RAPTOR, which stands for Robust Algorithm (using) Player Tracking (and) On/Off Ratings, is FiveThirtyEight’s new NBA statistic. We’re pretty excited about it. In addition to being a statistic that we bake in house, RAPTOR fulfills two long-standing goals of ours:
- First, we wanted to create a publicly available statistic that takes advantage of modern NBA data, specifically player tracking and play-by-play data that isn’t available in traditional box scores.
- Second, and relatedly, we wanted a statistic that better reflects how modern NBA teams actually evaluate players. NBA teams highly value floor spacing, defense and shot creation, and they place relatively little value on traditional big-man skills. RAPTOR likewise values these things — not because we made any deliberate attempt to design the system that way but because the importance of those skills emerges naturally from the data. RAPTOR thinks ball-dominant players such as James Harden and Steph Curry are phenomenally good. It highly values two-way wings such as Kawhi Leonard and Paul George. It can have a love-hate relationship with centers, who are sometimes overvalued in other statistical systems. But it appreciates modern centers such as Nikola Jokić and Joel Embiid, as well as defensive stalwarts like Rudy Gobert.
Before we get into more detail about RAPTOR, a few “getting to know you” basics about it:
- Like Box Plus/Minus (BPM) and Real Plus Minus (RPM), RAPTOR is a plus-minus statistic that measures the number of points a player contributes to his team’s offense and defense per 100 possessions, relative to a league-average player. For instance, a player with an offensive RAPTOR rating of +2.1 boosts his team’s performance by 2.1 points per 100 offensive possessions while he is on the floor. Likewise, a player with a defensive RAPTOR of +3.4 would improve his team’s defensive performance by 3.4 points per 100 possessions while he’s on the court.
- Plus-minus statistics have certain inherent limitations, and RAPTOR is subject to those, too. Namely, these statistics assume that player performance is largely linear and additive, that is, that you can roughly add up the ratings from individual players to project team performance. In other words, RAPTOR does not account for coaching, systems or synergies between teammates.
- RAPTOR consists of two major components that are blended together to rate players: a “box” (as in “box score”) component, which uses individual statistics (including statistics derived from player tracking and play-by-play data), and an “on-off” component, which evaluates a team’s performance when the player and various combinations of his teammates are on or off the floor.
- When applied to past data — for instance, in evaluating who the best players were in the 2018-19 season — RAPTOR is a descriptive statistic. Descriptive RAPTOR is based solely on a player’s on-court performance and the performance of the player’s teammates, as described above. It does not use priors based on a player’s height, weight, age or any other factor.
- However, RAPTOR can also be used to make team and player predictions, and indeed our NBA predictions are now fueled by RAPTOR. (We are retiring the CARMELO brand name from our previous projection system, although much of the code for RAPTOR projections is borrowed from CARMELO.) RAPTOR-driven predictions do use qualities such as height, age and draft position, and even whether a player recently appeared on an All-NBA team — that data improves the performance of the predictive measure. Predictions also weight variables slightly differently than descriptive RAPTOR does, as certain statistics are more subject to luck than others. We refer to this predictive version of RAPTOR as PREDATOR (PREDictive rApTOR).
- RAPTOR is based exclusively on publicly available data. There are other player-tracking statistics we believe could be highly helpful to RAPTOR, especially more detailed measures of on-ball defense, so we hope to be able to revisit RAPTOR as additional data becomes available.
- RAPTOR’s name (in addition to being a whimsical backronym in the tradition of CARMELO and DRAYMOND) honors the 2018-19 Toronto Raptors, which FiveThirtyEight’s previous projection system correctly predicted had an edge over the Golden State Warriors (even though we didn’t fully believe the projection ourselves at the time).
- The full-fledged version of RAPTOR is available for the 2013-14 season onward, as that’s when the NBA’s player-tracking data came on line. We also have a historical version of RAPTOR called Approximate RAPTOR dating back to 1976-1977, the first season after the ABA-NBA merger, but that uses a far more limited range of data.
RAPTOR ratings for players with at least 1,000 minutes played in a season since 2013-14 can be found in the table below. As you can see, RAPTOR generally loves perimeter players and wings, such as Curry, Harden, Leonard and Chris Paul, although some frontcourt players like Jokic, Anthony Davis and Draymond Green are also rated highly by the system. For more detail on past RAPTORs, including the breakdown of box and on-off components, you can download files that list the regular season and playoffs separately, or a version that combines a player’s appearances over the course of the entire season into one file.
RAPTOR ❤️s Steph, Harden, CP3 and Kawhi
RAPTOR ratings for players with at least 1,000 minutes played, regular season and playoffs combined
I’m not going to promise that it’s beach reading, but it does contain what we hope are some interesting insights about the NBA, plus more technical details.
Box RAPTOR
RAPTOR in many ways takes its inspiration from BPM, which was designed by Daniel Myers. BPM was designed by fitting a regression model for which the inputs are various traditional statistics (e.g., points, rebounds, etc.) and the dependant variable is long-term Real Adjusted Plus Minus (RAPM). The “box” component of RAPTOR does something similar, only using player-tracking and play-by-play data in addition to traditional statistics.
What is RAPM? It’s a measure of how many points a player contributes per 100 possessions based on his team’s performance when he’s on and off the floor, accounting for the quality of his teammates and his opponents. Adjusting for teammate and opponent strength can be tricky business, however. Mediocre players on great teams, such as JaVale McGee on the 2016-17 and 2017-18 Golden State Warriors, can have strong raw offensive and defensive ratings because they play with excellent teammates; it is obviously necessary to adjust for this when calculating McGee’s contribution to the team. Players with small sample sizes and rarely used lineup combinations can also create problems, so RAPM employs various techniques to regress their performance toward the mean. So in theory, RAPM is a truly comprehensive statistic, measuring all the tangible and intangible ways in which a player contributes to his team’s bottom line. It should also be an unbiased measure, not overvaluing or undervaluing any particular type of skill relative to its actual value on the court.
