A Basketball Philosophy

Summary

I’ve spent a lot of time thinking, playing, watching, and studying basketball. I have learned a lot and have plenty to learn. I am pausing momentarily to collect and summarize my thoughts, which cover a wide range of topics, including player development and NBA analytics. 

Playing Multiple Sports

Playing multiple sports at a young age is optimal. Each sport has different movement patterns. By developing multiple patterns, young athletes become more flexible and fluid movers. Participating in multiple sports reduces the risk of injury by developing movement patterns according to multiple sources because of increased strength and endurance and improved growth plates and muscles. Playing multiple sports also helps athletes adapt to various learning styles. It also helps athletes see basketball in a new way. When a basketball player plays soccer, they may pick up the importance of swinging the ball from quickly passing the ball, while playing football may teach a player how to shift their weight while running a route. Even a sport like tennis will teach players how to plant and change directions consistently. The ability to problem solve in team sports translates in ways that are difficult to predict. There is a time for specialization, but that doesn’t come until at least high school. Even after specialization, playing other sports once a week can help develop movement patterns. Even playing individual sports such as gymnastics or ballet can help athletes develop better flexibility and balance. NBA players such as LeBron James, Allen Iverson, and Ben Wallace played football in high school, while Joel Embiid and Hakeem Olajuwon grew up playing soccer.

Training 

Player training is a hotly debated topic. Many different trainers on social media show different drills to the point where it’s nearly impossible to determine which drills are effective. I love Bobby Whytle, Coach Ross, and Transforming Basketball (website/social media page) for the ideas they present and the drills they teach. Every basketball game is like a problem and needs to be treated like one. Players need to learn how to control the basketball, but soon after that, there should be stimulation while training. Anything giving a player a live obstacle would be classified as stimulation. For example, here is a solid ball handling workout: two minutes of stationary ball handling, moving pound dribbles, speed dribbling between the legs and wraps, acceleration into dribbles, dribble knockout, and dribbling with a defender in a confined space. Stationary ball handling to work on ball control, while the rest works on more game aspects such as speed with the ball, getting low, pounding the ball, and being creative against defenders. An entirely stationary workout is suboptimal due to a lack of stimulation. Shooting workouts should include closeouts or some form of defense when players are trying to optimize development.

Small-sided games (live games with less than ten players) are also excellent as they force players to play live but provide more reps. Some of my favorite drills include one-on-two in a set amount of time, two-on-two in a certain amount of time, three-on-three with the offense starting with an advantage, and one-on-one, where a player has to start by running around a chair or something similar. These drills pose problems that players have to solve on offense and defense. In my experience, one on two is an awesome drill to build creativity and aggression. Transforming Basketball wrote a great article (Why it's not about Fundamentals), which I recommend reading. Everything discussed above is great for player development. Still, there needs to be a focus on the live drills, prioritizing catch and shooting, moving without the ball, spacing, and attacking off advantage. Many players only practice with the ball rather than playing off the ball. To practice off-ball, teams should practice with no dribbles, limited dribbles, only certain players can dribble, etc., to help players learn the value of off-ball concepts.

Reforming AAU

AAU has its pros and cons, but it teaches many bad habits. The advantages include exposure to prep and college coaches and high-level competition. When I played AAU for the first time, I was shocked at the level of competition and the game's intensity. I matched up against players I didn’t even know existed and loved it. It wasn’t until later that I saw how turnover-prone and lazy the game could be. I started seeing more and more blowouts where teams were not evenly matched, and the games turned into track meets. AAU overemphasizes games over actual development and meaningful competition. Players switched teams frequently because they didn’t like how many minutes they got in a week, leading to chemistry issues. All of this leads to a not-so-optimal system that is a train wreck.

