CSS Button No Image Css3Menu.com
New! Search comments:
(NOTE: Relevance, Author, and Article are not applicable for comment searches)
I hear Carl Yakstrzemski is a favorite for the yak racing Triple Crown this year.
There are a number of second-career people working in the Astros front office, and several of us got our start in the baseball industry near or after age 40: Jeff Luhnow, Sig Mejdal, Kevin Goldstein, and me.
It's not common, but neither is it impossible.
Wait a minute...it wasn't stolen from your local library shortly before Christmas, was it?
You can find it at your local library using inter-library loan.
From an excellent (but out of print) book.
Carlos Corporan sees Chris Getz's face-plant slide and raises with his slide into second base in the 8th inning on October 3rd.
"Mike Fast would be spinning yarn in his Houston office if he saw this."
Ha! Good stuff, R.J.
You have the most comments (3450). Kevin Goldstein has the second most (2315).
Thank you all!
Good questions. As far as breakeven points, that's a good idea for further study.
In terms of year-to-year consistency, I'm not sure exactly which parameter you're asking about. If you asking about the average net run value for a hit-and-run attempt, I don't know that year-to-year consistency would indicate anything. The success of the play is very dependent on the situation. There were just over 8600 hit-and-run plays that I identified. The further you break down the sample from there, the more you have to be concerned about the situation. The samples with 1000 or more plays seemed to have a decent mix of situations, and the samples with 100 or so plays are quite situation-dependent.
In terms of various individual rates, like the success percentage for advancing baserunners with the play, or the caught stealing percentage when the batter swings and misses, those were pretty stable year to year, but I don't know if that's what you were asking about.
Joe Torre was at 321 hit-and-run plays in eight seasons, for an average of 40/year, which is why he didn't make the list of leaders.
The one table in which I probably have the least confidence in the whole study is the number of runs gained by each team. When I looked at the league as a whole, I spent a lot of time looking at all the numbers from different angles. For the teams, I mostly assumed that what applied to the league applied to each team, with the exception that I looked at the quality of hitters on each team that were asked to execute the hit-and-run.
So there are a number of situational biases that I accounted for at the league level that might not apply quite as well to a given team.
In particular, the performance of the batters following the hit-and-run attempt might have biased the results for a team. I looked into this some for the Yankees, and if you simply looked at the bases gained, rather than runs scored, and the RE24 values for those bases, the Yankees total is more like 40-50 runs gained rather than 85.
There wasn't a big difference between Torre's Yankee teams and Girardi's teams. If you use the runs-scored method, Torre's teams come out a bit better, and if you use the bases-gained method, Girardi's teams come out a bit better.
Differences in baserunner quality between H&R attempts and H&R situations with no attempts inflated the value by about .016 runs/play. If make that adjustment, it drops the overall hit-and-run advantage to about +.045 runs/play.
I divided the runners on first base into four equal groups based upon their stolen base attempt rate over the period 2003-2011. I defined the stolen base attempt rate as (SB+CS)/(singles+UBB+HBP).
The fastest runners do get more advantage out of the hit-and-run play than do the slowest runners.
Group SBA% range SBA% SB% H&R Adv. (runs/play)
A SBA>=17.5% 25.7% 77.4% 0.070
B 17.5%>SBA>=10.0% 13.2% 73.1% 0.043
C 10.0%>SBA>= 5.4% 7.3% 67.1% 0.060
D 5.4%>SBA>= 0.0% 2.7% 60.4% 0.008
MGL, as I had mentioned to you at the Book blog, I had looked into this preliminarily prior to publication, and I didn't see in the data much indication of what you claim. The good basestealers don't steal at their normal success rate in these situations (they are below 70 percent), nor do they make up a majority of the hit-and-run attempts, so I don't think it's going to move the results as much as you think.
But as a result I also disagree with how you're framing it. The only thing we care about is what happens with a hit-and-run play, regardless of the quality of the baserunner. If the baserunner quality is biasing the results, I need to adjust for that. My preliminary look said there was not a big bias there.
I believe that either the good-basestealer, straight-steal group is so small in these situations that it's insignificant to the results, or good basestealers on first base in these counts don't have a high enough run expectation to move the results all that much. But it's worth double checking.
There is no sizable difference in the rate of the hit and run attempts between visiting and home team until you get to extra innings, where the visiting team uses it more often.
However, I don't know whether the visiting or home team has an advantage in making it successful. I wouldn't see why, unless it was coupled to an advantage in personnel. It's very hard to study things by inning because various parts of the lineup come up more often in certain innings. If you have a big enough sample, you can split both by inning and lineup position, but I don't have the luxury of that sample size with this data set.
It must have been the game on August 10, 2003:
I love Pure Baseball. It's one of the best thinking-about-the-game books I've read.
