THE BOOK--Playing The Percentages In Baseball

Absolutely.  The world is full of talking head and lawyers, and they only confuse things.

When I was a kid, before knew enough to be right about every political question (like I am now), I would read the op-ed page of the newspaper.  There would be an essay on issue X.  It would make very reasonable arguments, and it was well-written, and it sounded like the guy knew what he was talking about.  So I would start to think he must be right.

Then, the next day, there would be a rebuttal.  It would make very reasonable arguments, and it was well-written, and it sounded like the guy knew what he was talking about.  But it took the exact opposite position of the guy the day before.

Moral: on a topic I know little about, anyone can sound really, really good while being completely wrong.  If you want me to believe you, show me evidence and logic.

My 2 cents, for what it’s worth on sports mechanics in general (I play tennis, not baseball):

I find in tennis there are players who have textbook mechanics, and there are players who don’t.  What does this mean?  Not much.  Steffi Graf’s incredibly high ball toss on her serve didn’t stop her from winning 22 grand slams.  And her slice only backhand, which is unorthodox, was particularly nasty.  That’s just an example of one great player, there are lots of other great players who all have their own quirks.  I’m sure the same applies to baseball players.

As far as baseball mechanics go, I think the release point data from MLBAM is really quite interesting.  That’s the only “mechanics” stat that I can think of.  It could be very telling when someone does a full season study on it.

We were doing an overhaul at our company on performance reviews. 

One email said: “It’s a GREAT idea to do performance reviews for the whole company at one point in the year, rather than spread out, because we can bunch the reviews together”.

Another email said: “It’s a TERRIBLE idea to do performance reviews for the whole company at one point in the year, rather than spread out, because we can bunch the reviews together”.

In fact, every time I hear a “policy change” whether at work or in the business world, I do this little exercise.  Take the complete opposite viewpoint.  If their argument for their originally policy change makes sense with the opposite viewpoint, it’s b.s. 

This was the case every time we implemented and deimplemented “casual Friday”.  For you youngsters who haven’t read Orwell’s 1984, read it.  We are living it.

***

Related to MGL’s blog entry, Will Carroll says this:
http://www.baseballprospectus.com/unfiltered/?p=337

UPDATE: The next person that suggests YouTube videos gets to sit through a lecture on copyright law. The next person that suggest that MLB doesn’t hold rights to video and that I should just do it gets the 2007 Aimster Award.

G-dd-mn are people afraid of copyright law.  The law is our FRIEND:
http://www.copyright.gov/fls/fl102.html

One of the more important limitations is the doctrine of “fair use.” Although fair use was not mentioned in the previous copyright law, the doctrine has developed through a substantial number of court decisions over the years. This doctrine has been codified in section 107 of the copyright law.

If Will wants to teach someone about a Gyroball, and puts up video of Matsuzaka throwing a Gyroball, a slider, and a curveball, it sure looks like he can do it to me.

If you ever notice, Howard Stern does this all the time.  He’ll play a copyrighted clip, then stop in the middle of it, and provide his witty commentary on the clip.  Then, he’ll continue the clip.  Or sometimes, he’ll paradody the clip.

These are the kinds of things that copyright law is there for.

For example, the quote of Will Carroll I have above?  That is in effect copyrighted by Will Carroll.  But, it is fair use for me to take that quote, paste it here, and call b-llsh-t on Will. (I’m “email friends” with Will.  Always very cordial.) He in turn can cut/paste anything I have in this post, attribute it to me, and then he can provide his commentary against what I said.

THE BIG PRINT
Since we live in a litigious society: I’m not a lawyer, so don’t take any legal advice from me.  Everything I said can be b-llsh-t.  When it comes to the law, if it hasn’t been exactly spelled out, judgement on a particular case will be made in a courtroom.

I’m hoping this is a good opportunity for me to get an issue cleared up that I’ve been struggling with.

It seems to me that the statistical argument with regards to slumps is that when a player is slumping it is just the random occurrence that is to be expect based on the probably outcomes for the situation.

If a player is expected to have a batting average of .300 and he hits .150 over a period of 10 games, it does not mean that there is anything wrong.  It is just part of the normal statistical distribution.

I have some understanding of probability and I think I understand the mechanics of the argument if not that actual math behind it.

What I have trouble with is it seems like that batter going to the plate and rolling dice.  Are players really just random output machines.  Are they wired at a particular talent level that determines their probabilities for different outcomes and when it is their time at the plate, the just spit out the next output?

When I saw Will’s post on using videos, he at first said it was a matter of cost, something about BP having to double their subscription rates if they did that. 

I immediately thought of Carlos posting on BTF, which anyone can see for free.

But the real problem with either Will or Carlos showing a gyroball is the resolution available on the internet.  To do the pitch justice somebody on ESPN is going to have to do it for their HD viewers.

Deryl/4: each player (pitcher, batter, fielders) has his own mean for a particular context.  Each player has his own standard deviation around that mean (though, likely follows the binomial or multinomial distribution).