But in practice, RAPM can be very noisy, taking several seasons to stabilize. It’s also fairly computationally intensive and can be sensitive to relatively subtle choices about exactly how it’s calculated. For these reasons, RAPM is not a great measure for use in a projection system, when our data needs are more time sensitive — e.g., if we want to see how much a player such as De’Aaron Fox improves from one season to the next.
The insight behind BPM — and now RAPTOR — is that we can use other statistics that stabilize much more quickly than RAPM to approximate long-term RAPM. More specifically, we fit a series of regression coefficients using a six-year dataset of RAPM as provided to us by Ryan Davis, with the six years matching the six seasons (2013-14 through 2018-19) for which player tracking data is available. (We made a few adjustments to RAPM from Davis’s version to make it more appropriate for our specific needs.)
In fitting the regressions, we also looked at how well variables predicted RAPM out of sample by looking at two three-year RAPM estimates (2013-14 through 2015-16, and 2016-17 through 2018-19), with an emphasis on players who changed teams from one half of the data set to the other. If a certain variable predicted RAPM well in the in-sample, six-year regression, but not in the out-of-sample, three-year regressions, that’s generally a sign that it reflects luck rather than skill or that it’s too noisy to provide for a reliable indicator of player value. For instance, data on how many 3-pointers opponents make when a player is the nearest defender is highly predictive of in-sample RAPM but not at all predictive of out-of-sample RAPM. Thus, variables like this were excluded from RAPTOR.
In addition, we used our basketball knowledge to inform our choices of parameters. For instance, 3-point attempts are a good proxy for creating floor spacing or having “gravity” — that is, drawing defenders toward you and therefore giving your teammates more open scoring opportunities. In our various regression specifications, it was ambiguous whether a better statistical fit was produced by using all 3-point attempts or instead weighting 3-point attempts based on how closely contested they were. In situations like these, we went with what made more “basketball sense”: in this case, that players who have a lot of contested threes are the ones who do more to create space. We also separately fit models for offensive and defensive RAPTORs, instead of combining them. Thus, for example, offensive rebounds contribute to a player’s offensive RAPTOR and defensive rebounds to a player’s defensive RAPTOR, rather than blurring them together. So while the regression specifications that follow might seem complex, there was quite a lot of basketball thinking behind them; it wasn’t just a matter of coming up with the best statistical fit.
Box RAPTOR Offense
The variables used in offensive “box” RAPTOR follow below. Although the list includes a few statistics, most of them fall into one of four major categories: scoring and usage; passing; rebounding; and space creation. Before being used in the regression, all variables are adjusted relative to league average. In addition, stats from the playoffs were adjusted to reflect the tougher competition in the postseason. Here are the categories in more detail:
Measures of scoring and usage
Points: This is just what it sounds like. Good ol’ points scored are in fact the highest-weighted category in offensive RAPTOR:
Usage rate: A “usage” is any shooting attempt, turnover or foul drawn that results in free throws, except for fouls (e.g., flagrant fouls and clear path fouls) that result in the team getting the ball back after the free-throw attempt. Heaves (shots from beyond half-court, which are almost always taken out of desperation at the end of the quarter) count as only a small fraction of a possession. Although this is complicated by the fact that RAPTOR contains a number of variables related to shooting, usage and scoring, overall it is calibrated such that players who score at average efficiency tend to improve their RAPTORs by doing so, as opposed to not taking any shots at all.
Time of possession: The value of a possession also decreases as time ticks off the shot clock. Thus, merely possessing the ball negatively predicts offensive RAPM, holding other factors constant.
Assisted field goals: In addition, assisted field goals are less valuable than unassisted ones. In some sense, this is a matter of basic accounting: If you’re giving players credit for assists (as RAPTOR does), you probably have to take some credit away from the player who benefits from the assist. More specifically, we find that the deduction for an assisted shot should be proportional to the expected value of the shot attempt. RAPTOR recognizes seven types of shots based on their location on the floor:
RAPTOR shot categories
While all players who rely heavily on assisted baskets are penalized by this statistic, it has a particularly large effect on players such as DeAndre Jordan who camp out at the basket and depend on assisted dunks. In fact, it’s essential to account for these players’ assisted field goals because they’re strongly correlated with other types of statistics, especially offensive rebounds. Failure to account for assisted field goals will bias the value of offensive rebounds downward, and some advanced stats such as RPM very likely understate the importance of offensive rebounds for this reason.
Measures of passing
Enhanced assists: Likewise, the value of an assist in RAPTOR is proportional to the expected value of the resulting shot. Assists on dunks and corner threes are considerably more valuable than assists on midrange jumpers. In addition, we give partial credit for what the NBA calls “free throw assists”: passes that result in a teammate drawing a shooting foul. However, we find that there isn’t much value in what the NBA calls “potential assists” that don’t result in baskets or free-throw attempts. We do, however, give players credit for …
Net passes: The NBA also keeps track of the number of passes a player makes and receives during the game, and a positive passing differential is associated with a higher RAPM in and out of sample.
Measures of rebounding
Enhanced offensive rebounds: Offensive rebounds are a tricky category. On the one hand, the value of an offensive rebound is intrinsically quite high: A team not only gets a new life on its possession after an offensive board, but it is also often in a premium position to score via a putback opportunity. (Although it depends on how the rebound is secured, the average value of a possession after an offensive rebound is around 1.2 points.) On the other hand, a lot of rebounding has to do with being in the right place at the right time. Rebounding can involve a fair amount of luck, and loitering near the basket hoping for rebounds can have negative consequences for a team’s spacing. But in general, offensive rebounds are becoming more valuable as offensive rebounding rates get lower, having fallen from 33 percent in the mid-1980s to about 23 percent in today’s NBA.
For both offensive and defensive rebounds, RAPTOR makes various fixes to the rebound statistics. Essentially, our goal is to calculate how much a rebound affects the expected value of a possession. For instance, after a missed shot, the expected value of a possession was around 0.28 points in 2018-2019 (a 23 percent chance of an offensive rebound times an average of 1.2 points scored conditional on securing the rebound). A defensive rebound would reduce this value to zero and end the possession; an offensive rebound would increase it to 1.2 points.
