The idea of formulating a numeric model that takes into account a player's worth based off of his current or historic statistics is hardly revolutionary, but the explanation of that data is what is rarely presented accurately. However, in this analysis a concerted effort will be made to present both the strengths and weaknesses of not only our techniques, but also lend insight into the processes that other media sources use to create their player rankings.
ESPN and the Wages of Wins Journal both publish statistical evaluations of NBA players and their statistical production. These two media sources calculate and analyze data using unique methods that define efficiency and a player's win value based on his statistical worth. ESPN writer John Hollinger is widely known for his work demonstrating a player's PER (Player Efficiency Rating). It is essentially a means of calculating a player’s performance and production given the chances he has to produce. It's a heavily marketed measure of efficiency endorsed by ESPN, but the question should be asked, does it provide an accurate description of a player’s actual efficiency? That answer is a resounding no. Hollinger himself has historically made mention that defensive efficiency is difficult to calculate based solely off of a player’s statistical production. Unfortunately for the PER and other efficiency measures, the same can be said about the offensive end of the court. Although most of the categories analyzed and calculated are offensive statistics, efficiency is difficult to diagnose because the toughest aspect of sports statistical evaluations is assigning an accurate number to a non-numerical value. For example, it's incredibly difficult to quantify whether a player makes the right pass in a given situation if the player he passes to fails to convert the opportunity.
Hollinger's formula is based in, or at least appears to be influenced by, baseball's accepted statistical formula referred to as Sabermetrics. While circumstances may exist in baseball where one player's correct play may not result in a positive result, the infrequency of plate appearances and the pace of play limit this inaccuracy from consistently occurring. However, in basketball, the pace of play and scoring frequency allows decision making and other important elements to be over-looked or inaccurately represented. Ultimately, the problem with the PER is it allows Player A's efficiency total to be affected and even determined by Player B's production. Would John Stockton have averaged 10.5 assists per game over his 19 year NBA career if he didn't have Karl Malone to pass to? Would Jose Calderon have averaged 11.6 assists per game if he, instead of Chris Paul, played in the New Orleans system? Although these questions are difficult to answer, they are impossible to answer correctly using result-based statistics. Another problem with a league-wide statistical model is its rigidity when factoring in different positional responsibilities. Proponents might claim that the formula is an unbiased determinant based solely on a players statistical merit, and while that might be true, it falls short of doing what it is intended to do and that's determining a player's efficiency.
Unlike Hollinger's PER, the Wages of Wins computations are designed to analyze a players on-court efficiency as a contributing factor in his team's win total, rather than the player’s individual ranking. It is an intriguing technique that can be used not only as a summation of a player’s previous performance, but also as a predictive tool to gage a players future potential. Obviously, predictions are not based in actual data, but if certain variables can be held consistent, their formula could be used to value a player’s future benefit to an organization. On the surface, it may be difficult to determine the difference between these two evaluations of efficiency, but there are some distinct fundamental differences. Within the Wages of Wins findings, it appears that starters maintain a higher value than do bench players largely because of their increased time on the court. This logical assumption that starters contribute and play at a higher level than bench players is not consistently maintained in the PER system.
As for our statistical formula, we believe that players who play different positions can not accurately be compared or analyzed using a singular fixed statistical formula. Although the nine major statistical categories remain constant in our formula, the weight or value placed on each statistical category varies from one position to the next. In simplistic terms, our opinion identifies assists as a more accurate measure of a point guard's skills than they are of a power forward or center. Transversely, the opposite would hold true when analyzing rebounds and block shots. Therefore, to establish a true measure of a players worth or statistical rank based exclusively from his statistical production, the variation between positional responsibilities should be considered.
Each position was investigated and analyzed using it's own specific formula, or weighted system of measurement. We have produced player rankings for the point guard position for each of the last seven seasons, and this year, the decision was made to expand our statistical evaluations to the shooting guard, small forward, power forward and center positions. Over the past seven seasons, the process has remained consistent. However, our statistical category weights have undergone subtle adjustments. In addition, this year we decided to scale back the total number of players evaluated from 50 to only the players who averaged over 30 minutes per contest.
STATISTICAL CATEGORIES
Assists Per Game
Blocks Per Game
Field Goal Percentage
Free Throw Percentage
Points Per Game
Rebounds Per Game
Steals Per Game
Three Point Percentage
Turnovers Per Game
These nine categories then received a value weight in accordance with our expectation of success for each position on the basketball court. This process was duplicated for each of five positions: point guard, shooting guard, small forward, power forward and center. Therefore, each position received its own unique set of weighted categories to establish a position specific ranking. Below is the weighted categorical scale for the point guard position.
