By Steve Shea, Ph.D. (@SteveShea33)
March 10, 2017
[There is a sortable table with CPR scores for 325 players, including all college players drafted in the last 6 drafts, at the end of the article.]
College Prospect Rating (CPR) is a formula based on college box score production that attempts to approximate prospects’ future NBA potential. It works from several foundational assumptions.
First, college players are inconsistent. In particular, freshman are inconsistent, and this is often the only year of production CPR has to evaluate. More than that, this inconsistency shouldn’t be held against players and using season averages has the potential to do that since particularly poor games on a small sample set can significantly alter season averages. (We wrote about this inconsistency in depth here, but note that the CPR scores produced there are from a very early version and not compatible with those presented in this article.)
To weed out the influence of inconsistency, CPR uses averages of top 10 performances in each stat. Unfortunately, this requires game log data, which is hard to come by for seasons even as recent as 8 years ago.
CPR’s second assumption is that the NBA game is evolving rapidly. The skills needed to compose a competitive NBA roster in 2017 are different from those in 2007 and very different from those in 1997. For example, today’s teams rarely depend on post up offense and instead, heavily depend on 3-point shooting.
So that CPR is more predictive of future success in the NBA than indicative of what would have been successful in the past, it does not use regressions. Instead, CPR weights skills according to what modern basketball analytics studies suggest are important today and predict will be vital in the future. In addition, these weights in CPR should be recalibrated regularly to reflect the most modern research. This year, we’ve decided to do just that, but before we go into the details, there’s one more very important assumption to discuss.
CPR cannot account for everything. CPR uses box score statistics, and these statistics tell an incomplete story of a player’s production. In addition, we can only measure qualities like athleticism and quickness indirectly as they translate to production in the box score. Finally, there is no jerk factor in CPR. Some prospects are jerks. 19-year-olds never come out of college ready to be superstars in the NBA. All of them must continue to work exceptionally hard and respect their coaches’ advice. Jerkitude tends to obstruct development.
What’s New in CPR V3
Previous versions of CPR weighted production by year in college. For example, a sophomore needed to out-perform a freshman in order to have an equal rating. This new version of CPR uses age. This was done for two reasons. First, previous versions seemed to be fooled by older freshman. 18-year-old production is different than 20-year-old freshman production, and in the old system, they weren’t considered as such. We’ve decided to go to age also because it eases the addition of foreign players, which we plan to include in the near future.
As CPR evolved two small-sample issues developed. First, some players didn’t play that many games (e.g. Kyrie Irving). To accommodate this, previous versions made case-by-case adjustments. For example, Irving might be judged on only his top 5 performances as opposed to his top 10 since he missed more than half of the season. We standardized this in the new model. Any player that played less than 20 games is judged on his top 7 performances.
In addition, we’ve added strict minimums for 3-point attempts in order to consider 3-point percentage. Players that take 20 or fewer 3s are considered to have a 10% 3P%. Players with between 21 and 40 are considered 20% shooters. We use actual 3P% for players with more than 40 attempts.
The most significant change to the model and the one we’re most excited about is that we’ve introduced stat groups built upon complimentary statistics. For example, we have a shooting stat that combines 3P% with FT% in a way that excellent production in both compounds more so than simply summing the two stats’ contributions. The data suggests that isolated stats can be misleading. For example, a player can shoot over 40% on 3s in 45 attempts while not really having the type of 3-point shot we’d expect to translate to the pro ranks. Instead of discounting that 3-point production altogether, we look at FT% as an indicator of the player’s shooting touch and consistency of mechanics. If the player was a 62% free-throw shooter, the new model suspects that his 3P% is inflated. If the player is an 85% free-throw shooter, the new model suspects the high 3P% is legit.
This grouping also allows players to not completely bottom out in a stat simply because they weren’t used in that capacity in college. For example, the previous model saw Karl-Anthony Towns as a weak 3-point shooter since he only had 8 attempts in college. The new version sees Karl Anthony-Towns as a respectable shooter because he shot 81% from the line.
In total, there are 6 stat groups—shooting, defense, versatility, scoring, playmaking, and interior production. (Playmaking is a group of one—assists—since the box score doesn’t provide any complimentary statistics in this category.)
In previous versions of CPR, player production was weighted by year in grade. In other words, a senior had to significantly out-produce a freshman to have the same rating. We’ve already mentioned that we’ve converted from year in school to age. There’s more modifications here.
