What Do Individual Win-Loss Records Mean?

Dateline: 11/10/98

IT IS DIFFICULT TO SAY WHETHER AN INDIVIDUAL player can "win" at a team game like basketball. Broadcasters like to say that Michael Jordan won a game by himself or that Patrick Ewing took over a game (usually with regard to a basketball game, not a game of chicken with owners), but no coach will ever say that any individual wins a game on his own. Even Wilt Chamberlain's 100 point game in 1962, which may have been the greatest individual game in NBA history, couldn't stand on its own to win the game for Philadelphia; the opposing New York Knicks scored 147, so Philly needed the extra 69 points thrown in by other guys.

Despite the impossibility of the basic concept, there are a lot of ways to associate wins and losses with players. Each of these ways implies different things about players and all have strengths and weaknesses. Sometimes they agree about the relative strengths of players; sometimes they don't.

...Method 1: Using Team Record...

The most conservative way to evaluate a player's win-loss record is to look at his team's record when he plays, something like they do for quarterbacks in football and starting pitchers in baseball. But this means that players like Ron Harper and Jordan had the same record in 1997-98, both playing 82 games for the 62-20 Bulls. If this continues for a few years, as it did for Kurt Rambis and Magic Johnson in the early '80's, you could have two clearly unequal players with quite equal long-term records.

The leaders in 1997-98 using this method are shown here:

Rank Player Team Wins Losses Win% G Above
0.500
1 Adam Keefe uta 61 19 0.763 42
2 Ron Harper chi 62 20 0.756 42
2 Michael Jordan chi 62 20 0.756 42
2 Shandon Anderson uta 62 20 0.756 42
2 Howard Eisley uta 62 20 0.756 42
2 Bryon Russell uta 62 20 0.756 42
7 Karl Malone uta 61 20 0.753 41
8 Toni Kukoc chi 57 17 0.770 40
9 Scott Burrell chi 60 20 0.750 40
9 Dennis Rodman chi 60 20 0.750 40
9 Eddie Jones lal 60 20 0.750 40
9 Greg Anthony sea 60 20 0.750 40
9 Jeff Hornacek uta 60 20 0.750 40
14 Derek Fisher lal 61 21 0.744 40
14 Rick Fox lal 61 21 0.744 40
14 Vin Baker sea 61 21 0.744 40
14 Hersey Hawkins sea 61 21 0.744 40
14 Gary Payton sea 61 21 0.744 40
19 Dale Ellis sea 59 20 0.747 39
20 Elden Campbell lal 60 21 0.741 39
20 Sam Perkins sea 60 21 0.741 39
22 John Stockton uta 51 13 0.797 38
23 Detlef Schrempf sea 58 20 0.744 38
23 Greg Foster uta 58 20 0.744 38
25 Kobe Bryant lal 58 21 0.734 37

Click here for full listing of numbers

...Method 2: Game-by-Game Offensive and Defensive Ratings...

A couple of the big developments of JoBS are the individual offensive and defensive ratings. These simply reflect a player's relative offensive and defensive efficiencies. Almost by definition, if a team has a higher offensive rating than defensive rating in a game, it wins. We can apply that definition to individual ratings -- if an individual has a higher offensive than defensive rating in a game, it is considered a win -- and arrive at the following win-loss records for individuals in 1998:

