Actually, it is. Mr. Lewis can spin a fine yarn, but as with much of his first major work (Liar's Poker), a good deal of detail, and outright truth, is sacrificed to back-up the original thesis and offer a more compelling reader narrative.
With hindsight, and much more data, long-term trending analysis of the Sabre-metrics examples given in Moneyball have not correlated with increased team, or even individual, performance in baseball.
Basketball is an entirely different kind of animal. The data miner in me looks forward to seeing if this sort of analysis will offer genuinely useful predictive performance models this sport.
Nay, basketball is not much different. Get the best players + good coaches and you'll win the most games. End of story.
"a good deal of detail, and outright truth, is sacrificed to back-up the original thesis and offer a more compelling reader narrative."
I think your comment is spot on, case in point:
The Grizzlies went from 23-59 in Battier’s rookie year to 50-32 in his third year, when they made the N.B.A. playoffs, as they did in each of his final three seasons with the team. Before the 2006-7 season, Battier was traded to the Houston Rockets, who had just finished 34-48. In his first season with the Rockets, they finished 52-30, and then, last year, went 55-27 — including one stretch of 22 wins in a row.
What Michael Lewis failed to mention is that in Memphis, Battier played with Gasol, the same Gasol who made the Lakers an instant contender and eventual champion. And in Houston he played with 2 perennial all-stars in Tracy McGrady and Yao Ming. This year Houston did not have either Yao or Tracy and they sucked, Battier is still there...
>Nay, basketball is not much different. Get the best players + good coaches and you'll win the most games. End of story.
Almost every sport is like that. Sabermetrics was an attempt to find an optimal set of players using statistics. That will never work in the NBA (you could certainly argue that it hasn't worked in the MLB, either).
Baseball is ultimately broken down into a series of 1-on-1 match-ups. A pitcher vs. a hitter, with a defense behind them. It's a team sport but the largest functions of the game are 1-on-1. No other major professional sport is like that. As such, major Sabermetrics stats like OPS and WHIP don't translate to basketball. You get something like Hollinger's PER rating, which is meaningful but not really telling.
Scoring is largely out of context. A good player may score 20 points a game on a bad team and only 10 on a great team. Matchups change throughout the game and what your team does has a huge effect on what you do.
In baseball, it's much more realistic to look at the statistics of a handful of players and figure out how you should organize your team. In basketball, crafting game plan strategies and rosters is still more of an art than a science.
"Baseball is ultimately broken down into a series of 1-on-1 match-ups. A pitcher vs. a hitter, with a defense behind them. It's a team sport but the largest functions of the game are 1-on-1. No other major professional sport is like that."
Cricket is a lot like that, actually. Its dynamics are different, though: whereas in baseball in a given matchup between a pitcher and a batter, the pitcher usually "wins", in cricket, between a bowler and a batsman, the batsman will usually "win" (or, at least, not "lose"). It's this difference that explains why cricket matches that aren't limited in overs or time last so much longer.
Including the often very dicey performance metrics. My favourite example is batting average(total runs/total dismissals), which is skewed in favour of lower-order batsmen, who are more likely to record not-out scores.
I'm more interested in their median score (or better yet, median partnership).
Agreed, except on the partnership bit, which I don't understand. Keeping track of partnerships always seemed like trivia to me. Can you explain a bit further?
Batsmen contribute to partnerships with, e.g. good running between wickets. Things like that aren't reflected in batting averages, or their effect is downplayed.
True. It still sounds uncomfortably like trivia, but maybe that's because nobody's figured out how to measure it properly yet.
It sounds a lot like evaluating fielding, in both cricket and baseball, because in addition to keeping track of whether a player drops the ball (both figuratively and literally) or not, there are many would-haves and should-haves to take into account.
But you're right: good running between the wickets, and also shot selection with a view toward getting the better batsman more of the strike, is a real skill which should be measured, even if it is difficult and nobody really knows how yet. :)
It's not that sabermetrics itself is without value, but the principles put forth in MoneyBall for "success" with sabermetrics in baseball have not held water for other teams in other markets.
Even arguing the other side of the coin, it's easy to make the case that with resources like Baseball Prospectus ( http://www.baseballprospectus.com/statistics/sortable/ ) available to all, any small sabermetrics arbitrage advantages which Beane & Co. may have once had are obviated.
"With hindsight, and much more data, long-term trending analysis of the Sabre-metrics examples given in Moneyball have not correlated with increased team, or even individual, performance in baseball."
Hmmm... what? The general principles Beane operated from have pretty much been adopted league-wide at this point. Of course, MLB is a zero-sum game, so once everyone adopts something, it's no longer an advantage. The Tampa Bay Rays are operating off the same statistical measurement philosophy (although they understandably don't share which) to make a small-market club operate efficiently, and they're pretty successful with it.