There is no easy fix, but I can offer a few suggestions. There needs to be one centralized league, and we need to eliminate all the different circuits (Nike EYBL, Adidas 3SSB. etc.). AAU should not start until middle school because any AAU younger than that is a money grab (99% of the time). In sixth through ninth grade, games, tournaments, and leagues should be local, and nobody should travel for more than a few hours for a game. Once a player is in tenth grade, we have a major circuit with multiple levels and stricter rules. Players can not transfer teams during the circuit period, and teams will only play twice weekly. This still does not fix the problem of practices, bad coaches, and poor game structure at times. I think dividing one circuit into multiple divisions based on quality is a great idea to centralize everything and cut back on the blowouts. While there is too much isolation at times, increasing the number of practices and roster continuity would help fix both these problems. I also don’t believe players should transfer to high schools more than once. Start at a local public or catholic school, and if the competition is not great or you simply want to play with the best, then transfer to a school like Montverde or IMG. Unfortunately, some of these schools are run like AAU programs without continuity. 

Ranking and Drafting

I also have problems with how players are ranked and drafted, which can be broken into two primary issues: How people understand potential and athleticism. I used to be a little kid who was in love with the NBA draft and enamored with the idea of potential. In the weeks leading up to the draft, I read as many profiles as possible, and the concept of potential came up constantly. Some “traditional” athletes were considered high-upside players, but why? I believe that there are only a few determinants of potential, but the most significant factor is current ability. One player isn’t more likely to develop than another player because they can jump higher. The reason most of the players with “potential” don’t pan out is because their athleticism isn’t translatable. The only traits that will lead me to say a player has more potential than another are age and length of playing basketball. Younger players will have more time to develop than their older counterparts, and the same will happen with players who have played for less time. When drafting players, you want to draft productive players as production translates. Some players I want to highlight because of the current production and age are Cameron Boozer, Collin Murray Boyles, Reed Sheppard, and Zach Edey. Boozer is ranked number two and yet is underrated by most people. He constantly produces, even if not flashy, and his skills are conducive to winning because they integrate well with other good players. Although entirely different playstyles, the other three were highly productive college players for their age. Edey is older but has been producing at a high level since his freshman year. He is often compared to players like Luka Garza (an older big man who dominated in college), which is unfair to Edey. Garza was a good prospect and has been solid (51% EPM in 2022, 90% EPM in 2023, 57% EPM in 2024) in the NBA in limited minutes. Garza had a BPM of 13.7 in his senior year, while Edey had a career BPM of 14.4 while he was younger. Edey’s sophomore year was comparable to Garza’s senior year when he was considered the country's best player (College Player of the Year in 2020-2021). 

This leads me to believe Edey will have a productive NBA career. The other problem is the idea of athleticism, which is part of the reason potential is misevaluated. I published an article (Challenging the Traditional Notions of Athleticism) about this topic a few years back, but I want to dive even deeper. I made a lot of good points about how people were off about the traditional notion of athleticism. Still, I never realized we would get dramatically closer to measuring these more functional traits. “James (Harden)' traditional performance numbers – the ones that get discussed at the Combine – were unremarkable. James’ metrics related to Deceleration, however, were beyond elite. Eccentric Force generated during the CMJ, the rate at which he generated this force, the kinematic values associated with his eccentric phase movement, and forces produced during change of direction testing…all broke our models. James has some elite athletic tools, they just aren’t the ones that we were accustomed to looking for. Excelling in Deceleration meant James had the perfect set of tools for step-backs and drawing fouls – traits that weren’t easily studied without detailed biomechanical data.” - P3 Sports Science. In the same article, P3 goes on to talk about how Doncic had a similar but better profile than Harden yet people still called him a poor athlete. On a tangent, Doncic has one of the best production profiles for any draft prospect ever and had a similar athletic profile to Harden, yet he wasn’t the first pick. I believe they also noted Anthony Edwards is an elite athlete, although he has a different athletic profile from Doncic and Harden. Edwards was one of the rare “bad” college players I believed in because of his functional athleticism and age. Functional strength is also an interesting concept because the idea of someone bench pressing and being strong is inaccurate. Functional strength is more about momentum and using your weight and core strength to carve out position for yourself. Jalen Brunson has amazing core strength and uses his weight to help himself create easy shots. To end this section, I believe NBA teams will have athletic models shortly, similar to how the Rams had advanced models that identified Puka Nacua.