Rank Player Team Wins Losses Win% G Above
0.500
1 Karl Malone uta 63 18 0.778 45
2 David Robinson san 58 15 0.795 43
3 Reggie Miller ind 62 19 0.765 43
4 Michael Jordan chi 61 21 0.744 40
5 Tim Hardaway mia 58 23 0.716 35
6 Eddie Jones lal 57 23 0.713 34
7 Wesley Person cle 57 25 0.695 32
8 Mark Jackson ind 56 26 0.683 30
9 Chris Mullin ind 56 26 0.683 30
10 Hersey Hawkins sea 56 26 0.683 30
11 Steve Kerr chi 39 11 0.780 28
12 Shaquille O'Neal lal 44 16 0.733 28
13 Danny Manning pho 49 21 0.700 28
14 Ron Harper chi 55 27 0.671 28
15 Tim Duncan san 55 27 0.671 28
16 John Stockton uta 45 19 0.703 26
17 Brevin Knight cle 53 27 0.663 26
18 Jeff Hornacek uta 53 27 0.663 26
19 Charlie Ward nyk 54 28 0.659 26
20 Gary Payton sea 54 28 0.659 26
21 Arvydas Sabonis por 49 24 0.671 25
22 Dikembe Mutombo atl 53 29 0.646 24
23 Kerry Kittles njn 50 27 0.649 23
24 Nick Van Exel lal 43 21 0.672 22
25 Avery Johnson san 48 26 0.649 22

Click here for full listing of numbers

Alternatively, since defensive ratings are more difficult to determine and since offense is more under individual control, one might calculate an offensive winning percentage by summing the number of times a player's offensive rating beats his team's defensive rating. This is shown below for 1998.

Rank Player Team Wins Losses Win% G Above
0.500
1 Reggie Miller ind 63 18 0.778 45
2 Michael Jordan chi 63 19 0.768 44
3 Tim Hardaway mia 61 20 0.753 41
4 Wesley Person cle 60 22 0.732 38
5 Steve Kerr chi 43 7 0.860 36
6 Detlef Schrempf sea 56 22 0.718 34
7 Eddie Jones lal 57 23 0.713 34
8 Avery Johnson san 53 21 0.716 32
9 Hersey Hawkins sea 57 25 0.695 32
10 Nick Van Exel lal 47 17 0.734 30
11 Dale Ellis sea 54 25 0.684 29
12 Karl Malone uta 55 26 0.679 29
13 Jeff Hornacek uta 54 26 0.675 28
14 Joe Dumars det 49 23 0.681 26
15 Steve Nash pho 50 24 0.676 26
16 Ron Harper chi 54 28 0.659 26
17 Mark Jackson ind 54 28 0.659 26
18 Gary Payton sea 54 28 0.659 26
19 Steve Smith atl 49 24 0.671 25
20 John Stockton uta 44 20 0.688 24
21 Danny Manning pho 47 23 0.671 24
22 Toni Kukoc chi 49 25 0.662 24
23 Kerry Kittles njn 50 27 0.649 23
24 Chris Mullin ind 52 30 0.634 22
25 David Robinson san 47 26 0.644 21

Click here for full listing of numbers

...Method 3: Pythagorean Comparison of Season Ratings...

The original way to model players' win-loss records was to interpret their numbers the same way that team numbers are interpreted -- applying the Pythagorean Method to individual offensive and defensive ratings. For example, if a player had an offensive rating of 110 and a defensive rating of 105, that player would be projected to have the same winning percentage as a team that had those ratings, which is also the same winning percentage as a team that scored 110 ppg and allowed 105 ppg.

The Pythagorean Method doesn't, however, evaluate the "number of games" played by an individual, just the winning percentage. This, frankly, is one of the biggest conceptual hurdles I have faced and still face; I cannot come up with a great theory for it. What I have done is to estimate "number of games" several different ways and combine them. The number of games a player is responsible for is evaluated by looking at

  1. how many minutes they play,
  2. how many games they start,
  3. how many defensive stops they have, and
  4. how many offensive possessions they are responsible for.
For example, if a player plays 3000 minutes out of a total 19705 played by his team, that player is given 15.2% of the 82 total games played, or 12.5 games. This type of calculation is carried out for all four factors; the fourth factor is weighted 50% and the others are weighted 16.7% each. Typically, one finds that players are responsible for up to about 18 games a year.

As already mentioned, it's hard to say what being "responsible for 18 games a year" means, but there is a nice little reality check in the estimate. By setting the number of games the way this way, the total number of wins and losses produced by all individuals on a team should closely approximate the team win-loss record, a constraint not placed on the prior two methods.