POINT GUARD WEIGHT
5.000 Assists Per Game
4.500 Turnovers Per Game
4.000 Two Point Percentage
3.500 Free Throw Percentage
3.000 Points Per Game
2.500 Steals Per Game
2.000 Three Point Percentage
1.500 Rebounds Per Game
1.000 Blocks Per Game
Finally, the last limiting factor used to restrict sample size is minutes played per game. This is important, since all of our usable data is in per game units. To accurately compare players within a specific group or position, a base level needed to be established to maintain as much statistical integrity as possible. With the player pool size set to include only the players who played 30 plus minutes a game, the players were then ranked from high to low in each of the nine categories. The leader in each statistical category was given a numerical value of 1.000. The remaining players in each category then had their actual statistical value divided by the leaders actual statistical output to determine a usable percentage. In simple terms, we created a curve based off of the statistical leaders output in that category, similar in nature to the curves employed in high schools and college classrooms. With the nine categories ranked, the next step was to multiply the positional weighting system to each of the nine statistical categories. For example, Baron Davis was the highest scoring point guard in the league last season, so he earned the 1.000 category rank. This 1.000 was multiplied by 3.000 (the weight, or importance quotient designated for the category; Points Per Game) to give Baron Davis the category lead at 3.000. This technique was applied for each player, in each statistical category with only the positional category weight or importance quotient changing.
What this formula lacks in complexity, it gains in simplicity and workability. The biggest factor is it presents the flexibility to adjust the weight of the nine statistical categories based on the positional production expectations. However, this formula is not perfect and does include certain statistical pitfalls. The biggest statistical problem with this particular analysis is the difference in scope between the nine statistical categories. For example, the difference between the player with the highest two point percentage (Jose Calderon 55.1) and the lowest percentage (Jason Kidd 38.7) is 16.7 percent. However, the difference between the player with the highest three point percentage (Steve Nash 47.0) and the lowest percentage (Andre Miller 8.8) is 38.2 percent. Although, the statistical integrity is maintained, the computations could be considered misleading. This discrepancy is further exaggerated if you consider the three point category as a player preference rather than a necessity. Players like the aforementioned Miller, Monta Ellis, Tony Parker and Dwyane Wade all prove that a player can be effective in scoring despite the inability to consistently knock down three point field goals. Also, much like ESPN's PER and the Wages of Wins calculations, our analysis is limited to actual statistical data. The same problems exist in our ranking system as does in theirs. However, we are not marketing our product as a true measure of efficiency. Rather, our system is simply a guide to compare a players historical production.
That leads us to our biggest complaint with ESPN and John Hollinger's Player Efficiency Rating. We have no problem with it's function, but rather the misrepresentation of the product itself. The PER is in basic terms a statistical formula that computes limited source data. It can not determine efficiency because efficiency by definition is: "The ratio of the effective or useful output to the total input in any system." By definition, the player or team's output can be categorized (although we have difficulty accepting that assessment as we discussed earlier), but in no way has, nor can Hollinger or ESPN give an accurate account of a player's input into the game or "system." Their analysis is inherently flawed because they do not have enough data to create an accurate model to gage efficiency. Also, it is imperative to understand that the slightest tweak, omission or addition of data to a formula can alter the final results in any statistical model. This holds true in PER calculations as well as in our 2008 NBA Player Rankings and all other player evaluations.
However, we do believe that such an assessment could be calculated, but to do so, the statistician would need to work closely with an organization to properly and accurately evaluate not only the end result, but also the decision-making that leads up to the factual end-point data. To do this on a league-wide scale would be nearly impossible, but on a smaller scale or on a team-by-team basis, it seems not only plausible, but a realistic possibility. Under the assumption that you could create a consistent, dependable and, most importantly, accurate evaluation formula, it would be conceivable to project a player's future value or success within an organization given that certain variables are also defined. This would also include psychological and behavioral tenancies.
