THE BOOK--Playing The Percentages In Baseball

Q: Is sabermetrics the Freakonomic analysis of baseball?

A: There are parallels. What I do was heavily influenced by the University of Chicago economists of the 1960’s. I think Freakonomics comes from the same tradition.

I usually put it the other way: that Freakonomics is the sabermetrics of non-sports.  I mean, sabermetrics was first by about 25 years.

Boy, that John F. Kennedy, I haven’t seen that much charisma since Barack Obama.

Yeah, but in the sense that freakonomics is “taking a closer, skeptical, analytical look at things that haven’t been looked at like that before” (my def), I think it’s an OK question.

I was definitely peeved by the whole

We haven’t figured out anything yet. A hundred years from now, we won’t have begun to have the game figured out.

I think we’ve already hit the point where analysis of offensive contribution (who’s “good”, and by how much) is becoming less and less useful because any new formula offers such a small improvement over previous ones that it’s probably not worth the time and effort.  Likewise with forecasting.  In 20-30 years, I’d bet we’ll have fielding down pretty well; same with pitch f/x.  100 years from now, there will still be stuff to analyze (what wood is best to use for bats?  which jock strap allows Jose Reyes IV to run the fastest?).  But to say we’ll know “nothing” is just dumb.  Sorry if I’m being nitpicky.

Other things I disagree with:

Q: Generally, who should have a larger role in evaluating college and minor league players: scouts or stat guys?

A: Ninety-five percent scouts, five percent stats.

I’d guess numbers can give us more info than that (even if it’s really 80/20) with things we’re learning about “young player skills”, etc.  Not to mention, it’s hard to really assign any weight here as we still, to my knowledge, don’t have any analysis on scout accountability and accuracy.

Lastly:

How high do you think Andruw Jones ranks among the all-time great center fielders, looking only at defense?

A: As to Andruw — I don’t honestly know, but he’s certainly somewhere in the top ten.

Really?  He’s never been a real thin guy, and while I realize that athletes today are in much better shape than those 30, 50, 70 years ago… I just can’t imagine his otherworldly “instinct”, “jump”, or whatever, makes up for his unimpressive speed.

And honestly … I’m not interested in the numbers. Never was. That’s your perception of what I do; it’s not mine.

stats as trivia vs. stats as analysis

Mike, IIRC, Andruw Jones rates very highly in WOWY.

UZR from 1999-2003 has him as one of the best CF of that time period:
http://tangotiger.net/UZR9903TT.html

UZR has him as average since.

I think there’s no question that Andruw, at his peak, was a great CF.  How great is another question of course. 

But, he’s in the conversation for sure, regardless how he looks in jeans.

And if Andruw is top ten, doesn’t that mean Mike Cameron is top three?

I very much agree with Mike in #2.  For a guy that seems to frown on hyperbole, he makes a lot of hyperbolic comments.

Just as he called someone else’s question/attitude “lame and useless,” the idea that we know “nothing about the game, and still won’t in a hundred years,” while making for a nice sound bite, is a lame and useless response.

I really tire of some of his responses.  When he was asked whether clutch hitting exists as a repeatable skill, how about he give the correct response, which is that, “Yes, we (the body of researchers that have looked into the matter) think that it does, but we also think that the skill is quite small, small enough to be almost practically useless, other than perhaps as a ‘tiebreaker’?”

For a guy that is so interested in a question (whether clutch skill exists and to what extent), perhaps he should pick up a book or article that addresses the issue competently.

I’m not trying to tout our Book, and of course I am biased, but that James has not read our Book, which I don’t think (but don’t know) he has, among other sabermetric books, is inexcusable.  I mean how many good sabermetric books are there?  He can read them all in one afternoon!  I would think he gets paid enough by the Sox to be required to keep up with the research.

Q: Shouldn’t in-game strategic decisions be made by a computer? Or, more to the point, isn’t there always a correct choice?

A: It is totally impossible to isolate the correct strategic choice in almost all real-life situations, for the simple reason that all real-life strategic situations involve dozens of variables, many of which have not been thoroughly tested by trial. People who think that they know when a manager should bunt and when a manager should pitch out and when a manager should make a pitching change are amateurs. People who have actually studied these issues know that the answer disappears in a cloud of untested variables.

Another really lame response which is absolutely untrue.  In fact, it is the other way around, at least in baseball.  Almost ALL baseball decisions can be made “correctly” with a computer, except for the close ones, in which case, it usually doesn’t much matter what you do.  If a computer made ALL baseball decisions in a game and the manager made (virtually) none, each team would pick up at least 1 game in WE, possibly as much as 3 (of course I don’t really KNOW the number of games).