The leaders in this category for 1997-98 are shown below:

Rank Player Team Wins Losses Win% G Above
0.500
1 Jordan, Michael Chi 16.0 1.8 0.897 14.2
2 Malone, Karl Uta 15.3 1.5 0.911 13.8
3 Robinson, David San 13.2 0.6 0.953 12.6
4 Duncan, Tim San 14.3 1.9 0.880 12.3
5 Payton, Gary Sea 13.7 1.9 0.877 11.8
6 Hardaway, Tim Mia 13.1 2.1 0.860 11.0
7 Person, Wesley Cle 11.4 0.8 0.937 10.7
8 Miller, Reggie Ind 10.9 1.0 0.919 9.9
9 Sabonis, Arvydas Por 10.8 1.0 0.914 9.8
10 Jones, Eddie Lal 11.0 1.3 0.892 9.6
11 O'Neal, Shaquille Lal 10.9 1.4 0.883 9.5
12 Mutombo, Dikembe Atl 11.1 1.7 0.867 9.4
13 Schrempf, Detlef Sea 10.9 1.5 0.880 9.4
14 Hawkins, Hersey Sea 9.7 0.6 0.938 9.1
15 McDyess, Antonio Pho 10.3 1.6 0.868 8.7
16 Ilgauskas, Zydrunas Cle 9.9 1.7 0.855 8.2
17 Ward, Charlie Nyk 9.1 1.1 0.897 8.1
18 Knight, Brevin Cle 9.4 1.6 0.855 7.8
19 Hornacek, Jeff Uta 9.0 1.2 0.879 7.8
20 Mullin, Chris Ind 8.1 0.7 0.919 7.4
21 Mason, Anthony Cha 9.8 2.5 0.794 7.3
22 Webber, Chris Was 10.7 3.6 0.747 7.1
23 Hill, Grant Det 12.1 5.1 0.704 7.0
24 Lenard, Voshon Mia 8.9 2.0 0.820 6.9
25 Kittles, Kerry Njn 9.2 2.4 0.792 6.8

Click here for full listing of numbers

As done above in Method 2, one can look only at an individual's offensive rating and the team's defensive rating, a method I have traditionally called the offensive winning percentage, following Bill James. In this case, you replace in the Pythagorean method an individual's defensive rating with his team's defensive rating. This doesn't incorporate the individual contributions a player makes on defense, hurting defensive players like Patrick Ewing, Alonzo Mourning, Gary Payton, etc. This was originally developed before I had individual defensive ratings and, because offense still appears to be the predominant factor in evaluating players, it still has validity. You'll notice that the list below is actually quite similar to the one above...

Rank Player Team Wins Losses Win% G Above
0.500
1 Jordan, Michael Chi 19.3 2.2 0.899 17.1
2 Hardaway, Tim Mia 15.2 1.9 0.890 13.3
3 Payton, Gary Sea 14.7 2.2 0.869 12.5
4 Malone, Karl Uta 15.9 3.6 0.815 12.3
5 Miller, Reggie Ind 12.8 0.6 0.957 12.2
6 Robinson, David San 12.7 2.2 0.851 10.5
7 Smith, Steve Atl 12.2 2.0 0.860 10.2
8 Person, Wesley Cle 10.0 0.5 0.951 9.5
9 Jones, Eddie Lal 10.4 1.1 0.907 9.3
10 Schrempf, Detlef Sea 10.6 1.4 0.885 9.2
11 Hornacek, Jeff Uta 9.7 1.1 0.899 8.6
12 Rice, Glen Cha 12.4 4.1 0.754 8.3
13 Duncan, Tim San 12.4 4.5 0.733 7.9
14 Van Exel, Nick Lal 8.3 0.6 0.936 7.8
15 O'Neal, Shaquille Lal 11.0 3.3 0.771 7.7
16 Lenard, Voshon Mia 8.7 1.0 0.896 7.7
17 Dumars, Joe Det 8.3 0.8 0.916 7.5
18 Stockton, John Uta 8.0 0.6 0.932 7.4
19 Jackson, Mark Ind 8.7 1.3 0.867 7.4
20 Johnson, Avery San 8.6 1.3 0.865 7.3
21 Kittles, Kerry Njn 9.6 2.3 0.805 7.2
22 Mullin, Chris Ind 7.9 0.7 0.920 7.2
23 Richmond, Mitch Sac 11.0 3.8 0.744 7.2
24 Murray, Tracy Was 8.8 1.6 0.842 7.1
25 Kukoc, Toni Chi 8.7 1.7 0.834 7.0

Click here for full listing of numbers

...So Which Is Best?...

