This is the Reference Site of the Quest, featuring all reference guides in one place where they are easy to access.

Thursday, February 26, 2009

[Historical, and Non-Current] User Guide for Real Player Rating Reports for the NBA, for NBA Teams, and for Games, February 2009

NOTICE: THIS USER GUIDE IS FOR HISTORICAL USE ONLY. FOR THE LATEST GUIDE, LOOK FOR THE MOST RECENT VERSION IN THE REFERENCE INDEX. This version applies for ratings produced from late February until the end of May 2009.

REAL PLAYER RATINGS BY TEAM USER GUIDE
Updated Feb. 25, 2009

INTRODUCTION TO THE CONCEPT OF REAL PLAYER RATINGS
The Real Player Rating (RPR) is a very carefully constructed all inclusive performance measure. Everything of value that a basketball player can do is recorded by official NBA scorekeepers who sit right along the edge of the court, mid-court, and who are trained to observe and record everything that happens in a game.

Since these days all of these counts are immediately input into continually updated public data bases online, such as at ESPN, it is theoretically possible to combine everything together into an overall performance measure for each player. This is what the RPR does.

Real Player Rating or RPR is everything tracked by scorekeepers that a player does, good and bad, added and subtracted (with negative things such as turnovers and missed shots being subtracted). Very carefully calibrated factors, or weights, are applied to the different elements. The calibration, as you would expect, is done to reflect the different value toward winning games that different actions on the court have. All of the good and bad combined together is divided by minutes, so we can tell the rate, which we need to determine the overall quality or value of the player.

REAL PLAYER RATINGS ARE ADJUSTED FOR DEFENDING NOT TRACKED BY SCOREKEEPERS STARTING IN 2009
Not counting purely subjective and abstract factors such as leadership, and not counting a few infrequently occurring actions on the court (not being counted or tracked by anyone yet) such as chasing down loose balls, the only thing a basketball player can regularly do on the court of any value that is not counted by NBA scorekeepers is preventing what would have been a score from being a score by defending against the shot or shots during a possession well enough to stop what would have been a score by the opposing team. In other words, what the player does to make the possessions of the opposing teams worthless other than what is already counted, which would be rebounds, steals, blocks, and personal fouls. These untracked or hidden actions would include effective man to man defending, effective rotation on defense off screens and picks, defensive recognition, and quickness of defensive reaction. These things would be counted by scorekeepers if it were possible. But, for example, there is no way to know exactly how many shots a good (or any kind of) defender has changed from being a score to a miss.

Quest for the Ring has developed a statistically valid way to accurately estimate the untracked or hidden aspects of defending. This is described in complete detail in the latter sections of this Guide.

SIMPLICITY, RELIABILITY, TRANSPARENCY, AND FOCUS ONLY ON "WINNING POWER"
Like everything statistical we do at Quest, we have kept this process as simple and reliable as possible, while at the same time spending as much time as necessary on design, quality control and performance evaluation. Unlike some other practitioners, we avoid what you might call layered complexity, which leads to formulas which can not be understood without studying them and which high traffic sites will not show on any of their web pages for fear that the public will rebel against the statistic. At Quest, we think that our rating systems can be understood and evaluated by most high school graduates, and we keep everything out in the open through User Guides such as this one.

Basketball statistical gurus frequently forget that no matter how intricate their formulas are, they are very heavily manipulating process items such as assists and rebounds while spending very little time on how these things fit together to produce wins and losses. We think that they are making the mistake, whether or not they are aware, of injecting value adjustments regarding how they think the game should be played and value adjustments about which playing styles are better than others.

Whereas, the primary objectives of the relative simplicity (small number of formulas, to be precise) of the Quest RPR is to avoid all how the game should be played and how players should play value judgments. We don't care about the styles, only about the results. The RPR is concerned first and foremost with the impact each player has on the potential for winning games.

Quest thinks it makes more sense to minimize the manipulation of process items, and to focus much more on coming up with the best possible estimation of how the process items impact points for and points against in games, which in turn of course determines wins. Whereas other "advanced statistics" might give you more depth and flavor regarding how a particular player plays, the Quest RPR is a way for the reader to, in a very quick and easy way, determine what the overall value of the player is with respect to producing wins or losses.

