THE BOOK--Playing The Percentages In Baseball

Well, if we’re not looking at the game situation, then of course you want to throw the one that will, on average, result in less runs being scored. 

However, we are looking at the game state (which you never mentioned we should ignore), then there may be certain situations in which it is prudent to throw the low outside fastball--for example, with a runner on first and less than two outs it would be best to throw whichever pitch is more likely to generate a groundball, which I suspect is the low-away fastball.

Low and away.

Your phrasing: “which pitch should a pitcher attempt to throw, and why?” The key word is attempt. If you make a mistake down and away, you probably won’t get burned as much as if you make a mistake going up.

If he has perfect control, then by all means take the one which the better value, but there is human error involved.

Low and away.  Zach nailed the explanation.

Jake, no game situation, no nothing.

#2 and #3, while it is not a given that the miss on the low and away as compared to the miss on the high and in will make up for the difference between the average run values of the two pitches (given those exact locations), your reasoning is exactly correct.

You CANNOT use the run values of pitch locations based on hit f/x data to make any decisions about what pitches to throw unless you consider what happens when you miss your exact location (and the distribution of those misses, location-wise), which will happen some non-trivial percentage of the time.

I was thinking about the pitch f/x article or two a while back that told us exactly what I told you - that the high inside fastball was a very effective pitch.  What the data and article did NOT tell you was the run value of a pitch that was ATTEMPTED to be thrown high and inside.  That may be (and probably is) a very different story.  And of course that is the number that most interests us in terms of figuring out what kinds of pitches a pitcher should throw and when he should throw them (and against what kinds of batters).

In general the reason why pitchers do NOT throw high and/or inside that much in this day and age is not because they are not man enough anymore as some broadcasters would have you believe, but it is not necessarily because a high inside fastball is a bad pitch (if it hits that location). It is because a miss on that pitch will more often result in a HR (or extra base hit) or a hit batter.  As well, batters will take a difficult to hit high and inside pitch more often now than they would in the old days when the strike zone was higher than it is now.

The pitchers who SHOULD throw more often inside (and high) on their fastballs are the ones with more control and the ones who throw harder.  That is a very important point, especially the part about having control in order to throw inside.  For example, Rivera with his cutter.  Besides the fact that an inside cutter is a very difficult pitch to hit for a LHB (and you can back door the inside part of the strike zone with a RHB), he can throw that inside because he has such good control that he rarely hits a batter and he rarely accidentally leaves it over the plate which would be a recipe for allowing a long ball…

Exactly right. I get pretty sick and tired of hearing some Royals fans complain that Zack Greinke is the only Royals pitcher who consistently throws his fastball inside. This is because Zack might be the only Royals pitcher with the velo and fastball command to effectively pitch up and in (and it’s a perfect set up pitch for his slider)

MGL, can you help me convince Marv at Sportvision that there is a big payoff to tracking the catcher’s target in the PITCHf/x data?

I’ve been asking him to do that since last year’s summit, where the Royals rep asked if they could do it.  Technically it is not challenging, but Marv’s current view is that there is not enough payback to make it a high priority item for them.  If I can convince him that enough clubs would be interested in that data, I might be able to get him to implement it sooner rather than later.

Mike, didn’t you do something about that in the THT annual (either 07 or 08)? I remember it sounded very work-intensive for one person to track by hand, but I would think that an expansion of your article could do the trick for Marv.

MGL- Not that I disagree with your reasoning, but don’t we actually have to know all of the information to make a good guess?  At a minimum we would need to know the odds that an average pitcher misses his spot, and the run values of the pitch in the the various missed spot.

Mike - Actually it is not a simple task for Sportvision to track the catcher’s target.  The video currently being captured for Pitch f/x do not currently contain images of the catcher’s target at the time he sets up.  The B camera, or high overhead, has the catcher’s glove blocked by the umpire and catcher.  The A camera, or 1st base side, cuts off the image right behind the batter, at least for the two Toronto games for which I have Sportvision camera views.  The C camera, or center field view, has a good image of the catcher’s glove, but not until the pitch is already on its way to the plate.  I looked at a few of those images and they probably would give a pretty good indication of where the catcher had set up, but on some of the images the catcher’s glove seemed to already be moving toward the eventual hit location.  One camera image wouldn’t be enough to give the exact location, however.  Whether it would be close enough for the kind of analysis you are proposing is uncertain.  It is also possible that the A camera on some fields is set up in a position that it is already capturing an image of the catcher’s glove, allowing you to fully triangulate the exact position in space even though it would still be later in time than you would like.