In the past, the weight was the same across all stats. That never seemed appropriate, and in this new version, we address it. The collective weight (or the sum of the coefficients) of the stat groups drops as the players get older, but the rate at which each one drops varies. Scoring starts with the bulk of the weight for the youngest players coming out. Young players can often lack the strength to post admirable interior numbers and the consistency of shot selection to shoot efficiently. Playmaking, defense and the ability to produce in a variety of areas (versatility) are also qualities that tend to take longer for players to develop. At this young age, scoring tends to be the biggest indicator of future potential. If an 18 or 19-year-old has put up a few 30+ scoring nights, that suggest two things. First, it shows that this young kid has the ability and confidence to take over a game. Second, it shows that the coach has the confidence and trust in him to allow the player to do so.
From 19-year-olds to 20-year-olds there is a significant drop in the weight on scoring and a shift in focus to interior production, playmaking and defense. For a 20-year-old to have the same contribution to his CPR from scoring as a 19-year-old, he will have to significantly out-perform the younger player. But the 20-year-old has more opportunities to contribute to his CPR in other areas. If a prospect is going to be a scorer in the pros, it’s expected that he will be scoring by his sophomore season in college. It’s expected, not extraordinary. To feel confident about projecting this player forward, we need to see excellent production in at least 1 other stat group.
As the players get older, the emphasis shifts to shooting and versatility. The reality is that each draft has very few star players. In Basketball Analytics (2013), we found that outside of the lottery, each 1st-round pick yields a star about once every 30 drafts. It’s not easy to land even a good role player. We found that picks 15-20 produce a good role player (weakest starter or 6th man) about a third of the time. But if teams can aim for the right player types, they can increase their odds of a successful draft.
With the direction the NBA is trending, teams need players that can space off the ball on offense, in other words, players that can shoot 3s. Teams also need defenders, and in particular, players with some positional versatility on the wings. By positional versatility, we mean the ability to guard multiple positions, which allows defenses to more easily switch screens. Arguably, the most important role player in today’s NBA is the guy who can space on offense and defend on D, the so-called “3 and D” players. (Think Shane Battier.)
For an older prospect to rate highly in CPR, he must either put up truly extraordinary numbers all around or he must profile as a 3 and D. We’ve found that our stat groups of shooting and versatility do this well.
Finally, the last major change to our model is that it’s been recalibrated to college players drafted in the last 6 drafts. After we average the top 10 performances in each stat for each player, we convert each average to standard deviations from a mean. In this case, the mean is the average over all NCAA players drafted in the last 6 drafts.
Sometime, analyzing basketball through statistics can feel like judging a movie by reading the script. When we’re limited to box score stats, it’s like reading a script where 4 out of every 5 words have been removed.
CPR is not a list to draft off of. It’s a perspective that is best used in collaboration with other means of assessing talent. Suppose a team’s scout presents his ranking of available college prospects prior to the 2011 draft, and let’s assume that order is the actual order in which these prospects were drafted. The following table presents the first 20 college players taken in that year (the top 20 in our hypothetical scout’s rankings).
In the table, we’ve also included each player’s CPR. The first thing we notice is that two picks in the teens rate well in regards to CPR. Those players are Klay Thompson and Kawhi Leonard. (Kenneth Faried looks pretty good too.)
In this instance, the comparison of the scout’s rankings and CPR should spark a conversation. The analyst should explain why CPR likes Klay and Kawhi. The scout should explain why he likes Tristan Thompson and Brandon Knight more. Neither the model nor the scout will always be right. Hopefully, the conversation enlightens all involved and leads to a collective decision that works best for the organization.
Here is another example. Suppose a team is drafting somewhere between pick 22 and 44 in the 2012 draft. The team has mapped out who they believe will go in the top 21 and their staring and the following scout’s rankings of the next 20 or so picks. (Again, we’ll use the actual order the players were drafted as our hypothetical scout’s rankings.)
Again, we see a couple players stand out in CPR. And again, this should start a conversation. In this case, Draymond Green and Jae Crowder rate well because they are older prospects, are particularly versatile, and CPR rates versatility highly in older prospects. In fact, Crowder and Green are the top two in versatility in this class. Here, the analyst would explain that CPR projects Crowder and Green to be good 3 and D forwards. (The high ratings for Crowder and Green are consistent with the previous versions of CPR.)
Let’s look at one more example. Suppose a scout presents the following top 10 college prospects from the 2014 draft.
Notice how CPR really likes Marcus Smart and doesn’t care for Aaron Gordon. It’s not hard to figure out why the model feels this way. Smart averaged 5.5 steals and 25.5 points per game in his top 10 performances in the respective stats. He was also a good rebounder for a guard, grabbing 9.3 a game in his top 10. The model’s biggest issue with Gordon is his shooting. He shot 42% from the free-throw line in college.
In an NBA that’s heavily invested in perimeter shooting, it is very hard for a wing player to be great without a decent 3-point shot. In this example, the discussion should begin with Gordon’s shooting. What is broken and can it be fixed?
