169 replies to this topic

[Reverend]

My concern is that his change is not the same pitch it used to be.

Meaning not that the total average spin is off. I mean the speed, spin, angle, horizontal movement and lateral movement are all off, as well as the spin. And it's off when you look at data month to month, and year to year; what's he's been throwing this year is different from every other sample I've looked at and I've looked at more than I really care to admit. (as per before before, it beats working.) I But it seems like the variances on given outings looks larger, though I have not attempted to quantify that.

In truth, I am loathe to disagree with you, and in truth again, I hope you are correct that this is a mistaken approach, but my concern is not that there is some single aspect off but that it is literally not the same pitch. If that is, in fact the case, if he's going to throw the change, he needs to find a new change.

But even all of that doesn't address my concerns with his fastball. Isn't the change effective because it's used against the fastball? Doesn't the fastball look rather different as well?

[StupendousMan]

A quick note on the accuracy of the spin numbers reported from pitch/fx: the input to these algorithms is a set of positions and times. Positions of the ball in three dimensional space, based on its appearance in video taken from several different angles, and time, based on the same video frames. A fastball reaches the plate in about 0.41 seconds, and the pitchf/x cameras take images at 60 frames per second, so the typical pitch might be measured in 25 or so frames.

In order to derive the spin rate from this information, one looks for deviations in the trajectory from the path that a ball without spin would take. That means applying Newtonian mechanics to ball -- easy peasy -- but also accounting for air resistance -- not quite so easy. After making a fit to the trajectory with these factors, one is left with deviations. Those deviations are due to the complicated effect of the spin of the ball and its interaction with the air.

If the deviations are huge, then one can make a relatively unambiguous reconstruction of the spin. For example, if a ball ends up 2 feet above the location of a spin-less ball, then by golly, there was a lot of backspin on it.

On the other hand, if the ball ends up in NEARLY the same spot in which a spin-less ball would end up, then it's harder to figure out what was going on. Maybe it had very little spin. Maybe it had a small amount of spin to the left, but was thrown with an unusual release. Maybe it had a small amount of spin down, but the measurements were slightly inaccurate. It will be harder to place firm numbers on the properties of a ball with little break.

Another way of saying this is -- following Alan Nathan's paper

http://webusers.npl..../MCAnalysis.pdf

that the uncertainty in the derived value of the break on a pitch is around 2 to 2.5 inches in each direction. If the total break of a pitch is small -- say, 4 to 6 inches -- then this random error can make a significant change in the derived break, which in turn can mean a significant error in the derived spin.

tl;dr: can't trust quite so much the derived spin on low-break pitches such as the changeup.

[smastroyin]

I realize getting into the nitty gritty of each pitch is a good way to dissect and look for root cause. But I just want to illustrate this problem.

Here is Buchholz K/9 and K/BB rates throughout his career, majors and minors. Really simple, but we can see the way his K rate plummeted after the initial tinkering in 2008. He has succeeded in cutting down walks as his K/BB rate has been consistent throughout (at least in the majors). But looking at the guy in the minors, I'm not sure why they felt the need to really change who he was. (I probably should have left the 2010 data off completely, it is only one rehab appearance, I just don't feel like going through the rigmarole of changing the chart, saving it to an image that I then have to upload to tinypic, etc.)



Just as a quick comparison, let's take a look at a similar chart for Lester.



We always heard about this being the model for the way they were going to rebuild Buchholz. Strip down his K-rate a bit in favor of getting him to work better in the strike zone, and then let his stuff build him back up. We see the step change with Lester from 08 to 09 in K-rate, you can also see that because his walk rates are now pretty consistent, the increased K-rate gave him a big boost to K/BB. Of course I am also going to try not to worry about the drop off this year due to small sample and April struggles. I also think that even if the data doesn't show it, Lester "needed" this rebuild a lot more than Clay, as he accumulated prodigious pitch count numbers even when getting good results. Regardless, as it pertains to Clay we can very easily see that he has never found a strikeout approach again. Whether this is strategic or a result of the effectiveness of his pitches is obviously in some question.

We'll never know what would have happened if they just let Clay throw. It is entirely possible that 2007 Clay was never ever going to happen again and that if they didn't try to change him he would have been permanently stuck in that first half of 2008 mode. But I still think the way they chose to develop him was a mistake, spoke to a little too much of either hubris or a need to fix something that didn't need fixing.

[Sprowl]

Did Clay get squeezed in the 7th inning? Not really, although there was one borderline ball called at the top of the strike zone. All the other ball-and-strike calls look to be technically correct, although certainly not generous. Buchholz got 8 swinging strikes in the game, but only one during his 27-pitch last inning. Even so, the performance was encouraging for more weak groundballs, fewer location mistakes on the fastball and cutter, and better deception on the changeup. Even Reddick's golf shot RH was on a well-located curve near the ankles.

[absintheofmalaise]

Have the Red Sox beat reporters asked Buchholz and McClure about the mechanical changes -- what the changes are and what they are supposed to accomplish? If they haven't put the question to them, I wish they would.

Taken care of.

[Van Everyman]

Did Clay get squeezed in the 7th inning? Not really, although there was one borderline ball called at the top of the strike zone. All the other ball-and-strike calls look to be technically correct, although certainly not generous. Buchholz got 8 swinging strikes in the game, but only one during his 27-pitch last inning. Even so, the performance was encouraging for more weak groundballs, fewer location mistakes on the fastball and cutter, and better deception on the changeup. Even Reddick's golf shot RH was on a well-located curve near the ankles.

Five runs in that inning was honestly the worst possible outcome given how Buchholz pitched. He did put the first two guys on, but should have gotten out of it on the doubleplay ball -- the runner was out of the baseline by a good 5 or 6 feet. And then, yes, the rather strict reading of the strike zone. And as noted, the ball Reddick hit out was a good pitch. Had any one of those things gone Clay's way, we are having a very different conversation about his performance.

[Reverend]

I think RPMs are very important for understanding the movement of fastballs and for curves, but less reliable as an indicator of hittability when it comes to changeups, cutters and sliders because their spin axis can be close to perpendicular to the strike zone. In such cases, PitchFX is missing the biggest part of the story. Alan Nathan mentioned that one firm now computes true spin rate in all dimensions (eg, how fast does a gyroball spiral, even if Pitchfx thinks it does not spin at all). It should be possible to get better insight into cutter and slider effectiveness using true spin rate.

The changeup still depends on the black box of deception.

Looking back over this very good point and czar's great post, I realized that I didn't post all of what I thought was interesting and, perhaps, provacative; I'm eager to hear thoughts on the rest.

I was discussing the change largely because that's how a couple of us initially started looking at the data, i.e. as the research question: Why isn't Clay throwing his change? This led to a consideration that the answer might be because it lost its effectiveness. This, in turn, led to a consideration that it's not the same change. Ultimately, though, the data began to suggest that Buchholz's has a comprehensively different composition as a pitcher altogether.

	Pitch	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Selection	Strike	Swing	Whiff	Foul	In Play
2010	FF	1492	53.4%	94.1	9.38	-6.9	217	2,389	53.4%	63.7%	45.0%	4.5%	19.7%	20.8%
DET: 4/8	FF	28	35.9%	91.8	9.61	-5.52	210	2,217	35.9%	67.9%	46.4%	3.6%	28.6%	14.3%
TBR:4/14	FF	44	42.7%	91.3	8.31	-5.06	211	1,932	42.7%	72.7%	61.4%	0.0%	29.5%	31.8%
NYY: 4/20	FF	45	43.7%	92.2	8.8	-4.6	208	1,992	43.7%	82.2%	60.0%	2.2%	28.9%	28.9%
MIN: 4/25	FF	47	43.9%	92.1	7.49	-4.99	214	1,822	43.9%	68.1%	38.3%	2.1%	14.9%	21.3%
OAK: 4/30	FF	47	48.0%	92.4	9.54	-4.54	206	2,108	48.0%	59.6%	34.0%	2.1%	14.9%	17.0%
	Pitch	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Selection	Strike	Swing	Whiff	Foul	In Play
2010	CU	235	8.4%	78.3	-8.45	5.57	34	1,685	8.4%	48.1%	31.1%	8.1%	11.5%	11.5%
DET: 4/8	CU	13	16.7%	75.4	-9.76	7.09	36	1,920	16.7%	61.5%	53.8%	30.8%	0.0%	23.1%
TBR:4/14	CU	16	15.5%	75.4	-8.99	9.34	46	2,066	15.5%	56.3%	31.3%	18.8%	0.0%	12.5%
NYY: 4/20	CU	19	18.4%	75.7	-8.61	9.7	49	2,084	18.4%	57.9%	31.6%	5.3%	21.1%	5.3%
MIN: 4/25	CU	30	28.0%	76.8	-8.53	9.23	48	2,062	28.0%	56.7%	36.7%	10.0%	10.0%	16.7%
OAK: 4/30	CU	18	18.4%	76.4	-8.76	9.45	47	2,086	18.4%	77.8%	50.0%	11.1%	16.7%	22.2%

I feel like I'm seeing a very different pitcher altogether. Also, as Sprowl and StupendousMan correctly predicted, the pitchfx data for the change has greater variance than the higher spin pitches and is therefore less reliable. The higher spin pitches, though, seem to further support that Clay has been very different, with differing outcomes and different spin rates (in different directions).

	Change	FF	CU
Mean	980	2,014	2,014
StDev	224.5	153.4	153.4
t-stat	-4.286	-5.464	4.799

Granted, that's only five games (though still passes significance test using the t-distribution), but I wonder when the Red Sox braintrust starts keeping an eye on such things. Also, given Buchholz's talent, as per above, I'm perhaps most interested if he can still be an elite pitcher just with a different makeup, or if it's necessary for him to revert to form to be elite.

It looks like his numbers the last two games are closer to his 2010 numbers, despite the pitches looking different. I hope that means he's becoming more effective--even if he's a new pitcher, he can be good at being that guy. (His highest game average RPM is 1.8 full standard deviations₂₀₁₂ below his 2010 average, and his lowest game average RPM for the curve is 1.5 standard deviations₂₀₁₀ above his 2010 average. [If I really feel like avoiding work later, maybe I'll plow through the 2010 numbers to get those deviaitons...]) The opposition is very different, though, and the concern there would be that that means his old mean is now his ceiling, which would be less good.

Thoughts?

Edited by Reverend, 01 May 2012 - 12:54 PM.

[ScubaSteveAvery]

Buchholz's changeup has never gotten its effectiveness from movement: it doesn't fade more than his fastball (as most pitchers' changeups do) -- it fades less. It often has practically no horizontal movement, though it usually has a slight vertical rise. Any pitch like that is likely to gyrate wildly in inferred spin, and the data could be thrown off badly by just a few inches of error in either the horizontal or vertical dimensions, or both. Buchholz's old plus-plus changeup seemed to work more by deception, especially arm action, than by movement, and I could never identify what made good when it was so good (check out the 2009 bash game for 12 whiffs on 25 changeups), or why it loses effectiveness when it is bad.

I think RPMs are very important for understanding the movement of fastballs and for curves, but less reliable as an indicator of hittability when it comes to changeups, cutters and sliders because their spin axis can be close to perpendicular to the strike zone. In such cases, PitchFX is missing the biggest part of the story. Alan Nathan mentioned that one firm now computes true spin rate in all dimensions (eg, how fast does a gyroball spiral, even if Pitchfx thinks it does not spin at all). It should be possible to get better insight into cutter and slider effectiveness using true spin rate.