In other words, the foundation of RPR is and will always be measurement of a player's power to help win basketball games, whereas the foundation of other, more complicated statistics may include preferences about how the game should be played and about the style of players, with winning power measured less accurately as a result of those focuses.

THE MAIN REASON REAL PLAYER RATING IS SO VALUABLE
Because it is per time, RPR is the best possible measure of the net quality of a basketball player, or simply "how good" the player is (on average) for each minute of playing time.

REAL PLAYER RATING REPORTS CAN BE FOR THE WHOLE NBA, FOR A TEAM, OR FOR A GAME
With a Real Player Ratings Report for the entire NBA, you can see very rapidly who the best players in the NBA have been during the course of the season.

With a Real Player Ratings Report for a Team for the Regular Season, you can see very rapidly who the best players on the team have been during the course of the season. You can use this information to investigate the possiblity that the coach is not perfect. Well, we know that no coach is perfect. So really, with the benefit of 20/20 hindsight, we can investigate and determine what mistakes the coach has apparently made with regard to rotations and playing times. Furthermore, by using the Ratings, basketball knowledge, a little creativity, and logical deduction, we can also investigate and perhaps determine whether the coach has made incorrect decisions regarding which strategies and plays are best for his team's offense and defense.

Real Player Ratings for games are the most important component of Reports called Ultimate Game Breakdown: Players Reports.

CAUTIONS
To be completely honest and clear, although it is the best possible overall real life measure, RPR is still not a perfect or absolute, "final word" measure on any player. In general, you must remember that all performance measures including this one for the NBA are relative rather than absolute measures. The ratings are relative to the team context. Players do not exist in a vacuum, especially in basketball.

Several specific cautions will now be described.

Because basketball is a team game, and more so than most other sports, players who are on really good teams might have their own performances "crowded out" to some extent by even better players. So, paradoxically, ratings of good players on good teams will generally have slightly lower ratings than they would have if they were on a bad team. Conversely, great players on bad teams will have slightly higher ratings than they would have if they were on a good team.

Players need not only playing time but possession of the ball in order to produce many of the things that count in the rating. So if, for whatever reason, a player does not get the ball as often as he would on a different team, or with a different coach, or with whatever other circumstances you can dream of, then his RPR will be lower than what it could or would be.

If a good player is on a good team where there are players even better than he is, than his RPR will likely be lower than it would be if he were on a not as good team.

If a good player plays a certain position for which his team has an even better player, then it's probable that the better player will crowd out the lesser player to one extent or another, so that the lesser player's RPR will be lower than what it would be if he were the best player at the position on the team.

The ratings are only for the current season. It has recently been discovered that many player's ratings often change up or down by more than 10% from year to year, and by much greater amounts over many years.

Those who think defense in basketball is much more important than offense may consider the magnitude of the defensive adjustment to be inadequate. They will contend that defensive specialists who are poor offensive players should have a higher rating. While we realized that we needed to adjust the ratings for defending not tracked by NBA scorekeepers, we continue to believe that players who are great defensive specialists but poor or undeveloped offensive players should in most cases rank no higher than the major role player level.

Do not forget that RPR is a per time measure. RPP and not RPR measures total impact of a player. RPR measures how valuable a player has been toward winning basketball games, per unit of time.

The classification scheme, like the ratings, is relative. A role player on a bad team might be a solid starter on a very good team. A star on a bad team might be just a major role player on a really good team. And so on and so forth. A player is a star, a role player, or whatever only in the contexts of the particular season and the particular team involved. If he was on a different team, or if it was a different year, his classification might be different.

So in conclusion, don't think of RPR as the ultimate gospel or bible on how good players are. But do think of it as an extremely accurate and reliable summary of how good the players actually have been in real life in the specific time (season or playoffs) and place (team) involved.

A NOTE ABOUT REAL PLAYER RATINGS FOR INDIVIDUAL GAMES
However, not as many breakdowns of individual game ratings are going to closely track the overall average for the roster as you might think. This is because one of the interesting things about basketball that makes it different from most other sports is that "how good" a player is from game to game varies radically. The best players have terrible games where they do almost nothing sometimes, while players who normally do not do much can every once in a while have outstanding games, at least if you measure it per minute on the court anyway. If you just looked at actual production, and never at a reserve player's Real Player Rating, you would hardly notice any of his unusually outstanding games, since players who normally do not do much will normally not have much playing time.