Unfortunately, I can’t measure these functional traits, but I have done testing with NBA combine metrics and college stats. After performing very small levels of testing, I wrote. 

“All of this comes with the goal of projecting future performance for players and teams. The first mystery is picking players in the draft, a crapshoot. For as long as the draft has been around, teams have had a tough time identifying the best player. Teams have still struggled even with improved data, computer analysis, and scouts. The first two important factors are production and sample size. Although players improve, the larger the sample size the more accurate the model. I predict the most accurate models would use the largest simple (all seasons available) to predict NBA performance. One example of this is Davion Mitchell who shot a ridiculous 44.7% from three as a senior. People thought Mitchell was a 44.7% 3P shooter although he had shot 31.2% from three in the previous two seasons. He had also only shot 65.7% from the FT line which doesn’t line up with 44.7% from three. Mitchell has shot 31.7% from three so far while having 87.3% of his threes assisted. People seem to underestimate production and that is why many players get undervalued. Production needs to be adjusted for situation and competition, but in the end, production should be the main factor when drafting. Some players have flaws hidden by a scheme in ways that are not possible in the NBA, but production is still the main indicator. 

The other important factor is development possibilities which are hard to predict. The main factors are age, limitations, and time playing basketball. Age is the main factor as the younger you are the more room you have for more development. This is probably not a linear curve as draft age*draft age might have a stronger relationship with NBA success, although there may be other issues with this. The second factor is athletic limitations as a defender and creator. To some extent, players max out physically but it’s also a lack of training functional athleticism just has flexibility, deceleration, and lateral agility. Lastly, players who have played basketball for less time have more room for development. Predicting development is impossible but age is the most important factor when thinking about development. The last part of predicting success is athleticism. In the past, I’ve written about athleticism and what athleticism means but how could it be measured? In short, you would need advanced tracking data to measure shin angle and other functional athletic traits. 

The last question is what stats correlate most with NBA success. Through a small amount of data collection, I have a few theories. ORBs, AST/TO adjusted for volume, and STKS(Steals+Blocks)/FL would all be good predictors because they would do a solid job predicting anticipation, pattern recognition, and understanding of the game. STK/FL is an underrated stat that is similar to AST/TO in some ways but is in no way perfect. UARM is another important statistic that likely does a good job predicting creation abilities and the ability to get paint touches. Another interesting idea is using EYBL/AAU stats because of the similarities between the type of game played in the EYBL and the NBA. It also increases the data size which always helps. The first basic model we can use is (1/Draft Age^2)*(WS/40)*(BPM). To make this easier to read we can multiply by 4284.9 to make the numbers easier to read so we come away the formula of  (1/Draft Age^2)*(WS/40)*(BPM)*(4284.9). This is flawed because WS/40 and BPM are not perfect but they are at least solid indicators. Advanced stats have started to come out tracking line-up data which would make it easier to make a better all-in-one stat. To predict three-point shooting, I think range, volume, and creation levels are underrated indicators of future success. Players who take a low volume of threes (on a per-possession basis) are not always safe bets.” - My prior work. 