The changeup still depends on the black box of deception.

There's also the issue of the movement on his change is just bad. A few years ago Dave Allen used PitchFX data to create heat maps and linear weights to find what type of movement, per type of pitch, was valued as most effective. The maps are a little old, but it shows that Buchholz' change has, as you mentioned, very little horizontal movement. It also has, per Allen's heat map, not very effective vertical movement. Last game, Buchholz' change was measured at -0.71 horizontal movement, and 5.80 vertical movement. Here is Allen's heat map for changes in terms of opposite handedness (changes are usually thrown to the opposite handed hitter). Of the 13 changes PitchFX caught, I counted 10 as thrown to LHB. Those horizontal and vertical locations put Buchholz' change right at the top of that big blue blob in the heat map on the right. Blue means positive outcomes to hitters.


Allen mentioned that this plot is missing a speed component, but it looks like the effective change ups have slightly less horizontal run than fastballs, but somewhat similar vertical movement. And this would make sense: a changeup looks like a fastball, but its slower. The only changes that don't follow this rule are those with tons of sink and similar horizontal movement. And its obvious why those would fool a batter - looks like a fastball in arm action, but slower and sinks. Buchholz' change has neither quality. This is somewhat of a longer way of confirming you point about deception. Clay's changeup isn't effective because it doesn't seem to resemble his fastball much in movement.

[ScubaSteveAvery]

Looking back over this very good point and czar's great post, I realized that I didn't post all of what I thought was interesting and, perhaps, provacative; I'm eager to hear thoughts on the rest.

I was discussing the change largely because that's how a couple of us initially started looking at the data, i.e. as the research question: Why isn't Clay throwing his change? This led to a consideration that the answer might be because it lost its effectiveness. This, in turn, led to a consideration that it's not the same change. Ultimately, though, the data began to suggest that Buchholz's has a comprehensively different composition as a pitcher altogether.

Thoughts?

You're on to something. His 2012 change is really a different pitch than his 2010 change for sure. Much less horizontal and vertical movement. And the spin data makes sense so far, although as mentioned a few times, the variance issue should be worked out.

One piece of information that would be useful is how RPM spin data correlates to vertical and horizontal movement? Is if just vertical? Or is it both? Lower RPMs this year on his changeup has resulted in less movement on both axis', so at a quick glance its possible that it affects both. But without that information its difficult to tell the full reach of the drop in RPMs on his change. Does anybody know anything more about this relationship?

[deconstruction]

Allen mentioned that this plot is missing a speed component, but it looks like the effective change ups have slightly less horizontal run than fastballs, but somewhat similar vertical movement. And this would make sense: a changeup looks like a fastball, but its slower. The only changes that don't follow this rule are those with tons of sink and similar horizontal movement. And its obvious why those would fool a batter - looks like a fastball in arm action, but slower and sinks. Buchholz' change has neither quality. This is somewhat of a longer way of confirming you point about deception. Clay's changeup isn't effective because it doesn't seem to resemble his fastball much in movement.

Good post. I agree that it's never been in the same class as Hamels's or Lincecum's changeups--which have lots of H-movement. But, with the bolded, are you talking about this year or in previous years? I agree that deception is part of its success, but in the past Clay's change seems to have the first quality you mention--slightly less fade, and somewhat similar vertical movement. From 2010, the first chart is his FB, the second is his change.



Not identical, but quite a bit of overlap. From the 2009 game Sprowl quoted:



Where's his change? Almost on the same plane as several of his FB. I'm not that well-versed in reading these things vis-a-vis the accumulated knowledge of pitching mechanics, effectiveness, etc. (had to throw in that self-handicap), and deception made it good, but part of what made it deceptive was the similar movement.

Right?

His 2010 CH plot would still cover some blue and mostly green on the heat map you posted, but, like you said, that doesn't take into account movement/speed relative to each pitcher's respective FB.

Edited by deconstruction, 04 May 2012 - 09:26 AM.

[czar]

There's also the issue of the movement on his change is just bad. A few years ago Dave Allen used PitchFX data to create heat maps and linear weights to find what type of movement, per type of pitch, was valued as most effective. The maps are a little old, but it shows that Buchholz' change has, as you mentioned, very little horizontal movement. It also has, per Allen's heat map, not very effective vertical movement. Last game, Buchholz' change was measured at -0.71 horizontal movement, and 5.80 vertical movement. Here is Allen's heat map for changes in terms of opposite handedness (changes are usually thrown to the opposite handed hitter). Of the 13 changes PitchFX caught, I counted 10 as thrown to LHB. Those horizontal and vertical locations put Buchholz' change right at the top of that big blue blob in the heat map on the right. Blue means positive outcomes to hitters.

While I appreciate the study Dave Allen did, I don't think it's close to a full representation of what makes a changeup successful. Taking those figures at face value, Clay has never had a "good" changeup (at most he's averaged -3 horizontal, +7 vertical-- 2010). However, in 2010 his CH was rated as one of the better pitches in baseball (by linear weights). So it's clear that movement alone cannot explain his formerly ++ changeup.

Allen mentioned that this plot is missing a speed component, but it looks like the effective change ups have slightly less horizontal run than fastballs, but somewhat similar vertical movement. And this would make sense: a changeup looks like a fastball, but its slower. The only changes that don't follow this rule are those with tons of sink and similar horizontal movement. And its obvious why those would fool a batter - looks like a fastball in arm action, but slower and sinks. Buchholz' change has neither quality. This is somewhat of a longer way of confirming you point about deception. Clay's changeup isn't effective because it doesn't seem to resemble his fastball much in movement.

Again, if this was a leading contributor as to why Buchholz's changeup is bad in 2012 compared to other years, I'd expect the deltas (diff between X and Z runs on FA and CH) to be dramatically different in 2012 vs. other years. And they are not (in 2012, his vertical movement is down the same (from 2010) within an inch on both pitches and his horizontal is positively shifted by the same amount as well).

---

I still think the two main explanations for a change's success (or lack thereof) is A) location and B) arm motion/release. The location has obviously been (subjectively) bad and the latter is not something that is within PitchF/X's domain.

One piece of information that would be useful is how RPM spin data correlates to vertical and horizontal movement? Is if just vertical? Or is it both? Lower RPMs this year on his changeup has resulted in less movement on both axis', so at a quick glance its possible that it affects both. But without that information its difficult to tell the full reach of the drop in RPMs on his change. Does anybody know anything more about this relationship?

Again, something Jnai or alannathan can provide more insight on but it seems logical that the greater the spin, the greater the Euclidean distance should be from 0,0. The spin angle will dictate whether that motion is preferred to go through the horizontal or vertical planes. Just thinking off the Magnus effect, the normalized movement (per unit spin) should be less the slower the pitch is.

Edited by czar, 01 May 2012 - 02:57 PM.

[ScubaSteveAvery]

I should have clarified that I'm talking about his current changeup having none of the qualities. His previous ones did, and they were effective. Hopefully this illustrates my point better. These are both from BrooksBaseball.

2010 pitch plot:


2012 Pitch Plot:


You'll notice how in 2010 his change slides right underneath his four seam fastball, but with less vertical movement. Aside from a few outliers it looks quite uniform (his cutter may be covering up some?). But now if we look at 2012, his change doesn't slide underneath his fastball, but to the right, almost under his cutter. My guess is that because there is little qualities in his current change that resemble his four-seam fastball, his current change has no deception relative to previous years changeups. Hopefully that helps.

[StupendousMan]

One piece of information that would be useful is how RPM spin data correlates to vertical and horizontal movement? Is if just vertical? Or is it both?

To first order, the magnitude of the spin determines the size of the break, and the direction of the spin determines the direction of the break. If you throw a ball so that it spins around a horizontal axis (as a pitcher with an over-the-top motion would have), it will break purely vertically. If you throw a ball so that it spins around a vertical axis (like an exaggerated sidearm pitcher), it will have a purely horizontal break.

Changing the arm slot will change the proportion of vertical to horizontal break, because it changes the spin axis.

[Eric Van]

Fastball effectiveness is a function of three things: velocity, movement, and location. In order of importance, it's probably location, location, location (the Retail Business Rule of fastball effectiveness), although velocity is obviously very important, much more so than movement. (In fact, it's been shown that the importance of velocity depends on location; IIRC, the velocity on an effectively located pitch is almost meaningless; but the harder you throw, the more you can get away with bad location.)

For every other pitch, deception becomes the fourth factor, and it might be #1 (a deceptive fastball simply adds to effective velocity in that hitters take a fraction longer to classify the pitch correctly. Deception on other pitches can result in the pitch being classified wrong.)

There's no way to measure deception; it can only be inferred by looking at results compared to expectations given the other three factors.

For a changeup, I believe that deception and location are far more important than velocity and movement (edit: and while I was writing this, czar offered the essentially identical opinion). I think the evidence is that Clay had a tremendous changeup in 2010 because he had unusually good deception and very good location. His location is definitely off this year, so the question might be whether his changeup has been also less deceptive (I think possibly so, just based on the lower swing-and-miss percentage) and why that could happen.

Edited by Eric Van, 01 May 2012 - 03:05 PM.

[deconstruction]

Hopefully that helps.

Yup, I was just pointing out that in 2010 his CH's V-movement wasn't much different from his FB's.

[joyofsox]

Buchholz pitched through a blister problem Monday night at Fenway Park, manager Bobby Valentine said Tuesday.

Valentine called the condition a chronic one that occurs when Buchholz is getting good snap on his curveball. Therefore it pops up when the right-hander is throwing well, which Valentine insisted he is, despite his less-than-stellar numbers.

http://espn.go.com/b...er-issue-monday

[ScubaSteveAvery]

I still think the two main explanations for a change's success (or lack thereof) is A) location and B) arm motion/release. The location has obviously been (subjectively) bad and the latter is not something that is within PitchF/X's domain.

Dave Allen also did a study using PitchFX data, linear weights, and location data to create heat maps looking at swing rate, contact rate, and value of the contact.

Here is the graphic that he made for right handed pitchers throwing changesups to left handed batters. This year Buchholz has thrown 46 changes to lefties and only 11 to righties, so Allen's observation of changesups being thrown to opposite handed hitters appears to be legitimate.


I'll post Allen's descreption of the three images so nothing gets lost in translation:

The run value of a pitch is determined by the outcome of four events.

• If the batter swings at the pitch or not.
• If no to 1, whether the taken pitch is called a ball or a strike.
• If yes to 1, whether the batter makes contact.
• If yes to 3, the run value of that contact.

Below I present a series of three images for each handedness combination that show how the outcomes of these four events vary by location for pitches. Reading left to right:

• The first image addresses events 1 and 2. The heat map is the swing percentage by location to address 1. On top of that are three contour lines where 75%, 50% and 25% of taken pitches were called strikes to address 2. So if a batter took a pitch inside the smallest circle it was called a strike over 75% of the time. If he took a pitch in doughnut between the smallest and middle circles it was called a strike between 75% and 50% of the time, and so on.
• The second image addresses 3 showing the contact percentage of pitches swung at.
• The final image addresses 4 showing the run value of a contacted pitch (including foul balls).

At the top of each image is the average value over all locations.

In the scenario of LHB vs. RHP, batters swing mostly at changesups slightly down and away, swing and miss 26% of the time, with a general run value at slightly negative (but essentially neutral). Even though batters make the most contact with changeups up in the zone, they don't swing at them the most. As for value, in usually produces a good outcome for the pitcher, with down and away producing the best outcomes. Pitches thrown on the inside half of the plate, from middle to middle-upper get abused by hitters. If Clay's location is bad, we should see most changes in the middle to inside part of the plate.

