THE BOOK--Playing The Percentages In Baseball

I have not read this piece yet, but this is one of the things I have been looking for a while now from the pitch f/x guys.  Unfortunately, many of them (with all due respect, as I love their work) are spending WAY too much time worrying about classifying pitches (which matters none).

Of course, one game is worthless.  One of the most important things a pitcher can do is to properly randomize his pitches, given the batter, count, game (score, inning, base/out state) situation, and perhaps most importantly, the quality of his own pitches.

My guess is that some pitchers (and catchers) are a little better than others at doing this, and that makes them, overall, considerably better pitchers.  And of course the beauty of that, from a team, or teaching perspective, is that you can teach a pitcher (and catcher) to “sequence” his pitchers better. Every team should be analyzing their pitchers to see who could benefit from a “class” in game theory, randomizing pitches, and sequencing of pitches, depending on count, game situation, the batter, etc.  In fact, I will volunteer, at no charge, to teach that class!

Of course, we need to start with a baseline of league average.  And it is probably best to start with count.  So, for example, we look at all 0-0 counts, and see how often each pitch is thrown by all pitchers combined.  Then you probably break that up according to base/out state. On and on (certain batters, etc.).

It is absolutely critical for a pitcher to randomize his pitches in every count/game situation, so the batter has the least chance of knowing what is coming.  Now, what those percentages are, given the count and situation, entirely depends on how good each pitch is for that particular pitcher, and how well he can command each one of those.  For example, if a pitcher had a 105 mph fastball which was virtually impossible to hit, there would be little reason to throw anything else, even if the batter knew that.  As far as commanding each type of pitch, for example, in a 3-1 count, the only reason that pitchers throw mostly fastballs, even if they don’t have a great fastball, is that they can’t get another pitcher over the plate often enough for it to be worth throwing that pitch, even if the batter is fooled.

A great clue that a pitcher is not good, for example, would be if his fastball were not that great in the first place, but that he threw 90% fastballs in fastball counts (meaning that he either is throwing them too often in those counts, or he can’t get his offspeed over enough or his offspeed pitches are so bad that it wouldn’t make enough of a difference if the batter were fooled by them).

In looking at pitch sequencing, we also have to be careful about testing for randomness.  A pitcher could, say, throw 50% fastballs at a 2-1 count, but that does NOT mean that he is randomly throwing 50% fastballs at that count.  It might be that he throws 70% fastballs to some batters and 20% to other batters, which may or may not be correct.  Or, more likely (if he was not randomizing properly), it might be that he was throwing 30% fastballs after the last pitch was a fastball or a curveball, or whatever.  A pitcher does not want to fall into the trap of following one pitcher with another, predictable one, like in Little League, where the pitcher tries to throw a curve, and bounces it 10 feet in front of the plate, and then never throws a curve on the next pitch no matter what the count.

Something similar to that would be the “setup” pitch in MLB.  How many times have you heard the announcer, after a high tight fastball on a 0-2 or 1-2 pitch, say that it was a setup pitch, and the next pitch was going to be an off-speed away?

Well, that is ridiculous, of course, from the pitcher’s perspective.  If the next pitch is going to be an off-speed, away, 85 or 90% of the time, then the batter simply knows what’s coming.  It is not like a major league hitter is going to bail out on the next pitch just because the last pitch was high and tight. The idea of a “set-up” pitch followed by a predictable one 90% of the time, is ridiculous.  Yes, after a high and tight pitch, you might be able to throw off-speed away a small percentage of the time more than you would normally, but not to the extent that anyone should “know” what is coming on the next pitch.

So is (ridiculous) the idea that a “smart” pitcher like Maddux can predict almost every pitch that the opponent pitcher throws, which you occasionally hear.  (And of course, if he could, he would imply relay that info to his batters.) No person should EVER be able to predict what a pitcher throws to any degree of accuracy, other than, “There is a 70% chance he throws this, 20% that, and 10% that,” or whatever.

Anyway, pitch sequencing and game theory is a critical part of pitch f/x analysis.  So far, the pitch f/x guys have been spending WAY too much time on esoteric (and some near worthless) things, like pitch classification, release points, spin rates, movement, etc., and not enough time on analyzing the things that make pitchers good or bad, so that we can evaluate pitchers in a much more reliable and granular level than their performance (in regular component stats).