The NBA’s player tracking data distinguishes between contested and uncontested rebounds. Contested rebounds are more valuable, although this makes less of a difference for offensive than defensive rebounds. The intuition behind this is as follows: Because 77 percent of rebounds are defensive rebounds, only defensive rebounds on which the offense has a serious shot at the ball (i.e., contested rebounds) have all that much value for a defensive player since his team would probably wind up with the ball anyway. On the other hand, in today’s NBA, any offensive rebound is rare, and therefore any offensive rebound is fairly valuable. Thus, players provide value through contested defensive rebounds (but not much through uncontested ones) and through offensive rebounds of any kind.
RAPTOR also evaluates the location of the shot preceding the rebound, as some shots are much more likely to produce offensive rebounds than others. For instance, missed free throws produce offensive rebounds only about 10 percent of the time, so defensive rebounds after free throws have very little value since the remaining expected value of a possession is already close to zero. Layups produce high rates of offensive rebounds, by contrast — so defensive rebounds are worth more in this case.
There are also a couple of more technical fixes to the rebounding stats:
- If a loose ball foul occurs on the rebound, but the rebound is not credited to a particular player, we give the player who drew the foul credit for a contested rebound. We also give the player who committed the foul blame for having yielded a rebound to his opponent (see below).
- The NBA also keeps track of deferred rebounds: when a player has a chance at the rebound but passes it up to a teammate. Empirically, the value of deferred rebounds is quite high, perhaps because it indicates overall selfless play. (Think about Steven Adams giving up rebounds to Russell Westbrook, for instance.) For simplicity’s sake, however, RAPTOR simply treats a deferred rebound as being as valuable as a regular rebound. In addition, we reduce the overall number of rebounds proportionately for all players on the team based on the number of deferred rebounds for the team.
Team offensive rebounds on missed shots: We also find that the shooter has a fair amount of influence on a team’s offensive rebound rate on his missed shots. As I mentioned, some types of shots produce more offensive boards than others; players who get to the rim for floaters and layups can produce particularly high offensive rebounding rates, for instance (see table below). In addition, big men who play away from the basket (Brook Lopez, for example) can cause rebounding problems because there’s often no offensive player in prime position to secure the rebound if they’re playing out on the perimeter. We give slightly more credit to rebounds that occur (i) in bounds and (ii) not after blocked shots, since these are associated with a higher expected value for the remainder of the possession.
Some shots produce far more offensive rebounds
Positional opponents’ defensive rebounds: RAPTOR attempts to figure out which player was matched up with which opponent on a given possession based on their positions as listed in our database. The idea is that centers are matched up against centers, power forwards against power forwards, and so forth. In practice, however, there’s rarely a clean one-to-one correspondence between players at different positions. Instead, in assigning players to positions for our depth charts, we deliberately draw from multiple sources to make most players eligible at multiple positions. The upshot of this is that in RAPTOR, player assignments are probabilistic, which likely makes sense anyway given the amount of switching in today’s NBA.
Despite this being a relatively noisy process, there is some predictive power (including in out-of-sample regressions) in seeing how many points and rebounds a player’s positional matchups secure. Opposing bigs get notably fewer defensive rebounds when playing against Embiid than against most other centers, for example, both because he’s effective at boxing out and because he can sometimes draw them away from the basket with his scoring ability.
Measures of spacing
Defended 3-point attempts: Although it’s possible to imagine more sophisticated measures of player gravity, for the time being, the best publicly available metric to measure spacing is simply 3-point attempts, with an emphasis on 3-point attempts that are closely contested by the defense. This is a little tricky, though: Even shots that the NBA’s data currently describes as “wide open” (no defender within 6 feet) likely involve some degree of defensive pressure. Based on players’ shooting percentages, we treat the various shooting categories as follows:
- Nearest defender within 2 feet: 100 percent covered
- Nearest defender 2 to 4 feet away: 80 percent covered
- Nearest defender 4 to 6 feet away: 57 percent covered
- Nearest defender 6-plus feet away: 31 percent covered
Isolation turnovers: Our research also found that some types of turnovers — which we call isolation turnovers — are more costly than others in terms of predicting in-sample and out-of-sample RAPM. In particular, turnovers that are associated with attempts to score — as opposed to attempts to pass or otherwise contribute to a teammate’s opportunity to score — are associated with lower offensive RAPMs and are therefore punished by RAPTOR. Isolation turnovers consist of the following categories:
- 100 percent of travels, charging fouls, 5-second violations, offensive goaltends and baskets from below.
- 75 percent of lost-ball turnovers, palming turnovers, double dribbles, discontinued dribbles and step-out-of-bounds turnovers.
Miscellaneous offensive metrics
Fast-break starts: Possessions that begin with steals or after certain types of blocked shots are often highly productive, so players deserve some offensive credit for these actions in addition to their value on defense. Specifically, we estimate that a steal increases the value of a subsequent offensive position by 0.2 points, and a blocked shot on which a team comes down with the rebound inbounds increases it by 0.11 points.
Nonshooting defensive fouls drawn: In RAPTOR, the main value of drawing fouls is in the points they create via free throws. But what about fouls that don’t result in free throws? These have a small amount of value also because they (i) reset the shot clock to 14 seconds and (ii) often allow the offense to inbound the ball from an advantageous position, such as along the baseline near the basket, depending on where the foul was committed (empirically, possessions that restart after a nonshooting foul have a fairly high expected value). Thus, we estimate that nonshooting fouls drawn are worth about 0.16 points.
Penalty fouls drawn: Some additional benefits to drawing fouls are hard to measure via RAPM. Because RAPM evaluates players by comparing how a team performs when the player is on or off the court, it struggles with situations where a player creates value for his teammates regardless of whether he’s on the court. In particular, fouls that contribute to the bonus/penalty can increase the value of possessions later on in the quarter by making the penalty (which results in free-throw attempts being awarded on nonshooting fouls) more likely to occur. In addition, drawing fouls can put opponents in foul trouble and yield worse opponent lineups going forward. Fortunately, we estimate these effects to be small: Combined, they’re worth about 0.04 points per foul that’s not reflected by RAPM.