Here is his changeup locations for 2012 (from Texas Leaguers):



His location isn't that bad. A big cluster down and away. Lots of pitches down and away outside the strike zone. There are two pitches near the sweet spot, inside and middle; however, one was taken for a strike and one called a ball. He has left some up on the outside part of the plate, which correlate to higher contact. I'd have to go through the game logs to figure out the exact outcome of those pitches. His location chart looks similar to his 2010 location chart (also from Texas Leaguers):



Again, there is a lot of pitches down and away, with a big cluster down below the zone. He definitely left some in the hitters sweet spot, but I cannot say what those outcomes were. But his trend in 2012 looks similar. He avoids the inside part of the plate and keeps things down and away.

I'll go through the game logs sometime tonight or tomorrow (or somebody else can ) to see what the outcomes are of those that are left up in the zone for his 2012 pitches. He may be getting unlucky, as the highest contact% was those thrown up in the zone (even if the run values were low for those who make contact on those pitches). Heck, he may be getting beat on pitches where his location is good. But his location may not be as bad as we think.

There are a few things to consider:

1. He is still throwing about half his changesups for balls. Hitters may not be chasing them as much if the pitch has little deception because of lower RPM and less movement.

2. Even those thrown in the zone are getting hit because there is a lack of deception (the "whiff" rate is 15% or so (Texas Leaguers has it 13%, whereas BrooksBaseball has it at 15.15%). This is a lower whiff rate than what Allen found.

3. There could be a SSS issue because he's only thrown 50 or so changesups this year and only 46 to LHB. I'm dubious because the spin and movement data show a different pitch, even if the movement isn't drastic.

Generally, there is a lot going on, but lots of smart people here to whittle down the noise to figure out what is going on.

Edit: I forgot to add that your observation about Buchholz' release point is probably correct and I think that has a lot to do with what is going on here.

Edited by ScubaSteveAvery, 02 May 2012 - 07:10 AM.

[Delicious Sponge]

I tweeted to Cherrington about Clay's mechanics on the webcast today, which he answered.

He said that he doesn't see his mechanics as being that much different from 2010. The rest was sort of generic.

It's neither up yet nor very interesting, but the video will be here: http://boston.redsox...at.jsp?c_id=bos

Edited by Delicious Sponge, 02 May 2012 - 11:30 AM.

[Reverend]

While I appreciate the study Dave Allen did, I don't think it's close to a full representation of what makes a changeup successful. Taking those figures at face value, Clay has never had a "good" changeup (at most he's averaged -3 horizontal, +7 vertical-- 2010). However, in 2010 his CH was rated as one of the better pitches in baseball (by linear weights). So it's clear that movement alone cannot explain his formerly ++ changeup.

. . .

Again, if this was a leading contributor as to why Buchholz's changeup is bad in 2012 compared to other years, I'd expect the deltas (diff between X and Z runs on FA and CH) to be dramatically different in 2012 vs. other years. And they are not (in 2012, his vertical movement is down the same (from 2010) within an inch on both pitches and his horizontal is positively shifted by the same amount as well).

For a changeup, I believe that deception and location are far more important than velocity and movement (edit: and while I was writing this, czar offered the essentially identical opinion). I think the evidence is that Clay had a tremendous changeup in 2010 because he had unusually good deception and very good location. His location is definitely off this year, so the question might be whether his changeup has been also less deceptive (I think possibly so, just based on the lower swing-and-miss percentage) and why that could happen.

Looking over the data (based on valuable feedback from Sprowl and Stupendousman I have made an attempt at gauging the reliability of a five game sample below for anyone who is interested) , I think it’s quite possible that “level of deception” of Buchholz’s changeup has changes because both his four-seam fastball and his change-up have each changed significantly. In fact, it looks like all three of the pitches that are constant in 2010 and 2012, i.e. the fourseamer (FF), the change-up (CH) and his curve ball (CU) all show significance variance in the data, both in terms of their individual characteristics (velocity, spin, vertical movement, and horizontal movement). Also, both the outcomes statistics and the usage statistics are different for all three pitches which offers credence to the conclusion. Thus, if the power of the Hand of God was in looking just like the FF (see animated GIF at bottom of the HardBall Times mini-article Sprowl linked too), if the CH and the FF are each altered, wouldn’t it no longer be safe to believe this was still the case and, rather, need to check tape or ask hitters to see if they look the same? Moreover, if they are different now and do not look the same, we no longer have reason to expect some regression to the mean but must figure out what the new mean is.

Here is the game by game data, the aggregate 2010 season data, and the 2010 data calculated in terms of “samples of games” (addressed this below as well, but the numbers appear to show that this is a valid approach, and it lets me offer some sense of standard deviations [I figure anyone interested in thinking of each game as a “sample of pitches” will realize that the a quick and dirty calculation for standard error for five games would be to just divide by two.].)

(This is apparently going to be a many-post post: the post editor just told me what I can go do with the tables in my post.)

[

Buchholz 2012: FF

Date		Opp	Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Apr 8	@	DET	FF	28	35.9%	91.8	9.61	-5.52	210	2,217	67.9%	46.4%	3.6%	28.6%	14.3%
Apr 14		TBR	FF	44	42.7%	91.3	8.31	-5.06	211	1,932	72.7%	61.4%	0.0%	29.5%	31.8%
Apr 20		NYY	FF	45	43.7%	92.2	8.8	-4.6	208	1,992	82.2%	60.0%	2.2%	28.9%	28.9%
Apr 25	@	MIN	FF	47	43.9%	92.1	7.49	-4.99	214	1,822	68.1%	38.3%	2.1%	14.9%	21.3%
Apr 30		OAK	FF	47	48.0%	92.4	9.54	-4.54	206	2,108	59.6%	34.0%	2.1%	14.9%	17.0%
2010			FF	1492.0	53.4%	94.1	9.38	-6.90	217	2389	63.7%	45.0%	4.5%	19.7%	20.8%
2012		Mean		42.2	42.8%	92.0	8.75	-4.94	209.8	2014	70.1%	48.0%	2.0%	23.4%	22.7%
		StDev		8.0	4.4%	0.4	0.89	0.40	3.0	153	8.3%	12.4%	1.3%	7.7%	7.5%
		Dev			-10.6%	-2.1	-0.63	1.96	-7.2	-375	6.4%	3.0%	-2.5%	3.7%	1.9%

Edited by Reverend, 03 May 2012 - 07:45 PM.

[Reverend]

Buchholz 2012: CH

Date		Opp	Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Apr 8	@	DET	CH	9	11.5%	78.9	6.83	-2.84	204	1,293	55.6%	55.6%	0.0%	22.2%	33.3%
Apr 14		TBR	CH	15	14.6%	78.7	5.59	-0.48	185	980	46.7%	26.7%	13.3%	6.7%	6.7%
Apr 20		NYY	CH	14	13.6%	78.0	4.83	0.87	168	920	50.0%	42.9%	7.1%	0.0%	35.7%
Apr 25	@	MIN	CH	6	5.6%	78.9	3.73	-0.25	182	669	50.0%	33.3%	0.0%	0.0%	33.3%
Apr 30		OAK	CH	13	13.3%	79.8	5.8	-0.71	184	1,037	61.5%	53.8%	23.1%	7.7%	23.1%
2010			CH	519.0	18.6%	82.0	7.08	-3.26	205	1410	63.0%	49.5%	23.1%	12.3%	14.1%
2012		Mean		11.4	11.7%	78.9	5.36	-0.68	184.6	980	52.8%	42.5%	8.7%	7.3%	26.4%
		StDev		3.8	3.6%	0.6	1.16	1.35	12.8	224	5.8%	12.6%	9.8%	9.1%	12.0%
		Dev			-6.9%	-3.1	-1.72	2.58	-20.4	-430	-10.2%	-7.0%	-14.4%	-5.0%	12.3%

[Reverend]

Buchholz 2012: CU

Date		Opp	Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Apr 8	@	DET	CU	13	16.7%	75.4	-9.76	7.09	36	1,920	61.5%	53.8%	30.8%	0.0%	23.1%
Apr 14		TBR	CU	16	15.5%	75.4	-8.99	9.34	46	2,066	56.3%	31.3%	18.8%	0.0%	12.5%
Apr 20		NYY	CU	19	18.4%	75.7	-8.61	9.7	49	2,084	57.9%	31.6%	5.3%	21.1%	5.3%
Apr 25	@	MIN	CU	30	28.0%	76.8	-8.53	9.23	48	2,062	56.7%	36.7%	10.0%	10.0%	16.7%
Apr 30		OAK	CU	18	18.4%	76.4	-8.76	9.45	47	2,086	77.8%	50.0%	11.1%	16.7%	22.2%
2010			CU	235.0	8.4%	78.3	-8.45	5.57	34	1685	48.1%	31.1%	8.1%	11.5%	11.5%
2012		Mean		19.2	19.4%	75.9	-8.93	8.96	45.2	2044	62.0%	40.7%	15.2%	9.6%	16.0%
		StDev		6.5	5.0%	0.6	0.50	1.06	5.3	70	9.0%	10.6%	10.0%	9.6%	7.3%
		Dev			11.0%	-2.4	-0.48	3.39	11.2	359	13.9%	9.6%	7.1%	-1.9%	4.5%

I get czar’s point about comparing the vertical (z) and horizontal (x) deviations of the FF and CH, although I’d point out that (assuming 5 game sample reliability) the difference in z run deviations is actually greater than an inch. I’m thinking a bit differently though; my thought is that if two things look similar, and then they both change in significant ways, there is little reason to assume that they will still look similar. And that, I think, is where there could be a lot of power in absinthemalaise’s approach of trying to look at batting independent information to guage if there may be differences in a pitcher.

I wish they didn’t stop letting people post video on youtube—I find it very difficult to get the slowmo and pause functions to work well. The outcome data suggests to me that they aren’t the same pitches though.

Which brings us to the reliability of judging off of five game samples...

Edited by Reverend, 04 May 2012 - 12:20 PM.

[Reverend]

I’ve spoilered this so that anybody who is not interested but may want to scroll around the thread doesn’t have to scroll by all this crap everytime they look at the thread.

Spoiler

So Sprowl’s posts, scientist that he is, prompted me to wonder how many games we would need to consider a sample of games credible in evaluating a pitcher. StupendousMan suggested a simple but elegant approach, at least as a preliminary approach: he proposed taking a pitcher known (or at least believed?) to be reliable and looking at some multiple game samples and assessing them—this is not unsimilar to the standard candle method referred to in this excellent Baseball Prospectus article on the reliability and errors associated with PitchFX.

To this end, I took from TexasLeaguers (I wanted to use BrooksBaseball but the cutting and pasting worked better with the way TL is formatted; if I were smart like Dan I could probably figure out a script that would get it from his site for me but I’m not, so that was the proverbial that.) all the 2011 game by game data on Roy Halladay; as RingOSU pointed out, the guy put up nearly identical WHIPs in two seasons (Sneak Peak at Results: The guy is a machine.). Also, his effective use of 5 pitches in 2011 makes the number of pitches in each sample which pushes the method harder. Finally, I wanted a guy with a reliable fastball who also threw an offspeed pitch, in this case, his CU, in order to look at differences in reliability of the approach for offspeed stuff, as referenced by Sprowl above. So I started with five game samples.

