Our Actionable Analytics squad has been working on a new feature to help coaches and scouts develop winning team strategies – learn all about how we are producing projected player stats and how they can be useful for coaches.

The transfer portal has become a huge part of recruiting in college basketball. Coaches want to find players in the portal who can help their teams win. Stats – like shooting percentages and play type efficiencies – can be used as filters to help quickly narrow down the pool of players available so that coaches can start watching tape and contacting the kids they think will fit into their programs faster.

One big challenge with college basketball stats, though, is that the season is short – just 40 games, at most – so the information provided by them can be noisy. That challenge is exacerbated in the transfer portal where some of the players won’t have had a chance to accumulate many stats. In fact, players often enter the transfer portal because they are unhappy with the number of opportunities they’ve had on their current team and are looking for a new team where they can get more playing time or more shots. These types of players may have “small sample sizes” for stats like 3-point percentage, ie. there won’t be much statistical evidence to show whether they are good or bad shooters. As a result, shooting stats and play-type efficiencies for these low-volume players can be misleading.

And really, rather than looking back at what players have done in the past, what coaches want to do is look ahead to what they’re going to do next! With that in mind, the goal of Synergy’s player projections is to estimate a player’s stats for the next game. These projected stats can be useful for team evaluation, opponent scouting, and player recruitment in the transfer portal, too.

For example, here’s a look back at the Top-25 leaderboard for projected 3-point percentage among the players in last year’s transfer portal, ie. this is how their projected stats appeared at the end of last season as they were announcing their intentions to transfer.

Finding shooters in the portal

Top-25 leaders in projected 3-point shooting among 2023 Division 1 college men’s transfers.

This represents the players who we thought were the best 3-point shooters available in the transfer portal last summer. It’s an impressive list featuring several fantastic shooters like last year’s Summit League Player of the Year, Max Abmas, 2023 All-Mountain West guard, Steven Ashworth, and All-Big 12 star and national champion, LJ Cryer.

Cryer topped the list with a projected next-game 3-point percentage of 38.3% at the end of last year. And, fortunately for the Houston Cougars, his effective long-range shooting has carried into this season, as he’s currently hitting 39.1% of his threes this year.

Of course, there were players in the transfer portal with even higher 3-point percentages than Cryer and the other guys on our projected 3-point shooting leaderboard. Omar Payne from the Jacksonville Dolphins went 1-for-1 (100%)! Elijah Buchanan shot 8-for-17 (47%). But the transfers on the projected 3-point shooting leaderboard had a combination of good 3-point shooting percentages and a large volume of 3-point attempts, which made us more confident that their shooting was real and that their stats would be sustainable going forward.

Is this guy’s shooting real?

Top-25 leaders in 3-point shooting among 2023 Division 1 college men’s transfers.

Cryer, Abmas, and the other players on the projected leaderboard attempted an average of 176 threes last year and they cashed them in at a 40% clip. Even though Payne, Buchanan, and the other “actual” 3-point percentage leaders made more of their collective threes last season (50.2%), they attempted far fewer of them, just 17 on average. That’s why the projections viewed their shooting as being less sustainable.

And now, having seen how the 2023-24 season has played out, we can look back and evaluate whether our projections were any good. The proof is in the pudding, as they say. Our projected 3-point shooting leaders have continued to shoot a lot of threes – 173 on average – and they have made 36.9% of them this season, which is pretty much exactly what was expected of them (the target was 37.2% for the group as a whole). Of course, individual players have exceeded expectations or fallen short of them, but, collectively, these guys have certainly proven their mettle again this year.

By comparison, our group of 25 straw men – the actual 3-point leaders from last year – have mostly continued to be low-volume shooters and, perhaps not surprisingly, they have mostly not replicated their sporadic success from last season. That group is shooting just 32.6% from three this year, just below the D1 average of 33.7%. This is what the projections expected from this group of low-volume shooters – that they would “regress towards the mean” for all D1 players.

Now, I know what you’re thinking to yourself. “I may not be a numbers guy, but I’m not dumb enough to recruit a kid who only took one stinking three during the entire daggum year and fool myself into thinking he will become a 3-point shooter for me!” Fair enough. It’s a no-brainer to dismiss 1-for-1 shooting. Toss it out. It’s meaningless. But that 8-for-17 is a little bit trickier. And how about 26-for-51? Would that be enough to convince you the shooting is real?

A common approach is to use a threshold for 3PA – say 50 attempts – to find a group of “qualified” high-volume leaders (this is the third group that’s shown in the table; the one that was between our other two groups in terms of 3PA and 3P% last year). These qualified leaders (50+ 3PA) are shooting better this season (35.6% on threes) than the unqualified leaders, but they still haven’t fared as well as the projected leaders.

Moreover, there’s another problem with using a minimum threshold to weed out low-volume shooters: it can cause us to overlook good players. What if our cutoff hid a really great shooter who just happened to have attempted exactly 49 threes?! A player who was injured for part of the year and missed a bunch of games, for example. The projected stats can help us balance out 3PA and 3P% in a more systematic way and they also incorporate information from previous seasons to help find more stable estimates of players “true shooting talent” without relying on any arbitrary thresholds for qualification.