MGL, what you wrote makes very little sense to me. 

You say that pitch classification is near worthless, and then you spend nine paragraphs talking about the importance of something (pitch sequencing) that relies on pitch classification in order to ever be done in the first place.

IMO, pitch classification and data correction are the two keys to the PITCHf/x data set, and we will like never get anywhere without them.  If data correction can be done close to perfectly, pitch classification gets a heck of lot of easier and maybe recedes into the background for some purposes.  Given the state of the data today, pitch classification is huge, and I don’t think you’ll get much out of the data without it.

I’m not arguing your point that a lot of people are just dabbling in the data for fun and not tackling the serious research issues, but I think your priorities for the order in which to tackle the serious research issues are out of whack.  I’m willing to be shown wrong, of course.

I agree that if you care about “sequencing”, then you need to worry about pitch classification.  Now, whether you need to worry about creating as many buckets as possible (4-seamer, 2-seamers, cutters, sinkers, sliders, curves, changeup) is another story (and maybe that’s MGL’s peeve).  After all, it’s what does the batter perceive, and I think that there are really only 3 or 4 types of pitches that he’s looking for.

At the moment, I’m happy with everything the PITCHf/x-ers are doing.

By pitch classification, I mean whether a pitch is a “cutter” or a “2-seam fastball,” or a “splitter” or a “changeup.”

What I meant was that you can call it a “grog” and a “flop” and a “zolch” for all I care.

For each pitcher, find out what he throws, and group his pitches “roughly” (by no means does it have to be perfect) into those buckets.  End of story.

And I certainly don’t care what the pitch f/x guys call them, unless you want to save yourself the trouble of classifying the pitches yourself.  Which would be nice of course, if you are the one doing the work.

So I meant, “pitch naming,” not “classification.”

And yes, without having to be politically correct (which I rarely am), I love the work that Mike and every other pitch f/x guy is doing, as I have said many times in the past, and I will continue to say, but I am not in the “business” of unduly stroking anyone’s ego either.  I think that a lot of time is being wasted on pitch naming, pitch classification, pitch movement, release points, etc. and not nearly enough time is being spent analyzing what makes pitchers good, bad, etc., and who are the good ones, bad ones, and why.

I mean, we’ve already had almost a year to do analysis, and there have probably been 2 dozen articles already.  Surely by now, we would have a list of the best and worst pitchers by “randomizing their pitches” (along with average baselines), by command of each pitch (some formula for in the zone, out of the zone at the various counts, plus in the heart of the plate, not in the heart of the plate), etc.

If the above two issues have been addressed then don’t take the above sentence literally.  Either way, I stand by my statement that too much time is being spent on NOT so important esoteric things and not enough time on really important issues pertaining to pitching talent.

Now, take that all with a grain of salt for two reasons.  One, I am not the one doing the research.  Two, it is just my opinion.  What may be important to me may not be important to other people.  For example, I hate anything to do with WPA and things that have little to do with a repeatable skill.  Hate them - think they are worthless.  But people seem to like them, so bully for them.

Let’s talk about the fastball and different variations (4, 2, cuts).  In order to know if there’s any value to distinguishing them, we have to figure out if they result in anything different.

For example, say you take a pitcher, the same pitcher, and he throws at 0-0 a 4-seamer, and at 0-1 a 2-seamer. And say he throws a 4-seamer at 0-0 followed by another 4-seamer at 0-1.

Now, if that results in the same BA/SLG in either case, with the same percentage of called balls, strikes and swinging strikes, with the same GB%, then we can safely presume that a 4-seamer and 2-seamer is really just a nuanced difference, since the batter thinks it’s really the same thing. If it looks like a duck and quacks like a duck…

So, that would be the next step in all these nomenclatures. Are we really needing to separate things into 7 or 10 different pitch names, or are we really only talking about 4 pitch types?

We’re at the mercy of sample size of course.  But, if we can’t distinguish with much statistical significance between a 4 and 2 seamer, then why bother separating them at all? Even if you need to separate them because you “know” there’s a difference, the end result is that if you look at the results of the PA, they are the same thing.

(All conclusions based on illustrations only.)