Finally, we believe that all rankings or ratings published by media outlets should be considered only slightly more informative than a sports journalist's opinion, but an accurate evaluation could be created and implemented if given full access to all of the necessary information.
| POINT GUARDS | RATING | SHOOTING GUARDS | RATING | |
| Chris Paul | 21.949 | Kobe Bryant | 22.043 | |
| Baron Davis | 19.840 | Allen Iverson | 21.168 | |
| Jose Calderon | 19.505 | Dwyane Wade | 20.306 | |
| Deron Williams | 19.339 | Manu Ginobili | 20.156 | |
| Steve Nash | 19.313 | Jason Richardson | 20.007 | |
| Jason Kidd | 18.563 | Vince Carter | 19.991 | |
| Jason Terry | 18.562 | Mike Dunleavy | 19.591 | |
| Chauncey Billups | 18.455 | Joe Johnson | 19.450 | |
| Maurice Williams | 17.723 | Brandon Roy | 19.163 | |
| Monta Ellis | 17.671 | Kevin Martin | 19.150 | |
| Devin Harris | 16.979 | Ray Allen | 18.769 | |
| Kirk Hinrich | 16.763 | Tracy McGrady | 18.715 | |
| Jamaal Tinsley | 16.670 | Mike Miller | 18.685 | |
| Raymond Felton | 16.602 | Michael Redd | 18.685 | |
| Andre Miller | 16.488 | Anthony Parker | 18.677 | |
| Gilbert Arenas | 16.405 | Richard Hamilton | 18.549 | |
| Tony Parker | 16.343 | Jamal Crawford | 18.490 | |
| Antonio Daniels | 16.135 | Raja Bell | 18.480 | |
| Mike Bibby | 16.100 | Kevin Durant | 18.149 | |
| Beno Udrih | 16.047 | Ben Gordon | 18.075 | |
| Rafer Alston | 15.960 | John Salmons | 17.287 | |
| Daniel Gibson | 15.934 | Cuttino Mobley | 17.178 | |
| Randy Foye | 15.807 | DeShawn Stevenson | 16.685 | |
| Stephon Marbury | 15.443 | |||
| Sebastian Telfair | 15.379 |
| SMALL FORWARD | RATING | POWER FORWARD | RATING | |
| LeBron James | 22.740 | Amare Stoudemire | 21.112 | |
| Caron Butler | 19.702 | Shawn Marion | 20.752 | |
| Carmelo Anthony | 19.253 | Josh Smith | 20.454 | |
| Andre Iguodala | 18.495 | Dirk Nowitzki | 19.884 | |
| Ron Artest | 18.399 | Kevin Garnett | 19.751 | |
| Gerald Wallace | 18.170 | Tim Duncan | 19.548 | |
| Paul Pierce | 17.903 | Chris Bosh | 19.491 | |
| Hedo Turkoglu | 17.782 | Pau Gasol | 19.116 | |
| Rudy Gay | 17.635 | Antawn Jamison | 19.105 | |
| Danny Granger | 17.577 | David West | 18.955 | |
| Josh Howard | 17.382 | Lamar Odom | 18.835 | |
| Andrei Kirilenko | 17.010 | Carlos Boozer | 18.662 | |
| Corey Maggette | 16.877 | Rasheed Wallace | 18.283 | |
| Stephen Jackson | 16.742 | Rashard Lewis | 17.534 | |
| Luol Deng | 16.629 | Elton Brand | 17.150 | |
| Grant Hill | 16.596 | LaMarcus Aldridge | 17.140 | |
| Richard Jefferson | 16.578 | Emeka Okafor | 17.069 | |
| Peja Stojakovic | 16.118 | Zach Randolph | 17.017 | |
| Shane Battier | 15.941 | Drew Gooden | 16.731 | |
| Tayshaun Prince | 15.783 | Kenyon Martin | 16.493 | |
| Ricky Davis | 15.223 | Udonis Haslem | 15.483 | |
| Marvin Williams | 14.876 | Tim Thomas | 15.042 | |
| Bruce Bowen | 13.275 |
| CENTER | RATING | |||
| Marcus Camby | 20.366 | |||
| Dwight Howard | 19.784 | |||
| Yao Ming | 18.775 | |||
| Al Jefferson | 18.539 | |||
| Chris Kaman | 18.315 | |||
| Brad Miller | 17.820 | |||
| Andrew Bogut | 16.741 | |||
| Mehmet Okur | 16.716 | |||
| Tyson Chandler | 16.313 | |||
| Zydrunas Ilgauskas | 15.989 | |||
| Samuel Dalembert | 15.836 | |||
| Al Horford | 15.036 | |||
| Ben Wallace | 14.737 |
1 comments:
Did Andrew Bynum get lost somewhere?
Post a Comment