What I found was:

• The game by game data (i.e. treating each game as a sample approximates aggregated pitch data very closely.
• The data on five game samples was even closer, as should be expected of a sampling distribution, and offers the association lower standard deviations and standard errors.

Here is the data summary, including minimum and maximums and interquartile range and stuff like that, a comparison of the sampling approaches (I did both a full set of every possible five game sample as well as the 6 possible samples of five non-overlapping games) as well as some numbers on how much his numbers deviated from his own season mean in each five game sample).

Halladay 2011: FF

			Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Games Samples		Mean	FF	28.7	26.4%	91.3	5.54	-4.63	220.5	1467.5	67.5%	48.1%	4.4%	22.1%	21.6%
		StDev	FF	13.4	11.3%	0.8	1.00	0.61	6.9	150.6	8.6%	10.5%	5.3%	8.9%	11.1%
Aggregate Pitches			FF	923	26.6%	91.3	5.69	-4.50	219.0	1474.0	67.6%	47.5%	4.8%	21.8%	20.9%
		Dev				0.0	-0.15	-0.13	1.5	6.5
		Max	FF	54.0	46.6%	92.8	7.57	-3.71	236.0	1764.0
		IQ3	FF	38.8	34.0%	91.7	6.23	-4.27	224.3	1545.0
		IQ1	FF	15.0	16.4%	90.8	4.88	-4.83	215.8	1336.8
		Min	FF	9.0	7.5%	89.9	3.30	-6.08	210.0	1189.0
			DevIQ3			0.4	0.54	0.23	5.3	71.0
			DevIQ1			-0.5	-0.81	-0.33	-3.3	-137.3
			IQR			0.9	1.4	0.6	8.5	208.3

			Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate
Aggregate Pitches			FF	923.00	26.6%	91.3	5.69	-4.50	219.0	1474
By Games		Mean	FF	28.72	26.4%	91.3	5.54	-4.63	220.5	1467.5
		StDev	FF	13.45	11.3%	0.8	1.00	0.61	6.9	150.6
5 Game Samples		Mean	FF			91.4	5.57	-4.63	220.39	1475.80
		StDev	FF			0.4	0.41	0.40	3.72	58.63
Non-Overlapping 5 Game Samples		Mean	FF			91.4	5.55	-4.62	220.47	1470.03
		StDev	FF			0.5	0.49	0.51	4.46	81.50
Deviation from Mean						VelDif	Vdif	Hdif	AngDif	SpiDif
5 Game Samples		Mean	FF			0.1	-0.12	-0.13	1.39	1.80
		StDev	FF			0.4	0.41	0.40	3.72	58.63

The guy really is a machine. Velocity and spin are remarkably consistent, especially looking how tight the interquartile range (within which 50% of all outcomes are cluteres).

Edited by Reverend, 03 May 2012 - 07:55 PM.

[Reverend]

Spoiler

How about the Curve? Not as consistent… but then I found something really interesting that also further suggests this might be an interesting approach: it seems something happened to him around August 8^th. Looking at the deviation data and there is a radical shift in the numbers describing his own deviation from his own mean in horizontal distance and the shift is remarkably stable.

In other words, all the horizontal deviations before that date were to one side of his season mean, while the deviations that date and after are on the other side of his season mean. So I checked and there is a similar change in the spin on the curve. Also, there is a radical change in his usage of the curve before and subsequent to this shift. (You can see the before and after on the TL site, which I looked up after I found it in the data while tracking down what I thought was an error but turned out, apparently, to be an effect.) Therefore, I compared all the pre-August 8 numbers and a lot of the discrepancies disappear; it seems the approach is highly effective for off-speed stuff too. Here’s the data.

Halladay 2011: CU

			Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Games Samples		Mean	CU	18.5	17.3%	77.1	-1.76	5.43	71.0	1002.2	67.7%	43.6%	17.1%	12.2%	14.3%
		StDev	CU	6.3	6.3%	0.8	1.09	1.28	10.9	195.9	8.2%	13.4%	9.6%	7.4%	8.6%
Aggregate Pitches			CU	592	17.1%	77.0	-1.75	5.88	72.0	1069	68.6%	44.9%	18.9%	11.7%	14.4%
		Max	CU	32.0	35.2%	78.7	0.58	8.28	94.0	1429.0
		IQ3	CU	20.0	19.4%	77.5	-0.95	6.65	80.3	1176.5
		IQ1	CU	14.0	13.3%	76.7	-2.21	4.53	62.3	869.0
		Min	CU	9.0	7.4%	75.4	-4.07	3.29	51.0	643.0
			DevIQ3			0.5	0.8	0.8	8.3	107.5
			DevIQ1			-0.3	-0.5	-1.4	-9.8	-200.0
			IQR			0.8	1.3	2.1	18.0	307.5

			Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate
Season			CU	592	17.1%	77.0	-1.75	5.88	72	1069
By Game		Mean	CU	18.47	17.3%	77.1	-1.76	5.43	71.0	1002.2
		StDev	CU	6.25	6.3%	0.8	1.09	1.28	10.9	195.9
5 Game Samples		Mean	CU			77.0	-1.78	5.37	70.68	992.96
		StDev	CU			0.3	0.32	0.86	4.32	126.13
Non-Overlapping Samples		Mean	CU			77.1	-1.70	5.31	71.07	981.80
		StDev	CU			0.2	0.31	0.88	4.29	124.13
Season through Aug. 3			CU	359	14.1%	77.1	-1.67	4.91	70	920
By Game thru Aug. 3		Mean	CU			77.1	-1.74	4.87	69.04	920.26
		StDev	CU			0.8	1.16	0.93	11.55	144.78
Samples thru Aug. 3		Mean	CU			77.1	-1.82	4.86	68.45	923.14
		StDev	CU			0.3	0.37	0.17	3.04	34.64
Non-Overlapping Samples thru Aug. 3		Mean	CU			77.0	-1.7	5.0	69.6	933.0
		StDev	CU			0.2	0.3	0.3	2.4	37.0
Deviation from Mean						VelDif	Vdif	Hdif	AngDif	SpiDif
5 Game Samples		Mean	CU			0.0	-0.03	-0.51	-1.32	-76.04
		StDev	CU			0.3	0.32	0.86	4.32	126.13
Samples thru Aug. 3		Mean	CU			0.1	-0.07	-1.02	-3.55	-145.86
		StDev	CU			0.3	0.37	0.17	3.04	34.64
Non-Overlapping Samples thru Aug. 3		Mean	CU			0.0	0.0	-0.9	-2.4	-136.0
		StDev	CU			0.2	0.3	0.3	2.4	37.0

So for Halladay, it looks like a five game sample could be a viable approach—five game sample numbers give pretty tight results in addition to lowering the random error inherent in the measurements (systematic error would still be a problem, but multiple game samples in different ballparks help alleviate that to a degree). It’s also provocative that it’s possible to identify an in-season shift in a pitcher; note that Halladay’s usage stats for the curveball change radically when the amount of spin changes.

Edited by Reverend, 04 May 2012 - 12:19 PM.

[Reverend]

Spoiler

So, how about Buchholz 2010? Much of the same:

Buchholz 2010: FF

			Type	#	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Games		Mean		53.3	53.5%	94.1	9.33	-6.83	216	2372	63.5%	44.9%	4.6%	19.9%	20.4%
		StDev		13.3	6.7%	0.8	1.4	1.0	6.3	235.5	6.2%	8.0%	3.5%	8.1%	5.2%
Aggregate Pitches			FF	1492	53.4%	94.1	9.38	-6.90	217	2389	0.64	0.45	0.05	0.20	0.21
		Max	FF	68.0	68.8%	95.3	12.24	-4.20	230	2858
		IQ3	FF	62.3	57.7%	94.5	9.71	-6.11	220	2482
		IQ1	FF	47.8	48.4%	93.5	8.63	-7.46	213	2220
		Min	FF	9.0	39.8%	91.9	6.42	-8.48	203	2011
			DevIQ3			0.4	0.33	0.79	2.5	93
			DevIQ1			-0.6	-0.76	-0.56	-4.0	-169
			IQR			0.9	1.36	0.56	8.5	208

Buchholz 2010: CH

			Type	#	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate	Strike	Swing	Whiff	Foul	In Play
Games		Mean		18.5	18.8%	82.1	7.01	-3.29	206	1404	62.7%	49.8%	24.4%	11.8%	13.6%
		StDev		4.8	3.2%	0.9	1.6	0.8	7.6	263.2	17.0%	15.6%	13.5%	7.8%	8.6%
Aggregate Pitches			CH	519	18.6%	82.0	7.08	-3.26	205	1410	0.63	0.50	0.23	0.12	0.14
		Max	CH	26.0	24.8%	84.2	10.03	-1.60	232	1902
		IQ3	CH	21.0	21.1%	82.7	8.00	-2.79	208	1608
		IQ1	CH	17.0	16.4%	81.5	5.98	-3.92	202	1186
		Min	CH	4.0	12.4%	80.1	4.34	-4.76	192	956
			DevIQ3			0.6	0.92	0.47	3.0	198
			DevIQ1			-0.5	-1.11	-0.66	-3.3	-224
			IQR			0.9	1.36	0.56	8.5	208

			Type	Count	Selection	Velocity	Vertical	Horizontal	Spin Angle	Spin Rate
Aggregate Pitches			FF	519.00	0.19	82.00	7.08	-3.26	205.00	1410
By Game		Mean	FF	18.54	0.19	82.10	7.01	-3.29	205.54	1403.54
		StDev	FF	4.83	0.03	0.93	1.58	0.84	7.64	263.2
5 Game Samples		Mean	FF			82.0	6.98	-3.21	205.1	1394.5
		StDev	FF			0.7	0.48	0.34	3.1	85.2
Non-Overlapping Samples		Mean	FF			82.0	7.08	-3.30	205.4	1414.0
		StDev	FF
Deviation from Mean						VelDif	Vdif	Hdif	AngDif	SpiDif
5 Game Samples		Mean	FF			0.0	-0.10	0.05	0.1	-15.5
		StDev	FF			0.7	0.48	0.34	3.1	85.2
Non-Overlapping Samples		Mean	FF			0.0	0.00	-0.04	0.4	4.0
		StDev	FF			0.8	0.65	0.45	4.1	108.3

Unfortunately, then, while it’s only five games, I see little reason to be optimistic that Buchholz’s numbers will regress back to his 2010 numbers unless something is changed. There is the basic random error involved in pitchfx, but even five samples of games smooths a fair amount of that out. Also, all the deviations from 2010 are consistent with one another in terms of spin, movement, outcomes and usage; it could be unlikely to see that level of confluence based on random error, which is why I think he is probably (this is all probabilistic, hope springs eternal and anyone who thinks I don’t want to see him spring back is nuts; this is just how I waste my time) not the same guy and that Clay Buchholz is not walking through that door. But on the hope issue, there are a few possible areas of hope:

• Buchholz’s greatest sample of five game deviations in 2010 was his first five games. (Unfortunately, the change doesn’t much look like this one from the PitchFX data*.)
• Clay is an athletic talent and he may yet find a way to find a new effectiveness changeup that looks deceptively like his fastball, or make his new composition work somehow.
• Maybe they can make adjustments; I have to assume the team is thinking about Buchholz as well.
• Um, I had another one earlier but I’ve forgotten. Bobby Valentine? I dunno…
• EDIT: I oh--velocity is still down. There remains the possibility that is a function not of changed delivery but arm strength, things could fall back into place or something. I am not sure what the relationship between arm strength and spin is, though it seems likely there is some kind of a connection. The reduced velocity on the CU with increased spin becomes interesting in this regards.

