Winval Stats

[aexchange]

Very Cool Article and good read. I've bolded the relevent parts to the mavs.

Calculating success for Bulls



May 6, 2003

BY CAROL SLEZAK SUN-TIMES COLUMNIST


They're not official members of the Dallas Mavericks, but Wayne L. Winston and his partner Jeff Sagarin might as well be. For the last three seasons, Mavericks owner Mark Cuban has used a player rating system designed by Winston and Sagarin to to help determine lineups and make personnel decisions. During that time, the Mavs have gone from a perennially bad team to one of the league's best teams. Not to imply that Winston and Sagarin deserve all the credit, or even most of the credit, for the Mavs' success. But they surely deserve some.

By beating the Portland Trail Blazers in Game 7 of their playoff series Sunday, the Mavericks advanced to meet the Sacramento Kings in a series that begins tonight in Dallas. Winston, a professor of decision sciences at Indiana (Cuban once took a class from him), stopped game-planning for the Kings long enough to answer a question for me: Could you (please) help the Bulls?

His answer was as intriguing as his system, which is called Winval and is based on a simple premise: Players are rated not by points or rebounds or assists, but by how their team performs while they are on the court. If a player's team outscores its opponent while he's on the court, he gets credit. If the opposition outscores his team while he's on the court, he loses credit.

Winston and Sagarin, a sports statistician, both majored in math at the Massachusetts Institute of Technology, so you'll understand why I am sparing you the details of the genius behind Winval. In truth, I'm not sure I could explain it if I tried. But I can explain that Winston and Sagarin have recorded every substitution in every NBA game for the last four seasons, allowing them to rate individual players while controlling for everyone else on the court. From this information, they can determine a team's most effective player, its best lineup, its best combination of any two or three (or four) players, its best fourth-quarter players, its most effective lineup for a particular opponent or anything else a team might want to know. (At the oft-fined Cuban's request, they also keep track of referees' tendencies.)

''We know more about the NBA than anybody else,'' Winston said. ''We can look at this information every conceivable way you want to look at it.''

If you watched Sunday's Mavericks game, you saw that Eduardo Najera started in place of Shawn Bradley. Why? Maybe because Winston advised the Mavs coaching staff that they needed the unsung Najera on the court.

''He makes the Mavs a better team,'' Winston said. ''The stats you see in the paper don't reflect what really is happening on the court. There's no stat for setting a pick, switching well or hustling back on defense. Najera always disrupts the opponents' offense.''

According to Winval, Kevin Garnett and Tim Duncan were the two best players in the league this season. No surprise there, but a few of the others who cracked the top 10--Dirk Nowitzki, Scottie Pippen, Richard Hamilton, Stephon Marbury, Kobe Bryant, Zydrunas Ilgauskas, Jason Terry and Jon Barry--might surprise you. And some of the names absent from the top 10, including Shaquille O'Neal, Allen Iverson and Tracy McGrady, might surprise you as well.

''McGrady has all the stats, but the Orlando Magic don't do that horribly when he's out of the game,'' said Winston, who correctly predicted the Magic would lose to the Detroit Pistons in their playoff series. ''The Magic give up too many points when he's in the game.''

Winston believes every team has an effective lineup, but coaches often don't know which lineup that is. Winston knows. Although he barely paid attention to the Bulls this season--who outside of Chicago did?--he made a Bulls lineup calculator for me. Sure enough, the Bulls had a couple of winning combinations last season. The trouble was, coach Bill Cartwright seldom played either combination.

Cartwright's favorite lineup in terms of minutes played--Tyson Chandler, Trenton Hassell, Donyell Marshall, Jalen Rose and Jay Williams--had a negative rating (with zero meaning average). In fact, Cartwright's top four favorite lineups had a negative rating, the worst being the lineup of Chandler, Eddy Curry, Hassell, Rose and Williams, which played 233 minutes over the course of the season and had a negative 21.94 rating (that's very bad).

There were lineups Cartwright used, however, that were above average compared to the rest of the NBA. Chandler, Jamal Crawford, Curry, Marshall and Rose was a winner, with a rating of 4.62, yet that combination played together only 150 minutes. Another winning combination, with a rating of 8.44, consisted of Crawford, Curry, Hassell, Marshall and Rose. But these five played together only 59 minutes.