Age was surprisingly linear although the highest correlation I could find was between 1/(Draft Age)^3. The correlation between that number and NBA win shares was 0.415, which is high for a simple stat like age. Age and NBA win shares have a correlation of -0.410, which is slightly lower, but it makes sense. The younger the player is, the more room for development. The correlation between 1/Draft Age was 0.414, and the 1/(Draft Age)^2 was closer to 0.415 but not quite there. Once you go past the third power, 1/(Draft Age)^4 was slightly lower than 0.415 as well. There was a correlation of -0.168 between minutes played and NBA win shares, likely explained by the correlation between age and minutes played. This can be said for every “total” college stat because older players will rack up more points, assists, etc. The correlation for assists/turnovers is surprisingly low, but high-volume creators will likely have lower ast/to. Unfortunately, I could only use per 40 minutes rather than per 100 possessions. Turnovers per 40 have a correlation of -0.149 with NBA success, which is surprisingly high, in my opinion. Turnovers are bad, but players with the ball in their hands will turn the ball over more. The correlation between NBA success and 1/(Turnovers)^2 is 0.191, which is interesting, but I guess it proves me wrong in the sense of turnovers being bad. In the 2024 NBA season, turnovers and offensive ratings have a correlation of -0.622, a relatively high negative correlation. This is interesting because defensive rating likely positively correlates with live ball turnovers because these often lead to easy looks. Rebounding had a decent correlation of 0.154 with success, but offensive rebounding had an extremely high correlation with success, although I didn’t have a ton of data. I assume the correlation is high but not quite as high as the correlation I got. There was a slightly negative correlation between ORB and NBA offensive rating which is shocking but is likely explained by the negative correlation between ORB and 3P% of -0.289. Teams with bigger players likely rebound the ball better and shoot worse. On a personal level offensive rebounding is likely a decent indicator of basketball feel. FTA and FTs have negative correlations with success which is surprising, and I don't have a theory as to why. TS% has a correlation of 0.141 with success which seems to checkout because it doesn’t account for self creation and volume. EFG% correlated more with success than TS% which makes sense after you know that success and FTA have a negative correlation. Regarding correlation with NBA success, the big NBA combine metrics are height (0.217) and standing vert (.209). Height correlating with NBA success makes sense. I am not surprised that standing vert has a positive correlation, although I am surprised at how the correlation compares to other metrics. Some of these correlations might be skewed by problems with how NBA win shares are judged, but this is all still interesting information. After this work, I developed a model to predict NBA win shares, which is a solid start. The model considers many factors, including age, production, and NBA combine metrics. In the draft classes between 2000-2002, the top seven qualifying prospects were Drew Gooden, Shane Battier, Brendan Haywood, Chris Wilcox, Gilbert Arenas, and Joe Johnson. 

I plan on using the model for the upcoming draft, but I will make significant updates in the upcoming year. I will start predicting Estimated Wins as I believe it is a better stat for reasons discussed later. Ultimately, I want to track RAPM wins, which is the best stat when evaluating value. I will also try to find stats like unassisted makes and rim makes to better predict NBA success. In terms of college production, a player like Reed Sheppard compares fairly similarly to Michael Redd, Etan Thomas, and Mike Dunleavy, while Isaiah Collier compares to Chris Wilcox, Dan Langhi, and DerMarr Johnson. I believe Collier will test much better, but it will be interesting to see how they both score in the end. I want to highlight a player who will likely be on the wrong side of my model. The first is Kyshawn George. If you read scouting reports, the word potential repeatedly, but why? George is the age of an older sophomore but only had a 3.6 BPM, 2.9 BPR, and a 0.7 RAPM. Alex Karaban is approximately one year older but had a 9.2 BPM, 7.3 BPR, and 6.0 RAPM. There is more to basketball than these numbers, but they paint some of the picture. You would have difficulty finding Karaban ranked, yet George is a “magical prospect”. These low indicators and older age for his class make it harder to buy the upside. When drafting, you want to draft the best player available rather than fit. You acquire the pieces and then fit them together as you see who is good and what works together..

Once you draft a player, you still have to be able to analyze players in the NBA where there is much more public data. Let’s start by analyzing different scoring metrics. Points per game is the amount of total points scored divided by the number of games played. Points per 40 possessions is total points multiplied by 40 divided by minutes played and points per 100 possessions is total points multiplied by 100 divided by total offensive possessions. Points per possession is the best way of comparing players because it takes out advantages in minutes played and pace. Field goal% is simply the amount of shots made divided by the amount of shots taken. “About a year ago I came up with a story. A (sells his lemonade for $1.00) and B (sells lemonade for $1.50) start lemonade stands. Let's look at them. A and B set up right next to each other where 100 customers will see each of them. A sells 51 cups of lemonade while B only sells 34 cups. 