Thoughts?

Edited by Reverend, 04 May 2012 - 02:16 AM.

[StupendousMan]

Wow. That's nice work, Rev. There's food for days worth of analysis there.

It strikes me as peculiar that the 2012 Buchholz has a lower spin on both the fastball and changeup -- pitches which involve backspin -- but higher spin on the curveball -- which requires topspin. The four-seam fastball, in particular, ought to be relatively simple to throw (disclaimer: I'm not a pitcher, nor a pitching coach): grip, throw, let it roll off the fingers. The decrease in spin rate on that pitch might be due in part to the small decrease in speed of the pitch: if it comes off the fingers with less speed, it will probably have less spin, too. But the 2 mph decrease in speed seems a bit small to explain the roughly 15 percent decrease in spin rate.

I wonder if Buchholz has changed his basic grip on the FF and CH in the past two years ...

As for the curve, a higher spin rate sounds like a good thing: more movement will cause the bats to miss. And although the stats aren't quite so secure, it does look like the 2012 curve is doing a better job of missing the bats. On the other hand, when it _is_ hit, it's going farther. The numbers there are pretty small, though.

[deconstruction]

Awesome, Rev. Thanks for the substantial work. The differences are alarming.

(Everyone: I'm not a quant guy, so if I've fucked any- and everything that follows, please call me on it and, if you're so inclined, do something more productive and valid with the data!)

I wanted to explore a different deviation: the game-by-game difference in movement and velocity between Clay's FF and CH in 2010 and 2012. We've been going back-and-forth in this thread about the qualities that make Clay's changeup effective: Is it deception? Velocity separation from his FF? Lots of X-movement? Similar Z-movement to the FF? ScubaSteve discussed this a bit vis-a-vis Dave Allen's work, but I'm not doing anything with batted-ball data here.

Inspired by the idea that Clay's change and fastball are different pitches than they were two years ago, I was interested in seeing if there were significant differences between the X-movement (Delta X), Z-movement (Delta Z), and Velocity (Delta V) of Clay's FF and CH in 2010 and 2012. That is, when we study the deltas of theses factors on a game-by-game basis, do we see significant differences between Clay's 2010 FF-CH movement/velo and his 2012 FF-CH movement/velo? Furthermore, which of these factors has most contributed to the (lack of) succes that Clay has had with his FF/CH? To answer this last question, I added the whiffs that Clay got on his FF + CH for each start in 2010 and 2012.

Rev questioned if a five game sample is reliable, and this question's germane to my analysis, too. Are the differences we see in FF-CH movement/velo in 2012 over only five games so different from those in 2010 that they could not occur by chance--that there is some underlying difference in how these two pitches relate to one another on a game-by-game basis?

First, for each start in 2010 and 2012 I calculated the Velocity differences, X-movement, and Z-movement differences between the CH and FF. Those values are below. I also added the swinging strikes on the FF and CH for each of these games (as a proxy for pitch effectiveness--other measures are welcome). These calculations are tabled below:

2010 and 2012 FF-CH movement and velocity

2010	Delta-V	Delta-X	Delta-Z	Total SwStr
1	12.93	-3.21	3.1	6
2	11.93	-4.32	0.55	7
3	12.91	-4.22	1.42	16
4	12.04	-4.12	0.94	10
5	12.77	-3.12	2.55	7
6	12.52	-4.58	1.45	5
7	12.47	-4.22	1.58	9
8	12.12	-3.8	1.46	6
9	13.17	-4.85	2.85	7
10	12.69	-4.25	2.15	7
11	10.13	-3.44	2.62	2
12	12.42	-3.18	2.09	3
13	12.57	-4.38	2.88	5
14	11.99	-3.78	1.75	10
15	11.6	-3.13	2.43	3
16	12.08	-4.51	2.51	7
17	11.61	-3.3	1.38	12
18	12.38	-3.77	3.88	5
19	11.85	-3.37	3.04	6
20	11.63	-3.09	2.67	8
21	10.95	-2.88	0.7	10
22	11.68	-2.5	3.33	7
23	11.74	-2.56	2.14	5
24	11.29	-2.79	3.4	2
25	11.92	-3.41	4.24	6
26	11.11	-3.14	3.56	8
27	11.13	-2.3	2.83	5
mean	11.99	-3.56	2.35
stdev	0.68	0.68	0.94
2012
1	12.86	-2.68	2.78	1
2	12.68	-4.58	2.72	2
3	14.13	-3.73	3.97	2
4	13.19	-4.74	3.76	1
5	12.58	-3.83	3.74	4
mean	13.09	-3.91	3.39
stdev	0.56	0.73	0.53

As you can see comparing 2010 and 2012, the biggest mean delta is in Z-movement (3.39 and 2.25 inches)--the difference in vertical movement between FF and CH. This year, the vertical separation between his FF and CH has been much larger than it was in 2010, but is this mean difference significant? To answer that, I ran two-tailed t-tests on all three variables, grouped by year.



I'm not a stat guy, but I can play around on SPSS, so there it is. Assuming equal variance, you can see that Delta Z and Delta V are both statistically significant (<.05). Therefore, we have evidence that, through just five starts, not only are Clay's 2012 FF and CH different from the ones he threw two years ago, the differences between vertical movement and velocity between these two pitches on a game-by-game basis are also significantly different. In other words, not only have the pitches changed, but the differences between the pitches have also changed, which could be a factor in his lack of success so far this year.

I next wondered which (if any) of these three factors correlated strongly with Buchholz's "effectiveness" throwing both--measured crudely by the combined swinging strikes on these two pitches. Here is the correlation matrix:


Here, we see that only Delta Z (difference in vertical movement) is significantly (negatively) correlated with the number of swinging strikes, which means that the larger the difference between the Z-movement of the FF and the Z-movement of the CH, the fewer swinging strikes. This means that Buchholz has gotten more swinging strikes on his FF + CH when both pitches have been on closer vertical planes--i.e., when there's not much vertical difference between the two as the approach the plate.

To me, this all means that the less Z-movement difference between Buchholz's fastball and changeup, the more successful he'll be. This year, he's seen a larger difference, and he's also been less successful.

Please, dig in, everyone.

Edited by deconstruction, 04 May 2012 - 07:04 AM.

[Reverend]

Wow. That's nice work, Rev. There's food for days worth of analysis there.

Thanks, and thanks for your help, and thanks for the kudos from D-Con (and thanks to abs, and Sprowl, and let's not forget phrenile for making these tables possible--if nothing else, I'm having fun, dammit.)

Just posting that I updated the posts because I accidentally left out deviation data for the Halladay 2011 curveball (which is key because it addresses the idea that there is a discernible alteration in his pitching, a change that seemingly might have possibly been noted through scrutiny of his pitchfx data in season, even though I can't find anything written about it in the webz) and the Buchholz 2010 change. (If anyone knows of something that happened with Halladay around august 3-8 2011, I'd love to hear about it; I'm curious what (if anything) went down.)

Also, the look at correlation between FF-CH deviation with swinging strikes directly looks seems like a potentially powerful approach, and I'm really intrigued by the idea of looking at not just whiff rates for the CH in this relationship but of both. Very cool.

FWIW, I've seen what appears to be some interesting associations between whiff, foul and in play, with different pitchers having different numbers with respect to whether or not fluctuations in foul% become balls in play or whiffs, but I haven't figured out anything more concrete, and I expect it's different for different pitchers for a number of reasons, but I thought I would throw it out there in case anyone else has ideas or want to go panning for gold, as... it's hard to say, but it looks like there are some interesting lines in getting the guy to foul it off versus put it in play when looking at on-day v. off-day data for some pitchers. Beyond BABIP's Will of course, that much is (never) given.

P.S. For anyone who wants to play along at home and drop the data into a spreadsheet, you will find that if you just highlight the data and drop it in the sheet, it comes up in only one line. BUT, if you highlight all the text and then highlight PAST the boundaries of the table, the copy function will pick up whatever computerese is for "this is a table" and when you post it into a spreadsheet, it will drop the values into different cells. This also works on other sites like BaseballReference and whatnot with tabular data.

Edited by Reverend, 04 May 2012 - 02:42 AM.

[pjheff]

Unfortunately, then, while it’s only five games, I see little reason to be optimistic that Buchholz’s numbers will regress back to his 2010 numbers unless something is changed.
Thoughts?

I think that there are two unknowns in this equation concerning Clay's mechanics and arm strength. Regardless of what Cherington tweets, Clay seems mechanically different, throwing more over the top as he did in 2008 as opposed to his 2010 iteration. Is this difference a deliberate one, as many have speculated, to take the strain off of his injured back, or is it a byproduct of rust, trying to find his groove after missing significant time last season? Similar questions can be raised about Clay's arm strength. It seems clear that his velocity is down from what we've seen in the past. Is this difference a reparable one that can be rebuilt after the layoff or is it simply the best we can expect from Clay in 2012? The answers to those questions are uncertain and therefore leave room for optimism. I will say, however, that the notion of throwing his 2010 repertoire with his 2008 mechanics, or of throwing his 2008 repertoire with his current arm strength, appears to be a flawed one, unlikely to produce positive outcomes.

[czar]

I think that there are two unknowns in this equation concerning Clay's mechanics and arm strength. Regardless of what Cherington tweets, Clay seems mechanically different, throwing more over the top as he did in 2008 as opposed to his 2010 iteration.

Does he? Camera height might be playing a small role here, but (on the surface) there seem to be only very subtle differences between 2010 and 2012. Remember, the only evidence we have for mechanics changes is that one (Globe?) story about McClure tweaking things to make it less stressful on his back. That and the P/FX release point *may* be lower by just a hair (but still within Clay's normally release scatter. Cherington and the few stills I've looked at seem to indicate this may not be as big a change as people in the game threads thought they were seeing (similar note: Even if they exist, now that I've looked at some of his 2012 starts and 2010 starts (thanks MLB.tv archive), I can't imagine how people would be able to pick up the subtle 2010/2012 differences at real speed).

[Sprowl]

Does he? Camera height might be playing a small role here, but (on the surface) there seem to be only very subtle differences between 2010 and 2012. Remember, the only evidence we have for mechanics changes is that one (Globe?) story about McClure tweaking things to make it less stressful on his back. That and the P/FX release point *may* be lower by just a hair (but still within Clay's normally release scatter. Cherington and the few stills I've looked at seem to indicate this may not be as big a change as people in the game threads thought they were seeing (similar note: Even if they exist, now that I've looked at some of his 2012 starts and 2010 starts (thanks MLB.tv archive), I can't imagine how people would be able to pick up the subtle 2010/2012 differences at real speed).

Something still looks different to me about Clay 2012's mechanics and release point. It doesn't show up as a different arm slot in the release point data, but I'm seeing a more coiled, less lanky motion, reflected in fewer high releases in the release point clouds for 2010, 2011 and 2012 that abs posted in #28. This gives the impression of a more overhand delivery. That is confirmed not in the arm slot but the arm extension. I remember Clay's 2010 motion having more east-to-west motion in it, with a more "slinging" or whip-like extended delivery, especially on the fastball. In principle, a north-to-south delivery, with all motion concentrated toward the plate instead of wasted motion side to side, should be more efficient, and might well put less stress on his lower back, which contributed some of the torque in the slinging motion I remember. Unfortunately, it also is implicated in the reduction in horizontal movement on his fastball.