INTERACTIONS BETWEEN PLAYING TIMES, PLAYER RATINGS, AND THE NEEDS OF TEAMS
There are certain things that only certain players can do very well, and if those things are crucial for the team, than those players will have to play more minutes than they might otherwise play. The extra minutes might tend to reduce the player's Real Player Rating, while his total production will rise with the additional minutes. So to fairly and completely evaluate any player, you must always look at both the Real Player Rating (RPR) and the Real Player Production (RPP).

Furthermore, it is strongly suspected that, in order to compete in the playoffs, a team must have as many players of as high a quality (RPR) as possible, while at the same time having at least one or two players whose actual production is among the highest in the NBA regardless of exactly how high the RPRs happen to be. (All high RPP players will be relatively high RPR players; some will be higher than others.) Specifically for example, LeBron James' actual massive amount of production is most likely just as important to the Cleveland Cavaliers as is his RPR or, in other words, as is his rate of production. Similarly, Kobe Bryant's quantity is probably at least as important to the Lakers as is his quality.

Whereas, teams such as the Denver Nuggets, who have instructed a possible huge producer, Carmelo Anthony, to "not worry about scoring," may have made a fatal mistake relative to the playoffs, because teams with no extremely high rate producers may be generally doomed to lose quickly in the playoffs even if they have an unusually large number of high quality players as shown by RPR. This is because extremely high RPP players can by themselves "dominate a game" to some extent, meaning they can by themselves possibly win the game for their team, without worrying about complications that come in to play if you need to coordinate several high RPR but ultimately and theoretically limited RPP players.

Players who over the course of a season appear to rank higher in RPR (quality) but lower in RPP (quantity) may not be getting enough playing time. Players who over the course of a season appear to rank lower in RPR (quality) but higher in RPP (quantity) may be getting too much playing time. But as alluded to earlier, you must not automatically conclude this, because some skills are needed out on the court most of the time, but yet may be available only from a small number players on the roster. Such players may have to get more playing time due to that critical skill in short supply, even if their overall quality does not seem to justify all of that playing time.

A relatively common reason for unusual playing time will be players who are either truly outstanding defenders (who get extra playing time) or truly bad defenders (who get their playing time reduced).

Another common reason for extra playing time will be if a team has a point guard who has many more turnovers than the average point guard has. Because the point guard is so important, a good coach has to play his best guard who can make plays at the position for a full set of minutes every game, pretty much regardless of how many turnovers that player makes. If you take out your designated point guard due to "too many turnovers," it may end up sort of like cutting your foot off because you have a bad case of athletes foot!

MINIMUM PLAYING TIME RULES
Only players who played at least 10% of the minutes of whoever has played the most minutes on the team are included in these reports. Any player who has played for less than 10% of the minutes of the player who has played the most minutes is not included, since he didn't play for long enough to be fairly or reasonably compared with the other players. Furthermore, as described previously in the adjustment for defending section, only players who have played at least 300 minutes can have a defensive rating, or an overall RPR given to him. Both the 10% and the 300 minutes minimums must be met for a player to be rated.

REAL PLAYER PRODUCTION
Of course, looking at actual production (everything positive added together and everything negative subtracted out) is something that is extremely important too. The total production (everything good and everything bad combined together) is simply called Real Player Production or RPP.

There is no methodology for including defending (other than rebounding, steals, blocks, and personal fouls) in RPP at this time.

SOURCE OF TRACKED BASKETBALL COUNTS
The sources for the raw counts of scores, rebounds, steals, turnovers, and so forth is ESPN.com and NBA.com.

THE FORMULA
For 2008-09, the RPR formula has been very carefully and accurately tweaked again and is set to be as follows:

POSITIVE FACTORS
Points 1.00 (at par)
Number of 3-Pt FGs Made 1.00
Number of 2-Pt FGs Made 0.60
Number of FTs Made 0.00

Assists 1.75

Offensive Rebounds 1.15
Defensive Rebounds 1.25
Blocks 1.60
Steals 2.15

NEGATIVE FACTORS
3-Pt FGs Missed -1.00
2-Pt FGs Missed -0.85
FTs Missed -0.85

Turnovers -2.00
Personal Fouls -0.80

DEFENDING RATING
A quality of defending rating of between 0 and .230 is added to "Base or unadjusted RPR". In most cases, the defending rating is between 0.050 and .150. See the User Guide for the Defending Components" below for a very detailed explanation of how we determine how to defensively rate the players.