A corollary there is that you cannot leverage in baseball like you can in financial markets to play at ever thinner margins and thus the theory becomes largely useless unless you can augment it with better information.
Also, people forget that "increased... performance" in baseball often means an increase in profit, not an increase in victories.
True. In financial markets, even small inefficiencies and impedance mis-matches can be exploited due to liquidity, fine-granularity, high transaction volumes, and uniform standards of measurement. The same principles don't down-scale to the individual athlete or small-group (team) dynamics.
Put another way, financial market strategies (in theory) can be constructed on matching/measuring against a continuous variance in capital efficiencies. Human performance is simultaneously less consistent, discrete, and difficult to measure.
Put another way, think of matching the torque/horsepower vs. rpm curves for a typical family sedan's engine versus a high-compression race engine. The former can get good performance with an automatic transmission, the later requires a stick shift and an experienced hand on the till.
I'm not sure that the point of Moneyball was that sabrmetrics will always produce the most successful team.
To me the point was that the team used statistical analysis to find undervalued players so they could compete with their opponents who often spend 2-3 times as much on player salaries.
I agree. He's such a superb writer, that no matter how long or convoluted the article is, he can keep you enthralled right to the very end. His piece at the now-defunct Portfolio magazine was one of the best articles on the financial crisis that I have read:
http://www.portfolio.com/news-markets/national-news/portfoli...
I voted it up anyway, that's how much I like this article. After reading this comment thread http://news.ycombinator.com/item?id=481557 on that page, I decided to read Moneyball, and it was one of the most riveting books I've read.
"The use of easy-to-obtain metrics with little or no predictive value has interesting parallels in all sorts of places."
I was just wondering if there were parallels for this in programming.
The obvious might be the various approaches and "methodologies" to programmer or team productivity based on dubious or non-existent hard data.
But I'm more wondering if there are tools or techniques that don't get any accolades (like, say, MVC or TDD or the framework of the week) but have a peculiar synergistic effect that ends up providing significant (and possibly counterintuitive) value.
What's the "Shane Battier" of software development practices?
I'd say good design. Refactoring has known benefits, but good design is similar to refactoring while skipping the middle step of having a cruddy design to start with. Developers who tend to write better designed code may not be appreciated for it. Consider that better design often requires considerable forethought and may take longer to implement than a typical "first thing that comes to mind" satisficing type design.
Also, the lack of future bugs and lack of future difficulty in expanding and adapting good code may result in a lack of attention for that code (people may not appreciate its subtlety and elegance and merely think it solves an easy problem to start with).
The end result may be that the developer takes longer to resolve fewer bugs and implement fewer bits of functionality than other developers while the benefits of higher quality design and code might be largely invisible.
Discerning developers may be able to recognize and appreciate higher quality coders, but in environments with short-sighted or overly bureaucratic management I'm sure this sort of problem exists.
Trouble is, "good design" is pretty hard to quantify. It's much much harder to quantify than things like LOC, or whether this milestone or that was reached.
Another difference is that a basketball game has a definite, qualitative outcome. That, at least, can be measured. With software development you don't even have that luxury.
Quite so, there's no objective measure for good software, other than perhaps sales (and it's enormously difficult to trace sales data back to individual contributions from one out of dozens, hundreds, or thousands of devs).
Which is why software development is still very much a craft. And a pretty esoteric one at that. It often takes the subjective judgment of a known good developer to determine the quality of another developer's contribution. The unfortunate side effect of this is that the really good dev shops (populated by talented devs even in management roles) are working at levels many, many orders of magnitude beyond what the run of the mill dev shops are capable of. There are a great many development projects managed or overseen by non-developers, and many of those people lack the basic skills necessary to tell the difference between gold and utter crap. It's no surprise then that so much "enterprise-y" development is little more than snake oil, VBA scripts and MS access duct taped together and sold for millions of dollars to big companies who don't have the skills to know they got taken for a ride.
Software development is only just barely scratching its way out of its alchemy and astrology phase. There are far more untalented, unskilled hacks out there than there are honest craftsmen.
No, I like Wodehouse, so I used Bertie and Oofy in my retelling of the joke. But I'm pretty sure it's much, much older. I don't actually recall it occurring in any of the short stories, novels, or Fry and Laurie dramatizations.
Ah, fair enough. I like PG Wodehouse too, which is why I asked. I can't claim to have read all the books though; I'm trying to space them out over a decade or so of long plane journeys.
For those who like to highlight text as you read, you can shut down the nytimes' messing with you, by adblocking all its js, from all its servers - and reloading. This filter worked:
I went to the trouble of finding the offending js a while back, but some aspect of its URL must have changed (server? directory? depth? name?), because the filter stopped working. So now I nuke everything; the loss of js hasn't harmed the text yet.