I should stress that this is what I think I'm seeing -- recognizing that this might be memory reconstructing a motion that was never there in the first place, working backwards from a preconceived diagnosis. Deduction is a dangerous business where mechanics are concerned.

One of the possibilities is that Clay v2010 obtained his slinging delivery and horizontal movement by small variations in foot positioning: ie, lining up on the left side of the pitching rubber, but landing on a spot more consistent with the center of the rubber. One thing I remember about McClure's work with Doubront is the concentration on how Doubront should land with his right foot -- on the heel, not the toe. If there were a way to focus in on how Clay is landing on his left foot, that might show a McClure mechanical signature.

[czar]

Something still looks different to me about Clay 2012's mechanics and release point. It doesn't show up as a different arm slot in the release point data, but I'm seeing a more coiled, less lanky motion, reflected in fewer high releases in the release point clouds for 2010, 2011 and 2012 that abs posted in #28. This gives the impression of a more overhand delivery. That is confirmed not in the arm slot but the arm extension. I remember Clay's 2010 motion having more east-to-west motion in it, with a more "slinging" or whip-like extended delivery, especially on the fastball. In principle, a north-to-south delivery, with all motion concentrated toward the plate instead of wasted motion side to side, should be more efficient, and might well put less stress on his lower back, which contributed some of the torque in the slinging motion I remember. Unfortunately, it also is implicated in the reduction in horizontal movement on his fastball.

I should stress that this is what I think I'm seeing -- recognizing that this might be memory reconstructing a motion that was never there in the first place, working backwards from a preconceived diagnosis. Deduction is a dangerous business where mechanics are concerned.

One of the possibilities is that Clay v2010 obtained his slinging delivery and horizontal movement by small variations in foot positioning: ie, lining up on the left side of the pitching rubber, but landing on a spot more consistent with the center of the rubber. One thing I remember about McClure's work with Doubront is the concentration on how Doubront should land with his right foot -- on the heel, not the toe. If there were a way to focus in on how Clay is landing on his left foot, that might show a McClure mechanical signature.

I don't know. You must be seeing something I didn't when going through the game archives, because any differences I noted are extremely subtle (I will agree that I did see a slight torque difference and mention that in post 44). However, the game thread griping about overhauling his mechanics in a Craig Hansenian fashion seem to be overstated to a degree.

Also, when you say you are implicating the more N/S delivery (i.e., the less torquey one) in less horizontal run, how is this manifested? Via a different spin angle, or are you arguing that this delivery forces him to have less velocity/RPM? (I'm not disagreeing with you in principle here, just trying to get a feel for what you think the physical mechanism is. My guess is it's a combination of both where the spin angle controls the H:V run ratio and the RPM/velocity affects the Euclidean distance but I am curious as to whether you have a different opinion).

-------------------

Reason I ask (about spin angle/velocity/RPM), is that Clay's final 2012 April numbers from PF/X are actually quite similar to his 2011 ones (which would predate any mechanical tweaks).

Buchholz FA, CU, CH -- April 2011

Pitch	Count	Frequency	H. Mvt	V. Mvt	Mph	H. Rel	V. Rel	Spin Θ	RPM
Fourseam (FA)	111	24%	-3.87	-12.28	93	-1.86	6.65	198	2,070
Curveball (CU)	56	12%	12.61	-52.66	77.36	-1.96	6.62	43	1,890
Changeup (CH)	88	19%	-1.21	-26.78	80.62	-1.96	6.56	193	1,050

Buchholz FA, CU, CH -- April 2012

Pitch	Count	Frequency	H. Mvt	V. Mvt	Mph	H. Rel	V. Rel	Spin Θ	RPM
Fourseam (FA)	116	21%	-5.09	-12.18	93.13	-1.92	6.31	203	2,095
Curveball (CU)	101	19%	13.57	-52.99	76.71	-1.99	6.19	46	1,954
Changeup (CH)	42	8%	-0.94	-28.08	79.5	-1.89	6.17	191	960

In fact, he has less horizontal run on his FB last April in addition to less velocity (a fact I've pointed out in game threads a couple times when people keep clamoring he's way down on his velocity). His spin angles are nearly identical and his RPM rates are pretty close (up a bit on CU, down a bit on CH). Interestingly, his release point is slightly lower this year compared to last (which abs showed visually in the Texas Leaguers data last page), but it's actually almost identical (vertically) to his release point in April 2010.

The reason I post this is... well, here are his May 2011 numbers:

Buchholz FA, CU, CH -- May 2011

Pitch	Count	Frequency	H. Mvt	V. Mvt	Mph	H. Rel	V. Rel	Spin Θ	RPM
Fourseam (FA)	65	11%	-5.59	-12.48	94.23	-1.96	6.52	206	2,076
Curveball (CU)	61	10%	10.32	-50.05	78.64	-2.09	6.51	40	1,657
Changeup (CH)	116	20%	-3.13	-25.63	81.17	-1.98	6.43	203	1,207

Big jump in FB velocity and CH spin rate (from April 2011). This correspond both to a ~2" bump in horizontal run on both pitches. His arm slot also actually was a tick lower in May than April last year and brings it about halfway to April 2012's release point.

People quickly forget that May 2011 was arguably Buchholz's best month as a major-leaguer peripherally (3.56 xFIP, 3.63 K/BB). So while there is a tendency to say that there are lots of negatives in Buchholz's 2012 start (and there are), I don't know if I'd be quite so sure about calling this "the new Buchholz" when he had a very similar (actually worse) April last year (5.69 xFIP vs. 5.08 this year) before regaining some velocity and a plus changeup last May/early June (before injury). Maybe McClure and Co. aren't so far off when they are saying "He'll be fine. He's throwing OK, he's just shaking the rust and finding his changeup."

EDIT: I should have mentioned that the data came from the Player Card feature over on Jnai's site (BrooksBaseball.net) in case people were curious.

Edited by czar, 04 May 2012 - 03:34 PM.

[Sprowl]

Also, when you say you are implicating the more N/S delivery (i.e., the less torquey one) in less horizontal run, how is this manifested? Via a different spin angle, or are you arguing that this delivery forces him to have less velocity/RPM? (I'm not disagreeing with you in principle here, just trying to get a feel for what you think the physical mechanism is. My guess is it's a combination of both where the spin angle controls the H:V run ratio and the RPM/velocity affects the Euclidean distance but I am curious as to whether you have a different opinion).

The major difference should be manifested primarily in horizontal run and spin angle. I think that the primary reason that it is not visible in the PitchFX data this year is because of some dodgy PitchFX fastball classification. So far in 2012, Buchholz is listed as throwing 211 fastballs and 12 two-seam fastballs; in contrast, in 2011 the division shows 487 fastballs and 116 two-seamers, so the ratio is quite different. It is further complicated by several horizontal movement errors in Buchholz's starts (Fenway is shifting all pitches to the right by a few inches), so a bit like the problem with identifying spin angle on the changeup, our diagnoses may be tainted by bad initial data as well as bad automated classification. If BrooksBaseball.net and THT were to do the same manual reclassification on the 2012 data that was done for 2011 and earlier, we would probably have a firmer basis on which to draw conclusions about movement and mechanics.

[czar]

The major difference should be manifested primarily in horizontal run and spin angle. I think that the primary reason that it is not visible in the PitchFX data this year is because of some dodgy PitchFX fastball classification. So far in 2012, Buchholz is listed as throwing 211 fastballs and 12 two-seam fastballs; in contrast, in 2011 the division shows 487 fastballs and 116 two-seamers, so the ratio is quite different. It is further complicated by several horizontal movement errors in Buchholz's starts (Fenway is shifting all pitches to the right by a few inches), so a bit like the problem with identifying spin angle on the changeup, our diagnoses may be tainted by bad initial data as well as bad automated classification. If BrooksBaseball.net and THT were to do the same manual reclassification on the 2012 data that was done for 2011 and earlier, we would probably have a firmer basis on which to draw conclusions about movement and mechanics.

This is probably a good point, as I left out the cutter (which Buchholz throws non-negligibly) because I wanted to just get the "top-end of the spectrum"data (and that should be the FA by PF/X's automatic detection), but this might skew up/down a little bit depending on how many of these pitches are misclassified in 2012.

That said, the point about him looking relatively similar to last April before he had a huge jump in May/June still should hold, even if some of the PF/X FA numbers are weighted slightly incorrectly through bad binning. It'll be interesting to see if his velocity (which has climbed steadily (albeit slowly) in each start) continues ticking upward and whether or not the enhanced CH location he showed at times in the MIN start as well as the OAK one continues into May.

Edited by czar, 04 May 2012 - 03:31 PM.

[Sprowl]

That said, the point about him looking relatively similar to last April before he had a huge jump in May/June still should hold, even if some of the PF/X FA numbers are weighted slightly incorrectly through bad binning. It'll be interesting to see if his velocity (which has climbed steadily (albeit slowly) in each start) continues ticking upward and whether or not the enhanced location he showed at times in the MIN start as well as the OAK one continues into May.

Yes, I think a little more velocity and better fastball command should go a long way toward bringing back an effective Buchholz, no matter which version we are getting. Regaining the changeup seems to be more about Clay's "feel" for the pitch. Given his 2008 performance, there's no reason why he can't throw an effective changeup regardless of how straight or fast his fastballs are. Even when his fastball was getting launched into orbit during July 2008, he could still throw the changeup in all circumstances, and he generally had better location on the changeup than the fastball. I can't imagine that he has somehow lost that ability to spot that pitch.

[deconstruction]

Yes, I think a little more velocity and better fastball command should go a long way toward bringing back an effective Buchholz, no matter which version we are getting. Regaining the changeup seems to be more about Clay's "feel" for the pitch. Given his 2008 performance, there's no reason why he can't throw an effective changeup regardless of how straight or fast his fastballs are. Even when his fastball was getting launched into orbit during July 2008, he could still throw the changeup in all circumstances, and he generally had better location on the changeup than the fastball. I can't imagine that he has somehow lost that ability to spot that pitch.

What's interesting about July 2008 is that he threw his change only 7.4% of the time, and it wasn't very effective (only 2 whiffs on 28 thrown). August 2008 was a bit better.

EDIT: but at least it wasn't getting launched, right?

EDIT 2: But, observationally, you remember it being pretty good?

Edited by deconstruction, 04 May 2012 - 04:26 PM.

[Sprowl]

What's interesting about July 2008 is that he threw his change only 7.4% of the time, and it wasn't very effective (only 2 whiffs on 28 thrown). August 2008 was a bit better.

I think that datum is an artifact of PitchFX misclassification -- back before Clay amped up the velocity on his slider, it overlapped with his changeup, and many of the changeups were labelled (incorrectly, I'm pretty sure) as sliders. Many of the whiffs on sliders should actually be credited to the changeup.

eg, against the Orioles, the Angels, the Mariners and the Angels again, most of those "sliders" are actually changeups.

The misclassification effect continues into August, and is noticeable against the Royals, but especially against the White Sox, where Clay got bombed on his fastball and curve, but threw his changeup to great effect -- even though early pitchfx has him throwing 9 sliders for 6 whiffs:



The difficulties that pitchfx had in separating early Buchholz's offspeed stuff was one of the reasons that the manual reclassification of later data was so helpful in improving data quality, but I think that did not apply to 2008, just to later years.

[deconstruction]

I think that datum is an artifact of PitchFX misclassification -- back before Clay amped up the velocity on his slider, it overlapped with his changeup, and many of the changeups were labelled (incorrectly, I'm pretty sure) as sliders. Many of the whiffs on sliders should actually be credited to the changeup.

eg, against the Orioles, the Angels, and the Mariners, most of those "sliders" are actually changeups.