This season, Gordon’s third year in the league, he is shooting 27% from 3 and 67% from the free-throw line. (Smart hasn’t been much better from 3, but he is shooting 79% on free-throws.)
What do the Numbers Mean?
We’ve already seen some sample outputs of the new CPR. The output is not in basketball terms. To interpret CPR, we need some loose guidelines.
A CPR ≥ 15 suggests a superstar, the kind of player that doesn’t come even once a draft. A CPR between 12 and 15 is still stellar and suggests an NBA All-Star. Those that fall between 9 and 11 are a mixed bag. A player in this range that checks all the other boxes (size, athleticism, character, smarts, and health) is a fantastic prospect and should be a top 3 pick. But in this group there are also players with bodies better suited to dominant the college game.
A score between 6 and 8 suggests either an older prospect that has one translatable skill (e.g. Buddy Hield and shooting) or a younger player that didn’t do anything extraordinary statistically at the college level (e.g. Harrison Barnes or Jaylen Brown). In the former case, we could be talking about a very good contributor off the bench. In the latter, there is potential, but plenty of work to be done.
A score of 5 or below says your drafting this player for reasons other than his statistical production. An example would be Trey Lyles, who played on such a loaded Kentucky team, that he didn’t get many touches (and arguably played out of position). It could also be DeAndre Jordan or Zach Lavine, a highly-touted player coming into college who didn’t quiet mesh with his college coach and system.
There are certainly some now very good NBA players that rated below 6 in CPR. However, more often than not, taking a player rated this low early in the draft turns out to be a regrettable decision. For example, Marcus Morris (CPR=4) was taken just before Kawhi Leonard (CPR=10) in 2011. In the same draft, Nolan Smith (CPR=3) was taken just before Kenneth Faried (CPR=10). In 2012, Dion Waiters (CPR=4) and Thomas Robinson (CPR=4) both went just before Damian Lillard (CPR=8).
The 2017 class is loaded with talent on the perimeter. There’s no Durant, Paul or Steph, but Malik Monk, Markelle Fultz, Lonzo Ball, and Dennis Smith all project to be All-Star caliber NBA guards. (These scores are also included in the table at the end of the article.)
Forwards Jayson Tatum and Josh Jackson both rated well and should be in consideration for top 5 picks.
The most interesting rating belongs to Caleb Swanigan, the sophomore from Purdue. Swanigan scored a 13 in CPR thanks to some other-worldly rebounding and shooting 45% from 3. He stands about 6’8 with a 7’3 wingspan which suggests power forward. However, he’s not an agile defender and could struggle to guard smaller 4s. The NBA is trending smaller at the PF, which means Caleb should prepare himself to play a fair amount of center.
The worst case scenario for Caleb is he can’t handle defending NBA centers on the interior. In that case, he could go the way of Jared Sullinger, who is struggling to find a role (and a team) these days. He’ll be a solid rebounder at the 4 with some ability to stretch on offense, but will only see the court in situations when the opposing team plays a slower forward he can defend.
The best case scenario is that Caleb can play a small-ball 5. It’s a lot of work to bang with the likes of Andre Drummond and DeAndre Jordan when you are only 6’8. So, Caleb may be limited to 20 minutes a night in that role. Toss in another 8-10 minutes a night as the 4.
If Caleb can handle the defensive duties of center, and he continues to shoot a decent percentage from 3, then he can be a real weapon for offenses. He can draw the opposing big out of the lane, perhaps to the corner and clear space for his teammates. As a pick and pop threat, he’ll get extra attention as the screener.
Right now, Caleb is projected to be a late first rounder. At that point, he could be a steal. And you have to root for a kid with this backstory.
The following table contains the CPR scores for every college player taken in the last 6 drafts. It also contains 25 of the top 2017 prospects including those mentioned above. Finally, we added 22 players from years prior to 2011 as reference points.
CPR has always been built to accentuate differences at the extremes. For example, Hassan Whiteside scores so well because in his top 10 performances, he averaged 8.4 blocks a game. (He was also a good rebounder.)
While Whiteside was exceptional in 1 or 2 categories, Durant was exceptional in almost everything. In his top 10 performances in the respective stats, he averaged 34.2 points, 15.7 rebounds, 3.6 steals, and 4.1 blocks. He did this with a FT% of 82% and a 3P% of 40%. And he was just a young freshman. All of this production compounds to produce a CPR rating so far above and beyond anything else we’ve seen (and may ever see again).
[All Stats on 2017 prospects are current through March 8th games.]
|173||Larry Nance Jr.||2015||27||5|
|202||Glenn Robinson III||2014||40||5|
|232||Roy Devyn Marble||2014||56||4|
|301||Johnny O'Bryant III||2014||36||2|