Introducing Adjusted Plus-Minus for College Basketball

This new data provides sharper evaluations of college players and draft prospects

Cooper Flagg (Harry How/Getty Images)

Today I want to provide a look at what I’ve been working on this offseason — and a preview of coming attractions.

While the NBA world is awash in advanced metrics, including my own family of adjusted plus-minus numbers, we have far fewer advanced metrics for college basketball.

That’s partly because the NCAA is way behind the NBA when it comes to making advanced data available to the general public. But things are improving to the point where things like lineup data become more and more reliable.

That allows us to calculate stats like plus-minus, on-off and their more accurate cousin, "adjusted plus-minus" (abbreviated below as RAPM).

Today we're putting out a free explainer on why this dataset matters, with some key takeaways for the 2025 draft class.

---

What is (R)APM and why is it important?

RAPM is an all-in-one (AIO) player metric.¹

AIO metrics are designed to determine a player's influence on the point differential of his team. Contrary to simple statistics such as points per game and doubles doubles, this helps identify players that contribute "empty calories" as well as the opposite: high-impact players who might fly under the radar because of their relatively small box-score contributions.

This type of information is valuable to NBA decision makers, gamblers and fans.

Most, if not all, of today's AIO metrics consist of two parts:

1. a metric derived from individual (box-score) stats, like Basketball Reference's Box Plus-Minus (BPM);

2. a lineup-based metric, such as RAPM.

RAPM tries to determine a player's influence by (a) incorporating points scored by the player's team and the opponents, while also (b) taking into account the strength of the player’s teammates, the strength of the opponents, and the state of the game.

A simple example:

Let’s say OKC picked up a player named Johnny Thunder for last season, and he posted an on-court rating of +10 points per 100 possessions. RAPM would not see Johnny as a positive-impact player because his teammates were even better — OKC recorded a +13 net rating last year.

Of course, play-by-play data allows RAPM to also know the exact lineups for every minute of every game. So Johnny’s exact rating would depend on how much he played with Shai Gilgeous-Alexander, how much he played against bench players, and so on.

On the other end of the spectrum, if Washington picked up Wayne Wizard for last season and outscored opponents when he was on the floor, Wayne’s RAPM would be significantly positive because his teammates were so bad.

---

Box-score metrics vs. lineup-based metrics

For college players, we already have a number of metrics built from box-score stats, including the previously mentioned BPM.

Box-score metrics are less influenced by noise and stabilize quickly. But they reward stat-stuffers — players who sacrifice team success for their own numbers.

And box-score metrics tend to have a weak grasp on defensive impact — we have fewer box-score stats for defense, and those can be misleading. Steals, blocks, and defensive rebound numbers can be inflated by players who play more selfishly without impacting the game.

RAPM, on the other hand, is specifically engineered to help us get a stronger grasp on a player's defensive impact. It also helps us identify empty-calorie stat stuffers.

Another area where RAPM shines is adjusting for opponent strength. Box-score metrics don't distinguish whether someone scores 20 PPG playing against starters or primarily against bench units. For RAPM, that’s an important part of the equation.

And here is a more technical but important benefit of RAPM: Metrics like BPM use RAPM to have an objective way to weigh individual stats — for instance, to determine the value of an assist or a turnover. Without multiseason RAPM, this becomes a very difficult problem that's also much more computationally expensive.

---

What NCAA RAPM can and cannot tell us

While critics can point to the noise in the NBA version of RAPM, consider that single-season RAPM rates SGA, Nikola Jokić and Giannis Antetokounmpo as the league’s top three players. (To be fair, RAPM’s NBA rankings do often diverge from the consensus.)

For college, the situation isn't quite as elegant, for two reasons:

• There are far more college players, making it more difficult for the algorithm to divvy up credit.

• College teams play fewer games which are also shorter in duration and played at a slower pace. All of that significantly reduces sample size.

Still, one thing seems clear: When NBA teams are preparing for the draft, they should focus more on players who were difference-makers in college. Draft prospects should typically outperform other college players in positive impact, by a good distance — enough to overcome the noise in the data. When they don’t, that should be a major red flag.

And if that player is younger — like Cooper Flagg, who turned 18 during his freshman year — all the better.

While no metric is the be-all and end-all, RAPM can provide an objective view of every college prospect and provide a crucial complement to scouting while also reducing the number of players to evaluate.

---

---

How much does this analysis help with drafting?

This is something we are still working on — more statistical analysis is necessary.

Preliminary analysis reveals that defensive RAPM — as compared to offensive RAPM — adds more valuable information to the equation when predicting impact players in the draft.

That’s for three reasons: Defensive stats are misleading, as noted above; scouts tend to have a harder time identifying defensive impact accurately; and evaluating defense requires more video work, which makes it more tedious and time-consuming.

---

---

Other use cases for RAPM

AIO metrics for NCAA already exist. But essentially all of these metrics were originally tuned for the NBA. In other words, the implicit assumption is that, for example, a turnover in a college game is exactly as costly as a turnover in an NBA game. The same goes for offensive rebounds and other counting stats.

But the probability that these stats should truly be weighed the same across such disparate leagues is low.

With long-term RAPM being the backbone for NBA box-score metrics such as BPM, long-term NCAA RAPM, then, should allow for creating a box-score metric tailored to the college game.

That will lead to more accurate impact estimates, which will be helpful for everyone from NBA teams to bookmakers and bettors.

---

Ace Bailey (Harry How/Getty Images)

NCAA RAPM for the 2025 draft class

Let’s wrap up by looking at some of the NBA’s rookie class members who played college ball last season.

What you will get here is just an overview because I’m still working to incorporate all of the available data — it’s a multistage process.