ACTUAL COMBINED AWARD OR PENALTY BY TYPE OF SHOT
3-Pointer Made 4.00
2-Pointer Made 2.60
Free Throw Made 1.00
3-Pointer Missed -1.00
2-Pointer Missed -0.85
Free Throw Missed -0.85

ZERO POINTS: PERCENTAGES BELOW WHICH THERE IS A NEGATIVE NET RESULT
3-Pointer 0 score % 0.200
2-Pointer 0 score % 0.246
1-Pointer 0 score % 0.459

This means that if a player has a lower percentage than any of the three above, then his RPR would be lower rather than higher as a result of his shooting that type of shot.

ASSISTS VERSUS TURNOVERS ZERO POINT
Assist/Turnover Ratio That Yields 0 Net Points: 1.143

This means that any player who has an assist/turnover ration of less than 1.143 is losing RPR rating when assists and turnovers are considered. He would have to either increase assisting or reduce turnovers to turn the combined effect from assists and turnovers positive.

QUALITY (RPR) AND QUANTITY (RPP} SUMMARIZED ONE LAST TIME
RPR reports show for each player the RPR (Real Player Rating) which tells you how good a player did (all the good things minus all the bad things) out on the court per unit of time. The RPP (Real Player Production) report tells you how much in total (the sum of the of the good things minus the sum of the bad things) a player did out on the court, without regard to playing time.

Many and maybe most sports watchers and an unknown but probably disturbingly large number of sports managers make the mistakes of exaggerating the importance of quantity and overlooking to some extent quality. These reports allow you to expand your horizons. These reports put quantity and quality side by side, which is extremely valuable, because both are roughly equally important in explaining accurately why and how the team is playing the way it is.

======== DEFENDING AND OFFENSIVE SUB RATINGS ======================

THE DEFENDING ADJUSTMENT TO REAL PLAYER RATINGS AND THE DEFENDING SUB RATING

THE DEFENDING COMPONENTS OR SUB RATINGS OF REAL PLAYER RATINGS--NEW AS OF JANUARY 2009
As of January 8, 2009, The Quest is proud to announce to you that the second major improvement to Real Player Ratings (RPR) in less than half a year is now fully up and running. The first major improvement were some needed changes in the factors used for RPR. The second major improvement (series of improvements, actually) is so far as I am aware the first ever effort to rate the defensive efforts of players that are hidden unless you watch all that player's games, because they are not scored or tracked by scorekeepers.

I have been talking about and working for and expecting the breakthrough in evaluation of defending for almost two years. Now that the breakthrough has come, I am now even more certain that RPR is the best overall rating system in existence, and that it is now roughly as good as it will ever or can ever be.

I recently developed a statistically valid way to rate the defending of players, that is, what they do to prevent scores other than rebounding, blocks, steals, and fouls, which were always included in RPR. This would include man to man defending, zone defending, pick and roll defending, defensive recognition, and defensive rotation.

Although the technique used had to be indirect and subject to a very small amount of statistical error, it validly awards the better defenders with bigger RPR bonuses. It has been validated by comparing results obtained with the defensive ratings shown on three different "advanced basketball statistics" web sites. Our results were shown to be extremely highly correlated with the results shown on the other sites. Where there are small differences, I believe mine are better, if only because mine uses simple, bedrock statistical theory rather than involved formulas.

HIDDEN DEFENDING
Before revealing what we do to reveal it, let's define "hidden defending." Exactly what is hidden defending? It's defending not tracked by the NBA. It's every action that helps to prevent the other team from scoring other than rebounding, stealing, and blocking, and fouling. So it would include man to man defending, zone defending, rotating in general, defensive recognition, and quick defensive response to various offensive tactics, such as pick and rolls. Obviously, if a defender is good at these things, the other team doesn't score as many points than if the defender is lousy at these things.

