 |
| Photo by statixc on Flickr |
Specifically, in this week when the Oakland A's return to the playoffs for the second time in two years and the seventh time in the last 13, I want to talk about the A's, sabermetrics, and why "Billy Beane's shit doesn't work in the playoffs".
I could have written this ten years ago, but the extra years have actually made it more apt. And I want to ask a question I haven't seen answered elsewhere - or at least not directly. I'm not a statistician, but a lot of sabermetrics started with guys like me raising questions or providing hypotheses, which smarter guys then tried to quantify with numbers.
First, if you haven't actually read Michael Lewis' "Moneyball", you should. (Though you don't have to in order to understand this post.) The movie version's good but it's more about the man than the concept; the book is the opposite. The concept is using objective, advanced statistics to exploit inefficiencies in the player market, which allows a team like the A's to afford players that help them score runs. The "inefficiencies" arise because most teams value the wrong attributes in a player; attributes like speed or body type that do nothing to predict success in scoring runs. Those advanced statistics come from "sabermetrics", the catch-all term for a movement that's turned baseball on its head - using math and "big data" to challenge conventional wisdom and prove that "everything you know about the game is wrong." The A's have been the biggest proponent of this approach in the major leagues for probably the last 20 years now, from their previous General Manager Sandy Alderson (now with the Mets) to their current longtime GM Billy Beane.
Sabermetrics can be (and is) also used on the field to manage in-game strategy. It's the ultimate expression of "playing the percentages" - it's just that the percentages often suggest you do things completely against conventional wisdom. For example, there is almost no situation in which you would intentionally walk a batter (not even Miguel Cabrera with an open base). Almost no situation in which you'd sacrifice bunt, except with very light-hitting pitchers or with no outs and a runner on second when playing to tie in the late innings (yeah, sabermetrics gets incredibly specific). The numbers say you have a better chance to prevent and score runs if you don't do these things.
While the effect is not totally consistent, overall most casual watchers and baseball old-timers would probably call a sabermetric gameplay style conservative. (The lack of intentional walks is one of the few exceptions, along with taking the extra base on hits, which is more important than steals to scoring runs.) Traditionalists criticize it as "sitting back and waiting for the big hit" or "not making the defense work". But this is what the math says works over a full season.
"Moneyball" is a massively interesting book, even if you only have a passing interest in the game. And you can't argue with Billy Beane's approach during the regular season - his A's have gone to the playoffs seven times since 2000 (including this year), with a payroll that's always among the lowest in baseball. He has shown poor teams how to compete, and he has shown all owners how to maximize profits. (Not all of them have quite caught on yet.)
For example, this year the Yankees spent $2,345,238 per regular season win, and missed the playoffs. The Oakland A's spent $576,797 per win and made the playoffs. The results have been similar for every year they've made the postseason - the A's never have anything approaching a $200 million payroll. (I'm using this site as a source for payroll info.)
It's taken a long time for sabermetrics to really catch on, and in fact I still hear TV commentators complaining when teams don't sacrifice outs for bases, or when they don't sign the latest overpriced "superstar" free agent in the offseason. Most casual fans know nothing of sabermetrics, even if they wonder what the "OBP" or "OPS" stats are that they see popping up more often these days. But a lot of other teams do use it now, or at least parts of it, when selecting players and setting strategy, because it gets results as the math always proved it would.
But 11 years after "Moneyball" was written, the A's have now lost six straight playoff bids - five of them before reaching the ALCS. The first question is why? (The real question I want to ask is a bit further down - keep reading!) Why does Billy Beane have so much success with such a small payroll in the regular season, and almost none at all (to this point) in the post-season? Is there something wrong with the sabermetric approach that's specific to the playoffs?
To the oldschool baseball guys, it's because the playoffs are a different beast, and are even more slanted towards "manufactured" runs - playing small ball - something sabermetrics proves (with math) is counterproductive over the long run.
Baseball Prospectus' "Baseball Between the Numbers" has a chapter actually entitled "Why doesn't Billy Beane's shit work during the playoffs?" (co-written by Nate Silver - the guy whose mathematical model correctly predicted all 50 states in the last presidential election - and Dayn Perry.)
Long story short, they make a convincing mathematical case that, all else being roughly equal between good teams in the playoffs, the only real correlations between team attributes and playoff success are closer quality (expressed in WXRL, or Win Expectancy over Replacement Level), pitcher strikeout rate/opponent batting average (in sabermetrics, these are two sides of the same coin) and team defense (expressed in FRAA, or Fielding Runs Above Average). They're all weak correlations (about 0.2), but they're there. They found no statistically significant correlations for playoff success and any other offensive or defensive attribute, including offensive home runs, sacrifice bunts, walks, stolen bases or anything else.
When they then went back and compared the relative strengths during the regular season of the last 180 playoff teams in these three specific attributes, they found that 9 of the top 10 teams in the combined rank went to the World Series, and 8 of those 9 won it (up to 2005, the year the book was written). Of the bottom 10 playoff teams in these three combined attributes, none even came close to the World Series, and some were swept out of the division series in 3 games.
So, to have the best chance of winning the World Series, you ideally want a top closer, great defense, and pitchers who can strike guys out. What you don't need is a bunch of bunts - the sabermetric approach to offense doesn't suddenly stop working in the playoffs.
They also give several reasons why this might be the case (in typical Nate Silver fashion, he asks the question if the playoffs really are different, and answers "yes, probably"), although the reasons themselves aren't fully understood - only the results are. But the extra days off do give relief pitchers and closers more rest. When every team has great hitters, defense will be the differentiator. Managers use closers, typically a bullpen's best pitcher, more innings in the playoffs. Etc.