Opponents’ defensive rating: Finally, we calculate the average defensive rating of the opponents that the player faced (excluding possessions against the player himself). This is another way to account for the degree of difficulty of a player’s competition.
Box RAPTOR Defense
In measuring offense, RAPTOR is relatively elegant. The different aspects of an offensive possession — scoring, rebounding, passing, spacing — are well-represented, and the values assigned to various types of offensive statistics are reasonably intuitive.
Defense is more of an uphill battle. Some of the statistics RAPTOR uses to rate defensive performance are really more like proxies for other unmeasured statistics. We expect that the state of publicly available defensive metrics will improve in future years, and RAPTOR will improve along with them.
Nonetheless, we think RAPTOR majorly moves the ball forward on defense. The R-squared of our defensive RAPTOR regression in predicting within-sample RAPM is about 0.6, as opposed to only about 0.3 using traditional defensive statistics (steals, blocks, defensive rebounds, fouls committed) alone. This brings us a lot closer to capturing major parts of defense that have traditionally gone unmeasured.
Specifically, RAPTOR uses the following variables in its defensive regression:
Steals: Steals are an example of how defensive statistics can serve as both direct and indirect measures of player value. In our defensive RAPM regression, a steal is worth 1.49 points on defense. This is almost certainly more than the direct value that a steal provides, since the average NBA possession is worth around 1.08 points, meaning that the value of terminating a possession with a steal probably isn’t worth much more than 1.08 points. However, steals are also a proxy for overall defensive activity, some of which is currently going unmeasured.
Offensive fouls drawn: The same holds for offensive fouls drawn. In fact, they’re worth even more in the RAPM regression. Drawn fouls are rated highly by the regression both because they end a possession (often when the opposing team is in a strong position to score) and because they serve as a stand-in for stout overall on-ball defense. Players who are adept at inducing offensive fouls include Kyle Lowry, Ersan Ilyasova, Marcus Smart, Patrick Beverley and J.J. Barea. These types of players often have higher defensive RAPMs than their traditional defensive statistics would imply, and some of the reason for that is that they’ve been producing a lot of “hidden” defensive value by inducing offensive fouls.
Opponents’ field goals made and attempted: Earlier this year, we introduced DRAYMOND, a measure of on-ball defense based on the NBA’s opponents’ shooting statistics. In some ways, DRAYMOND was a first step in the creation of RAPTOR, our first foray into incorporating player tracking data into our projections. But it left two major things to be desired:
- DRAYMOND placed too much weight on how many shots a defender’s opponents made, which can involve a lot of luck, as opposed to how many shots the player defended.
- DRAYMOND essentially treated all types of shots equally. Our subsequent research, however, suggests that the current publicly available data on opponents’ 3-point shooting is largely noise. Many 3-point shots are relatively open shots, so which player happens to be the nearest defender is largely random in these cases, and whether the opponent makes the shot adds additional randomness. The opponents’ shooting data is quite a valuable indicator of rim protection or 2-point defense, on the other hand.
Thus, in RAPTOR, the different components of opponents’ shooting are weighted as follows:
- Each missed 2-point shot when the player is the nearest defender: +1.05 points.
- Each made 2-point shot: -0.33 points.
- Each attempted 3-point shot, whether the shot is missed or made : +0.17 points.
As an aside, RAPTOR defensive ratings do not use blocked shots. We find that there is no additional predictive power in using blocks when projecting RAPM, once you’re already accounted for opponents’ field goals.
Enhanced defensive rebounds: RAPTOR handles defensive rebounding as it does offensive rebounding. Contested defensive rebounds are worth considerably more in RAPTOR than uncontested rebounds. And defensive rebounds after shots that produce a high rate of offensive rebounds (such as layups and other shots near the rim) are worth more than rebounds on shots that don’t.
Positional opponents’ points scored: As mentioned earlier, attempting to infer positional matchups — and counting how many points and rebounds a player’s positional opponents secure — provides helpful information. Since 2013-14, the best and worst players based on positional opponents’ points allowed are as follows:
How NBA players vary in allowing their opponents to score
Positional opponents’ points per 100 possessions for players with at least 10,000 possessions played, 2013-14 through 2018-19
One advantage of this metric is that it can capture players who produce lots of blocks or rebounds at the rim — such as Kenneth Faried or Mitchell Robinson — but who aren’t very mobile defenders and might allow opposing centers and power forwards (especially stretch bigs) to score at high rates. Likewise, players who gamble on steals are sometimes punished by this statistic if they aren’t playing sound fundamental defense.
Positional opponents’ offensive rebounds: RAPTOR also accounts for how many offensive rebounds a player’s positional matchups secure. Some players such as Hassan Whiteside are effective at producing their own rebounds but also allow opposing bigs to secure offensive rebounds at relatively high rates. Others like Adams are both skilled at getting their own rebounds and at boxing out opponents from getting theirs.
Distance traveled, for perimeter defenders only: As mentioned, current publicly available defensive metrics are more effective at measuring interior/rim defense than perimeter defense. One metric that helps a bit on the perimeter is distance traveled per 100 defensive positions. RAPTOR uses this metric only for defenders that spend a lot of time on the perimeter, based on their ratio of 3-point shots to 2-point shots defended. (It’s not particularly helpful to have a rim protector like Rudy Gobert running all around the backcourt.) This stat can pick up on some additional defensive value for Avery Bradley or Iman Shumpert types who are pesky, active perimeter defenders. It can sometimes also detect players like Harden who take their share of defensive possessions off. This metric is a good candidate to get swapped out for more precise measures of defensive activity in future versions of RAPTOR.
Opponents’ free throws made: RAPTOR deducts value for free throws made on fouls committed by the defensive player. This is its main way of punishing defenders for committing fouls. However, the deduction for a made free throw is relatively minor (0.19 points). This is because fouls, although costly to the team, are at least a sign that the defensive player is challenging shots. (If we had better measures of defensive activity, in other words, the coefficients associated with fouls and free throws would probably be more steeply negative.) In fact, in the predictive formulation of RAPTOR used in our projection models (PREDATOR), fouls are handled slightly differently: A defensive player still gets a deduction when an opponent that he fouled makes a free throw, but the defender actually gets a small amount of credit for committing a foul.