HOW TO REVEAL HIDDEN DEFENDING IN FOUR STEPS
STEP ONE: CALCUATION OF RAW HIDDEN DEFENDING RATINGS
Unlike most "advanced statistics" that are published on the internet or in print, we give you all the details about how we do ours, so that you can evaluate the evaluations, so to speak. The following is specifically what we are doing to be able to accurately and fairly compare players' defending:

Where do we start to discover what is hidden? We keep it as simple and yet as accurate as possible. We use the most official and therefore presumably the most reliable data as the building blocks for rating the defense of NBA players. We start with the player minutes and points scored by the other team while the player was on the court that are shown in the plus/minus statistical section at NBA.com.

There are no value judgments made regarding a player's defending style, or regarding a team's defending style for that matter. We don't care about style. Using points allowed per minute is looking at results, nothing more and nothing less.

After simply dividing points allowed by minutes on the court, we adjust (we standardize, to be more precise) that rate for the pace of the team and for the quality of the team's defense. The two adjustments are needed so that the ratings of players who are on different teams can be fairly compared.

Players who are on teams with faster paces give up more points per minute through no fault of their own. Similarly, players who are on teams with less efficient defenses give up more points per minute, regardless of how well they defend, everything else held constant. You could not fairly compare players on two or more teams with different paces and different team defense qualities unless you standardized, or in other words controlled for those differences for all NBA players.

USE OF BASIC STATISTICAL SAMPLING THEORY
What we are doing is using an indirect and inexact yet accurate and statistically valid way to discover who the better defenders are. No two players are out on the court for all the exact same minutes. So although for every player, what the other players out on the court do defensively while they are out on the court is a very large factor determining what that player's points per minute allowed will be, when you look at many, many hundreds of minutes, what the individual player does, or does not do defensively, as the case may be, will eventually show up in that particular player's points allowed per minute statistic.

In other words, what any individual player does defensively has to sooner or later show itself in a differentiation from other players of his points allowed per minute. As the number of minutes rise above 500, and then 1,000 and then, for many players, above 2,000 and even 3,000 for a regular season, what a particular player does or does not do defensively becomes more and more exactly shown by the points allowed per minute number. This is very basic statistical sampling theory in operation. Statistical sampling theory is the easy to understand bedrock theory of statistics.

Due to the necessity of a large sample of minutes, we will not do defending estimates for any player who has played for fewer than 300 minutes. Quality of defending estimates will be slightly less accurate for players who have only played between 301 and about 600 minutes than they will be for players who have played for more than 600 minutes. We believe that the estimates are going to be extremely accurate for all players who have played 750 minutes or more. The idea is relatively simple: as the number of hundreds of minutes played goes up, the accuracy of this system improves, to the point where it gives you the same information you would have if you knew exactly how many possessions of the other team each player ruined with his defending.

For your information, all players allow between 1.85 and 2.18 points per minute; most allow between 1.94 and 2.11. The overall NBA average is about 2.03 points per minute allowed.

STEP TWO: CONVERSION OF RAW HIDDEN DEFENDING POINTS ALLOWED PER MINUTE TO FILTERED HIDDEN DEFENDING POINTS ALLOWED PER MINUTE
Since different players have different breakdowns between how much of their defending shows up in tracked statistics such as defensive rebounding and how much of it does not, in order to improve accuracy we need to have a method to filter, or in other words, separate, the two categories of defending. If we didn't do this, we would still have a useful statistic, but it would be biased in favor of players whose defending is counted in tracked statistics more so than other players. There would be in effect some double counting of defending for players who have most of their quality defending tracked by scored statistics.

The filter used is to multiply the raw hidden defending ratings by the percentage of the real player production that is offensive. In other words we take the inverse of the percentage of a player's real player production that is defensive and multiply the raw hidden defending ratings by that. The rationale to do this is that although the exact relation is unknowable, we know that for a given raw hidden defending performance level, there will be an inverse relation between scored defending and hidden defending. The more defensive rebounds, steals, and blocks a player is making for any raw level, the less he is relying on hidden defending to achieve the raw level. And vice versa. So multilying by the inverse of the percentage of all contributions that are defensive (in other words, multiplying by offensive contributions) filters out much of the bias that is in the raw hidden defending rating.

To be even more specific, we first extract out defensive rebounding, steals, blocks, and personal fouls, the sum total of which is called "Scored Defensive Contribution". All of the other components combined constitute "Scored Offensive Contribution". Now we can determine the percentages of the RPP that are offensive and defensive, and then we can use the offensive percentages to convert the raw hidden defending ratings to filtered hidden defending ratings.