Oh, yeah, wow. Thanks for that clarification.

[Sprowl]

EDIT 2: But, observationally, you remember it being pretty good?

Yeah, I remember him being able to spot the changeup perfectly while he was wild in the strike zone with the fastball, so much so that I posted something about how he should nibble with the fastball, while throwing the changeup and curve for strikes. It would have been pitching backwards in two ways, but his fastball was so bad that he had either to change his approach or get yanked from the rotation. Ultimately, Tito decided on the latter.

[Reverend]

Reason I ask (about spin angle/velocity/RPM), is that Clay's final 2012 April numbers from PF/X are actually quite similar to his 2011 ones (which would predate any mechanical tweaks).
The reason I post this is... well, here are his May 2011 numbers:
. . .
Big jump in FB velocity and CH spin rate (from April 2011). This correspond both to a ~2" bump in horizontal run on both pitches. His arm slot also actually was a tick lower in May than April last year and brings it about halfway to April 2012's release point.
. . .
People quickly forget that May 2011 was arguably Buchholz's best month as a major-leaguer peripherally (3.56 xFIP, 3.63 K/BB). So while there is a tendency to say that there are lots of negatives in Buchholz's 2012 start (and there are), I don't know if I'd be quite so sure about calling this "the new Buchholz" when he had a very similar (actually worse) April last year (5.69 xFIP vs. 5.08 this year) before regaining some velocity and a plus changeup last May/early June (before injury). Maybe McClure and Co. aren't so far off when they are saying "He'll be fine. He's throwing OK, he's just shaking the rust and finding his changeup."

I thought I would drop this in here because it might be useful, from FanGraphs.







This next one confused me for a moment.

The major difference should be manifested primarily in horizontal run and spin angle. I think that the primary reason that it is not visible in the PitchFX data this year is because of some dodgy PitchFX fastball classification. So far in 2012, Buchholz is listed as throwing 211 fastballs and 12 two-seam fastballs; in contrast, in 2011 the division shows 487 fastballs and 116 two-seamers, so the ratio is quite different. It is further complicated by several horizontal movement errors in Buchholz's starts (Fenway is shifting all pitches to the right by a few inches), so a bit like the problem with identifying spin angle on the changeup, our diagnoses may be tainted by bad initial data as well as bad automated classification. If BrooksBaseball.net and THT were to do the same manual reclassification on the 2012 data that was done for 2011 and earlier, we would probably have a firmer basis on which to draw conclusions about movement and mechanics.

Isn't there something weird about the Sabre-happy Red Sox having one of the more messup up pitchfx installations in the league?

I think there's a conspiracy here.

Speaking of, I've watched a bunch of Buchholz in some of the condensed games across the last three years and I have come to a powerful conclusion: whoever is in charge of condensed games at MLB doesn't give a shit about watching pitching.

Edited by Reverend, 04 May 2012 - 05:05 PM.

[Sprowl]

Isn't there something weird about the Sabre-happy Red Sox having one of the more messup up pitchfx installations in the league?

Fenway has persistently been on the high side in reporting velocity (the Fenway inflation factor), but the horizontal error for 2012 isn't a multi-year problem, so I'm hoping it will go away when the whole system is reset during the current road trip. It's just my gut feeling, but I think that Fenway's movement data has been quite good over the long term. There are a few parks where problems seemed to linger for years -- Yankee Stadium's horizontal error (4" or more to the left) seemed

If you ever want to see a sinker on camera steroids, check out Justin Masterson's horizontal movement in the Bidet from a few years ago. It was like Rich Hill's purported 17" horizontal movement on his fastball -- literally off the charts.

[ScubaSteveAvery]

Out of all the changes Buchholz threw today, one stood out to me as indicative of his previous dominant change. He threw it to Adam Jones in the 3rd inning today. The third pitch of the at-bat before Jones grounded into a double play. It had good movement, breaking in to Jones and sinking a tad. Jones swung and missed and was completely fooled.

I looked it up after the game to see what the movement was like and the vertical and horizontal movement was above his averages this year:



It looks like he got around 6.5" of vertical movement and 2" of horizontal movement and came in at around 80mph. Both are about an inch more than he's been averaging this year. What I found curious was that the spin graph shows a still low amount of spin:



At 1,100 RPMs it is still above the 950 or so he's been averaging this year, but not by a lot. The pitch was great and quite visibly different than the other change he threw in the at bat (the only ball thrown) Any thoughts on this? I would have expected a larger jump in RPMs.

[Reverend]

With respect to the deceptivity issue, it occurred to me to think about how hitters are fairing and approach Clay in a kind of inverted xFIP kind of way. The idea that a great change’s effectiveness comes about in its association with the fastball is intriguing, particularly in terms of some conversations I've had about pitch deception with abs and redsox13. To start, check out Buchholz's linear weights by pitch:

Total Runs by Pitch (Standardized)

Season	wFA/C	wFT/C	wFC/C	wSL/C	wCU/C	wCH/C
2008	-2.98			1.17	-0.66	-0.14
2009	-0.44	14.43		1.59	0.84	-1.03
2010	1.14			0.76	-1.86	1.22
2011	-1.00	0.80	1.07	-1.94	0.36	0.60
2012	-2.26	2.69	-2.40		-2.24	-6.02

Not terribly conclusive, as it could just mean he pitched better when he’s good than when he’s not. It is interesting to see how the numbers go up and down, though, and how both the fastball and change have appeared to be changing in conjunction with one another since 2009.

It seems worth looking into what hitters are doing to those pitches, respectively, and if there is any relationship there. Specifically, what happens once a hitter decides to swing at the pitch?

FF: Swing & Contact

Agg	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	FF	53.4%	63.7%	45.0%	4.5%	19.7%	20.8%	90.0%	46.2%
2011	FF	41.5%	61.2%	42.2%	4.8%	15.0%	22.4%	88.6%	53.1%
2012	FF	42.5%	69.4%	45.9%	2.1%	21.5%	22.3%	95.4%	48.6%

“Agg” here refers to “aggregate pitches, i.e. the numbers for all the pitches of that type he threw that year without reference to which game is which (this may matter). (Also, the %(Con|Sw) and %(Play|Sw) notation denotes "% contact given a swing" and "% put in play given a swing" respectively.) What’s of interest here is that once the batter decides to swing at the FF, he usually makes contact, and that has been true over all of the past three seasons. Moreover, the batter has also consistently put that ball into play about half the time, also fairly consistently.

Some have wondered if how this start compares to past years, as Buchholz has been known to start slowly in the past, so here are the numbers for his first six starts for each season, both in terms of aggregate pitches across those six games and as unweighted averages of samples of games:

FF: Swing & Contact

Agg 6	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	FF	56.4%	63.8%	44.7%	5.6%	19.7%	19.4%	87.5%	43.4%
2011	FF	43.4%	62.0%	43.6%	3.6%	14.8%	25.2%	91.7%	57.8%
2012	FF	42.5%	69.4%	45.9%	2.1%	21.5%	22.3%	95.4%	48.6%
Games 6	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	FF	56.7%	63.7%	44.8%	5.7%	19.7%	19.4%	87.3%	43.4%
2011	FF	43.5%	62.5%	43.9%	3.8%	14.7%	25.4%	91.4%	57.9%
2012	FF	42.1%	69.2%	45.4%	2.2%	21.6%	21.6%	95.1%	47.5%

So far so consistent; Clay’s contact numbers may be up a bit this year, but the in play isn’t bad.

Now, what about the change?

CH: Swing & Contact

Agg	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	CH	18.6%	63.0%	49.5%	23.1%	12.3%	14.1%	53.3%	28.5%
2011	CH	18.5%	63.1%	49.4%	19.7%	15.7%	14.1%	60.3%	28.5%
2012	CH	12.5%	53.5%	40.8%	11.3%	8.5%	21.1%	72.5%	51.7%

Methodological note:

Spoiler

Both contact numbers and in play numbers are way up. This also appears to hold for the first six games with a caveat:

CH: Swing & Contact

Agg 6	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	CH	19.7%	63.8%	44.7%	5.6%	19.7%	19.4%	87.5%	43.4%
2011	CH	20.7%	61.3%	45.4%	16.8%	13.4%	15.1%	62.8%	33.3%
2012	CH	12.5%	53.5%	40.8%	11.3%	8.5%	21.1%	72.5%	51.7%
Games 6	Type	Selection	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
2010	CH	19.6%	60.7%	50.2%	24.2%	12.6%	13.4%	51.7%	26.7%
2011	CH	20.3%	57.6%	39.5%	14.8%	12.1%	12.6%	62.4%	31.8%
2012	CH	12.7%	53.5%	41.3%	9.6%	8.5%	23.2%	76.7%	56.1%

In terms of aggregate pitches, it looks like Buchholz’s contact numbers in 2010 were pretty bad. When you run the numbers by unweighted game samples, though—a number which appears to converge to the season parameters for aggregate pitches across a season—then his first 6 games of 2010 look much more like his season averages. I went and checked and saw that this kind of stuff occurs because of differences in approaches to other teams and their approach to Buchholz. For example, Texas hacked at the change at a 44% whiff rate, so Clay threw a few more that game, and then left the usage up against Toronto who started knocking it into play. Results such as these tend to distort the aggregate averages for various outcomes in the short term in ways that appear to be “corrected” by using games as samples; in other words, the aggregate pitch data is obviously what actually happened, but if you want to know what is “going on” in a “how is he pitching game to game” sense, it is useful to compare unweighted averages of games. Or not—I report, you decide.

It looks to me, then, that batters can tell the change up when they see it. When they see fastball, the approach has been the same and contact and in play numbers are basically unchanged. The significantly higher contact and in play numbers on the change suggest that when they are swinging at the change, they have correctly identified it as the change up at a much higher rate.

In fact, there are two different ways we can look at how batters are beating it:

CH: Swing & Contact

Date		Opp	Strike	Swing	Whiff	Foul	In Play	%(Con|Sw)	%(Play|Sw)
Apr 8	@	DET	55.6%	55.6%	0.0%	22.2%	33.3%	99.8%	59.9%
Apr 14		TBR	46.7%	26.7%	13.3%	6.7%	6.7%	50.2%	25.1%
Apr 20		NYY	50.0%	42.9%	7.1%	0.0%	35.7%	83.2%	83.2%
Apr 25	@	MIN	50.0%	33.3%	0.0%	0.0%	33.3%	100.0%	100.0%
Apr 30		OAK	61.5%	53.8%	23.1%	7.7%	23.1%	57.2%	42.9%
May 6		BAL	57.1%	35.7%	14.3%	14.3%	7.1%	59.9%	19.9%

On three occasions this season, teams have put it into play at about a 60% clip or better; 55.5% and 46.2% were the game maximums in 2010 and 2011 respectively.

On the flip side, teams seem to be better at laying off it; 41.2% and 43.6% represent the bottom quartile numbers for swing% in 2010 and 2011.

So teams are better both at holding off on hacking at the change and better at knocking it into play. I obviously can’t be sure that this means batters are correctly identifying the change better than in the past, but I find this much more suggestive than the whiff% numbers alone.

And I see that phren and out fine Dope overlords have got lines working in the tablizer, which is fantastic, IMO.

Thoughts?

• Back to top

---

#93 Reverend

• 



B.P.I.W.