For my part, I was surprised to see the Molina brothers among the leaders at executing the hit-and-run play. The other names made more sense to me.
I spent a little time looking at whether contact hitters and groundball hitters got more advantage from the hit-and-run play, but I wasn't able to deal with the selection bias issues there to my satisfaction prior to writing this up.
Your observation would be interesting to test even outside the context of the hit-and-run play.
MLB clubs are welcome to come knocking. I'll answer the door.
Thanks, Randy. Yes, foul balls are included.
The most popular Rangers hit-and-run combos 2007-2011 were Young behind Andrus (10 times), Andrus behind Kinsler (7 times), Young behind Kinsler (5 times), and Borbon behind Andrus (5 times).
Washington used the hit-and-run a lot with the Rangers the last two years, but not as much before that. Some people on Twitter mentioned that Washington liked to hit-and-run with the Kinsler-Andrus combo. I have the data on the combos he used; I'll have to take a look at that.
Anyone can just waltz right in. You wouldn't have access to some media events and other functions, but the lobby crowd where much of the talking is done is completely free.
There is a job fair for internships that requires registration, but otherwise there are many people who just come on their own and hang out with the crowd. I even saw a couple kids there looking for autographs.
I appreciate your thought, Nathan, and agree with a lot of it. However, there is not much evidence that the strike zone depends on the reputation of the pitcher, though that is commonly claimed. There is evidence that the strike zone depends to some extent on whether the pitcher hits the catcher target. Pitchers who can do that also tend to have reputation as better pitchers, and deservedly so, but that may be where the claim arises about a pitcher's reputation mattering for the strike zone calls.
Yes, there is a sequel to BBTN in the works. One of the other staff might be able to give more detail on a timeline.
That is Colin's smile.
I was skeptical of how the comments would work when they were first added (I was not yet a columnist at that point), but I am a big fan of them now. I really enjoy the conversation I get on many of my articles, as well as participating in the comment threads for other articles.
I'm not sure there's a good or precise definition of a hanging curveball. The best definition I've been able to come to is one that is left up in the zone. But there's not a huge difference in results between high curveballs and low curveballs in terms of contact. More home runs are hit off of high curveballs, so perhaps that's it, but that's true of other pitch types to some extent, too.
BABIP on low curveballs is .295 and on high curveballs in .301. HR/contact on low curveballs is 2.2% and on high curveballs is 3.8%. But the difference is similar to that between low pitches of all types (2.4%) and high pitches of all types (4.1%).
I did look into difference on performance on curveballs based upon how fast they were descending when they crossed the plate in some research that Matt Lentzner and I did for the 2010 PITCHf/x Summit. I'll have to dig that up and see if I found anything, but if it were dramatic, I think I'd remember.
1) That makes some sense, but I haven't looked at what the data shows.
2) LD rate is problematic because it includes a lot of bloopers. If you remove the bloopers and focus on the hard-hit line drives, LD rate has some persistence. Some of it is still luck, or perhaps due to other characteristics we don't fully understand (e.g., ground ball rate).
3) I don't know. The data is available here in the sortable stats for pitchers if you want to check.
4) I think I've linked to this or similar work before, but Jeremy Greenhouse answered that question here. John Walsh, Dave Allen, Harry Pavlidis, and others have probably also put out similar data.
It's that speed is more predictable/controllable than angle.
For most batted balls, more speed is a good thing. (That's not true for bloopers, which is a big reason BABIP is inconsistent for pitchers.) So if the pitcher can fool the batter and make him swing defensively or not get good wood on the ball, he's going to win out more often than not.
Launch angle matters a lot right around 12 degrees, where you're dropping it over the infielders heads. For any individual batted ball, the launch angle matters more than speed. But there doesn't seem to be a skill for pitchers to be able to control launch angles within narrow windows. There is a lot I have yet to learn about this, but my impression is that the basic idea that has been advanced before is correct: that even if a pitcher gets lots of groundballs, he'll allow some hits in the line-drive range of angles, and that even if a pitcher gets lots of popups and harmless flies, he'll allow some hits in the line-drive range of angles. It's not true to conclude from that that pitchers have no control over their line drive fraction, but it's probably why they have limited control.
Even though Mariano Rivera gets way more weakly-hit balls than anyone else, and he also gives up fewer line drives than most other pitchers, he still gives up a fair amount of line drives.
Part of this has to be that a batter can hit a line drive even if he mistimes the pitch or misjudges its trajectory. If the timing error and spatial error work in opposite directions, the batter can still get solid wood on the ball.
I'm not sure I understand your question.
Do you mean why Felix Hernandez isn't listed among the pitchers with best or worst adjusted BABIP allowed? His adjusted BABIP was right at the league average of .300 for 2008.
