Wednesday, March 17, 2010
The robots are coming, and they're cranky
By Henry Abbott
When the topic of Moneyball-style stat-geekery comes up, that's what people are scared of, I think: The automation of every damned thing. They're scared that if spreadsheets drive decisions, something important and human will be lost forever, to the detriment of us all.
It's a fear that goes far beyond hoops. Do you want a doctor to treat that disease, or an unfeeling flowchart of decision-making run by some low-level employee who might as well be a robot? Do you want restaurant food made to computerized specifications from the national chain's test kitchen, or by a human you can see from the seating area?
The human touch appeals and is, to many of us, essential. I am 100% on board with the idea that humans should run the show, with all their compassion and insight and even gut instincts. I cheer when Dean Oliver writes, in "Basketball on Paper:"
My null hypothesis is usually traditional coaching or management wisdom. So a hot hand exists, defense wins championships, and statistics are irrelevant until I prove otherwise (which I think I've done in many cases). Others may choose a different null hypothesis, but I think mine makes sense because I work with coaches and management and I'm not Billy Beane -- it is my burden to prove things, not theirs.
My real attachment to the statistical movement comes not out of wanting to place spreadsheets and analysis above humanity, but out of wanting to place smart analysis over bad.
In short: Basketball teams have always used statistics. Statistics like points per game and rebounds per game have had a huge effect on who gets drafted where, who gets paid how much, and even who gets minutes. Stats have always been a huge part of the game.
But compared to what's out there, points per game may be pleasingly easy to understand, but as of 2010 that's far from the best at helping teams make the smartest decisions. And as they're using numbers to determine important stuff anyway, they might as well get it right, even if that's a little complicated and overwhelming at times.
That's not to say statistics can't, in addition to being smarter, also evolve into a bigger role. The best analysis may well change how those humans think, and even reverse some of sports' conventional wisdom.
But it's hardly the case of a hostile invasion, where the cyborgs are coming to eliminate the humans.
Stumbling on Wins
David Berri and Martin Schmidt, two of the three authors of the 2006 book "The Wages of Wins," have just released a new book, from FT Press, called "Stumbling on Wins." It is just now available online and will be in stores next month.
While most stat experts are, like Oliver, a little timid in defining the impact of their work -- almost nobody calls for statistical models to replace general managers -- Berri and Schmidt are adventurous in asserting that the people running the show seem to have little idea how to do their jobs.
"We wish to emphasize," Berri writes, in a letter that accompanied a review copy, "that our purpose in writing this book is not to argue that coaches and general managers are 'stupid.' As we argue in the book, we think people employed in sports are at least as smart as people in the general population."
Important though that distinction may be, either way, just about every front office employee is, after reading that line, itching for a fight.
If you're a general manager in basketball, baseball, football or hockey, going about your business the same old way things have always been done, this book is that clock radio in your hotel room. You didn't even set an alarm, but there it is, annoyingly blaring to life. Aghast, you don't know how it turned on, nor how to turn it off. You just want it to go away. But guess what, unless you can figure out how it works, it's not going to shut up.
These professors (Berri is an associate professor of economics at Southern Utah University, Schmidt a professor of economics at the College of William and Mary) may not be qualified to run a pick and roll, but they're not at all convinced you're qualified to run your team, either. They're here, and they're packing spreadsheets.
March, when front offices fall in love
An NBA insider once told me that if you want to be drafted high, there's nothing like playing well in the NCAA tournament. That's intuitive, but also a bit of an insult to the NBA decision-makers. In what wacky world would a few flashy, luck-infused wins in a single elimination tournament overwhelm the combined insight of the years of scouting, interviews, measurements, evaluations and the like that teams tell us they perform?
Nevertheless, Berri and Schmidt have found the claim to be amazingly true. "A player who appears in the Final Four," they write, "can improve his draft position by about 12 spots."
12 spots! That's a ton! Other factors that they find affect draft position include points scored, shooting efficiency, assists, steals, blocked shots and height (relative to position). Interestingly, staying in school, but not improving as a player, was cause to fall an average of five spots in the draft -- which would seem to be an argument for coming out earlier, especially when you consider the author's evidence that even years after the fact, draft position has a heavy influence on playing time.