A: 50% “Conversion Rate”, $0.50 per customer 

B: 34% “Conversion Rate”, $0.51 per customer 

Let us make this a bit more complex and add in donations. We’ll call them donation attempts where the seller can receive $0.50, and every 2.25 donations attempt is worth one customer walking by. C (sells lemonade for $1.00) decides he wants to compete and makes his own stand. C gets 45 donation attempts which is equivalent to the chance of selling to 20 customers. He successfully gets 36 of these donations. With the rest of his 80 customers, he sells 40 cups of lemonade.

A: 50% “Conversion Rate”, $0.50 per customer 

B: 34% “Conversion Rate”, $0.51 per customer 

C: 50% “Conversion Rate”, 80% “Donation Attempt Rate”, $0.58 per customer

“Conversion Rate '' is the same as FG%, and per customer is the idea of TS% and EFG%. 33.33% from 3 is just about the same as 50.00% from 2. Top mid-range shooters will shoot 55% if they’re elite while that is the same points per possession as 36.67% from 3, which is only a little bit above average. FG% is simply how often a shot will go in with no context as to how much that shot is worth. EFG% ((FGM + 3PM x 0.5)/FGA) makes 3PM 1.5x more valuable than a 2PM. TS% (PTS/(2(FGA + FTA x 0.44)) gives a value to a FTA and now you have a better picture of scoring efficiency. The goal of basketball is to maximize points per possession! Getting FTA, in general, is very efficient so drawing fouls leads to good offense. James Harden, from 2014-2015 to 2019-2020, had a 44.2 FG%, 36.1 3P%, and 52.9 EFG%. In all honesty, these seem similar to the league average, but once we dig a bit deeper we see how far from the truth this is. Harden’s TS% would constantly be 11-13% (TS+ ranged from 111-113) higher than the league average. Harden represented C in the lemonade example. On the opposite end, we see Andre Drummond. Throughout the same time span, he had a 54.0 FG% (FG+ ranged from 115-116, which means FG% was 15-16% above league average) which is well above league average yet only has a 54.0 TS% (TS+ ranged from 92-100, which means TS% was between league average and 8% below league average). Drummond is meant to represent A. Here is another example. 

Player X: 44.2 FG%, 36.8 3P%, 87.9 FT%

Player Y: 50.5 FG%, 42.6 3P%, 92.8 FT%

Player Z: 47.2 FG%, 43.7 3P%, 91.6 FT%

How would you rank these players on pure efficiency? Y, Z, X? These numbers would be given in a basic box score but not always give the complete picture of efficiency. The right answer is Z (64.1 TS%), X (61.6 TS%), and Y (61.4 TS%). Just a small difference but FG% is very misleading if it is used to assess scoring efficiency, especially when comparing players with different shot diets.” - My prior work. 

For the most part I was spot on with the statistical analysis and the next part, but it is the final part where my views have changed. “It is clear that TS% and PTS are important and much better indicators of efficiency than FG% and other basic stats. While TS% and PTS are important in judging a player’s scoring value, we still face the flaw of lacking context. If we just used these two statistics, our evaluation of a player would be severely lacking and we wouldn’t get the complete picture. 

To begin with, we have to understand not every two points are created equally. Every time two points are recorded we have no clue how those two points were scored. Just because you scored two points doesn’t mean you created the same value as somebody else’s two points. While this sounds weird in theory, when you break it down it’s not. If a player scores an assisted layup where he did nothing except finish the layup, it isn’t as valuable as somebody else creating their own layup. In the first situation, the player who passed the ball is likely doing most of the work, although that isn’t always the case. In terms of future projection, scoring a higher percentage of points on unassisted looks will lead to more attention and therefore more open looks for teammates. We also have to access the talent around a certain player to determine their value. A player’s teammates are going to play a major role in scoring volume and efficiency. Surrounding a player with elite horizontal spacing, vertical spacing, and advantage creation will make it much easier for a player to score. On the flip side, playing with high talent might also limit the volume a player will score with. 