Even though players don’t really exert any control over whether their opponents make their free throws, free throws made nevertheless outperforms free throws attempted as a measure of the cost of committing fouls because players do exert some control over who they foul. Big men tend to make free throws at lower rates than wings and guards, so fouls committed by big men (usually against other big men) tend to be less costly. In addition, some very smart defenders (e.g., Green or Gobert) show indications of being selective about who they foul, based in part on which opponents make free throws at a high rate.
Fastbreak turnovers committed: Just as generating turnovers that result in fast breaks help a team’s offense, committing turnovers hurts a team’s defense. Thus, live-ball turnovers (i.e., steals) result in a 0.2-point deduction to a player’s defensive rating, while field-goal attempts that result in blocked shots where the defense rebounds the ball inbounds result in a 0.11-point deduction.
Penalty fouls committed: As described earlier, fouls have some costs (potentially putting the opposing team in the bonus and creating foul trouble) that aren’t well-measured by RAPM, although these effects are small. Thus, players get a 0.04-point deduction for every foul they commit that counts toward the bonus/penalty.
Opponents’ offensive rating: RAPTOR calculates the average offensive rating of the opponents that the player faced as a defender and adjusts his defensive rating accordingly as a way to account for the strength of his competition.
RAPTOR On-Off
In comparison to “Box” RAPTOR, calculating a player’s “On-Off” RAPTOR rating is relatively simple. To calculate it, we undertook essentially the same process as for “Box” RAPTOR, regressing various offensive and defensive ratings against Davis’s six-year RAPM estimates. RAPM can be replicated quite effectively using three types of on-court ratings.
- The player’s offensive and defensive ratings — that is, how many points the team scored and allowed per 100 possessions while he was on the floor, adjusted for strength of opposition.
- The player’s courtmates’ weighted average offensive and defensive ratings when they weren’t sharing the floor with the player. The average is weighted by the number of possessions that the courtmate shared with the player, multiplied by the number of possessions that the courtmate had without the player.
- Finally, the player’s courtmates’ other courtmates’ weighted average offensive and defensive ratings. Whereas a player’s raw offensive and defensive ratings (step 1) are associated with positive coefficients (It’s good if a player’s team is outscoring its opponent while he’s on the floor!) and the player’s courtmates’ ratings (step 2) are associated with negative coefficients (It’s a bad sign for a player’s if his teammates are doing well even when he’s not on the floor!), the ratings of his courtmates’ other courtmates are a positive indicator. (I know this part is a little confusing.) Essentially, high courtmates’ courtmates’ ratings mean the teammates who seemed like they were doing well without the player on the floor may only have been doing well because they were paired with other good teammates.
We find that further iterations (i.e., looking at a player’s courtmates’ courtmates’ courtmates’ ratings) don’t contribute toward predicting RAPM.
We also find that this comparatively simple way to evaluate a player’s on-court/off-court impact not only replicates RAPM extremely well in sample but also predicts out-of-sample RAPM as well or slightly better than RAPM itself, depending on the regression specification. In other words, RAPM doesn’t appear to add much value as compared with computationally simpler approaches to evaluating on-court/off-court ratings.
Combining and Adjusting Box and On-Off Ratings to Create Overall RAPTOR
Overall RAPTOR is a blend of the “Box” and “On-Off” component ratings. We determined the respective weight assigned to “Box” and “On-Off” RAPTOR ratings by testing how well they predicted RAPM out of sample. Specifically, overall RAPTOR is equal to roughly 85 percent of “Box” RAPTOR, plus 21 percent of “On-Off” RAPTOR. A couple of fairly obvious observations about these figures:
- First, note that the combined value of “Box” and “On-Off” RAPTOR is slightly greater than 100 percent because they provide some nonredundant information. If a player’s “Box” rating is +3.0 and his “On-Off” rating is also +3.0, we’d exepct his overall RAPM to be slightly greater than +3.0, in other words.
- Also, note that “Box” RAPTOR is quite a bit more predictive of out-of-sample performance than “On-Off” RAPTOR. While on-court/off-court ratings are sometimes treated as though they’re the holy grail of NBA statistics, they’re very noisy. Conversely, many areas of player performance that were once thought to be intractable to statistical analysis can now be measured through player-tracking or play-by-play statistics. Other analysts may differ, but we think the medium-term future of NBA analytics is probably more about assigning value to players based on discrete actions they take on the court and less in trying to perfect an RAPM-like approach.
After combining “Box” and “On-Off” ratings, RAPTOR is then adjusted in two ways. (These are the same adjustments that are made by BPM, so we are again indebted to BPM and Daniel Myers for inspiration.)
The score effects adjustment
If you’re about my age (41) and played a lot of NBA Jam as a kid, you’ll remember computer assistance, which was how the software helped teams who trailed by significant margins by magically making their shots more likely to go in. It turns out that there is something vaguely analogous to this in the real NBA! Relative to the personnel they have on the floor, teams perform substantially worse when they have large leads and substantially better when they trail by significant margins. These tendencies, which we call score effects, can have profound effects. As Jeremias Engleman writes, when a team is behind by 20 points, it’s expected to score around 6 points per 100 possessions more than it does in a tied game, which is like “replacing an average offensive player with LeBron [James].”
In nontechnical language: You need to adjust “junk time” statistics. When a team is way ahead, it tends to be less efficient, and its opponents tend to be more efficient. As a result, unadjusted statistics will tend to underrate players on good teams and overrate players on poor teams because players on good teams are more often playing with significant leads and lollygagging their way through games, especially in the regular season. In crunch time, these teams may have a bigger advantage than their raw stats imply.