Or are you asking about something else?
The incoming pitch angle ranges from about -4 to -12 degrees at the common extremes (a fastball at the shoulders to a curveball at the knees).
You could, of course, do a really high eephus lob that came basically straight down on the plate, but most pitches have to be thrown in that range I mentioned in order to make it across the plate at typical MLB game speeds.
Here's a crude illustration of what I mean by speed in the 12-degree plane. Perhaps showing it graphically will help some people understand it better than explaining it with more words.
That's a very good suggestion, Alan.
The best launch angle for home runs is around 30 degrees, and the best launch angle for a base hit is 12 degrees. I think that's what you meant, and that's correct. Having a launch angle by definition means that contact was made.
Your question is a good one. What I know about that from this data, I have not gotten permission from Sportvision to share. However, if you look at the BABIP data by pitch type, you can see that fastballs have the highest BABIP allowed.
Of the two things, though, the one that seems to impact BABIP the most by far, when large samples are considered, is the speed, which batted ball data is very poor at capturing.
For any individual batted ball, the angle off the bat matters a lot, but for large samples this effect is very muted. There is still some effect based on the typical vertical launch angle, but don't give it equal credit with the speed of the ball.
Pitchers do have a line-drive skill (which is mainly where launch angle comes into play), and it's important to consider this for additional accuracy, but I don't feel like I understand the origin of that skill well enough yet to build it into a metric. I'm still working on that.
However, I did show how my approach worked for pitcher BABIP based upon dividing every batted ball into either hard-hit or weakly-hit. I did not compare to a batted-ball BABIP predictor. Which one did you have in mind? Something based upon Graham's tRA? Something based upon Brian's findings in the THT Annual?
When I measured in the horizontal plane, it was a measure of how quickly the ball got to the fielders. Popups can be hit hard, but if all the speed is going up, that doesn't make it a very hard ball to field.
The tweak to the 12-degree plane is just an acknowledgment that it helps to get the ball over the head of the infielders, so a little bit of vertical speed is a plus (but not so much vertical speed that it's a popup or a can of corn to the outfield).
If you think of it simply as the measurement I made in the previous article, which was in the horizontal (0-degree) plane, you will still get the basic idea.
With the hSOB, it's how fast the ball travels toward the base of the outfield fence. With the 12-degree SOB, it's how fast the ball travels toward a point about 80 feet in the air above the outfield fence.
The 12-degree speed-off-bat is pretty similar to the horizontal SOB I defined in the previous article. You can still think of it as hSOB if that's easier to conceptualize. The difference is small, but I wanted to include in my calculation the advantage to getting the ball in the air a little bit. So I tilted the speed measurement plane by 12 degrees.
I calculated the speed in the 12-degree plane for every batted ball, regardless of the angle at which it was launched. I did a vector projection of each initial speed onto a plane inclined 12 degree above vertical.
Take three batted balls, all leaving the bat at 70 mph. A high 70-mph pop-up has most of its speed going vertical, and will have a very low speed in the 12-degree plane, maybe 10 mph or so. A 70-mph line drive just out of the reach of the shortstop will have a speed in the 12-degree plane of about 70 mph. A ball pounded into the dirt in front of the plate at about 70 mph will have a low speed in the 12-degree plane maybe also around 10 mph, depending on how steep of an angle it was hit at.
My next article is actually on how hSOB and vertical launch angle interact to affect BABIP. It's important to understand that before moving on to determining what the pitcher and batter are doing that determine hSOB and VLA.
I agree, though, that the kind of thing you mention is where we are headed with this, though I don't know that a linear regression is the best tool for the job. I prefer to develop a physical model for what is happening, if I can. Linear regression can play a supporting role in that process, but ultimately, we want to know why and how the batter and pitcher do what they do.
Yes, the HITf/x data should be illuminating in that regard. We already know something about the most effective speed range for changeups (See Dave Allen's piece here.), but we don't know nearly as much as we could about the how and why.
We also don't know much about whether it is helpful for a pitcher to vary the speed on his fastballs by 2-3 mph. That turns out to be a very difficult question to study properly because speed changes are related to differences between four-seam and two-seam fastballs (and in some cases pitchers use cutters as fastballs, too). Those pitch types tend to be used in different locations, different ball-strike counts, etc., which complicates the analysis.
It's always helpful for me to review the past literature on a topic when I am studying it, and I also think I owe the reader and the previous researchers a mention of the work that I am building on.
I don't think I can say any more than I said about that at this point. Sportvision gave me the data under NDA, and I don't want to go beyond the bounds of what I told them I would write about without getting permission from them.
