ideas or suggestions, feel free to use the contact form to let me know.