If you're curious, Beane's early playoff teams didn't come close to the top 10 in this combined rank. (I'm actually not sure of his recent ones.) So while their lack of playoff success is still somewhat down to bad luck, you can sum up at least some of their playoff woes as mediocre closers (at least relative to other top teams) mixed with bad defense.
So why do they win so much in the regular season? Because defense is far less important. In fact, the math proves it's so unimportant that Billy Beane all but ignores it. When he signed Scott Hatteberg, he had never played first base before in his life, and had no feeling in his throwing hand due to nerve surgery. When he signed Jeremy Giambi, he did so knowing he was one of the worst outfielders in the league. When he signed David Justice, he knew that he could no longer run. The thing is, everything can be measured and compared, so if the math says a player will contribute more runs offensively than he will cost the team defensively, that's still a net positive, and that overall result in net runs created is what's used to compare players. Really good offensive players can more than make up for terrible defense over a 162 game season, and a guy who's great offensively and terrible defensively will create more net runs than a guy who's mediocre offensively but great defensively. Defense is less important than offense in the regular season, and more important in the post-season.
This has nothing to do with defensive errors - that's a stat that sabermetricians hate. There's no other stat based on an assumption of what a player should have done, so don't bother looking up fielding percentage to see how the playoff teams this year stack up defensively. It's a completely subjective stat. Sabermetricians instead look at objective statistics in actually getting guys out (put-outs, assists, etc.) and combine them in ways that can tell them whether a player at a particular position is performing above or below average compared to other players at that position around the league. It's by these measures that the early 00's A's, at least, were bad defensive teams. (The 2013 team is a little better, so we'll see if that helps them.)
So there are quantifiable, objective statistics by which the A's teams that did/do so well in the regular season can be said to not be competitive in the post-season. But I think there's more to it than that, and so do Silver and Perry. Heck, so does Beane. There's luck. Silver and Perry admit that with a small sample size (not many games), this is a large part of winning in the post-season. Beane says in "Moneyball" (and the interview linked above) that it's pretty much the only part - he believes there's nothing different about the playoffs, and the small sample size means it's all down to luck. But after six failed tries (and a sample size of 30 games now), this raises the question I'd love to see quantified:
Do winning teams in the postseason play to get lucky?
If you walk into a casino and head to the blackjack table, at some point the house is going to pick your pockets - that's just the way the odds are set up. It's going to happen. But if you play just 2 or 3 hands, it's entirely possible you could stop there and walk out a winner. The odds are still against it, but it's possible.
However, if you don't even sit down at the table - if you don't try to get lucky - there is no chance you will walk out a winner. You also won't lose, but you won't win.
This is equivalent to the playoffs vs. the regular season in baseball. Over just a few games, a team can get lucky enough to win through seemingly random events. However, you have fewer chances to get lucky in the playoffs if you don't play to give luck a chance. The big difference between the blackjack analogy and baseball is that in baseball, not playing at all to get lucky doesn't keep you from losing, it is losing. Only one team needs to get lucky to win the World Series, and you're playing against nine other teams in the playoffs.
Again, using sabermetrics to manage a baseball game results in a conservative offensive strategy, more or less. Very little base stealing, basically no sacrifice bunts. Take as many pitches as possible, wear the pitcher down, walk a lot. Wait for big hits. This is what the math says to do, and the math is not wrong over the course of 162 games.
However, with a small sample size, luck plays a much bigger role. How did the 1986 Mets win the World Series? They were down 3 games to 2, losing 5-3 in the 10th inning of game 6, with a 2 strike count on Kevin Mitchell. They ended up with 3 straight singles, a wild pitch, then the infamous ball through Buckner's legs. Nothing about that was anything but lucky. Game 7 was then rain delayed by a day, which gave the Mets' Ron Darling an extra day of rest. Again, luck. But game 7 wouldn't have been necessary, and that booted ball and wild pitch would never have happened had the Mets themselves not put themselves in the position to take advantage of it. They did that by swinging the bat aggressively and being aggressive on the basepaths. (Let's not forget that the free-swinging Mookie Wilson fouled off six pitches in his at-bat, not many of them strikes, giving Bob Stanley the chance to throw his wild pitch before Wilson's grounder up the line.)
I don't know of a sabermetric statistic that would account for the 10th inning of game 6 - this is the kind of thing where even sabermetricians shrug their shoulders and say "I dunno, they got lucky". But is there a way to at least account for the number of lucky chances in a game (say, "optional" plays that have a chance of either succeeding or failing), and assign a score for a good or bad outcome to the results of those chances (based on number of expected runs created or lost), thereby determining mathematically not just how much luck accounted for a team's World Series victory, but whether that was a result of an aggressive style of play?
To look at it another way, with roughly equal skill, a team with no luck is going to have an awfully hard time beating a team having a run of good luck that they're riding through aggressive play. (Even if 9 other teams are trying to do the same and suffering through a run of bad luck. Again, there's only one World Series winner.)
Winning teams seem to embrace randomness in the playoffs. Luck cannot be controlled, but if you play for it and it smiles on you, it can separate the winner from all the losers. If you play against randomness, then you can never benefit from it the way winning teams do. Or at least, that seems intuitive to me - but so much about baseball that's intuitive has been proven wrong by sabermetrics. So I'm really just posing the question.
I don't follow the A's closely enough to know if Billy Beane has learned from the past and is now consciously preparing for the playoffs by stocking better defensive players, or pitchers with a high strikeout rate. But I'll be watching them this year to see if they've made any adjustments to their post-season strategy to try to get lucky.
No comments:
Post a Comment