Preseason Player Projections With Bayesian Performance Rating
Top 30 reveal, plus methodology breakdown and interesting trends
With only a few days remaining till the start of college basketball, I wanted to highlight this year’s top players in preseason projected Bayesian Performance Rating at EvanMiya.com. BPR is, in my opinion, the best single metric for capturing a college basketball player’s value that he brings while on the court, on a per-possession basis. The metric incorporates a player’s individual efficiency stats and on-court play-by-play impact all in one. It also adjusts a player’s rating to account for the strength of all other players on the court for every possession played. You can find the full historical leaderboards for BPR on the Player Ratings page.
Instead of every player starting a new season at a value of 0 (representing a D1 average player), each player begins the year with a preseason “prior distribution” which represents a range of possible outcomes for the player based on historical data and other relevant information. The “mean”, or “average” predicted BPR value for a player is the projection shown on the site before any games are played.
The preseason top 50 players in Bayesian Performance Rating are below:
Summary of the Methodology
The algorithm that constructs these preseason projections is complex, but here are the main two factors that weigh into the final calculation:
Stats from previous college seasons: Several advanced stats, including previous season BPR and Box BPR (my version of a Box Plus-Minus) play an important role in predicting the next season's rating for a player.
High school recruiting rating: For freshmen who have not played college basketball, the most accurate indication we have of their college ability comes from composite high school recruiting rankings. Even then, these are pretty flawed, since they often value NBA potential more than NCAA impact. Freshman projections are constructed by looking at how previous players with similar high school rankings have fared in their first year at the D1 level. For players who have played in college before, we also use recruiting profiles to help us know more about their potential, especially if they have only played one or two years in college (e.g. Kel’el Ware, Harrison Ingram).
Though only a single number is shown for each player’s projected Offensive and Defensive BPR, an equally important variable is the amount of uncertainty attached to each projection. Players with more collegiate playing data have more certainty around their projection because we know more about them. By contrast, freshman typically have larger uncertainty attached to their projection, which means that their in-season ratings may move more quickly based on whether they perform differently than preseason expectations.
The table below shows the players with the most uncertainty around their preseason projection who are inside the top 150 in our preseason rankings (“prior standard deviation” is the technical term):
Trends
Most seasons, you’ll notice that the best projected freshman in the preseason falls in the top 15-25 range. That might seem too low, but even freshmen with the highest expectations sometimes don’t make a big impact in college. Here’s a link to a study I did recently showing how high school recruiting ranking corresponds to freshman year BPR. There are as many misses at the top as there are slam dunks. A freshman or two typically end the year in the top 10 nationally, but we can’t accurately predict who they will be.
At the top of the preseason leaderboard, there are plenty of recognizable players who will be dominant scoring forces for their teams, such as Zach Edey, Kyle Filipowski, Hunter Dickinson, Baylor Scheierman, Armando Bacot, Tyler Kolek, and Wade Taylor. However, you will equally see players who aren’t being discussed as possible All-Americans because they play more of a secondary offensive role, such as Anton Watson, Braden Smith, Zakai Zeigler, Reece Beekman, Kevin McCullar, and Dajuan Harris. This is part of what makes Bayesian Performance Rating unique in the space of rating players: The model doesn’t care if you score 20 points a game or 6, as long as you are having an impact on your team’s performance. The individual stats do still matter a lot. There’s a significant emphasis placed on each player’s individual box score efficiency stats. But players who contribute in other ways besides scoring, such as facilitating for other teammates, crashing the boards, or being hyper-efficient on attempted shots, can also shine.
Bayesian Performance Rating also cares about defensive impact. Measuring defense is a very difficult task in basketball, but BPR does a great job at doing the best it can to determine who the most valuable defenders are, both through box scores stats like steals and blocks, but also through how well the defense performs while a player is on the floor. Players who are predicted to have a high impact defensively based on BPR get some emphasis at the top of the preseason leaderboard: Donovan Clingan, Kyle Filipowski, Kadary Richmond, Kevin McCullar, Adem Bona, Josiah Jordan-James, and Johni Broome. The best offensive players are still predicted to impact the final score more than the best defensive players, but BPR isn’t as offense-biased as most human rankings are.