I am sure that Marv would lend you the captured camera clips from a couple of games so that you could calculate the position by hand to use for some analysis.  If you could demonstrate that the existing Pitch f/x camera set up was capturing enouogh information for analysis and that the analysis was yielding interesting results on pitcher’s command, I am sure that including catcher’s target position would move higher up on the list of Sportvision priorities.

Mike, I know STATS, LLC does keep track of catcher targets when they chart the games.  While I would love to have this information available publicly like the rest of Pitch f/x, it is something to which clubs already have access.

But your point remains - it would be great if Sportvision kept track of it as well, if possible.

Mike, of course it is valuable information - hugely valuable.

“At a minimum we would need to know the odds that an average pitcher misses his spot, and the run values of the pitch in the the various missed spot.”

From my last post:

“You CANNOT use the run values of pitch locations based on hit f/x data to make any decisions about what pitches to throw unless you consider what happens when you miss your exact location (and the distribution of those misses, location-wise), which will happen some non-trivial percentage of the time.”

And of course the run values of all the pitches, so you can include the run values of the “misses” in your analysis.  The holy grail of knowing the value of a certain attempted pitch is knowing the distribution of all the locations when that particular pitch is attempted - which is what the point of this thread was.  Just knowing the run value of a particular pitch in a particular location does not really tell you much of anything.

This reminds me a little of sac bunt attempt analysis that for years forgot about the fact that a sac bunt attempt is a completely different animal from a sacrifice hit (out and runner(s) advance).  I am hoping that we don’t go down the same path with pitches and pitch locations…

The real problem is that the analysis is flipped on its head, which means your question is flipped on its head.

It’s easy to classify things as “sacrifice bunt attempts” because the event is discrete and is noted in the data logs.

But it’s nearly impossible to classify “pitching at a particular location”. So rather than binning “pitches that were attempted to be thrown at location X”, you have simply “the value of a pitch that lands in location X”.

What we’re really doing is charting outcomes without being able to chart intent. I mean, to take your example MGL, how would you go about quantifying the value of a bunt attempt if you could ONLY identify the outcomes in the data and not the act of attempting the bunt itself? For example, rather than having “Bunt Attempt” in the data logs, you just had “Single”, or “Man Moved to Second, Man Out At First”, etc. It would be impossible to identify the value of a bunt attempt in that dataset.

On the other hand, you could still get a good idea of what kinds of things are good and bad outcomes. And, I think, what this type of analysis is telling us is what happens on particular outcomes (not on particular attempts).

I agree, tracking the location of the catcher’s glove would be a huge asset. But I wonder if there’s a broadcast story in it for them, because the catcher is almost always in the frame and easy for the viewer to track while the pitch is being thrown.

If the purpose is to tell Cole Hamels where to throw his changeup, you MUST know how often he hits his spots, and when he misses, where does he miss, and does he also miss on speed and movement, etc.

To just say that his changeup is most effective when it lands in the corner, but that when he intends to throw to the corner he actually lets it sit down the middle 40% of the time, that’s not good.

Intent is absolutely what you need, and if all you show is outcome, then the analysis should come with a huge asterisk if you are telling Cole Hamels where he intends to throw.

"This reminds me a little of sac bunt attempt analysis that for years forgot about the fact that a sac bunt attempt is a completely different animal from a sacrifice hit (out and runner(s) advance).”

Yes, I remember you tearing into a commenter at BtB for doing that same thing.  It was a good rant to say the least

Anyway, the question I have, is that given the information we have (IE, fastballs up and in are good, down and away are bad, relatively speaking) and that we also have no other solid information on the subject, wouldn’t it be better to go with the one that we know for sure is better?

"wouldn’t it be better to go with the one that we know for sure is better?”

Absolutely not.  Unless you can make NO inferences whatsoever about the effect of missed locations, which is never going to be the case.  For example, in the case of the inside fastball, if nothing else, you know that the batter will get hit more often.