I'll just say that it's not inconsistent with what you can find about the effect of pitch types and location on BABIP from the public PITCHf/x data.
I did not remove bunts from this analysis, though in retrospect that would have been a good idea.
Bunts make up about 2 percent of batted balls in MLB, and large portion of those are by pitchers, but for a few batters it's much more significant.
Taveras, for instance, had 12 percent of his batted balls as bunts in 2008, and Bonifacio 11 percent.
It's true that if the ball strikes the bat in exactly the same way, that the faster it came in, the faster it will go out. A two-mph increase in incoming pitch speed will result in a little less than a one-mph increase in outgoing batted ball speed.
However, it also seems to be true that the faster the pitch comes in, the harder it is for the batter to square up the bat on the ball.
These two effects seem to roughly cancel out in the MLB population of batters and pitchers, though the latter effect may be somewhat more important.
No, that's not what I'm saying. How hard the pitcher throws has very little to do with how hard the ball is hit, at least in MLB. (It may have a little bit to do with it, but to the extent that it does, it appears that the harder the pitch, the slower the resulting batted ball.)
The pitcher and batter BOTH control quality of contact. The batter has a little bit more control over that than the pitcher, but the pitcher has a lot more control than people have thought since the acceptance of DIPS.
The pitcher presumably controls the quality of contact by deceiving the batter as to where and when he should swing.
Mo Rivera is one of the best, probably THE best, in baseball at this, and he throws hard. But he locates extremely well, and this makes it difficult for the batter to make solid contact with the ball.
Dave, it makes intuitive sense to me that both the pitcher and the batter have some influence over the quality of contact. I'm not sure it's possible to intuit accurately who would have more influence. The pitcher controls the location of the pitch and which way it's moving, which limits the possibilities that the batter has, but the batter is the one who actually swings the bat and determines how the bat contacts the ball. I don't know any way other than observation to determine which one is more important.
Correlation between pitch speed and hSOB is not strong, at least not at major league game velocities. Pitch types and locations make a bigger difference than pitch speed itself. That's not to say that fastball speed has no effect, but it's a lesser effect, and it's not trivial to disentangle from pitch movement and from selective sampling effects (i.e., pitchers that throw slower are in MLB because they are above average at other things).
I'm not planning to directly address your last question in the next piece. It's something I've previously investigated from the April 2009 HITf/x data, but I don't intend to publish the results from the batter-pitcher model I developed from that.
HITf/x measures the speed of the ball and its direction. From that, it is easy to calculate the various components of the batted ball speed.
For example, take a fly ball that is in the air for four seconds before it is caught at the 375-ft sign against the left-center field wall. Ignoring the effect of drag that would have slowed the ball slightly, it traveled 375 feet horizontally in four seconds, for a speed of 375/4 = 93.75 ft/sec (equivalent to 63.9 mph).
HITf/x doesn't measure the whole flight of the ball, just the initial portion, but the idea is the same.
Take another example, a popup that is skied over the infield and caught half way down the third base line after seven seconds in the air. The popup may have come off the bat going really fast, maybe 70-80 mph, but most of that speed was vertical. The horizontal component of the speed was much less. Again ignoring air resistance effects, the ball went only 45 feet horizontally in 7 seconds, and 45/7 = 6.4 ft/sec = 4.4 mph.
The horizontal component of the speed tells you more about how solidly the ball was hit than the total speed (including the vertical component). It also tells you more about how difficult the ball was to field because it tells you how quickly the ball got to or past the fielder (how long they had to react, as you said.)
Thanks, Brian. I looked at within-pitcher variation in hSOB, and I found something I didn't understand. The standard deviation in hSOB tends to go down as average hSOB goes up. That was true somewhat for batters but especially for pitchers. (SD on the order of 20-25 mph).
The only thing I could think was that there's practically/physically an upper limit on hSOB around 100-110 mph that is closer to the mean than is the lower limit at 0. Also, the distributions are typically not normal (peak above average with a long lower tail), so I don't know how well standard deviation describes the distribution in that case.
The short answer is lots of fly balls (which produced the home runs) and popups (which cut down on his average hSOB).
Thanks, Sharky! As far as batter/pitcher control over the result is concerned, I looked at that both at a 100 batted ball and 300 batted ball threshold and found similar results.
In terms of whether those who hit weaker balls (as batters) or allow harder-hit balls (as pitchers), I wouldn't be surprised if they get weeded out earlier, perhaps very quickly for the many fringe-MLB quality players who only get a brief chance to establish themselves. Tom Tippett's study of BABIP indicated as much. There is, of course, a selective sampling issue in that future playing time is allocated partly based upon the past outcomes for that player as opposed to their actual skill (we learn their skill partly from their outcomes).