Really? I don't like sport and I found I couldn't understand it at all.
For instance, you have to read to paragraph 4 to find out that "Houston" is the same team as "The Rockets". I never quite figured out whether Battier and Ming are on the same team.
And "The game drew a huge national television audience, which followed Bryant for his 47 miserable minutes: he shot 11 of 33 from the field and scored 24 points." -- I have no idea what that means or whether it's a lot.
Winning teams in professional basketball usually make around 45%-50% of their shots. During an average game, a team might attempt 80-90 shots. In this case, Kobe Bryant, a career 45% shooter, was held to 33% shooting and still attempted a large proportion of the teams overall shot attempts.
However, like in reviewing individual statistics from almost all sports, missing 4 additional shot attempts than he normally would could just have been an expected statistical fluctuation.
I'm sorry but if you don't have a clue about professional basketball you really can 't appreciate this article. Not every piece of writing/reporting can be framed in a "gee-wiz" manor. Go do your "more important" things and we'll all enjoy the article.
There is a difference between "not liking" sports and not knowing anything about sports. Not that I fault you for it, obviously, this article assumes the reader has some basic understanding of the NBA and the rules of basketball.
> There is a difference between "not liking" sports and not knowing anything about sports.
Yes, and that's even an argument in favour of explaining more. E.g. I like some sports, but I don't know anything about basketball. (It's just not a big sport where I come from.)
Either you typoed "sports" or you are not from the U.S. (England?) I'm not a huge sports fan myself but this sort of jargon sort of penetrates by osmosis where I live.
As a side note, why do you have to login to email this article to a friend? My friend and I are huge Rockets fans, and I know he'd love to read this, but forcing me to sign in to email it to him is ridiculous.
that's called face guarding, and this is a widely known legal maneuver. Perhaps it's not prevalent because you don't pad your stats with it, but I bet it's easy for the shooter to draw a foul from the defender.
55 comments
[ 4.2 ms ] story [ 103 ms ] threadWith hindsight, and much more data, long-term trending analysis of the Sabre-metrics examples given in Moneyball have not correlated with increased team, or even individual, performance in baseball.
Basketball is an entirely different kind of animal. The data miner in me looks forward to seeing if this sort of analysis will offer genuinely useful predictive performance models this sport.
"a good deal of detail, and outright truth, is sacrificed to back-up the original thesis and offer a more compelling reader narrative." I think your comment is spot on, case in point:
The Grizzlies went from 23-59 in Battier’s rookie year to 50-32 in his third year, when they made the N.B.A. playoffs, as they did in each of his final three seasons with the team. Before the 2006-7 season, Battier was traded to the Houston Rockets, who had just finished 34-48. In his first season with the Rockets, they finished 52-30, and then, last year, went 55-27 — including one stretch of 22 wins in a row.
What Michael Lewis failed to mention is that in Memphis, Battier played with Gasol, the same Gasol who made the Lakers an instant contender and eventual champion. And in Houston he played with 2 perennial all-stars in Tracy McGrady and Yao Ming. This year Houston did not have either Yao or Tracy and they sucked, Battier is still there...
Almost every sport is like that. Sabermetrics was an attempt to find an optimal set of players using statistics. That will never work in the NBA (you could certainly argue that it hasn't worked in the MLB, either).
Baseball is ultimately broken down into a series of 1-on-1 match-ups. A pitcher vs. a hitter, with a defense behind them. It's a team sport but the largest functions of the game are 1-on-1. No other major professional sport is like that. As such, major Sabermetrics stats like OPS and WHIP don't translate to basketball. You get something like Hollinger's PER rating, which is meaningful but not really telling.
Scoring is largely out of context. A good player may score 20 points a game on a bad team and only 10 on a great team. Matchups change throughout the game and what your team does has a huge effect on what you do.
In baseball, it's much more realistic to look at the statistics of a handful of players and figure out how you should organize your team. In basketball, crafting game plan strategies and rosters is still more of an art than a science.
Cricket is a lot like that, actually. Its dynamics are different, though: whereas in baseball in a given matchup between a pitcher and a batter, the pitcher usually "wins", in cricket, between a bowler and a batsman, the batsman will usually "win" (or, at least, not "lose"). It's this difference that explains why cricket matches that aren't limited in overs or time last so much longer.
Including the often very dicey performance metrics. My favourite example is batting average(total runs/total dismissals), which is skewed in favour of lower-order batsmen, who are more likely to record not-out scores.
I'm more interested in their median score (or better yet, median partnership).
It sounds a lot like evaluating fielding, in both cricket and baseball, because in addition to keeping track of whether a player drops the ball (both figuratively and literally) or not, there are many would-haves and should-haves to take into account.