In 2021-2022, Shai Gilgeous-Alexander had a 55.7 TS% which equated to a 98 TS+. When we dig deeper we see that Gilgeous-Alexander produced more value as a scorer than somebody would think looking for multiple reasons. To begin with, the talent around him was poor as his second-best offensive player was either a rookie Josh Giddey, or Lu Dort. The team also only shot 32.3% from three with little vertical spacing. Another important piece of context was the level of creation of Gilgeous-Alexander’s creation levels. Only 13.7% of Gilgeous-Alexander’s 2P FGs were assisted and 30.3% of his 3P FGs were assisted. Most impressively, he had a 27.7 Iso Frequency% (third highest in the league). A large number of these points came not only unassisted but from a standstill without any previous advantage. He was creating a ridiculously high % of shots for himself making his scoring more valuable. The ability to create shots for yourself is not only important as a scorer but also as a playmaker. Gilgeous-Alexander was also able to score from all four levels (rim, short mid-range, long mid-range, three) which put even more pressure on the defense. Although you typically don’t want your team to take mid-range shots, it is a valuable skill to have for players who create their own offense frequently. Specifically, his ability to consistently get into and score from the paint (led the league in drives per game and second in points per game off drives) was valuable for the offense as a whole and helped create open looks and advantages for teammates. 

In 2011-2012, Rajon Rondo averaged 11.7 assists per game, a very high number, but how valuable was Rondo as a playmaker and passer? The answer is not as valuable as one may assume. When someone says assists, it has no set definition as not every assist is created equally. Just like when someone says scoring two points, it gives no definition of how much creating the player did. To truly answer how valuable Rondo was, we have to determine what playmaking is. We can go with the simple definition of creating plays for others, but what creates plays for others? Passing does create plays for others but in specific cases. If James Harden beat his man off the dribble, drew help, and set up Clint Capela for a layup, Harden would get credited for an assist and Capela would get credited for a layup. Harden is creating the two points, but nobody would get that from the box score. How valuable is Capela in this play? He could be replaced by pretty much anyone in the league and the two points would still be scored. Say Capela missed the layup, nobody would know Harden created a great look, probably around 1.99 points on average, because it didn’t show up in the box score. On the other hand, say Rondo dribbled the ball up while his defender sagged off, and passed the ball to Paul Pierce on the wing who took a highly contested one-dribble pullup. Pierce could make the shot and Rondo could still be credited with an assist. Rondo could potentially create negative value as his defender sags off and helps contest the shot but still gets credited with the assist. In another situation, Stephen Curry comes off an off-ball screen, and two defenders reacting to Curry’s shooting, follow him while Kevon Looney slips to the rim and receives a pass from Jordan Poole. Curry is creating the value but will receive no credit in the box score. We have to consider Rondo’s value off the ball. Does he cut at the right time? Can he make open catch and shoot threes? In Rondo’s case, his weak three-point shooting limits his playmaking. His lack of shooting and vertical spacing threat let defenders provide more help. While it may not be possible to quantify how valuable one’s playmaking is, it is clear Rondo’s assists don’t line up with his playmaking value. His passing value is different. Passing is a part of playmaking and one that Rondo does well. Rondo had the ability to find holes that others couldn’t find. 

After understanding why assists are flawed statistics, we start to understand that we can create better stats by adjusting for teammates. The Flarescreen defines Modified Assists as “A player's assists adjusted for league average conversion rates and weighted for 3-point makes being statistically 1.5x more valuable than 2-point makes, using a model similar to eFG.” This is an improved version of assists and does a better job of adjusting for teammates and shot location. Another interesting statistic is Box Creation which tries to add in points and three-point shooting to quantify other aspects of playmaking. So what are all the aspects of playmaking? In my opinion playmaking as a whole would include passing (quality of passes, ability to find high-value shots), scoring (volume, efficiency, versatility, creation), finishing (pressure on the rim, proficiency at the rim), shooting (range, accuracy, volume), and off-ball ability (moving off the ball effectively). Playmaking is built on these pillars and great playmakers are players that can create shots for themselves and leverage these into open looks for others. Another aspect of offense is being able to finish advantages. If a play is created, being able to maintain (finding the open player) or finishing the advantage (making the shot) is also valuable. 