I'd probably need multiple seasons of HITf/x data, or minor league HITf/x data, before I could tease out that effect better than was done in Tippett's study, for example.
Thanks. Foul balls are not included. I wish that Sportvision and/or TrackMan would track data on foul balls, too. I believe that data has almost as much analytical value as the data for fair balls.
That's a good suggestion.
I gave the average hitter performance for each of the 4 zones in the 2x2 grid in the tables above.
I haven't calculated the average performance for lefties in the 9 zones, but here are the average TAv numbers for RHB:
Up+in = .248
Up+middle = .279
Up+away = .244
Middle+in = .289
Middle+middle = .315
Middle+away = .245
Down+in = .295
Down+middle = .279
Down+away = .205
If we wanted specifically to investigate how lefties and righties did in various parts of the zone, it might make sense to break that out by pitch type. Once we throw the whole league into the sample together, our sample sizes get much, much bigger, and we can safely break it down into additional categories without worrying about sample size being too small.
Dave Allen had a good series of posts on that topic:
Run Value by Pitch Location
Run Value by Pitch Type and Location
Home Run Rate by Pitch Location
Deconstructing the Fastball Run Value Map
Deconstructing the Non-Fastball Run Maps
I am not advocating the idea that the pitcher should just go up there and throw whatever pitch he wants wherever he wants and ignore scouting the hitters. That probably works for some pitchers, but I don't knock the value in scouting the opposition for those who find it helpful.
There are certainly more sophisticated approaches to this problem, both qualitatively and quantitatively. Quantitatively, I ultimately believe the right approach is to build a swing model and test it against the empirical results (and the qualitative observations) rather than simply reporting the empirical results in zones and ignoring the question of why/how it happens.
I chose to use only PA-ending pitches for this evaluation because that is what most existing hot/cold zone graphics do. When Brandon McCarthy was talking about hot/cold zone info, I'm pretty sure he was talking about the kind of charts that you can get from Inside Edge. That's what most of the teams use, and video game makers, too, for that matter.
I definitely agree that as a next step, you could evaluate hot/cold zones based upon performance on all pitches, and of course you would need to baseline the performance on each pitch to the ball-strike count.
Especially if you spell it wrong and have to start over.
That's something we want to do here at BP sooner rather than later. There's still a few important loose ends to tie up. The most notable one I can think of is figuring out how to separate the effect of the pitcher from the catcher in the running game; and following close behind that, separating the effects of the hit-and-run from SB/CS.
So, ironically, the running game is probably one of the least-well understood parts of catcher fielding at the moment.
Beyond that, pitch calling is probably the hardest part of catcher fielding to quantify at the moment, and I don't hold out any hoping of cracking that one in the immediate future.
Bill, I'm not claiming that everything perfectly evens out for LHB and RHB in terms of the strike zone, just that it mostly does so.
The horizontal points at which 50% of taken pitches are called strikes is -1.20 and +0.81 feet for LHB and -1.03 and +1.00 feet for RHB.
Now, of course, as you point out, the pitch distribution matters, too, not just the 50% point. But the average pitch location that LHB see is about 2.4 inches farther outside than the average pitch location to RHB, and the strike zone is shifted by about 2.2 or 2.3 inches. There's a bit of a chicken and egg problem there for determining cause and effect, but for simply describing what we observe, the strike zone shift follows almost exactly with the pitch distribution.
As to whether this information is available anywhere, I don't think it is, other than in the raw PITCHf/x format. (There have been a lot of articles written about the strike zone, so I might be forgetting one where someone has shared some of the pertinent data, but all those authors were basically doing what I'm doing, which is downloading the raw PITCHf/x data and putting it into their own database for analysis.)
By the way, if you look at the reference articles at the end of the post, particularly the first one, I discuss fairly extensively some possible reasons for the LHB strike zone shift. And it is mostly just a shifted zone, not a bigger zone. So I don't know that LHB are disadvantaged in having to defend the outside of the plate more and the inside of the plate less. LHB stand a couple inches closer to the plate anyhow, so it more or less evens out.
Yes, I'm sure the data are right. I'm also far from the only person to observe this. John Walsh was the first, as far as I know, back in 2007:
The Eye of the Umpire
CONT = contact
I agree we're a long way from being finished with our understanding of the strike zone, batter plate discipline, and how to measure them accurately, consistently, and in ways that have useful baseball meaning.
As we go down the road, hopefully we'll make this more user friendly, better formatted, easier to sort, etc.