But here's the amazing part: Rebounding ability did not, they report, significantly impact draft position.
John Hollinger has written several times that rebounding happens to be one of the skills that carries over best from the NCAA to the NBA. (My theory on why comes from David Thorpe's Train Like a Pro -- his main rebounding instruction focused on chasing down balls that fall far from where you're standing. Players who chase rebounds all over the floor -- including well out of their area -- get rebounds all over the floor. But most players just don't do that. Players who have the vision and mentality to launch all over the court chasing rebounds in college will keep that advantage in the NBA.) Berri and Schmidt, in fact, say rebounding skill tops the list of skills that contribute to NBA winning.
If "Moneyball" was all about identifying value where others don't see it, it would seem that in the draft, rebounding has long been one such sweet spot. And it's worth noting that some of the draft's better values of recent years -- Paul Millsap and DeJuan Blair come to mind -- were elite college rebounders.
And here's the kicker. The authors found that playing for an NCAA champion in the year drafted is a statistically significant predictor that you will be less productive in the NBA.
When they're paid to shoot, guess what they do
Berri and Schmidt do something marvelous in citing Tommy Craggs' excellent 2007 New York magazine profile of Stephon Marbury:
“If I didn’t play the way how I played, I wouldn’t have gotten no max contract,” [Marbury] said. “They can talk about whatever they wanna talk about me, because I got maxed. I’m a max player. Don’t get mad at me, because I’m telling you what’s real. One plus one is two, all day long, and it’s never gonna change. And that’s factorial.”
Marbury, the book demonstrates, is dead right. It's trendy and fun to describe Marbury as a flawed player, who could have been far better player, who won a lot more, if he got his teammates more involved or played better defense. But let's be clear: If Marbury's fixation was on a maximum contract, he was as sane as they come. They even explain how assists are valued in NBA contracts -- points are so much more valuable that almost any amount of shooting will result in enough points to make it worthwhile.
The numbers in "Stumbling on Wins" demonstrate that NBA front offices are obsessed with scoring and will overpay handsomely for it. That being so, who's crazy -- the vaseline-eating Marbury, or the allegedly genius GMs and owners who create a market that incents Marbury to help his team less than he could?
Berri and Schmidt go on to make a case that in his time running the Knicks, Isiah Thomas (who acquired Marbury) didn't demonstrate that he had poor taste in players, compared to the rest of the League. They say essentially every GM overvalues scoring. That Thomas brought in scorers is a matter of historical record. That every team values scoring is demonstrable in their research showing statistically significant factors in getting NBA players paid more. It's also evident in this one crazy little example: By the author's metric, in his rookie year, Adam Morrison was literally the worst player in the NBA. Yet, he was a scorer. And guess what? He ended the season just one coach's vote shy of the being a first-team All-Rookie.
Thomas' great flaw, in such a context, may not have been desiring scorers like Marbury, Jamal Crawford, Eddy Curry, Quentin Richardson and Zach Randolph. His flaw, this book argues, was in having a nearly unlimited budget to acquire them. (Think of it like children set free in a candy store. How many wouldn't make themselves sick?)
There's no doubt that NBA fans, teams, owners and front offices generally treasure scoring ability -- in many cases perhaps beyond all reason. "Stumbling on Wins" is valuable in bolstering that case with evidence. But in their zeal to stick it to the status quo, Berri and Schmidt may have stretched the ensuing analysis beyond its breaking point, especially when they conclude:
The failure of the New York Knicks to win not only cost Isiah his job, but perhaps more importantly, severely damaged his reputation as a basketball expert. Yet our examination of the data suggests Isiah's only crime was implementing what every basketball expert knew to be true. Scorers are considered the most valuable players in the NBA. Therefore, a team of scorers should be the most successful team.
I'd be surprised if you could get any basketball expert to sing along with that song. To summarize expert thinking so ham-handedly is one thing. To deploy it as proof of their idiocy defies logic. I mean, we've all heard current coaches and GMs talk. They have subtle theories about the game, honed over time. Sure, they may have some misperceptions here or there. But you're not going to get them that easily.