Using Box Creation as a reference, I came up with McGuire Offensive Efficiency Rating. This is not perfect, but uses (scoring + estimation of shots created)/(true shooting attempts + turnovers) to estimate offensive efficiency. Instead of using AST, we use ModAST + SecAST to better to quantify passing value. I also fiddled with weighting unassisted makes more than unassisted makes. We arrived with this work in progress.

(PTS + (1.2 x ((ModAST + SecAST) x 0.1843 + (PTS + TOV) x 0.0969 - 2.3021 (UA3PA x 1.2 + A3PA (3P%)) + 0.0582 x (ModAST+ SecAST x (PTS + TOV) x (UA3PA x 1.2 + A3PA (3P%)) - 1.1942))/(FGA + FTA x (0.44) + TOV)).” - My prior work. 

Again this is still a good analysis of offense and the problem with offense statistics. An interesting note is the correlation between assists and offensive rating was 0.412. I have consistently gone back to the correlation between different offensive stats and offensive rating, but I wonder what changes in the playoffs. This is an interesting question in general as teams strive to win in the playoffs. Is isolation more important in the playoffs? Does experience really matter? Do smaller teams struggle? These are all interesting questions I would love to answer in the coming years. 

In Baseball, a player's value can be put into one stat with 99.9% accuracy. Basketball is a little bit harder because there is a lot more going on, but what else goes on besides scoring and playmaking? Can it all be quantified in one number? 

The first mainstream all-in-one stat was Player Efficiency Rating (PER). While PER is a solid stat, it is far from perfect and is very flawed. It weighs some basic box score stats and some slightly more advanced stats such as Defensive Rebound% (DRB%). On a team level, defensive rebounds are valuable but similar to points and assists, just because the player records the rebound doesn’t mean he brought the value on the specific play. If Steven Adams and the rest of the 2017 Thunder box out their assignments, but Russell Westbrook records the rebound, you would assume Westbrook created the value. This doesn’t depict the full picture as Westbrook could be replaced in this situation. Contested Rebounds (CRBs), ORB (ORBs), and on/off rebounding numbers do a better job showing rebounding value compared to DRB and DRB%. Another major issue is the weighting of DRB%. We have already established there are better stats to use, but the general weighting of rebounding is too high. In 2021-2022, Montrezl Harrell, Hassan Whiteside, and Javale McGee ranked inside the top 20, and Daniel Gafford, Clint Capela, Daniel Gafford, Andre Drummond, Jakob Poetl, and Jonas Valanciunius ranked within the top 40 alongside James Harden and Shai Gilgeous-Alexander. This could also potentially be influenced by an overweighting of FG% which we have already established is a flawed stat that favors big men who take a high volume of unassisted layups. Another flaw is an overweighting of Blocks and Steals. Creating turnovers are valuable but they don’t fully quantify defense. The goal of a rim protector is to force low-percentage shots or prevent shots. DFG% > 6FT is a starting point, but to fully quantify rim protection we would need to consider the full value a player creates. This would be hard to do as you would have to consider a wide number of variables such as teammates and defensive scheme. To start adding STL, BLK, DFL, and CHRG would give a solid idea of understanding “defensive playmaking.” Combining this with on/off defense rating, help defense, matchup difficulty, and fouls would be a good continuation. Regardless, defense is hard to judge. Other problems with PER include an overweighting of PTS and ASTS which are flawed stats as stated before. VORP and WS are similar to PER in the way they are reliant on general counting stats, so they face the same general problems. 