• 
• 14,557 posts

Posted 07 May 2012 - 04:41 PM

The above is, of course, a kind of indirect approach to looking at some facet of the problem in the context of location, which is always key, so for those who would like (a) point(s) of reference:

Edited by Reverend, 07 May 2012 - 04:42 PM.

• Back to top

---

#94 Kull

• 



wannabe merloni



• 
• 1,072 posts

Posted 09 May 2012 - 01:32 AM

Small sample size, obviously, but one noticeable difference from 2010 to 2012 is that he seems to be planting a lot further away from the rubber. If that's happening consistently, it may cause two things, both of them bad. If he's starting from the same set spot as in 2010 but lengthening his motion, that seems like it would reduce the max height from which he could release a ball, something we might be seeing in the lower release point showing up in the graphs posted upthread. The other effect is that home plate is now, from the look of it, 8-10 inches closer when the ball is released. Pitching is a "touch" skill if ever there was one, so balls that in the past would have been dropping in at the lower end of the front of the strike zone are now going to be higher up, deeper into the plate area. Usually not a good thing, especially if it's not what you're arm is telling you should be happening.

You'd think this would be something easy to spot and correct, so maybe I'm reaching. But even if the photos are not indicative of plant foot results on every pitch, that's still looks like a lot of variance.

• Back to top

---

#95 Reverend

• 



B.P.I.W.



• 
• 14,557 posts

Posted 11 May 2012 - 06:04 PM

I initially opted not to bump the Buchholz thread, but Sprowl thought it was sufficiently interesting, so here it is. I came upon quite the statistic over on FanGraphs: Clay's HR/FB%

It's 27%.

One in four fly balls put in play have left the park this season. One in four times the batter gets a hit where he really gets under the ball, it leaves the park.

His fly ball rate is around 32%, so about 8% of hits or so has left the park.

I wouldn't have thought that was even possible.

9.5 % is league average, by the way, and 13% is considered awful. (Beckett is at 18% i believe). There are two ways of looking at this. One is bad, as this is consistent with the greater contact numbers posted above; batters know what's up. There is a hopeful angle, though, in that HR rates arevery "noisy" and not always consistent with an honest evaluation of how a guy is pitching. From the HR/FB% explanation on FanGraphs:

Taking a glance at a pitcher’s HR/FB ratio can help tell you if a player had an over- or under-inflated ERA. Pitchers with HR/FB ratios much higher or lower than league average will normally regress towards league average in the future, which will have a corresponding effect on their ERAs and FIPs.

In fact, this variation is the reason xFIP was developed:

Home run rates are generally unstable over time and fluctuate around league-average, so by estimating a pitcher’s home run total, xFIP attempts to isolate a player’s ability level. A pitcher may allow home runs on 12% of their flyballs one year, then turn around and only allow 7% the next year. HR/FB ratios can be very difficult to predict, so xFIP attempts to correct for that.

So whoever Buchholz is this year and regardless if there are serious problems or not, Clay may yet have an inflated ERA due to random variation stemming from HR/FB stuff.

Edited by Reverend, 12 May 2012 - 12:42 PM.

• Back to top

---

#96 kieckeredinthehead

• 
• 
• 1,924 posts

Posted 11 May 2012 - 11:27 PM

I initially opted not to bump the Buchholz thread, but Sprowl thought it was sufficiently interesting, so here it is. I came upon quite the statistic over on FanGraphs: Clay's HR/FB%. It's 27%.

That is literally higher than many home run derby pitchers can manage.

• Back to top

---

#97 budcrew08

• 
• 
• 990 posts

Posted 12 May 2012 - 10:14 PM

I think the planting angle is interesting. I think the back is still an issue, no matter how slight. He's not getting the nice break on the curve as he did a couple years ago, IMO.

• Back to top

---

#98 EpsteinsGorillaSuit

• 
• 
• 216 posts

Posted 12 May 2012 - 10:45 PM

9.5 % is league average, by the way, and 13% is considered awful. (Beckett is at 18% i believe). There are two ways of looking at this. One is bad, as this is consistent with the greater contact numbers posted above; batters know what's up. There is a hopeful angle, though, in that HR rates arevery "noisy" and not always consistent with an honest evaluation of how a guy is pitching. From the HR/FB% explanation on FanGraphs:

Another way to interpret this information is that pitchers who go through a period of giving up a large number of home runs often make an adjustment to fix the problem. There could be a big selection bias in this data, as pitchers unable to make such an adjustment will not remain MLB pitchers much longer.

I'm not suggesting that Buchholz is cooked by any means, but it is astounding how hard he is getting hit right now. If EV were running an SSS regression on him, the numbers would be so alarming that he'd have to conclude that Clay is demonstrating a new (awful) true talent level.

Edited by EpsteinsGorillaSuit, 12 May 2012 - 10:47 PM.

• Back to top

---

#99 Reverend

• 



B.P.I.W.



• 
• 14,557 posts

Posted 13 May 2012 - 12:12 AM

Another way to interpret this information is that pitchers who go through a period of giving up a large number of home runs often make an adjustment to fix the problem. There could be a big selection bias in this data, as pitchers unable to make such an adjustment will not remain MLB pitchers much longer.

I'm not suggesting that Buchholz is cooked by any means, but it is astounding how hard he is getting hit right now. If EV were running an SSS regression on him, the numbers would be so alarming that he'd have to conclude that Clay is demonstrating a new (awful) true talent level.

I don't have s clue what you're talking about.

Edit: OK, I think I sort of have a clue. But the thing is Buchholz obviously already has a track record, so if the high home run rates are a structural problem, they indicate either that there is something with what he is doing now that is different from what he was doing before, or he has somehow been solved or is tipping pitches or something.

Nobody can see anything different in what he is doing, but there are observed differences in outcomes both in terms of his own pitch movement, the behavior of batters, and his own pitch selection. So we have an observed change, but nobody can figure out what has changed to cause it.

Edited by Reverend, 13 May 2012 - 02:28 PM.

• Back to top

---

#100 absintheofmalaise

• 



too many flowers



• 
• 8,699 posts

Posted 13 May 2012 - 12:08 PM

I'm using this since it's the last post in the thread to get all the data on the next page. Don't delete it.

• Back to top

---

---

• Prev

• Page 2 of 4
• 1
• 2
• 3
• 4

• Next

Back to Red Sox Forum

[Reverend]

The above is, of course, a kind of indirect approach to looking at some facet of the problem in the context of location, which is always key, so for those who would like (a) point(s) of reference:

Edited by Reverend, 07 May 2012 - 04:42 PM.

[Kull]

Small sample size, obviously, but one noticeable difference from 2010 to 2012 is that he seems to be planting a lot further away from the rubber. If that's happening consistently, it may cause two things, both of them bad. If he's starting from the same set spot as in 2010 but lengthening his motion, that seems like it would reduce the max height from which he could release a ball, something we might be seeing in the lower release point showing up in the graphs posted upthread. The other effect is that home plate is now, from the look of it, 8-10 inches closer when the ball is released. Pitching is a "touch" skill if ever there was one, so balls that in the past would have been dropping in at the lower end of the front of the strike zone are now going to be higher up, deeper into the plate area. Usually not a good thing, especially if it's not what you're arm is telling you should be happening.

You'd think this would be something easy to spot and correct, so maybe I'm reaching. But even if the photos are not indicative of plant foot results on every pitch, that's still looks like a lot of variance.

[Reverend]

I initially opted not to bump the Buchholz thread, but Sprowl thought it was sufficiently interesting, so here it is. I came upon quite the statistic over on FanGraphs: Clay's HR/FB%

It's 27%.

One in four fly balls put in play have left the park this season. One in four times the batter gets a hit where he really gets under the ball, it leaves the park.

His fly ball rate is around 32%, so about 8% of hits or so has left the park.

I wouldn't have thought that was even possible.

9.5 % is league average, by the way, and 13% is considered awful. (Beckett is at 18% i believe). There are two ways of looking at this. One is bad, as this is consistent with the greater contact numbers posted above; batters know what's up. There is a hopeful angle, though, in that HR rates arevery "noisy" and not always consistent with an honest evaluation of how a guy is pitching. From the HR/FB% explanation on FanGraphs:

Taking a glance at a pitcher’s HR/FB ratio can help tell you if a player had an over- or under-inflated ERA. Pitchers with HR/FB ratios much higher or lower than league average will normally regress towards league average in the future, which will have a corresponding effect on their ERAs and FIPs.

In fact, this variation is the reason xFIP was developed:

Home run rates are generally unstable over time and fluctuate around league-average, so by estimating a pitcher’s home run total, xFIP attempts to isolate a player’s ability level. A pitcher may allow home runs on 12% of their flyballs one year, then turn around and only allow 7% the next year. HR/FB ratios can be very difficult to predict, so xFIP attempts to correct for that.

So whoever Buchholz is this year and regardless if there are serious problems or not, Clay may yet have an inflated ERA due to random variation stemming from HR/FB stuff.

Edited by Reverend, 12 May 2012 - 12:42 PM.

[kieckeredinthehead]

I initially opted not to bump the Buchholz thread, but Sprowl thought it was sufficiently interesting, so here it is. I came upon quite the statistic over on FanGraphs: Clay's HR/FB%. It's 27%.

That is literally higher than many home run derby pitchers can manage.

[budcrew08]

I think the planting angle is interesting. I think the back is still an issue, no matter how slight. He's not getting the nice break on the curve as he did a couple years ago, IMO.

[EpsteinsGorillaSuit]

9.5 % is league average, by the way, and 13% is considered awful. (Beckett is at 18% i believe). There are two ways of looking at this. One is bad, as this is consistent with the greater contact numbers posted above; batters know what's up. There is a hopeful angle, though, in that HR rates arevery "noisy" and not always consistent with an honest evaluation of how a guy is pitching. From the HR/FB% explanation on FanGraphs:

Another way to interpret this information is that pitchers who go through a period of giving up a large number of home runs often make an adjustment to fix the problem. There could be a big selection bias in this data, as pitchers unable to make such an adjustment will not remain MLB pitchers much longer.

I'm not suggesting that Buchholz is cooked by any means, but it is astounding how hard he is getting hit right now. If EV were running an SSS regression on him, the numbers would be so alarming that he'd have to conclude that Clay is demonstrating a new (awful) true talent level.

Edited by EpsteinsGorillaSuit, 12 May 2012 - 10:47 PM.

[Reverend]

Another way to interpret this information is that pitchers who go through a period of giving up a large number of home runs often make an adjustment to fix the problem. There could be a big selection bias in this data, as pitchers unable to make such an adjustment will not remain MLB pitchers much longer.

I'm not suggesting that Buchholz is cooked by any means, but it is astounding how hard he is getting hit right now. If EV were running an SSS regression on him, the numbers would be so alarming that he'd have to conclude that Clay is demonstrating a new (awful) true talent level.

I don't have s clue what you're talking about.

Edit: OK, I think I sort of have a clue. But the thing is Buchholz obviously already has a track record, so if the high home run rates are a structural problem, they indicate either that there is something with what he is doing now that is different from what he was doing before, or he has somehow been solved or is tipping pitches or something.

Nobody can see anything different in what he is doing, but there are observed differences in outcomes both in terms of his own pitch movement, the behavior of batters, and his own pitch selection. So we have an observed change, but nobody can figure out what has changed to cause it.

Edited by Reverend, 13 May 2012 - 02:28 PM.

[absintheofmalaise]

I'm using this since it's the last post in the thread to get all the data on the next page. Don't delete it.