I really tire of his schtick.

Right, I realize he was really good, and I guess I overemphasized the portion of my intuition that relied on his weight, but James has a pretty good understanding of baseball history… and saying someone is one of the top 10 of all time, pretty much means they were one of the top 1 or 2 in their decade, I’d think.

And the fact that he was about the 7th best player in a 5-year period that included his prime is sort of calling into question his spot in a top ten that includes another 95 years of baseball.

Bill James definitely has read The Book, or at started to.  He also said that while he never makes notes in books, he was making notes on every single page he was reading of The Book. 

(I don’t know if that means that The Book had so many mistakes that he was correcting them, or if it means it provided so many insights, that he had to highlight them. I didn’t ask, and I’m proceeding on the “highlight” assumption.)

Last I asked (this was when The Book came out), he was at around page 56 or so.  He never said anything else about it, and I never asked.

My guess is that he had some issues with it, and doesn’t want to get in a “debate” about it.  His choice of course.  My preferred path is to debate anything, as long as you hit above the belt.

***

(John Dewan was very complimentary, insisting on calling me personally, to tell me that it was a great book, and it was the book he wanted to write.  Very nice thing to do, considering that he didn’t know me at all.)

***

James had the opportunity to name drop during that Q&A, to at least say “David Pinto” or “MGL’s UZR”, or “Hardball Times”.  He didn’t, other than to drop his own and Dewan’s name.  I was disappointed in that.

"And the fact that he was about the 7th best player in a 5-year period”

should read:

“And the POSSIBILITY that he was about the 7th best player in a 5-year period”

UZR is not fact, but an estimate.  And there’s enough uncertainty in the UZR estimate that he could really be a true 2nd place.  Darin Erstad is on his own planet, according to UZR.

But, good point that if you are one of the 10 best all-time, it pretty much means that you are the best for your generation, notwithstanding the bad luck of say of someone like Schmidt/Brett having overlapping careers.

MGL, what computer?

I think it might be possible to construct a computer manager that can make better decisions than humans, but so far it has not been done, at least to my knowledge.  I’ve never met a simulation game that I think I’m at a disadvantage managing against.  If you had your choice between APBA or me with evenly matched teams, you’d be a fool to pick the computer.  Anybody know if Diamond Mind or Out of the Park does a better job with computer managing?

Not to mention that a computer might think a given relief pitcher is the right pitcher to bring in, but a manager knows he got trashed last night and is hung over for the game.

Of course you could build that into the computer, input how many drinks he had so many hours ago and come up with his BAC.  Or account for nagging injuries, the kind that don’t disable a player but could reduce performance.

Maybe it could be done but I don’t think it exists now.

The other point is that once you’re managing by computer, as soon as anyone figures out what the computer is doing you’re vulnerable to opposing teams managing to the computer.

I think I’d compare it to poker - there are two games going on at once, the card game and the mind game. The best poker players are the ones who can keep track of the cards in play and run the odds in their head, as well as reading the other players and figuring out when they can fudge on the odds productively.

Now, most current MLB managers don’t know how to play cards at all, which clouds the issue - they mostly use a very conservative set of strategies that ALL of them know; APBA baseball might well outmanage Dusty Baker when it comes down to nuts-and-bolts strategy.

What annoys me is that strategy is only one part of a manager’s role - if a manager is bad at strategy, why can’t he simply admit it and punt the question over to someone that is good at strategy? The obsession with the idea that managers have to do everything themselves bugs me.

Colin, you are referring to game theory, and that was covered extensively my MGL in The Book.

As for letting the computer control things, it’s all a matter how much “human beings psyche” impact the the specific decision-making process.

What obsession with managers having to do everything?  Seems like everybody has a bench coach these days.  And managers not knowing how to play cards?  I’d be careful about making generalizations like that.  There’s a lot of time to kill on the plane and bus rides.

The best computers can beat the best humans in chess.  I’m not sure where they are in poker or blackjack.  I think it would be possible to make a computer somewhat unpredictable enough that you couldn’t manage to the computer, not that it’s been done yet though.

I was continuing on with the metaphor - I really don’t know or care how good baseball managers are at poker.

Let’s take as an example Joe Torre. He has a reputation for being bad at handling bullpens, thus his over reliance on certain pitchers in set roles. (I have no idea whether or not this is true, so take it as an illustration only.) It should be possible for Torre to admit that, and lean on his bench/pitching coach to handle that aspect of the game for him. But it doesn’t seem to be possible the way the role of the manager is defined these days.