''Why would a coach play a bad lineup more than a good lineup?'' Winston said. ''You can't tell me [Cartwright] knew the lineups he played were that bad.''

There were a few positives that emerged about the Bulls. For instance, during the last five games of the season, Curry was rated among the top 15 players in the league. (And don't forget, Winval doesn't care about how many points a player scores, but about how his team fares while he's on the court.)

Winval's impact rating, which measures whether a player's contributions came when the game was close, ranked Fred Hoiberg, Corie Blount, rookie Roger Mason Jr. and Crawford as the Bulls' best. Rose and Marshall finished at zero, or average.

Several players, including Rose, Curry, Crawford and Williams, had positive fourth-quarter ratings. Crawford also had a big in/out differential, which Winston said can be used as a quick check of a player's effectiveness. When Crawford was on the bench, the Bulls were outscored by a significantly wider margin than when he was in the game.

While Cartwright resisted the idea of playing Crawford and Williams together, Winval indicates that it's a good idea.

''By far the best tandem the Bulls had was Crawford and Williams in the same backcourt,'' Winston said. ''They didn't get a lot of time together, but they were really good together. The Bulls were nine points better than the average team when those two played together.''

And the trio of Rose, Crawford and Williams? Another winner, with a positive 15. But that combination played only 95 minutes together.

''I guarantee the Bulls don't know how well [Rose, Crawford and Williams] played together,'' Winston said.

Consider them told. And there are plenty more details Winston can share with general manager John Paxson, if he's interested. The service isn't free, but what price playoffs? At a cost of less than $200,000, it's worth a look. And it could be the best money Paxson ever spends.

[MavKikiNYC]

Why are all the great combinations the ones that hardly played together?

Is that some kind of insufficient data phenomenon?

[Murphy3]

in most cases, probably so.
and, you never know the context in which the lineups were played..whether it was after a game was out of reach or whatever

[nowitzki_prophecy]

how could he not include Shaq?remember the lakers at the beginin of the season with Kobe as their guide?!it was pathetic,they were one from last at their Division,second only to the Nuggets.
even if not on the court,apparantly shaq adds some much needed morale support,of course besides his 27.5/11 per game,second in the NBA in FG%.

[Marv Albert]

I think its funny how they cite these guys predicting the Pistons (#1 seed) would beat Orlando (#8 seed) as evidence of how smart they are.

And those lesser used lineups being better in the ratings doesn't surprise me. The Bulls are a bad team, whichever lineup they use the most is going to be the one they lose the most points (and games) with. They don't have any players who can put up 35+ minutes of great play and they lose almost twice as many games as they win, that basically guarantees a bad rating for whoever plays the most minutes for them.

As mentioned by someone else, context is everything. If their highest rated lineup got most of their action in against the opposition's 2nd team during blowouts, then that data is worthless because thats not anywhere close to being as relevant as what happens in a close game where you keep a certain lineup out there because they give you the best chance to win. The system is greatly flawed because it doesn't appear to factor in context, 5 minutes of crunch time appears to be worth the same as 5 minutes of garbage time. If the Bulls are down 30 with 5 minutes to go and Jay Williams hits 4 meaningless three pointers in that period, he could be rated higher than a game where Jalen scores 8 hard earned points in the last 5 mins to bring them to a win.

[Fidel]

<< As mentioned by someone else, context is everything. If their highest rated lineup got most of their action in against the opposition's 2nd team during blowouts, then that data is worthless because thats not anywhere close to being as relevant as what happens in a close game where you keep a certain lineup out there because they give you the best chance to win. The system is greatly flawed because it doesn't appear to factor in context, 5 minutes of crunch time appears to be worth the same as 5 minutes of garbage time. >>


If the system is really in use to evaluate players or lineups I think they´d somehow try to correct the effects you describe, or at least factor them in. After all the Mavs for example have such a huge coaching staff and those guys are no dummies. I´m sure they would notice these effects and find a way to work with the data in a usefull way.