STEP THREE: CONVERSION OF FILTERED ALLOWED POINTS PER MINUTE TO FILTERED HIDDEN DEFENDING RATING
We need to translate the adjusted or filtered points allowed per minute into numerical terms that are the most useful with respect to RPR. So with a very carefully designed translation scale, we amplify the very small differences in different player's points allowed per minute numbers into much larger different hidden defending ratings for each player. Then we simply add the hidden defending rating to the Base RPR to yield RPR.

STEP FOUR: USE OF HIDDEN DEFENDING RATING
We now have added in a reasonably good estimate of the value of actions of players that are not even kept track of by scorekeepers! The filtered hidden defending ratings are added to the Base, Unadjusted, or Scored RPR to give RPR. Technically, you could call the final result "Ajusted RPR," but we are trying to avoid that terminology because of how important we think it is to include the hidden defending in the performance measure.

SIZE OF THE DEFENDING ADJUSTMENTS
Base RPR's for most NBA players range between .400 and 1.000. The range of possible defending adjustments to the base RPRs is from 0 to about .230. In most cases, however, the adjustment will be between 0.020 and .170.

THE DEFENDING SUB RATING: PUTTING THE HIDDEN AND THE UNHIDDEN TOGETHER
Aside from the Hidden Defending Rating we can find out how well each player does in terms of unhidden or scored defending, can't we? Of course se can.

Aside from the hidden there is of course unhidden defending, which would be rebounding plus steals plus blocks minus personal fouls. If we extract the combination of those four out of the same counts that underlie the RPR as a whole, we get what we are going to call the Scored Defending Contribution. This could also be thought of as Tracked Defending Contribution if you prefer. Then if we divide this by minutes, we can have a Scored (or Tracked) Defending Rating.

Finally, if we combine Hidden Defending Rating (HDR) with Scored Defending Rating (SDR) we can have an Overall Defending Rating (ODR). I am for now going to simply multiply the HDR by two and add that to the SDR to yield the ODR. To combine them this way is more arbitrary than my usual standards allow; I am doing this because there is as of yet no non-arbitrary way of doing it. The formula of two times HDR plus SDR brings HDR almost up to par with SDR in terms of the actual numbers and the averages of those numbers involved.

In other words, I am saying for now that hidden defending is almost as important as scored defending. There appear to be many coaches and not a few hardcore basketball fans who believe that hidden defending is actually more important than scored defending, but I am very likely never going to agree with that. I think that although hidden defending is important, and plausibly almost as important as tracked defending, that it is like a quicksand in that there seems to be a tendency for a substantial minority of basketball people to get carried away with estimating the importance of it and then become more and more trapped by their error in terms of how they look at basketball or in terms of how they coach their team if they are coaching.

THE OFFENSIVE SUB RATING
The Offensive Sub Rating is all tracked actions other than the defensive ones (defensive rebounding, steals, blocks, and personal fouls) combined together using the RPR weights, divided by minutes. In other words, it is Total Offensive Production divided by minutes. For the list of all tracked actions and the weight factors assigned to each, see the secion titled "The Formula" above.

======== SUMMARY OF PRIMARY FORMULAS =================
Real Player Production or RPR = (All tracked or scored actions weighted according to best available analysis of importance / minutes) + Filtered Hidden Defending Adjustment

Real Player Production or RPP = Total Offensive Production + Total Defensive Production. (All tracked or scored actions weighted according to best available analysis of importance.)

Offensive Sub Rating = Total Scored or Tracked Offensive Production / Minutes

Defensive Sub Rating = (Total Scored or Tracked Defensive Production / Minutes) + 2 X Filtered Hidden Defending Adjustment

Filtered Hidden Defending Adjustment = Raw Hidden Defending Adjustment X Percentate of RPP that is Offensive

Raw Hidden Defending Adjustment = Assigned value based on chart. the objective of which, is to amplify seemingly minute differences in points allowed per minute.