What separates the player from the machine is that his “own mean for a particular context” includes himself in that context.  If a player is feeling particularly s-cky one day, he has a new mean, even if the rest of the context is unchanged.

However, if it’s true that each player has a dynamic mean, then the true spread of outcomes for that player must be, by definition, wider than expected from the binomial distribution around a static mean.  And what we do find is that a player’s mean in any particular context if far more static than dynamic. 

In short, given the choice between choosing to give a player a static mean rather than dynamic, choose the static.

And to the extent that a player, being human, does indeed have a dynamic mean, you’d be hardpressed to figure that out.  Who the heck really know if Pujols today is a true .435 wOBA and not .427?

Tangotiger-

Thank you.  That was awesome.

MGL, I pose some inquiries about your statements on slumps in the BTF thread.  I assume you’d prefer to respond to them there (if you’re so inclined, obviously, you’re not obligated), but I thought I’d let you know here.

To summarize:

1.  Why would we expect a player to have diminished performance after a slump?

2.  If every slump had an identifiable cause, would that data look different from what we have now?

The slump period is predefined for *all* players as the exact same amount.  So, we are not talking about 20 or 25 or 50 PA, etc.  We are talking about exactly one of those numbers.

If you look at the worst performances during 7 days, what do you think will happen on the 8th day?  Presumably, the worst performances during days 1 through 7 included bad luck, as well as some injuries or “mechanical” changes.  On the 8th day, all the bad luck players resume their normal talent levels, while the injured players continue being injured.  Overall, therefore, you expect below typical mean performance.

That makes sense, Tango, but why would we expect every person’s slump to last the same amount of time? 

In your example of the 8th day, when all the bad luck players resume their normal levels, why would we expect all injured players to continue being injured?  Or, in the possible example of a “mechanical” problem, why would we expect that problem to last into the 8th day?

When MGL says, in the BTF thread, “if indeed some slumps were due to mechanical, mental problems as opposed to just statistical fluked, the percentage of slumping players who had those problems would have to be infinitesimal, otherwise they would cause the numbers to go down,” that presumes that we would expect those problems to out-last the period defined as the slumping period; I don’t know that that’s a safe presumption.

These issues may be addressed in the chapter, which I haven’t read in awhile, but may re-visit now that this has cropped up.

If we assume that slumps are real, then a player’s overall batting line will be less than his true ability when he is not slumping.

So we have some players continue to slump, and some return to their higher, non-slump hitting, and the result in day 8 will be the same as their overall stats, which include hot streaks, slumps, and everything in between.

If some significant percentage of slumps are due to some mechanical or mental problem, the researcher still has no idea how long it will last (until the problem is corrected) of course.  The presumption is that that will be different for everyone.  For some, it is 1 day.  For others 2 days.  For others 1 week.  For others 2 weeks.  Etc.

So we merely say (when we do the research) that a “slump” is when a player has done extremely poorly for a week straight at any time in the season.  Admittedly, even if these problems did exist and persist, we are also capturing some bad luck for no reason.  But, presumably we are also capturing lots of slumps due to mechanical problems.

Now, what are the odds that for all the players who were indeed having problems, those problems are going to be over exactly 7 days after a bad period?  Surely some (actually most) of those players are still having problems.

Heck, even though we are isolating 7 day bad periods, the problems may have started a day or two before - the first 5 or 6 days of bad performance may have been bad luck only.  So certainly some (again, most) of those problems will persist for one more day.

Again, if it is true that more than an ifinitesimal percentage of 7 day bad performances are caused by “problems” and if more than a small percentage of those problems will persist for at least one more day (again, why should they all of a sudden get corrected exactly 7 days after a bade spate of performance?), then BY DEFINITION (what I mean is that it is a mathematical certainty) if we aggregate the performances of all these batters in a 7-day (or we can do the same thing with 3 or 4 day, or whatever) slump, we would find that their collective performance on the next day (or next 2 days, or whatever) would be LESS than their overall performance, or baseline true talent level.

We DON’T find that to be the case, so we have to reject the proposition that a significant (more than a tiny fraction) of players who have had a terrible spate of performance for at least 7 days have something wrong with them and that this will persist for at least one more day.

It is the same things, but to help you understand, think of the study as being about batters who are IN THE MIDST OF a slump which has gone on for 7 days, rather than HAVING HAD A SLUMP for 7 days.

The best studies mimic the conditions of real life that exist when the questions pop into our heads (or some taking head commentator suggests the question).  For example, imagine all the times when a player comes to bat, and the announcer says, “So-and-so is 2 for his last 25 (a bad spate of performance over the last 7 days), he is mired in a terrible slump.” Now the announcer starts talking about how he has been pulling off the ball, swinging at bad pitches, has no confidence, etc., with the presumption that he will hit worse than, say, his career average, for some period of time from now on, at least until his problems are corrected (could take one day, one week, etc.).