Our score effects adjustment is a little different than some of the other ones we’ve seen. Instead of inferring how far a team was ahead or behind based on its average final score, we calculate it directly by evaluating how far it was ahead or behind in an average possession throughout the season. In some cases, this can make a fairly big difference. For instance, the 2018-19 Philadelphia 76ers had a lower average victory margin (+2.7 points) than the Indiana Pacers (+3.3 points). The 76ers frequently had established large leads by the fourth quarter, however, while the Pacers did not — so they actually led their opponents by a larger margin on average throughout the game. Adjusted for score effects, they were a better team, in other words. Furthermore, in examining the impact of score effects on individual players, we evaluate them only for possessions when the player was on the court, rather than the team’s rating for all possessions in the game.
Another important difference in RAPTOR’s score effects adjustment is that it recognizes that the effects become larger in later quarters. A team will coast more with a 15-point lead in the fourth quarter than in the second quarter, in other words. In addition, score effects are considerably larger in the regular season than in the playoffs. This should make intuitive sense: a team is less likely to step off the gas pedal in the postseason when where is more on the line.
How score effects impact NBA efficiency
For every 10 points that it leads by, its scoring margin is affected by ___ points per 100 possessions, controlling for the personnel it has on the floor:
The team effects adjustment
Finally, RAPTOR adjusts individual players’ ratings so that they sum up to reflect the team’s overall performance, adjusted for score effects and strength of competition. If the Golden State Warriors score 7 adjusted points per 100 possessions more than the league average, for instance, then the Warriors’ players’ offensive RAPTOR ratings should also add up to +7.0, weighted by playing time. In reconciling team and player ratings, we make bigger adjustments to players with higher offensive and defensive usage rates. Colloquially speaking, this means that if a team was better or worse than the sum of its parts, we give more of the credit or blame for that to the players who were most heavily involved with the offense or the defense, respectively.
Note that we do not apply the team effects adjustment in the predictive version of RAPTOR, PREDATOR, as it does not appear to improve out-of-sample performance. This implies that the differences between a team’s overall scoring margin and the sum of its statistical components may actually be due mostly to luck rather than necessarily reflecting any intangible or hard-to-measure skills.
Individual Pace Impact
RAPTOR also attempts to evaluate an individual player’s impact on his team’s pace. Because pace is partly a function of a team’s coach and system, these ratings were derived from an analysis only of players who switched teams, and seeing which factors were persistent in predicting pace from one team context to the next. The resulting pace impact estimates reflect a combination of essentially an on-court/off-court pace rating — how much, empirically, a team’s pace changed when the player was on or off the floor — plus various statistical inputs that correlate with pace. Both inducing and committing turnovers tends to increase pace, for instance, as does commiting and drawing fouls, and taking open shots. We estimate that the following players had the biggest impact on their team’s pace in 2018-19 (minimum 1000 minutes played):
Which players had the greatest impact on team pace?
RAPTOR Individual Pace Impact ratings for 2018-19
Westbrook had the highest Individual Pace Impact in 2018-19, speeding up the Thunder’s pace by 2.7 possessions per 48 minutes while he was on the floor, while the Nuggets’ Monte Morris did the most to slow down his team’s pace.
Overall, we find that about half of a team’s pace is a result of the players it has on the floor, while the other half reflects the coach and system.
Replacement Level, WAR and Market Values
RAPTOR calculates wins above replacement level using a replacement level of -2.75 points per 100 possessions. The replacement level estimate is derived from evaluating the historical performance of players on two-way contracts, who are quite literally on the fringes between the major and minor leagues (the NBA and the G League), a status that reflects the traditional definition of replacement-level players.
In contrast to our previous system, RAPTOR uses the same overall replacement level (-2.75) across different positions, although note that replacement-level guards will tend to be terrible defensively and tolerable offensively, while the reverse is true for replacement-level bigs. This is because, unlike in many other advanced stats, RAPTOR ratings tend to be fairly even across the five traditional positions. The main exception is that point guards are slightly more valuable than shooting guards in RAPTOR on average, which makes sense to us since the league’s best point guards (think of a player like Curry) often have all the skills that off-guards do, but they also have additional ball-handling and passing abilities that off-guards sometimes lack.
RAPTOR ratings are relatively even across positions
And RAPTOR replacement level is set to -2.75 points per 100 possessions….
The precise formula that RAPTOR uses to calculate WAR is as follows…
… where the WAR multiplier is 0.0005102 for the regular season and 0.0005262 in the playoffs.
On our player projection pages, you’ll also find estimated market values — for instance, a certain player is worth $120 million over the next five seasons. These are designed to be slightly nonlinear rather than being a straight-line extrapolation of WAR. That is, a 10-win player is slightly more than twice as valuable as a 5-win player. The reason is that having superstar players makes a team considerably more likely to advance far in the postseason. These market values reflect how NBA teams value both regular-season performance and championships, in other words.
PREDATOR and RAPTOR Projections
As mentioned, RAPTOR now fuels our team and player projections. Or more technically, PREDATOR does, since that’s the version of RAPTOR we use for projecting future performance. The variables in PREDATOR are essentially the same as those in RAPTOR, but they use coefficients calculated with out-of-sample rather than in-sample RAPM. However, because we also avoided variables that performed poorly in out-of-sample tests in constructing RAPTOR, it and PREDATOR are extremely well-correlated. PREDATOR and RAPTOR have a 0.98 correlation on offense, and 0.95 on defense.
Our projections also use a variety of biographical inputs apart from RAPTOR and PREDATOR ratings that help in projecting performance going forward:
- Age
- Draft position
- NBA career length
- Height
- Weight
- Position
- Recent All-NBA, MVP and All-Star appearances
For college players making their NBA debuts, we also use variables related to the strength of their college program and the strength of their college program’s schedule. And for international rookies who did not play in the NCAA, we use variables related to both their country of birth and the country where they played professionally before coming to the NBA. In general, players who come from wealthy countries and who play in higher-quality international leagues start out of the gate faster but do not necessarily show as much improvement following their first few NBA seasons. Conversely, players who played in worse leagues and who come from poorer countries start out slower but show steeper improvement.
Otherwise, RAPTOR projections are essentially the same as our previous projection system, CARMELO, which is described here at some length. These projections basically involve a three-step process:
- Create a regression-based baseline projection of a player’s future RAPTOR ratings for the next seven seasons using his PREDATOR ratings from the past three seasons, plus his age and the other biographical variables that I described above.