YEAR ZONE SWING CONT Z_SWING O_SWING Z_CONT O_CONT SW_STRK
2008 50.7% 45.8% 81.4% 63.0% 28.0% 87.9% 66.1% 8.4%
2009 51.0% 45.2% 81.3% 61.8% 28.0% 87.9% 65.6% 8.4%
2010 51.0% 45.4% 80.8% 61.7% 28.4% 87.8% 64.7% 8.6%
2011 50.8% 46.0% 80.8% 62.5% 28.9% 87.9% 64.7% 8.7%
The strike zone definition used for these stats is the one I described here:
A Zone of Their Own
(after the fourth paragraph)
You may be right. We don't have a control group any more to test the theory, as far as I know. I suspect that you're not right, and that it's more driven by umpires calling the zone relative to the catcher target.
I say that for two reasons. One, I looked at where a few umpires stand, and it had no obvious effect on their zone:
Home Plate Umpire Positioning
Two, if you look at the difference in zone between RHB and LHB, which I think is what bugs a lot of people, the zone for LHB is not actually wider, it's just shifted toward the outside (on both the inside and outside edges). This makes sense if it's due to the catcher target, which is shifted outside by 2-3 inches for LHB. But if it's because the umpire is in the slot and can't get a good view of the outside edge, why wouldn't he call the inside edge the same for RHB and LHB?
I blame the neutrinos.
I looked at another Gary Cederstom pitch about a month ago, here:
Framing Ball Four to Cano
Are you talking about the Lucroy pitch? (Braves-Brewers, solid blue background on the wall behind the umpire)
If so, that's Gary Cederstrom.
The PITCHf/x calibration adjustment is a park-related effect, though it doesn't operate strictly like most traditional park corrections do (i.e., it isn't constant across the season).
I can't think of any other park-related effect that would be significant. If you have ideas about some other park-related effect that might be in operation, please share.
Soto and Hill both rate well in blocking pitches, according to Matt Klaassen's numbers.
I know anecdotally that some organizations do coach framing-- the Rays, for instance--but whether their principles line up with my findings is a different question. At least one scout indicated to me that my findings were quite different than how catching is currently scouted and taught.
I, too, really enjoyed the discussion by Len Kasper and Bob Brenly. They talked about it in the top of the fourth inning on Friday.
Thanks, Russell. It's a big enough effect to make a difference in who should play. It's not big enough to play a bad bat over a good bat, generally speaking.
I want to refine the run estimates that I gave in this article with further work, but they should give you an idea of how much value should be ascribed to this skill.
People are welcome to continue to post comments and questions here. I'll check in every day or two. Also, I'll be chatting here at BP on Monday at 1pm ET, so bring your questions for me about catcher defense or other topics.
Thank you, and I'm pleased you took the time to read the linked articles, too. I'm "standing on the shoulders of giants" in all of this.
According to this measure, Thole saved 29 extra strike calls in a half-season of 2010 and lost 18 extra strike calls in 2011, though he wasn't playing quite full time in either 2010 or 2011.
If you simply look at it as a rate stat, it's a pretty big contrast, but as a counting stat, it doesn't seem so big to me.
I suggested an amount that the observed data could be regressed toward league average to find the persistent skill, and if you apply that to Thole, you get him at +4 runs / 120 games in 2010 and -2 runs / 120 games in 2011. That doesn't seem unreasonable to me, but I have not looked into any details for Thole about the actual pitch locations where he gained or lost strikes.
I also don't have as firm a handle as I would ultimately like to have on quantifying the uncertainty level on these numbers. I certainly have a decent idea about the uncertainty from the year-to-year correlations, but I'd like to get a better sense of the impact of catcher teammates, among other things.
In general, I view that we are in the middle of the investigation on this topic, with the work that Max and I have done this summer. We're no longer at the beginning, as we were with Dan and Bill's work, where the findings looked promising but didn't make baseball sense. But we're also not in the end game yet where we've sewed up all the important knowledge and are only making tiny incremental changes. We've gained a significant handle on this effect, but there is significant work yet to be done, as there is with much of baseball analysis.
So, that was a good question, and this was my long-winded way of saying, "Maybe, I think so, but we'll know better as we refine this."
The earliest the PITCHf/x data exists for this analysis is back to the 2007 season.
Maddux was not particularly good at getting calls off the plate in 2007-2008. It's possible he was better at that further back toward his prime, but we don't have the detailed data available before 2007.
Glavine, on the other hand, is at the top of the charts.
I did not account for the variations between umpires in this analysis. My belief is that over a season-size sample, that catchers would be paired with enough different umpires that it shouldn't make a big difference.
I would definitely, though, at some point like to investigate the second part of your question, which is whether some umpires are more susceptible to these sorts of techniques than others.
Well, there is a difference of 30-40 runs, conservatively, between Mathis and Napoli in batting performance. So that's a pretty big hill to climb with fielding performance.