The new generation of stats started with BPM and Plus-Minus. BPM, a stat similar to PER, tries to estimate how much value (in points) a player adds per 100 possessions. The stat is flawed but the way of quantifying players is an interesting start. Plus-minus is a basic stat that measures how much your team outscores the opponent when you’re on the floor. It doesn’t factor in teammates, opponents, and other potential factors. One variation from Plus-Minus is Net Plus-Minus, which measures how good your team is when you’re on the court compared to how good your team is when you’re off the court. If Draymond Green played most of his minutes with Stephen Curry and Klay Thompson his Net Plus-Minus might not fully represent his value. After BPM and Plus-Minus, we have the stats that are currently used such as EPM and RAPTOR which combine more advanced versions of Net Plus-Minus and BPM. They factor in different types of tracking data and plus-minus adjusted for opponents and teammates. These stats are a major step up from any previous stat, but they aren’t perfect.” - My prior work. 

RAPM

After all this discussion, is there a single stat that best determines value? Yes. Regularized Adjusted Plus Minus (RAPM). Ben Taylor introduced me to the interesting concept of impact vs. production. He uses two examples, the first being a player we will call ABC. ABC was among the most productive players in the NBA based on points, assists, and other basic score stats. He had a great BPM, PER, etc., but it seemed like his team was always okay when he was off the floor, so was he impactful? The other example used was Bill Russell and Wilt Chamberlain. Taylor thought the optimal way to measure impact would be to ignore the box score because of preconceived biases, and the more I thought about it, the more I loved this idea. He claimed Russell was more impactful than Chamberlin based on Wins With You Vs. Wins Without You which led to RAPM. So what is RAPM? It looks at three factors: success with you on the floor, your teammates, and your opponents. In a long and messy linear programming equation, it assigns each player an estimate as to how valuable they were to their team. In short sample sizes, this is too noisy, but over a full season and multiple seasons, this is the best way to look at value. 

For the rest of this article, I will refer to two RAPM estimates. The first is RAPM 2023-2024, which I will refer to as R. The second is Time Decay RAPM, which also uses RAPM from other seasons, with the past season having the most weight, which I will call TDR. First, we’ll highlight several highly overrated guys, discuss why, and then do the opposite. The first player we’ll highlight is… Paolo Banchero. He is often considered as a top-25 player in the NBA, but I am here to tell you he is nowhere near that good. His TDR was 32.3%, and his R was in the 77.6%. He is a low-end NBA starter, even if we look at the more optimistic estimate. Why? My best guess is due to his overrated offensive output. He scores at an inefficient rate that is easily replaced by ball movement and less isolation play, while his defense is average at best. His off-ball contribution, or lack thereof, also contributes to these numbers. Players who score below 96-97 TS+ rarely contribute positively offensively, especially when holding the ball for long periods. Cade Cunningham is considered as a top 50 player but is again nowhere near that ranking with a TDR in the 35.8% and an R in the 65.7%. Like Paolo, his offensive output is relatively replaceable because of his 94 TS+. Other overrated players include Cam Thomas (5.5% in TDR, 67.1% in R) and Brandon Ingram (73.4% in TDR, 82.0% in R). To summarize, mid-low efficiency on a lot of isolations, not great playmaking, poor off-ball offense, and questionable defense usually equate to an overrated player. These guys tend to be floor raisers rather than ceiling raisers because isolation scoring can take a bad team to average but not a good team to great.

On the flip side, Franz Wagner (98.5% in TDR, 95.3% in R) and Derrick White (98.4% in TDR, 96.6% in R) are some of the more underrated players in the game. While RAPM is a perfect concept, there are still flaws over small amounts of minutes. 82 games still isn’t perfect, but it is very good. The data has led me to some unpopular opinions such as Derrick White and Franz Wagner being better than Paolo, Fred VanVleet being better than Jamal Murray, and Donovan Mitchell being better than Devin Booker and Anthony Edwards. On the historic side, Manu Ginobili being underrated and Draymond Green being more important to the Warriors than Klay Thompson. RAPM is estimating your impact in points per 100 possessions, so players within one to two points of each other, there is room for debate, but otherwise there is usually not room for debate unless there are special circumstances. Using Time Decay RAPM which weights current performance more heavily, but still uses a small portion of past years is a good aid to RAPM.