- Identify similar or comparable players using a number of statistical and biographical categories; for instance, Damian Lillard’s top comparables include Chauncey Billups and Ray Allen.
- Tweak the player’s projection, and develop a range of uncertainty around the forecast, by seeing how the comparable players performed relative to their baselines when they were the same age as the player is now. For instance, if Allen had a better age-29 season than his baseline projection expected, that would favorably impact Lillard’s RAPTOR projection for this season.
Other than the adoption of RAPTOR rather than BPM and RPM as the basis for our projections, changes to our projection methodology this season are relatively subtle. They include the following:
- We now calculate separate projections for the playoffs and the regular season. Differences between regular-season and playoff performance are mostly attributable to luck and strength of competition. However, there is some persistence in which players tend to perform better or worse in the playoffs than in the regular season. Thus, RAPTOR projections evaluate the difference between a player’s regular season and playoff performances in the past and carry a small amount of that gap forward to future seasons. For 2019-20, players with the largest positive playoff adjustments are Draymond Green (1.4 points per 100 possessions better in the playoffs), LeBron James (+0.9), P.J. Tucker (+0.8), Kawhi Leonard (+0.7), Nikola Jokic (+0.6), Joel Embiid (+0.6), Tristan Thompson (+0.6), Terry Rozier (+0.5) and Paul George (+0.5). And players with the largest downward adjustments for the playoffs are Damian Lillard (-0.9 points), DeMar DeRozan (-0.6), Eric Gordon (-0.6), Klay Thompson (-0.5) and JJ Redick (-0.5).
- Several of the biographical variables that we employ this year are new. The most interesting one is probably awards received in the past three seasons, which is somewhat helpful for projecting out-of-sample performance. That is to say that MVP, All-NBA and All-Star voters can sometimes pick up on subtle aspects of player quality that RAPTOR misses.
- In weighting performance over the past three seasons, our projections previously assigned 60 percent of the weight to the most recent season, 30 percent to the second-most-recent season and 10 percent to the third-most-recent one. While this is a good rule of thumb for players in the middle of their career, it’s too conservative a weighting scheme for very young or very old players. Thus, the weights assigned to past seasons now depend on a player’s age. For a 23-year-old player entering his fourth NBA season, for instance, the program assigns around 76 percent of the weight to the player’s most recent season. What this means is that breakouts for young players (or declines for old players) mostly tend to “stick,” whereas you should expect more mean-reversion if a player shows a sharp apparent improvement or decline in mid-career.
Team Projections
As compared with our player projections, our process for calculating team projections is more straightforward. We create depth charts for each team and project playing time using a combination of algorithms and human inputs. Namely, we tell our depth charts program in which order the team prioritizes its players and (based on recent news accounts) which players are injured and for how long. The program then uses RAPTOR playing time recommendations to estimate how much each player will play at each position given these inputs. Players are allowed to slightly exceed their RAPTOR-recommended number of minutes per game, but if a player is playing significantly more minutes than recommended because the team is short-handed, our projections apply a penalty to his efficiency.
Once we have projected playing time, we can essentially just take a weighted sum of RAPTOR ratings to forecast the number of points a team will score and allow in a given game. We can then use Pythagorean expectation to estimate a team’s winning percentage. In the Pythagorean equation, we use an exponent of 14.3 for the regular season and 13.2 for the playoffs.
One important wrinkle is that in summing up individual RAPTOR projections to the team level, we need to account for score effects. Since RAPTOR ratings reflect a player’s efficiency in a tied game, but good teams often play with a lead — which reduces efficiency — good teams will perform slightly worse than the sum of their RAPTOR ratings, and bad teams will perform slightly better than them. To account for this, we multiply the sum of a team’s player projections by 0.8 in the regular season and by 0.9 in the playoffs.
Approximate RAPTOR ratings for historic players
Since our player projections use data since the 1976-77 NBA season (the first year after the ABA-NBA merger) we also have to approximate RAPTOR ratings for past seasons, even though modern player tracking and play-by-play data wasn’t available then. This requires a few tricks that we don’t have to use on current data. For instance, to do a good job of replicating RAPTORs using older data, we have to adjust for position, giving a boost to shooting guards and small forwards and penalizing centers. We also make heavier use of a team’s overall offensive and defensive ratings than our current RAPTOR ratings do. For seasons from 2000-01 onward, we also use RPM (which accounts for a player’s on-court/off-court impact) as an input.
An interesting philosophical question is whether these Approximate RAPTOR ratings are an optimal reflection of which players were the best of their eras given the (somewhat limited) data available to examine their performance — or, rather, since RAPTORs are calibrated using only data since 2013-14, whether they essentially reflect which past players would have been best under modern conditions. Either way, they help to reveal something about how RAPTOR thinks about players. Here, for example, are the 500 best RAPTOR and Approximate RAPTOR seasons of all time, ranked by combined regular season and playoff WAR.
The 500 best RAPTOR seasons of all time
Using actual RAPTOR (2013-14 onward) and Approximate RAPTOR (1976-77 through 2012-13); all statistics reflect the regular season and playoffs combined for players with a minimum of 1,000 minutes played.
Although LeBron James’s 2008-09 is the top season on a rate basis, when he had an Approximate RAPTOR rating of +12.6 per 100 possessions, it’s Michael Jordan who dominates the list by WAR, both because he got a ton of playing time and because he did all the things that RAPTOR loves: create shots, play defense and so forth. In fact, working on RAPTOR has convinced me that Jordan’s peak was probably a little higher than LeBron’s, something I didn’t necessarily believe before. We’ll save that discussion for another time, though, as well as the conversation about how RAPTOR feels about players such as John Stockton (loves) and Patrick Ewing (hates). For more detail on Approximate RAPTORS, you can find a files here that lists each player’s rating in the regular season and playoffs separately, or a version that combines a player’s performance over the whole season.