While these numbers should provide a rough tendency where things might be heading, we are aiming to increase the model's accuracy for the next draft class.

The rankings you see for each player are within the draft class (40 draftees played more than 2,400 possessions in the NCAA last season), and within the entire sample of college players (1,368 players last season).

Players that RAPM likes

Cooper Flagg (2nd in draft class, 4th overall)

Kon Knueppel (5th in draft class, 8th overall)

Tyrese Proctor (6th in draft class, 10th overall)

It's no surprise that several Duke players rank highly, given the team’s strong performance throughout the season. And all three players sported a net rating around +30, indicating their dominance over (almost) the entire field.

And while no one was doubting Flagg, the fact that he came out so strong in a metric that's typically dominated by older players has to count as an additional positive.

Yes, Proctor played three years of college ball, but he rated highly in RAPM each season — a very positive indicator for his NBA career.

Ace Bailey (3rd in draft class, 6th overall)

Both Bailey and Rutgers teammate Dylan Harper received criticism because the Scarlet Knights posted a losing record.

This method sees Bailey as a highly positive player last season, and blames his teammates.

The fact that Rutgers defended eight points (per 100 possessions) better with Bailey on the floor should give Utah fans reason for optimism.

Will Riley (10th in draft class, 19th overall)

I list Riley here because these new numbers appear to dispel my concerns about him somewhat. As I’ve noted, Riley's steal and block rates indicated that he was one of the weaker prospects.

But now we have data that points in a more positive direction, with Illinois outscoring opponents by 13 points more when Riley was on the court (vs. off the court).

Thomas Sorber (15th in draft class, 50th overall, 1st overall in defense)

Sorber, as you can see, ranked No. 15 in this draft class — and No. 15 is exactly where OKC selected him.

But since defensive RAPM appears to be the slightly more informative part of the equation, Sorber's No. 1 ranking really catches the eye, especially given his age — Sorber won’t turn 20 until Christmas Day.

With Sorber on the court, Georgetown allowed a very impressive 88 points per 100 possessions, making them one of the best defenses in the country. Without Sorber on the floor, that number blew up all the way to 114, a whopping 26-point differential.

Derik Queen (14th in draft class, 42nd overall)

Whether you see Queen's ranking as a positive depends a little on your perspective.

Critics, me included, have pointed out that Queen can look lazy on defense. But the numbers say Queen was a crucial part of Maryland's success.

If he can cut down on his defensive lapses — by boxing out, for instance — Queen’s defensive basketball IQ might be enough to allow him to eventually be a positive contributor for the Pelicans.

Kam Jones (1st in draft class, 2nd overall)

Nique Clifford (4th in draft class, 7th overall)

Jones and Clifford are two of three non-freshmen on this list — in fact, they played four and five years of college ball, respectively. But both players deserve a shoutout, as their RAPM numbers are simply stellar.

Marquette outscored opponents by 16 (per 100) with Jones on the court, and got outscored by 16 when Jones sat. (I should note that there isn’t a lot of “off” data for Jones, making the numbers less reliable.)

As a productive combo guard, Jones might be just what the doctor ordered for a Pacers team that probably will try to win, even without Tyrese Haliburton.

Clifford’s strong showing in summer league has already raised expectations for the shooting guard, selected No. 24 by Sacramento.

As with Jones, Clifford's Colorado State team went from strongly positive with him on the court (+15) to very negative when he sat (-13).

---

---

Players that RAPM doesn’t like

There is one saving grace for this group: Young, highly athletic players can have positive NBA careers despite posting underwhelming college RAPM numbers.

That applies here because the three players below were freshmen. So while they might have a bumpy transition to the NBA, a bad ranking here is far from a death knell.

VJ Edgecombe (33rd in draft class, 352nd overall)

Edgecombe might be the player for whom a disappointing rating is the least worrisome: The more athletic a player, the more likely he is to raise his game over the following years.

That said, Baylor scoring a whopping 12 points more per 100 without Edgecombe on the court — that should serve as a warning to Philly fans to keep immediate expectations in check.

Tre Johnson (34th in draft class, 372nd overall)

As a Johnson skeptic, this ranking doesn’t surprise me. At the same time, like Edgecombe, Johnson might have what it takes to overcome his college track record.

The low ranking comes from his inefficient shot diet in college — including a lot of midrange jumpers — and his below-average defense. That’s a recipe for bad impact metrics.

Johnson’s fans are enamored by his shooting touch and ability to make tough jumpers. If he can convert difficult shots at an above-average rate, that gives him a path to positive impact for the Wizards.

Johnson played so much for Texas that we don’t have a large “off” sample. But I don't love the fact that Texas outscored opponents by 15 points more when Johnson sat.

Asa Newell (39th in draft class, 841st overall)

Newell gets the dubious honor of being the only 2025 first-round pick rated as a negative-impact player in college. It’s worrisome that Georgia defended eight points per 100 better without him on the floor.

As an NBA big man, Newell projects to be a bit of a tweener, which also raises questions about his future defensive impact. According to RAPM, Hawks fans should temper their hopes for Newell to contribute a lot as a rookie.

---

Still to come

For this coming college basketball season, expect us to fine-tune our algorithms to make them more accurate by incorporating age adjustments, individual stats and other statistical tricks, such as better Bayesian priors.

This should allow us to identify true difference-makers in college even better than we can now. And it should help us create an NBA draft model that's likely to outperform what NBA teams actually do in the 2026 draft.

1. "The "R" in RAPM stands for "regularized” — a technique that makes RAPM a machine-learning-powered, more accurate version of APM through the introduction of Bayesian priors.