Monday, February 9, 2009

[Historical and Non-Current] User Guide for the Defending Components or Sub Ratings of Real Player Ratings, January 2009

THE DEFENDING COMPONENTS OR SUB RATINGS OF REAL PLAYER RATINGS--NEW AS OF JANUARY 2009
As of January 8, 2009, The Quest is proud to announce to you that the second major improvement to Real Player Ratings (RPR) in less than half a year is now fully up and running. The first major improvement were some needed changes in the factors used for RPR. The second major improvement (series of improvements, actually) is so far as I am aware the first ever effort to rate the defensive efforts of players that are hidden unless you watch all that player's games, because they are not scored or tracked by scorekeepers.

I have been talking about and working for and expecting the breakthrough in evaluation of defending for almost two years. Now that the breakthrough has come, I am now even more certain that RPR is the best overall rating system in existence, and that it is now roughly as good as it will ever or can ever be.

I recently developed a statistically valid way to rate the defending of players, that is, what they do to prevent scores other than rebounding, blocks, steals, and fouls, which were always included in RPR. This would include man to man defending, zone defending, pick and roll defending, defensive recognition, and defensive rotation.

Although the technique used had to be indirect and inexact, it validly awards the better defenders with bigger RPR bonuses. It has been validated by comparing results obtained with the defensive ratings shown on three different "advanced basketball statistics" web sites. Our results were shown to be extremely highly correlated with the results shown on the other sites. Where there are small differences, I believe mine are better, if only because mine uses simple, bedrock statistical theory rather than involved formulas.

HIDDEN DEFENDING
Before revealing what we do to reveal it, let's define "hidden defending." Exactly what is hidden defending? It's every action that helps to prevent the other team from scoring other than rebounding, stealing, and blocking. So it would include man to man defending, zone defending, rotating in general, defensive recognition, and quick defensive response to various offensive tactics, such as pick and rolls. Obviously, if a defender is good at these things, the other team doesn't score as many points than if the defender is lousy at these things.

HOW TO REVEAL HIDDEN DEFENDING IN FIVE STEPS
STEP ONE: CALCUATION OF RAW HIDDEN DEFENDING RATINGS
Unlike most "advanced statistics" that are published on the internet or in print, we give you all the details about how we do ours, so that you can evaluate the evaluations, so to speak. The following is specifically what we are doing to be able to accurately and fairly compare players' defending:

Where do we start to discover what is hidden? We keep it as simple and yet as accurate as possible. We use the most official and therefore presumably the most reliable data as the building blocks for rating the defense of NBA players. We start with the player minutes and points scored by the other team while the player was on the court that are shown in the plus/minus statistical section at NBA.com.

After simply dividing points allowed by minutes on the court, we adjust that rate for the pace of the team and for the quality of the team's defense. The two adjustments are needed so that the ratings of players who are on different teams can be fairly compared.

Players who are on teams with faster paces give up more points per minute through no fault of their own. Similarly, players who are on teams with less efficient defenses give up more points per minute, everything else held constant. You could not fairly compare players on two or more teams with different paces and different team defense qualities unless you standardized, or in other words controlled for those differences for all NBA players.

USE OF BASIC STATISTICAL SAMPLING THEORY
What we are doing is using an indirect and inexact yet accurate and statistically valid way to discover who the better defenders are. No two players are out on the court for all the exact same minutes. So although for every player, what the other players out on the court do defensively while they are out on the court is a very large factor determining what that player's points per minute allowed will be, when you look at many, many hundreds of minutes, what the individual player does, or does not do defensively, as the case may be, will eventually show up in that particular player's points allowed per minute statistic.

In other words, what any individual player does defensively has to sooner or later show itself in the points allowed per minute. As the number of minutes rise above 500, and then 1,000 and then, for many players, above 2,000 and even 3,000 for a regular season, what a particular player does or does not do defensively becomes more and more exactly shown by the points allowed per minute number. This is very basic statistical sampling theory in operation. Statistical sampling theory is the easy to understand bedrock theory of statistics.

Due to the necessity of a large sample of minutes, we will not do defending estimates for any player who has played for fewer than 300 minutes. Quality of defending estimates will be slightly less accurate for players who have only played between 301 and about 600 minutes than they will be for players who have played for more than 600 minutes. We believe that the estimates are going to be extremely accurate for all players who have played 750 minutes or more. The idea is relatively simple: as the number of hundreds of minutes played goes up, the accuracy of this system improves, to the point where it gives you the same information you would have if you knew exactly how many possessions of the other team each player ruined with his defending.