But you're right: good running between the wickets, and also shot selection with a view toward getting the better batsman more of the strike, is a real skill which should be measured, even if it is difficult and nobody really knows how yet. :)
It's not that sabermetrics itself is without value, but the principles put forth in MoneyBall for "success" with sabermetrics in baseball have not held water for other teams in other markets.
Even arguing the other side of the coin, it's easy to make the case that with resources like Baseball Prospectus ( http://www.baseballprospectus.com/statistics/sortable/ ) available to all, any small sabermetrics arbitrage advantages which Beane & Co. may have once had are obviated.
Hmmm... what? The general principles Beane operated from have pretty much been adopted league-wide at this point. Of course, MLB is a zero-sum game, so once everyone adopts something, it's no longer an advantage. The Tampa Bay Rays are operating off the same statistical measurement philosophy (although they understandably don't share which) to make a small-market club operate efficiently, and they're pretty successful with it.
Also, people forget that "increased... performance" in baseball often means an increase in profit, not an increase in victories.
Put another way, financial market strategies (in theory) can be constructed on matching/measuring against a continuous variance in capital efficiencies. Human performance is simultaneously less consistent, discrete, and difficult to measure.
Put another way, think of matching the torque/horsepower vs. rpm curves for a typical family sedan's engine versus a high-compression race engine. The former can get good performance with an automatic transmission, the later requires a stick shift and an experienced hand on the till.
To me the point was that the team used statistical analysis to find undervalued players so they could compete with their opponents who often spend 2-3 times as much on player salaries.
I voted it up anyway, that's how much I like this article. After reading this comment thread http://news.ycombinator.com/item?id=481557 on that page, I decided to read Moneyball, and it was one of the most riveting books I've read.
Thanks, HN.
I was just wondering if there were parallels for this in programming.
The obvious might be the various approaches and "methodologies" to programmer or team productivity based on dubious or non-existent hard data.
But I'm more wondering if there are tools or techniques that don't get any accolades (like, say, MVC or TDD or the framework of the week) but have a peculiar synergistic effect that ends up providing significant (and possibly counterintuitive) value.
What's the "Shane Battier" of software development practices?
And how would you know?
http://blogs.msdn.com/larryosterman/archive/2004/04/20/11699...
Also, the lack of future bugs and lack of future difficulty in expanding and adapting good code may result in a lack of attention for that code (people may not appreciate its subtlety and elegance and merely think it solves an easy problem to start with).
The end result may be that the developer takes longer to resolve fewer bugs and implement fewer bits of functionality than other developers while the benefits of higher quality design and code might be largely invisible.
Discerning developers may be able to recognize and appreciate higher quality coders, but in environments with short-sighted or overly bureaucratic management I'm sure this sort of problem exists.
Another difference is that a basketball game has a definite, qualitative outcome. That, at least, can be measured. With software development you don't even have that luxury.
Which is why software development is still very much a craft. And a pretty esoteric one at that. It often takes the subjective judgment of a known good developer to determine the quality of another developer's contribution. The unfortunate side effect of this is that the really good dev shops (populated by talented devs even in management roles) are working at levels many, many orders of magnitude beyond what the run of the mill dev shops are capable of. There are a great many development projects managed or overseen by non-developers, and many of those people lack the basic skills necessary to tell the difference between gold and utter crap. It's no surprise then that so much "enterprise-y" development is little more than snake oil, VBA scripts and MS access duct taped together and sold for millions of dollars to big companies who don't have the skills to know they got taken for a ride.
Software development is only just barely scratching its way out of its alchemy and astrology phase. There are far more untalented, unskilled hacks out there than there are honest craftsmen.
What's up?
I lost my favourite collar stud.
Right here?
No, somewhere down the street.
Then why are you looking for it here?
The light's better over here.
*
That being said, Bayesian Search works a little like this:
http://en.wikipedia.org/wiki/Bayesian_search_theory
http://www.youtube.com/watch?v=x4aDM5d27_w
For instance, you have to read to paragraph 4 to find out that "Houston" is the same team as "The Rockets". I never quite figured out whether Battier and Ming are on the same team.
And "The game drew a huge national television audience, which followed Bryant for his 47 miserable minutes: he shot 11 of 33 from the field and scored 24 points." -- I have no idea what that means or whether it's a lot.
However, like in reviewing individual statistics from almost all sports, missing 4 additional shot attempts than he normally would could just have been an expected statistical fluctuation.
With a few dozen more like it I could probably puzzle my way through this article, but I'm sure we've both got better things to do.
Yes, and that's even an argument in favour of explaining more. E.g. I like some sports, but I don't know anything about basketball. (It's just not a big sport where I come from.)