Wayne Winston is a professor at Indiana University and for the last nine years he has been Mark Cuban's stat guru for the Dallas Mavericks. Winston's recently published book "Mathletics," explains much of his work -- complete with formulas and spreadsheets. This is the first in a series of TrueHoop posts in which Winston explains the surprising things he has learned about what works and what doesn't in the NBA.
Imagine you live on an island, with 13 people, and one of them is murdered.
Murderers are usually found (or not) by assembling all the available clues and seeing if they point to anybody. Sometimes they do, and sometimes they don't.
Another way to look at this one, however, would be to say, look, we know one of these 12 people did this thing. Let's try to find out what everybody was doing at the time of the murder, and then we can start making smart guesses at who was responsible.
This is a messy analogy for the state of basketball statistics.
A team loses a game. That's your murder.
The box score is the trail of clues. John Hollinger's PER is the embodiment of what can be learned from that.
But there are some cases where PER doesn't tell it all -- maybe you have no suspect! We all know that some chunk of what matters in basketball doesn't make it into the box score.
So there are people like Winston who instead favor saying we know all the players from both who were on the court while the loss occurred. Let's try to break the game apart, into little pieces, to see who gets the blame.
The result is adjusted plus/minus. Winston is one of many -- others include Dan Rosenbaum, Aaron Barzilai, Stephen Ilardi -- who basically look at the scores of games, and then use complex formulas to assign credit and blame for that happened to individual players.
It's often derided as an imprecise process, but it's worth noting that Winston, Barzilai and Rosenbaum all work for NBA teams. Winston has his detractors, but he's adamant, and often convincing, that such work can yield fascinating results:
The many new kinds of basketball statistics tend to fall into two groups. There are things that we can, with certainty, ascribe to individual players. Those things are mostly in the box score, or PER. But you put them all together, and a chunk of the game is missing. Then there's stuff we know the team does -- the final score, and the increments of it we see in +/- and adjusted +/-. The trouble there is that it's hard to know how to assign what the team does to an individual player. It's murky in both camps. But you're an adjusted +/- guy, right? Why?
Basketball is half offense and half defense. I don't think I have to prove that mathematically. It's got to be true. The box score is not half offense and half defense. I think that's where the box score breaks down.
The nice thing about adjusted +/- is that it's half offense and half defense. I think if we can estimate offense accurately, and most of the adjusted +/- stuff that is out there for offense agrees with the rest of the world.
It's on the defensive ratings I think that we disagree with whatever people think. And defense is half the game, I would argue even more, because you're only as strong as your weakest link. If you've got a guy who can't guard somebody, they'll just go at him all night long. In that sense, defense may be more important than offense.
We're trying to measure how you help a team win. There is noise in that system. But during the season, you can't change your roster very easily.
But I have an infinite number of stories about lineups and how it can help you.
The best example is about the Spurs/Mavericks series. Del Harris came to me before Game 2 (of the 2006 Spurs vs. Mavericks series). I love him to death, he's a wonderful person.
Boy, he's a genius. When he was working with the Mavericks, he'd always ask me questions. He always knew the right question to ask. The numbers, by themselves, mean nothing.
In the regular season, Adrian Griffin was terrible against the Spurs. They had a terrible offensive rating, which means they couldn't score when he was in.
So Devin Harris had a great rating against the Spurs, and Tony Parker had a lousy rating in those games. The coaches sort of knew that Devin Harris could handle Tony Parker, but this gave them a metric to prove that.
So they started Devin Harris in Game 2 and they won by 20.
Then we can do head-to-head -- when one guy is on the court against another guy. When Marquis Daniels was on the court against Manu Ginobili, the Mavericks lost by a point a minute. So in Game 7, they didn't play Daniels. Del Harris told me "we don't know why this happens, but since you tell me Marquis Daniels is getting creamed, we didn't play him."
This is where there's a really old debate with scouts and the data people, that's in Moneyball and everything else.
I don't think either person is right, by themselves. Well, the data is one factor that you should look at.
The flaw with adjusted +/- is that there's noise in the system. But there are flaws in any system. Red Auerbach said K.C. Jones' team won every scrimmage. His PER sucks. There has got to be something missing.
Kevin Martin always has a fantastic PER meanwhile, but every year his defense is terrible.
So, I don't mind looking at PER. If we mess up, PER would probably get it. But PER messes up a lot because it just doesn't do defense.
So we're saying if you're talking box score based stats, you're going to miss defense ...
A lot of it. Not all of it, a lot of it. You're also missing things like taking the charge. Saving the ball going out of bounds, the pass that leads to the assist. Nobody knows what percentage of basketball is not in the box score, but that determines which side of the debate you're on.
But looking at the lineups, you can see a lot.
For instance, in every playoff series, there's what I call the team's kryptonite. There's two or three players on the
court that the other team can't handle.
For the Mavericks against Denver last year, it was Chris Andersen and J.R. Smith. When those two were on the court, the Mavericks got killed.
So what we do is we play detective. We look at every minute those guys are on the court. What worked?
That's the type of stuff that we do.
My prediction is that the Bulls are going to stink this year. Ben Gordon and Brad Miller were their best players. They let Ben Gordon go to the team they need to beat for the playoffs? Why'd they do that?
He wanted a lot of money.
Well, he's worth it.
Letting him go is just beyond stupid. It's ridiculous. And who'd they pick up to replace him? Jannero Pargo? I looked at their lineups, and I guess that they're expecting that Luol Deng can play his position. If he's healthy -- and I don't know if he's healthy.
You gotta mine the data. Because sometimes you're helpless. Denver -- I knew that would be bad for the Mavericks last year.
And Golden State [when the top-ranked Mavericks famously lost to the upstart Warriors, in 2007], I knew that would be bad for the Mavericks. The only hope the Mavericks had was to go small, and they did in Game 1 and lost that game. They got a lot of heat for that, but it was probably the best thing they could do.
Dampier is on the team so you can beat the Spurs and you can beat Shaq. And they beat the Spurs really easily. They had no trouble. But against the Warriors, small was better.
Did you advise the Mavericks to go small against the Warriors?
I show the numbers to the coach and they make the decision.
Against the Hornets, I would have certainly gone small. Against Golden State I would have gone small.
In Game 1 they just came out flat. Baron Davis hit like two half-court shots in that series. The Mavericks played horrible. And that series ... I do think it had a long-term effect, the hangover from that. The team didn't go back to being as good as it used to be.
How can you possibly fix something like that?
You can't. That's the whole thing. One of the holy grails of stats is predicting how well a player will do next year.
More from Winston on TrueHoop tomorrow.