Dan, even if you have no specific information regarding “attempts” you have to be able to make some inferences otherwise the information is practically useless.

There are so many things about the pitch f/x data that has to be taken in context to have much practical use and meaning.  For example, let’s say that we want to use the run value of a particular pitch to compare pitchers to see who has the “best” curveball or fastball.  As with location, unless we control for other pitches, and the game situation and batters, that information is practically meaningless.  If I have the exact same fastball, for example, as another pitcher, but I throw it 80% of the time and he throws it only 60% of the time, who do you think is going to have the better run value on that fastball, in general.

I have to say that location intent, which can be proxied by catcher glove, location is a HUGE missing piece of information.  Now, in lieu of that, if I were in charge of pitch f/x analysis for a team, I would probably have someone manually charting location intent for all my pitchers.  In fact, no doubt about it.

Peter, in my (albeit limited) manual analysis of catchers’ targets from video, I found that the catcher mainly set three targets: either inside or outside for fastballs, changeups, and sliders and over the middle of the plate for curveballs and the occasional fastball with a three-ball count.  The catcher didn’t seem to vary his target up and down very much.  This is from a sample of four different catchers, who all behaved pretty much the same way.

I found that watching the catcher’s feet was sufficient to identify the placement of his target with respect to inside/middle/outside.  The catcher usually starts with his glove centered between his feet left-to-right.  Presumably his feet would be easier to track from the center-field camera than his glove, and his feet stay stationary longer than his glove does, too.  A catcher typically only moves his feet during the pitch if he has to block a ball in the dirt, and that usually doesn’t happen for a couple tenths of a second after the pitch leaves the pitcher’s hand, which is long enough for the pitch to be detected by the Pfx tracker and the position of the feet noted.

Mike, didn’t you do something about that in the THT annual (either 07 or 08)? I remember it sounded very work-intensive for one person to track by hand, but I would think that an expansion of your article could do the trick for Marv.

Dan, yes, I talked about it in my Cliff Lee article in the 2008 THT Annual. Here’s an excerpt:

When measuring the effect of command, it’s important to know where the pitcher was aiming. A good way to measure this is from the catcher’s target for each pitch. Unfortunately, this information isn’t yet captured by PITCHf/x, so we must consult game video if we want to record the location of the catcher’s glove. Recording the location of the catcher’s glove is complicated by several facts. The television camera angle is generally slightly offset from straight-up center field, and it may shift perspective between left-handed and right-handed batters. In addition, the camera lens has significant spherical distortion. Moreover, the catcher catches the ball a few feet behind home plate, and some pitches may drop more than half a foot from where they crossed the plate to the catcher’s glove.

I watched several games for Cliff Lee and another pitcher and discovered some facts that simplify the process, at least for those two pitchers. (Someone with more experience with baseball may inform me that these facts are not generally applicable, but I believe they apply to Lee, at the minimum.) First, the catchers mainly moved their target left and right. Rarely did the catcher move the target vertically; typically it was set near the bottom of the strike zone. Second, the catchers had roughly three distinct target positions, and keying off the movement of the catcher’s feet was sufficient to indicate which position the catcher was in.

For curveballs, the catcher took a wide stance with his feet and set the target in the middle of the zone left and right. For all other pitch types (fastballs, change-ups, sliders, cutters), the catcher would shift one leg toward the direction the pitch was supposed to go and position the target near the low -outside corner or low-inside corner of the strike zone. On a 3-0 count, the catcher might set up down the middle for the fastball, and he might shade that way on 3-1, but otherwise all four catchers I watched seemed to have three basic target positions.

Mike: If the catcher’s feet work, would the top of the catcher’s head also serve as a reasonable proxy? You might be able to capture that from the over-the-plate camera installed in some parks, no?

I’m serious about you guys with affiliations or connections with clubs to tell them to talk to Sportvision or MLBAM about getting this included in the data set.  Marv has told me that he doesn’t think it’s technically difficult, so the fact it’s not currently in the data set is because he doesn’t think his consumers are interested.  His two primary paying consumers are broadcasters and MLB (MLBAM and the clubs); neither of those include me.

I was hoping my Cliff Lee analysis in last year’s annual would spur some interest in that direction among the clubs, but it doesn’t seem to have, at least not the extent that they talked to Sportvision about it.

