This marks our first public unveiling of our new rating system [link to the rating system explainer post]. Our goal will be to update these rankings every Monday evening / Tuesday morning.
First and foremost, huge thanks to Professor Dwight Wynne for doing the heavy lifting and getting this system off the ground. I’m excited to see the predictive power of this work and how it can influence downstream metrics that rely on team / opponent strength to set appropriate expectations.
These rankings are built on top of publicly available datasets and open-source statistical packages as a means to democratize this process, increase the transparency of how models like this are calculated, and ultimately serve to inform our Point Win Probability models as we attempt to account for team strengths.
For those playing along at home, the latest calculated inputs such as the HCA as well as the Serve Adjustment and Scaling Factors are shown below:
Using the rankings calculated in the first section, we are able to leverage the volleysim package designed by Dwight Wynne and Ben Raymond. This package takes a variety of inputs to run simulated matches, aggregate the outcomes, and summarize the findings to reveal predicted winners. In our case, we focus on Point Scoring % (likelihood to score a point when you serve) as the main input to the simulations. We export the results to our Tableau dashboard and you can scroll through various matchups from each week.
Here are all the games this week featuring two teams with ratings over 2000 (top 50…ish of the rankings), as well as our Predicted Winners and Expected Match Win Probability. Quick reminder that these are probabilities, not guarantees – a Win Probability of 70% means that during our simulations, the team lost 30% of the time.
Prediction breakdowns by game
For those who want a little more detail about the predictions – or are looking for an edge in their favorite VolleyTalk polls – here is a breakdown of our predictions at the set-level.
Will be doing our best to keep this data up to date as we move through the 2021 season – as well as tracking how the model performs at predicting the future using only historical Point Scoring numbers and Opponent Rating.