Acknowledgements: Thanks to Ryan Davis, Steve Ilardi, Ben Taylor, Seth Partnow, Charles Rolph and Evan Wasch for their advice and assistance on RAPTOR.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
The differences are that PREDATOR accounts for the number of defensive fouls committed and assigns it a positive coefficient, whereas RAPTOR does not, and that PREDATOR accounts for minutes per game while RAPTOR does not.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
The differences are that PREDATOR accounts for the number of defensive fouls committed and assigns it a positive coefficient, whereas RAPTOR does not, and that PREDATOR accounts for minutes per game while RAPTOR does not.
Since Lillard will be 29 this season.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
The differences are that PREDATOR accounts for the number of defensive fouls committed and assigns it a positive coefficient, whereas RAPTOR does not, and that PREDATOR accounts for minutes per game while RAPTOR does not.
Since Lillard will be 29 this season.
We use a 5-point scale for past awards, where a player gets 5 points for winning the MVP, 4 points, 3 points and 2 points for making the 1st, 2nd and 3rd All-NBA teams, and 1 point for making the All-Star team. The categories are exclusive rather than additive, so if a player gets 5 points for winning the MVP, he does not get additional credit for being an All-Star or on an All-NBA team.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
The differences are that PREDATOR accounts for the number of defensive fouls committed and assigns it a positive coefficient, whereas RAPTOR does not, and that PREDATOR accounts for minutes per game while RAPTOR does not.
Since Lillard will be 29 this season.
We use a 5-point scale for past awards, where a player gets 5 points for winning the MVP, 4 points, 3 points and 2 points for making the 1st, 2nd and 3rd All-NBA teams, and 1 point for making the All-Star team. The categories are exclusive rather than additive, so if a player gets 5 points for winning the MVP, he does not get additional credit for being an All-Star or on an All-NBA team.
Weighted by playing time.
That is, reduce how many points the opposing team scores; positive defensive ratings are good in RAPTOR and negative ones are bad.
Regular season and playoffs combined.
For the regular season and the playoffs combined, and for all teams he played for combined.
Namely, we removed the mean-reversion from RAPM; we also recalibrated RAPM such that the average score was zero for each regular season.
However, a player is charged for part of a usage a for free-throw attempt after a technical foul, since a team can choose which player takes the technical free throw.
Although there isn’t quite a 1-for-1 tradeoff: Being either the assister or the assistee is better than having nothing to do with the basket.
It may be that the subjective element of an assist is actually worth something and official scorers who give credit for assists are picking up on higher-quality passes. By contrast, the NBA’s potential assist category is determined algorithmically.
Indeed, most rebounds that occur amidst loose ball fouls are scored as team rebounds, not individual rebounds.
That is, in predicting RAPM.
Here’s one interesting comparison: Players make about 51 percent of 3-point attempts in the 3-point contest, when there is literally no defensive pressure (but there is time pressure). Under game conditions, the players who participated in the 3-point contest hit so-called wide-open 3-pointers at a 44 percent rate instead. It’s likely that these players hit truly wide-open shots — no defender closing out or threatening the shot in any way — at a rate greater than 44 percent.
For instance, through setting screens.
Note that while isolation turnovers are more costly to a player’s offensive RAPTOR because they indicate a lack of spacing, they’re actually slightly better from a defensive standpoint because they tend not to be live-ball turnovers.
That is, all fouls except for offensive fouls, which don’t count toward the penalty.
More precisely, we calculate each individual defender’s defensive rating and average them together — not the defensive rating for the five-man unit.
Steals do create additional value on offense, but this is covered by the offensive regression.
2.28 points.
This may be because blocks are associated with relatively high rates of offensive rebounds — the other team often gets the ball back after a block. It also may be because defenders who frequently go for blocks are causing other problems for their defenses — e.g., being out of position — that current stats are not measuring properly. Note that blocks do provide some offensive value in RAPTOR when they initiate fast-break possessions.
To some extent, this statistic is also capturing a team’s overall defensive performance while a player is on the floor. But that’s not necessarily a bad thing since RAPTOR is looking at team performance in various respects anyway. Basically, this statistic allows us to punish guards and wings more if opposing guards and wings are doing most of the scoring and to punish bigs more if opposing bigs are doing most of the scoring, as opposed to punishing all players equally.
The one plausible exception is if hard fouls reduce the opponent’s free-throw percentage by hurting them physically or psychologically.
Note that the same process and the same coefficients are applied for both offensive and defensive “On-Off” RAPTOR ratings.
Again adjusted for strength of opposition.
Note that evaluating the performance of a player’s courtmates provides for a more precise and direct way to evaluate a player’s impact than looking at his team’s overall rating while he was off the court. If, for example, Giannis Antetokounmpo never shared the floor with Donte DiVincenzo, then nothing about how DiVincenzo performed with Antetokounmpo off the floor is very informative about Antetokounmpo.
Adjusted for strength of opposition.
We calculate defensive usage rate by adding the number of possessions where the player induced a turnover, plus possessions where he committed a foul that results in free-throw attempts, plus possessions where he was the nearest defender on a field-goal attempt. It’s the mirror image of offensive usage, in other words.
In projecting a team’s pace going forward, we retain a share of its residual pace rating — that is, how much faster or slower it played than you’d expect from the sum of its players’ Individual Pace Impact ratings.
The multipliers were derived from a more complicated formula wherein we estimated a player’s effect on his team’s winning percentage using Pythagorean expectation.
The differences are that PREDATOR accounts for the number of defensive fouls committed and assigns it a positive coefficient, whereas RAPTOR does not, and that PREDATOR accounts for minutes per game while RAPTOR does not.
Since Lillard will be 29 this season.
We use a 5-point scale for past awards, where a player gets 5 points for winning the MVP, 4 points, 3 points and 2 points for making the 1st, 2nd and 3rd All-NBA teams, and 1 point for making the All-Star team. The categories are exclusive rather than additive, so if a player gets 5 points for winning the MVP, he does not get additional credit for being an All-Star or on an All-NBA team.
Weighted by playing time.
The lower exponent in the playoffs reflects the fact that score effects are less profound in the playoffs.