Common sense, or a peek at their public comments would reveal that every basketball expert thinks every team needs things like rebounding, passing and defense, even if they buy them on the cheap. The proof that 29 other teams wouldn't have done what Thomas did is that so many of them willingly traded their scorers to Thomas' Knicks. Even in real time, articles were published explaining how just about the smartest thing a GM could do was make as many deals as possible with Thomas, and more than a few teams did. It's questionable that even Thomas dreamed of a roster of 12 pure scorers, it's impossible that thinking drove the whole league.
At moments like that, it's easy to see that Berri and Schmidt's book -- packed with insight that should be heeded-- would have far greater impact if they laced their thinking with Dean Oliver's null hypothesis that the people running the show just might know a thing or two. That many of us may over-value human insight is no reason to discard it wholesale in favor of the lessons of the spreadsheet. That's especially true when that data -- we're still in the early stages of the statistical analysis of basketball -- has a decent shot at being wrong.
In defense of coaches
When I say that basketball statistics are young, I mean that I'm pretty certain that a lot of what we think we know now will look stupid in a decade or two. There's a ton of stuff that matters to the game that hasn't yet showed up in statistics.
The most glaring example of this is defense. While team defense is fairly easy to measure, who could doubt that some players are really good at defense, while others are terrible? Yet it's exceedingly tough to quantify the difference between the league's various Quinton Rosses and Nicolas Batums. The box score, read carefully, is pretty darned informative about who does what on offense, because of the coincidence that things that are easy to count (shots, passes, free throws) tend to happen there. The commonly charted stats are almost no help in helping coaches decide who to stick on Carmelo Anthony with the game on the line, however.
As there are plenty of wins, and therefore dollars, riding on individual defense, there are efforts afoot to measure it. That requires teams of people charting things that aren't in the box score. It's the holy grail of basketball statistics. One day, we'll know a ton more about it, and we'll see our current efforts along those lines (including ignoring defense entirely, charting deflections and the like, or extrapolating a lot from small sample sizes in adjusted plus/minus) as laughably basic.
Berri and Schmidt make a reasonable-for-2010 decision to ignore individual defense in their statistical models. As they explain in the appendix (shouldn't this be mentioned up front?) they look at how an offense performs against a certain team, and then give everyone playing D the same rating. This is understandable -- what other option is there? -- but it's horribly imprecise. All you have to do is watch that play from the other day when four Clippers played reasonable defense, but Baron Davis didn't even make it down the floor and offered no resistance at all. Vince Carter's uncontested layup surely ought not be equally applied to all five Clippers. And saying that all five players generally deserve the same defensive rating makes as much sense as saying they all deserve the same offensive rating. It can't be accurate.
Given that there's a lot we know we're not measuring in hoops, you might think Berri and Schmidt would feel the need to bolster any findings with other evidence -- from video, anecdote, other statistics. This is especially true if they're going to engage in the rather weighty exercise of telling sports fans that their off-court heroes are making systemic and constant mistakes. But in at least one instance, you'd be wrong. Using nothing but those "defenseless" numbers, Berri and Schmidt declare that NBA players' productivity peak at about 24 years of age, and yet those silly coaches (stumbling, as they do, on wins) insist on skewing their minutes to players with an average age of 28. "Coaches," they write, "are exaggerating the contributions of older players."
Maybe they are. But maybe not. If it's true, as many coaches will assert, that defense is a veteran's game, then it could well be that Berri and Schmidt are simply not identifying elite defenders as such. I can't help but notice that NBA champions often have remarkably old rosters. Show me an elite NBA team with an average age of 24, and I'll show you five with an average age far older.
As the people who picked the fight -- they're the ones who poked the coaches and GMs in the chest -- Berri and Schmidt have an obligation to come strong. And on many points they do. It's a little shameful that GMs and owners are drafting heroes of the NCAA tournament, for instance. Anyone fascinated about what works and what doesn't in sports should read this book for fresh thinking based on the evidence.
But at the moments when the book's points are balanced on this or that narrow piece of novel data -- like in declaring coaches wrong for favoring slightly older players -- it remains to be seen who is really doing the stumbling.
Until the robots are walking steadily on their own two feet, I'd say there's little risk one of them will take over your front office anytime soon.