I have Mathis making up about 16 runs per season over Napoli with framing performance. We need to regress that a bit toward average if we're going to estimate actual skill, but then we inflate again a little bit because Mathis and Napoli are being compared partially against one another.
Jeff Mathis does not grade out particularly well at stopping the running game or blocking pitches, so he doesn't gain much on Napoli there, if anything.
If Mathis gains on pitch calling, I haven't tried (and don't know how) to measure that directly, but Sean Smith's study suggests it wouldn't be enough to make up the rest of the offensive difference between Mathis and Napoli.
If you're really generous about leaning toward Mathis as far as possible in all the areas of uncertainty in our fielding measurements, you can get them close to comparable territory, but I wouldn't necessarily recommend that viewpoint.
Thank you. Max Marchi's work addressed the effect of the ball-strike count. I've tried to learn more about the mixed-model regression technique he used, but that's something I don't yet know how to replicate.
I think that as a general rule, the things like batters and umpires, that a catcher is paired with fairly randomly, will tend to wash out with larger samples. Things like pitchers and parks (and the specific PITCHf/x camera system in a park) are tied more closely to the catcher's identity, and we have to take more care in adjusting for them.
Things like ball-strike counts and pitch types probably fall somewhere in between. There shouldn't be a wide variation among catchers in what counts they're in or what pitch types they see, but the small variation that is there could be tied to the team, and thus a persistent bias for the catcher.
I don't know any way to adjust for it simply without killing the sample size, though Max's mixed-model approach does appear to be a potential candidate if you want to adjust for these additional factors.
That's about my feel for the situation, too, but I would have more confidence in the size of the pitch/game-calling portion of catcher ERA if I had a way to measure it directly like we do with catcher framing.
I looked at many more pitches than the four clips I presented here. What I presented here is representative of what I saw across all the pitches I viewed.
Other things like ball-strike count and pitch type can also have an effect on the size of umpire zones, but I looked at a mix of counts, and some mix of pitch types, though mostly fastballs.
The specific pitches shown above are as follows:
Lucroy: May 4, 2001, 6th inning, 2-2 pitch to Alex Gonzalez, Milwaukee trailing Atlanta 4-2
Varitek: July 17, 2011, 12th inning, 1-1 pitch to Ben Zobrist, Boston and Tampa Bay tied 0-0
Molina: May 21, 2011, 8th inning, 1-1 pitch to Brian Bogusevic, Toronto leading Houston 6-4
Doumit: May 21, 2011, 2nd inning, 0-0 pitch to Alex Avila, Pittsburgh and Detroit tied 0-0
Thank you, Alex and TJ.
Yes, it is a problem, for some catchers moreso than others. I discussed it in the section "Problems with Catcher Comparisons."
In that analogy, sabermetricians also have to be careful about assuming they can run the restaurant without having more advanced cooking experience than being able to boil pasta. It's not impossible to do, but it does require extra effort to stay in tune with what the cooks are doing.
I don't have a written guideline for either MLBAM or BIS. What I have is my observation of the data, which is from a fairly large pool for MLBAM (several seasons) and a smaller pool for BIS (partial season).
What I see is that, leaving line drives aside here, MLBAM basically codes anything in the air that is of a depth that could be reasonably caught by an infielder at some position to be a popup. It's not about whether the infielder actually catches it or lets it fall (ideally). My impression is that MLBAM drew a line that was basically along the boundary where an outfielder racing in and an infielder racing back would meet.
On the other hand, I have seen balls just on the outfield grass be coded by BIS as outfield flies. Because I have much less BIS data to go on, I don't have as firm of an impression as to where or how they drew the boundary. I had thought it was at the edge of the outfield grass (or the equivalent line painted on the turf), but you seem to indicate that's not the case, so I don't know.
I should clarify that my understanding is that the BIS definition is that anything on the outfield grass is an outfield fly (or line drive, of course). If that's not right, please correct me.
Because a little bloop or a sky-high, well, I don't know what you'd call it besides a pop-up, that happens to land or be caught a few feet on the outfield grass does not seem to me to be like a high can of corn out to an outfielder. It seems much more like a little bloop or a sky-high pop up that is caught on the infield dirt.
The MLBAM definition is not really about who caught it, it's about distance from the plate, and 160-180 feet or so from the plate makes more sense to me as a popup boundary than the varying 127.6-155.5 feet boundary that BIS uses.
Ultimately, of course, I think a lot of things about the GB-LD-FB-PU division are screwy, but the MLBAM definition just seems a little less screwy here than the BIS definition.
I have not, generally speaking, been able to see a quality difference between BIS and MLBAM data. I would say that for popup data, too, to the extent that I've compared the two in that. I would also say that the MLBAM definition makes a lot more baseball sense to me than the BIS definition, even if the BIS definition is somewhat more carefully applied (and I wouldn't actually assume that it is).