I thought it was a good interview. Certainly, 100 times better than the 60 Minutes interview (I do understand the reasons why this is so). As far as the critics like Mike and MGL--I won’t dispute the correctness of your specific nit-picks (I agree with them)--but I wonder, if MGL was asked the same wide range of questions as in this interview, how many questionable responses he might have given. I thought that, given the specific questions he was asked, James probably did better overall than almost any other ‘analyst’ would have done. Plus, in the ‘stupid’ responses he gave--’we won’t even know it all in 100 years’, or ‘a computer can’t make correct decisions’--he was probably engaging in some hyperbole to make a point--don’t get too confident in what you think you know…

In this interview, or maybe it was the 60 Minutes one, James repeated his old statement that ‘reaching base vs making an out is the largest difference in baseball. If you keep getting runners on, runs will score. Everything else is trivial’. If teams varied over the entire OBA range of 0 to 1, this would be true. But in the actual OBA range of MLB, there is just as much variation among teams in the % of runners scoring, as there is in the % of batters reaching base…

It is as simple as the computer knows the right answer to all the questions that the manager does not (which are a lot) and then the manager can add anything he wants to that.

When I said that a computer can manage a game and pick up 1 to 3 wins, I wasn’t really referring to bullpen decisions, although a “computer” (we don’t really need to a computer) can certainly help the manager with certain bullpen decisions, as we all know (I’m referring to managers not bringing in closers unless it is a save situation in the 9th and other such things).

When I say “a computer” I simply mean the correct answer, including all the variables.  All of the “intangibles” can be programmed into the “computer” if necessary.

There are plenty of things (actually almost all strategic decisions) in which the “computer” would outplay the average manager (actually ALL managers) by a mile, among them being:

1. IBB or not
2. pinch hit or not
3. when to sac bunt or not
4. when to attempt a steal or not
5. moving your outfield in or out a little, depending upon the relative value of the single and extra base hit
6. where to play the defense, including for a possible bunt
7. when and if to guard the lines or not
8. when and if to execute a hit and run
9. when to take or have a red light on a 3-0 and sometimes a 3-1 count
10. when to play infield in with runner on 3rd

Obviously most or all of these things have a “human” element to them.  Don’t use that as an argument against the “computer” out-playing the manager.  It won’t work.  Either the human element can be put into the algorithm, the result of the algorithm can be tweaked by the human element, the human element, regardless of what it is, is not enough to overcome a clear choice otherwise, or the computer simply says, “It’s too close for me to call, you make the call, using your intuition or what have you.”

The problem with managers making decisions based on what they know that others may not, is that in order to still make the correct decision, using that “extra” information, you still have to know how to process all the information.  A manager (actually, no human being) is not capable of doing that.  Not to mention the fact that half (or more) of the information that managers think they know that you don’t is nonsense anyway.

BTW, a good poker “engine” (computer that plays poker), IIRC, will beat, but just barely, a world class poker player, as well it should.  The only thing that a human being can do better than a computer in poker is to “read” other players (and even then, a computer can do a darn good job, given a history on a player - obviously the computer cannot read “tells"), and perhaps process other information to determine the odds of a player or players having certain hands, based on how they played that hand thus far in the betting (and again, a computer can do the same thing, but it is difficult and requires a lot of sophisticated AI in the program).

David makes a good point about the quantity and breadth of questions.  He certainly showed some patience with some of the more inane questions.

What would you estimate is the probability that a team will allow a computer to make in-game decisions within the next 10 years? The programming is certainly possible.

My understanding of the current situation is that teams have reams of data available but that humans make the in-game decisions. I wouldn’t be surprised to see a team try the Deep Blue approach in a meaningless late-season game in, say, 2015.

Estimating those kinds of probabilities have enormous confidence intervals, at least for me.  But I think that it is certainly a reasonable possibility at any time, now or within 10 years.  If I owned a team, there is no doubt in my mind that I would have a guy with a computer and a sim following the game and making the decisions for the manager.  I take that back.  While I think (I know actually) that the computer would make much better decisions than the manager, it wouldn’t work unless the manager were 100% (or 90%) behind it.  And to some extent the players.

It’s not like all of a sudden you would see bizarre plays.  It’s not like that at all. No one would even know from watching the games.  I also guarantee that people would realize that certain things that are NOT done but are correct are clearly the way to go, just by watching the dynamic of what goes on on the field.  For example, how tough is it to realize that if you randomize your sac bunts and non-sac bunts in situations where it could go either way, that you confuse the defense, and force them to either play wrong or agnostically?