Like MGL says, you can have someone record that data manually, and that’s better than nothing, but automated and digitized data has so many advantages over manually collected data, particularly when the number of data points is in the millions.

Dan/#18, I would guess that it would, but the Sportvision already collects and stores the center-field camera footage and has it for every park.

I really don’t think the challenge is mainly a technical one, it’s a question of perceived interest and value by the parties of influence.

I sent an email to Marv asking him exactly the protocols here.  I’ll do whatever I can to influence our needs (if I even have any influence).

I’ve been on the “hang time” bandwagon since before Dudek’s groundbreaking article, and his article should have been the nail in the coffin.  And still, we had to wait four years before the operations folks saw the light.

This is the same b.s. I have to put up with at work.  The users out on the field know exactly what they want.  But, their requirements do not necessarily get translated by the time I, as the technical implementer, gets to see them.  The idea is that everyone between us are the ones best positioned to translate their needs, and prioritize them.  Yes, it works most of the time.  There are gaps. 

I’ve always believed that after all the middle players do the bulk of the work that it would be advantageous for me to actually sit with the end users and see them in action, and I’d be able to fill those missing gaps.

Mike - I took a look at my Sportvision clips after reading your Posts 16 and 17 and they showed the same catcher shift of feet positions that you noted.  I didn’t realize that inside, over the plate, outside would be enough for the analysis that you were proposing.  The feet positions are easily seen on the C clips from the center field camera.  They are sometimes visible for left handed batters in the B clips (high home), but not consistent enough to be of use.  The pitch f/x operator could probably click on a foot position for each pitch, but that’s one more thing he or she has to do.  I have no idea how hard it would be to program to automatically recognize the foot position like Pitch f/x is able to do for the ball position.  My impression was that the ball position algorithm compared changes from frame to frame to pick out the ball.  It seems like you would have to use some other method to recognize the catcher’s foot.  You also would have to have some static registration points to compare to in the C clip frame.  Probably doable, but more work than Hit f/x was for example.

Peter, thanks for taking a look at that.

I didn’t realize that inside, over the plate, outside would be enough for the analysis that you were proposing.

I am asserting that inside/middle/outside is all the catcher typically does.  The catcher does not typically set a more precise target than that, i.e., he does not set a two-dimensional target, he sets a target in one dimension and at limited positions in that one dimension, at that.  I may very well be wrong about that as it is an assertion based on a small sample size (two pitchers, four catchers, over several games).

In those games, I noticed only a small handful of exceptions: once or twice the catcher set a target noticeably higher than the bottom of the zone, and a few times the catcher shaded his target for a changeup from the inside/outside edge toward the middle of the plate.

I’d be interested in comments either confirming or denying my assertion.  Probably I need to do more legwork on this, but maybe there are some catchers or other experienced people out there who can save me some legwork with a comment.

I’ll also mention that the pitcher’s ability (or lack thereof) to hit the target dwarfs any small changes the catcher may make from pitch to pitch.  For example, if the catcher calls twenty inside fastballs during a game, his target position for them probably moves around by a few inches.  However, the pitcher is typically going to miss the target by a foot or more on average.  Missing by two feet is not unusual even for pitchers with good control.

Mike: Jason Varitek will often “stand” for high fastballs that he wants out of the zone, and catchers often motion “down, down” with their glove before the pitch if they want something low.

But, on the average, I think you’re probably right, in that it’s mostly 1-D.

I watch thousands of games, and Dan is basically correct.  It is practically impossible to set a low target, so, as Dan says, the catcher often motions “down, down” with his glove. Not all the time, but some of the time when he really wants a low pitch, like an 0-2 or 1-2 off-speed pitch. 

Some catchers, like Varitek, almost stand up and/or hold the glove really high for a fastball out of the zone.  In general though, it is assumed that an outside pitch is always wanted low.  An inside fastball can be high or low, though, although the high inside pitch is more common at least for a fastball (some pitchers throw sliders or curveballs low and “under the hands").

But in general, I agree that most of what you get from the target is inside, outside, or middle of the plate (middle of the plate could be a 3-0 fastball, it could be a “get me over” curveball, it could even be a low splitter or changeup, or it could be any pitch at any count from a pitcher with little or no control like a Zumaya fastball).