I frankly wouldn't have presented all three methods above if I didn't think they all were good. "Which is best" is a question with no answer. It's like asking a guy whether a hammer is better than a screwdriver.

We can look at all methods and see that certain players are on all the lists: Jordan, Malone, Payton, and Eddie Jones. These are some of the best players in the league, Jones not quite in the class the other three are. Other players that almost make all lists include John Stockton, Reggie Miller, David Robinson, Tim Duncan, Tim Hardaway, and Shaquille O'Neal. Scottie Pippen doesn't make it on any lists, but is close for all of them.

Each of the methods has its strengths and weaknesses, some of which are obvious from the numbers and explanations above. But for those dolts who don't see them and for those smarter people looking for subtleties, I've summarized what I think are strengths and weaknesses below. I've also included a "constraint" column; constraining methods like these to have some measurable form of reality is a vital thing in my mind and something I strive for in much of what I develop.

Method Advantages Disadvantages Constraints
1. Team Record Unambiguous to calculate, longterm value probably pretty good. Possibility that good and bad players with same record, not good for short periods of time or for differentiating players that always play together, difficult to calculate without game-by-game numbers, doesn't add up to team win-loss record, poor representation of "number of games" a player is responsible for. Maximum number of wins and losses for any player equal those of the team.
2. Game-by-Game Comparison of Ratings Interesting concept, incorporates individual efficiency and number of possessions used by individuals, can be approximated with season stats. Hard to calculate precisely, doesn't add up to team win-loss record, poor representation of "number of games" a player is responsible for. Not clear. There is no obvious rule for checking these numbers as there is for the others.
3. Pythagorean Comparison of Season Ratings Easy to calculate with just season stats, adds up to team win-loss record, theory of "if this player were a team, its record would be…" is convenient Unclear how many "games" played, not clearly measurable for individuals (though team total is), uses Pythagorean Method that is empirical for teams (though can be extended to use Correlated Gaussian Method). Total number of wins and losses by teammates should approximately equal the win-loss record of the team, or at least the Pythagorean Projection or Correlated Gaussian Projection for the team.
...Some Final Comments...
  1. Method 1: When I get this data for the last 15 years, I will take a close look at it to see how well it does over the long term. For the time being, it serves as a nice piece of measurable evidence when people argue about players' abilities. For example, the Minnesota Timberwolves did better without Stephon Marbury in his rookie year than with him. For example, the 1998 Atlanta Hawks were 3-5 without Christian Laettner and 8-5 without Alan Henderson. For example, the Sacramento Kings were 26-44 with Mitch Richmond and 1-11 without him; the Washington Bullets were 39-32 with Chris Webber and 3-8 without him. Hmmmm.
  2. Method 2: I personally like this one, even though there is no real constraint on the values. The sum of teammates' totals doesn't equal the team totals, which is an important goal, but this win-loss number tells you how often a player is really doing their part for the team. I am afraid that it highlights individualism, but I honestly believe (perhaps naively) that the goals of most players are to win first, have fun second, and showcase themselves third. I say this in the face of labor strife where greed is at the forefront -- you don't need to remind me.
  3. Method 3: Since it has been easy to calculate, this has always been my method of choice. Until I complete a database with boxscores going back through time, this will likely continue to be the method of choice. Method 2 can be approximated with season stats and adjusted to be more like Method 3, meaning that it may eventually overtake this method, but it won't be real soon.

Acknowledgements

The game-by-game statistics used in this column are courtesy of Doug Steele, who provides a tremendous service.