I have always wondered why managers did not figure out that when the defense knows that you are bunting 90%+ of the time, it is difficult to be successful and that much easier to sneak a ground ball (or even a pop fly) hit through the infield.

Another interview:
http://thephoenix.com/Article.aspx?id=59029&page=2

Second time now that he has mentioned PITCHf/x.  He used to talk about some other macro-level analysis as being the next big thing.  I, and most of you here I’m sure, thought of PITCHf/x as the next big thing.  Not sure if that’s a shift in his thinking, but at least he’s acknowledging the obvious power here.

Re #18/19

I’m not a fan of computer input during the game. Let them prepare as much as possible/desired before the game--but once it starts, you’re on your own. I realize that the next question to this line of thinking is whether the manager should have ANY written info (such as a few index cards). But, it doesn’t have to be all or none. Imagine a couple of chess grandmasters in a match with computers at their chairs. Sort of defeats the purpose…

Bill James on his site: “Extreme groundball pitchers do reduce the batting average against them on balls in play”

Me: The batting average on balls in play is lower for flyballs, and higher for groundballs.  The SLG on balls in play is higher on FB than GB.  The DP on balls in play is higher on GB than FB.  OVERALL, the linear weights run value of GB and FB is virtually identical (both around -.10 runs per ball in play).  So, even though a pitcher controls a great deal the extent to which a ball is a GB and FB, overall, it works out almost dead-even.  I have supporting evidence if you want me to look for it. 

Note however: I am removing HR from all this.  Because of the HR, generally speaking, it’s better to be a GB pitcher than FB pitcher.

Perhaps James meant that the hit% on GBs only is lower in the case of extreme GB pitchers?

Actually, he wrote me, and he was including LD with FB, as “air ball”.

I’m not a fan of computer input during the game.

I love the idea, because that is my thing.  I played blackjack in casinos with hidden computers for several years when it was legal to do so.  But I can see and appreciate your point of view.  If they banned computers from the dugout or something like that, I would not be opposed, but until they do, teams pay a lot of money for an extra win, and fans want virtually nothing but to see their team win.  And if a team used a computer for strategy decisions, it is not like anyone would know it and would not have any impact on the integrity of the game.  None, whatsoever.  No one would even notice.  That is the beauty of optimal decision-making in baseball.  No one would think you were doing anything unconventional, except perhaps bringing in your closer a little early now and then and not using him in the 9th as much (OK, I would also never let my 4th and 5th starters bat unless the leverage were low, as we suggest in the book). If a team actually thought it would work, I can’t imagine them eschewing it.

#24, so someone figured out that a pitcher who allows fewer line drives has a lower BABIP?  What a revelation!

If they banned computers in the dugout you can always just have the computer guy communicate with the manager through a headset.

It’s incredible what perceived “culture” does to potential rule changes.  Outfitting a QB (and now LB) with a headset is ok, putting a headset on every coach is ok, but letting a cranky old manager have one of those computer-thingies with keyboards while spittin’ tobaky?  You’d think we’re up against the NRA trying to ban guns.

My manager: “Tom, I’d like to import 8 million Retrosheet records.”

Me: “No problem, I’ll need database software, and...”

My manager: “Now hold a goshdarn minute there… in my day, we didn’t have any database software.  We had punch cards if we were lucky, otherwise, we had everything on paper, and stored in the archives. {spit}”

***

As for David’s purpose of the grandmasters: good point.  After all, if the manager has the computer, then Brian Bannister will be looking at Mike Fast’s PITCHf/x charts for every batter due that inning.

The difference, I find, is that you *can* program a computer to move for you.  In a real sport, the most you can do is supply information to the player, and he has no choice but to exert his physical attributes.

So, while a grandmaster can optionally choose to ignore the computer, it’s possible that he can let the computer make every single move.  In baseball, the player still has to time his swing.

Tango, it has nothing to do with the players executing. It has to do with one manager being smarter and better prepared than another, with this difference largely negated by computer input during a game. The manager, in the area of in-game moves, is like a 10th player. I want to preserve this difference, not eliminate it with computers.

I would think a coach in the NFL is far more an extra player than the manager in MLB. Or, at least as much.  And, there’s no issue with the NFL head coach having a headset talking to all his coaches, who at least one can be in front of a computer, and another in front of video replay.

Why the difference here?

Well, who said the NFL has it right?

Anyway, this is essentially an ‘esthetic’ question--and therefore a matter of opinion, about what makes the game of baseball better.

I gave my opinion, and I stand by it.