In order to test that proposition, all we have to do is to put our historical data in the exact same position and look at the results.  Do they support the “will continue to hit poorly for SOME period of time,” or do they not?  We find that they do not.

It is like my oxygen example.

Keep in mind that like the clutch debate, that does NOT mean that there are not some small percentage of players who are not experiencing problems that will tend to make their poor performance persist for some time.  It also does NOT mean that a manager or coach cannot identify these small subset of players, although I doubt they can.  If they could, also keep in mind that they would literally have to look at all players in one week terrible slumps and conclude that only 1% (or whatever the number is - it has to be small) are really having problems - the other 99% are fine and have just had a flukey, unlucky, week of bad performance.

MGL, that all makes sense, and I don’t think we substantially disagree.

MGL:  I agree with your basic argument.  But I wonder if there may be a small “injury exception” that wouldn’t get picked up in your general analysis.  For example, if we learned that over the 10 days prior to going on the DL (or 10 days after coming off the DL), injured pitchers generally allowed a wOBA-against .030 higher than their norm, that would be pretty strong evidence that pitching while hurt had reduced their performance.  Have you ever looked at this issue?

For one thing, we actually see a small hot and cold streak effect for pitchers.  I speculate that it is associated with health and cite some DL data.  You’ll have to re-read the chapter in The Book.  I have no doubt that there are a subset of pitchers who pitch really poorly for a while and then go on the DL.

I just re-read the chapter in The Book.

One thing I find interesting is the choice to use absolute thresholds for “hot” and “cold” streaks instead of looking at performance relative to a player’s mean.  (For those unfamiliar with the chapter, a hot streak is defined as five games in which a hitter has a wOBA equal to or greater than .525, and a cold streak as equal to or less than .195.)

I understand the rationale offered up for the choice, but I do wonder what we might find if some threshold were used that was relative to each player’s established performance.  I would guess that hot streaks would not be very predictive (regression to the mean), but I’m not so sure that cold streaks would be.  However, I have no idea what those thresholds would be (I don’t know what the SD for wOBA is over 20 PA) or if any trends would be detectable in all the noise of day-to-day variation in performance.

The fact that considerably more of a hot streak or slump “hangover” is detected amongst pitchers than hitters is certainly intuitive, however.

The SD for wOBA is calculated similarly for OBP.  Instead of OBP*(1-OBP), make it wOBA*(1.1-wOBA).

More mechanical analysis:
http://baseballanalysts.com/archives/2007/04/as_the_arod_tur_1.php

I share the concern as to trusting whether one set of mechanics are “right or wrong”, which I suspect is player specific anyway, but rather it’s interesting, possibly useful, to see the mechanical differences between two players (a good one and bad one, for example), or the same player when they’re playing well and when their not.  I think the problem is making predictions based on someones mechanics, based on these articles.  It’d be really interesting for someone to breakdown the same player, over a given year, week, game, etc, to see how consistent players are with their own mechanics.  One of the problems I see with these articles is that the author will cherry pick two (or more) swings or pitches and compare only those sequences.  I think it’d be more useful to do the same thing for many, or every pitch or swing (or do the same thing for fielding).  Simply looking at two swings is not enough to do any kind of analysis.  Perhaps a computer program could do the analysis.  I’m sure it’s possible, since we have stuff like facial recongition, just not easy, and would be prone to error.  I could see something like automated scouting developed within the next few years.

Well, honestly, if the absolute threshhold does or does not capture a streak effect, I don’t think that a relative one will either.  The threshholds are so extreme that it will include almost all of the relative ones and vice versa.  An important think for everyone to remember in these kinds of studies looking for an effect of a certain phenomenon is that it doesn’t really matter how you define the phenomenon.  If an effecdt exists, it will show up no matter how you define the phenomenon as long as your definition is reasonable.  If an effect shows up for one definition and not for the other, it is likely that that effect is very small or non-existent and that the result if an artifact of sample error.  It irks me when someone criticizes the conclusion of a study on clutch hitting, for example, by criticizing the definition of clutch hitting.  If clutch hitting exists as a SIGNIFICANT skill, it will show up no matter how you define clutch as long as your definition is reasonable and as long as it reflects what the batter is likely thinking (e.g., there is no reason to include as clutch a situation where the batter does not think it is clutch).

I have no quarrel with the observation sof pitchers and batters mechanics and how they differ from day to day or from season to season.  It is the conclusions based on these observations (if indeed there are any) that I question until we see evidence that they are true or likely.  For example, to say, “Look at this pitcher’s mechanics while he was doing poorly and look at it while he was doing well,” is a step in the right direction, but by no means does it tell us about mechanics relate to effectiveness or to future health, unless we start conducting “real” PROSPECTIVE experiments with the observations and the data.  For example, it may in fact be one of the variables associated with what we perceive as random fluctuations in performance by both pitchers and hitters is day-to-day (or whatever time period) variabliity in mechanics.