We are producing player projections for several other stats in addition to 3-point percentages including a new on-ball points-per-possession metric. On-ball PPP is a stat we haven’t surfaced on the Team Site before – it’s an amalgamation of a few familiar play types: a weighted average of a player’s scoring as the ball handler in isolation, on picks, in the post, or in transition. Like shooting percentages, on-ball PPP numbers are noisy for players who don’t get a lot of on-ball opportunities.

Perhaps you’ve had the disappointing experience of sorting a play type leaderboard only to find it littered with bench players who have those dubious “2.000 PPP” marks right at the top, unhelpfully denoting somebody with exactly 2 points on exactly 1 possession, one lucky shot. But when you sort the on-ball PPP player projections leaderboard from the 2023 transfer portal you find nothing but studs: Tylor Perry, Zyon Pullin, Great Osobor, Jaylin Sellers, Isaac Jones, etc.

And who do we find at the very top of the projected on-ball PPP leaderboard? Consensus All-American, Hunter Dickinson. Last year, Dickinson scored almost exactly 1 point per possession on over 300 on-ball possessions (304 points on 309 possessions). The projections expected more of the same from him this year, predicting a juicy 1.00 PPP from his on-ball possessions heading into this season. So far, he’s done even better than that, scoring 1.13 PPP over nearly 200 on-ball chances this season at Kansas.

Just like we saw for the projected leaders in 3-point shooting, our projected on-ball PPP leaders from last year’s transfer portal had likewise produced on a higher volume of opportunities than the actual leaders and, thus, their stats were deemed to be more sustainable.

Is this guy’s scoring real?

Top-25 leaders in on-ball points-per-possession among 2023 Division 1 transfers.

And like we saw before, the transfers who were at the top of the player projections leaderboard have been more successful this season than the actual leaders (with or without using a 50-possession threshold to identify qualified scorers). Our projected leaders produced 1.03 PPP on-the-ball in 2022-23 on 151 possessions each. As a group, we expected them to notch 0.95 PPP this year and so far, they are coming in just a tad under that mark, at 0.91 PPP. In contrast, there was a group of lower-volume on-ball scorers who were fortunate to tally a higher scoring average (1.43 PPP) on a smaller number of on-ball possessions (8 total, on average) last season and, as expected, they have been less successful this season (0.84 PPP).

If you want to get into the nitty gritty, we’re using a mixed modelling approach with so-called “random effects” for each player. The more information we have about a player (more minutes played, more shots attempted, more possessions used), the larger those player-specific effects can be, in either a positive or negative direction, allowing the player’s projections to float higher or sink lower than league average. In contrast, a player with less information (fewer minutes played, fewer shots attempted, fewer possessions used) will have his projections pulled back towards the mean.

Each new game feeds new information to our player projection models, and our models can incorporate multiple seasons of data, for as long as a player has been playing in his current league (throughout his entire college career, for example). A game played yesterday will always be weighted more heavily by the model than a game that was played a year ago, but the precise weighting depends on the stat in question and how quickly it stabilizes. Each stat that we’re producing has a different “learning curve”. A stat like 3-point percentage takes a long time to stabilize, so it can be useful to consider a player’s career shooting percentages when trying to estimate his next-game performance. In contrast, a stat like points-per-game stabilizes much more quickly, so recent games will have a much greater impact on our estimates than games from previous seasons.

We have back-tested our projections against past years of data to optimize predictive power. We have also validated our projections by comparing their accuracy against DARKO, a leading public projection system for NBA stats. We found that our model was more accurate than the leading projection system for some stats, and that we fell slightly behind DARKO in others, always with a close margin of error.

A unique part of our projections is that, thanks to Synergy’s vast database, we can use a similar methodology to produce player projections for many of the leagues and levels that Synergy covers, including women’s college basketball, men’s DII and DIII, and several international leagues as well. The model accuracy for those leagues is comparable to the accuracy of our NBA projections, after accounting for the fact that NBA players play more games and thus have more reliable sample sizes. For college basketball, we train the player projection models at the division level.

For now, our player projection models provide projected stats for the player’s “next game.” They are not an attempt to predict a player’s “career trajectory” and they won’t anticipate changes in true talent level due to a player’s off-season development, injuries, or aging. The player projections are not opponent-adjusted either. The “next game” projection represents what is expected from a player going forward in a generic way (you can think of them as “remainder of the season” per-game projected stats, like something you’d find in your favorite fantasy basketball app). Likewise, our player projections won’t “know” about and, thus, cannot account for specific strategies you might employ against an upcoming opponent, like double-teaming a scorer to get the ball out of his hands.

Synergy’s projected player stats will be helpful tools for finding winning players in the transfer portal. Coaches can also use projected player stats for team evaluation and opponent scouting, too. The new Player Projections Leaderboard is live on the Team Site now for Insights customers. We’d love to hear what you think of this new tool and how you’re using it, so get in touch with us!

The development of our new projected player stats was led by Jonathan Lewyckyj. Jonathan is building tools to help coaches, scouts, and players find winning team strategies as part of Synergy’s Actionable Analytics Team.