For your information, all players allow between 1.87 and 2.16 points per minute; most allow between 1.94 and 2.11. The overall NBA average is about 2.03 points per minute allowed.

STEP TWO: CONVERSION OF RAW HIDDEN DEFENDING POINTS ALLOWED PER MINUTE TO FILTERED HIDDEN DEFENDING POINTS ALLOWED PER MINUTE
Since different players have different breakdowns between how much of their defending shows up in tracked statistics such as defensive rebounding and how much of it does not, in order to improve accuracy we need to have a method to filter, or in other words, separate, the two categories of defending. If we didn't do this, we would still have a useful statistic, but it would be biased in favor of players whose defending is counted in tracked statistics more so than other players. There would be in effect some double counting of defending for players who have most of their quality defending tracked by scored statistics.

The filter used is to multiply the raw hidden defending ratings by the percentage of the real player production that is offensive. In other words we take the inverse of the percentage of a player's real player production that is defensive and multiply the raw hidden defending ratings by that. The rationale to do this is that although the exact relation is unknowable, we know that for a given raw hidden defending performance level, there will be an inverse relation between scored defending and hidden defending. The more defensive rebounds, steals, and blocks a player is making for any raw level, the less he is relying on hidden defending to achieve the raw level. And vice versa. So multilying by the inverse of the percentage of all contributions that are defensive (in other words, multiplying by offensive contributions) filters out much of the bias that is in the raw hidden defending rating.

To be even more specific, we first extract out defensive rebounding, steals, blocks, and personal fouls, the sum total of which is called "Scored Defensive Contribution". All of the other components combined constitute "Scored Offensive Contribution". Now we can determine the percentages of the RPP that are offensive and defensive, and then we can use the offensive percentages to convert the raw hidden defending ratings to filtered hidden defending ratings.

STEP THREE: CONVERSION OF FILTERED ALLOWED POINTS PER MINUTE TO FILTERED HIDDEN DEFENDING RATING
We need to translate the adjusted or filtered points allowed per minute into numerical terms that are the most useful with respect to RPR. So with a very carefully designed translation scale, we amplify the very small differences in different player's points allowed per minute numbers into much larger different hidden defending ratings for each player. Then we simply add the hidden defending rating to the Base RPR to yield RPR.

STEP FOUR: USE OF HIDDEN DEFENDING RATING
We now have added in a reasonably good estimate of the value of actions of players that are not even kept track of by scorekeepers! The filtered hidden defending ratings are added to the "Base or Scored RPR" to give RPP. The range of possible defending adjustments to the base RPR is from 0 to about .230. In most cases, however, the adjustment will be between 0.030 and .150.

STEP FIVE: OVERALL EVALUATION OF DEFENDING
Aside from the Hidden Defending Rating we can find out how well each player does in terms of unhidden or scored defending, can't we? Of course se can.

Aside from the hidden there is of course unhidden defending, which would be rebounding plus steals plus blocks minus personal fouls. If we extract the combination of those four out of the same counts that underlie the RPR as a whole, we get what we are going to call the Scored Defending Contribution. This could also be thought of as Tracked Defending Contribution if you prefer. Then if we divide this by minutes, we can have a Scored (or Tracked) Defending Rating.

Finally, if we combine Hidden Defending Rating (HDR) with Scored Defending Rating (SDR) we can have an Overall Defending Rating (ODR). I am for now going to simply multiply the HDR by two and add that to the SDR to yield the ODR. To combine them this way is more arbitrary than my usual standards allow; I am doing this because there is as of yet no non-arbitrary way of doing it. The formula of two times HDR plus SDR brings HDR almost up to par with SDR in terms of the actual numbers and the averages of those numbers involved.

In other words, I am saying for now that hidden defending is almost as important as scored defending. There appear to be many coaches and not a few hardcore basketball fans who believe that hidden defending is actually more important than scored defending, but I am very likely never going to agree with that. I think that although hidden defending is important, and plausibly almost as important as tracked defending, that it is like a quicksand in that there seems to be a tendency for a substantial minority of basketball people to get carried away with estimating the importance of it and then become more and more trapped by their error in terms of how they look at basketball or in terms of how they coach their team if they are coaching.