I didn't cover it in this article, but yes, LHB do stand a little closer to the plate. I talked about that in more detail in my article on hit batters.
I had thought that maybe what you suggest was a factor in the shifted strike zone. My speculation was that because the left-handed batter stood closer to the plate, the umpire had to shift closer to the middle of the plate, shifting his strike zone in that direction along with it. But I didn't see that in operation in the positioning data for these three umpires. It's possible that (1) it does work that way, but this sample of data is too small to show it, or (2) the batter position relative to the plate affects the strike zone in some other manner than the umpire positioning.
I discussed those possibilities in my original strike zone (linked in the second paragraph of this article) and concluded that catcher target was the most likely explanation. For one thing, I observed a number of exceptions to the "stands close to plate, strike zone shifts outside" rule. But batters who saw a lot of outside pitches almost all had their strike zones shifted outside. In that article, I discussed some of the questions of cause and effect around that.
I'm sure hitters would do fine with a smaller strike zone. Pitchers maybe not so much. I'm paraphrasing from memory here, but one of the umpires (or ex-umpires) interviewed in the Weber book said something like, "The bat is 32 or 34 inches long. If the batter can't cover a few inches off a 17-inch plate, he's doing something wrong."
A number of umpires also spoke about how they are urged by the league to "find strikes" and "hunt for strikes" in order to keep the games moving.
Handedness of the pitcher and pitch type do not have a large effect on the zone, not to the extent of the 2-3 inch horizontal shifts that we are talking about here.
However, they do have some small effects. Properly measuring those effects, though, I have found to be nearly impossible with the PITCHf/x data. It is very hard to distinguish pitch type effects, for example, from effects due to the catcher target and the ball-strike count. The ball-strike count can be eliminated or adjusted for, but without catcher target data, that effect is tougher to adjust for.
Umpires call a few inches above the belt, but you are correct, they do not call up to the rulebook top of the zone, which would be somewhere around 3.8 feet.
I'm not sure this reshaping of the zone is necessarily a bad thing for the game, though it would of course be much better if the rulebook and reality were in sync with each other. Bruce Weber has a discussion of this topic in his excellent book on umpiring, "As They See 'Em", in which he talks about how umpires basically call the zone in the region where hitters can hit the ball. The rulebook zone, on the other hand, doesn't reflect that reality very well.
My point on bringing up PITCHf/x data calibration issues was not to cast aspersions on that data, though I do think it is a healthy reminder for people who want robot umps or the equivalent thereof. PITCHf/x data is still much more accurate than the judgment of a TV viewer, and most of the time it's probably also a more accurate judge of pitch location than the umpire.
The two PITCHf/x tracking cameras are mounted so that they have a downward-looking angle. They are probably at about the optimum placement for that, unless you suspended a camera on a special mount above the playing field, where it would be subject to being hit by popups.
Even as such, the PITCHf/x cameras have some error in tracking the ball, as discussed in the article, though I believe there are techniques for significantly improving the calibration that MLBAM and Sportvision are not currently availing themselves of.
That's an interesting thought. It seems to me that if ocular dominance were the primary cause, we'd have basically two groups of umpires, one much bigger than the other. Instead, the distribution of left-right shift of the umpire strike zones roughly follows a normal distribution. Also, I would think that shifts in head position on the order of a foot would have a greater effect than dominance of eyes that are a few inches apart. But it is possible that ocular dominance is playing a role here, along with other factors, and I appreciate you bringing it up.
Quentin has an extreme crouch, so that puts his upper body closer to the plate where it's more likely to get hit, and it probably also makes it a little tougher for him to bail out of the way.
I played around with some ways to potentially identify intentional HBP, and I could not find any helpful statistical markers. One would assume that the intentional HBP would be fastballs at the torso, but so are a large number of the unintentional HBP.
For example, I looked at HBP following home runs, HBP on 0-0 counts, and HBP that resulted in ejections, but I couldn't come up with a helpful set of markers from those.
Yeah, on second thought that's probably reasonable, especially considering that Greg had subtracted out the effect of positioning to get to that graph.
Graham, very interesting piece.
I wonder, though, if the primary factor that is driving the "command delta" for fastballs is pitch height. The catcher usually holds his glove low whether or not the pitcher aims low. So pitchers that throw high fastballs end up looking bad according to "command delta" even if they are hitting their spots.
If I take the ten pitchers that you listed for fastball command delta and compare to average pitch height, this is what I get:
Fastball Command Delta vs. Average Pitch Height
The data I had for average pitch height included all pitch types, but the correlation is pretty dramatic, still (r=.96).
That movement on a cutter is quite typical, actually. The more the cutter is fastball-like, the more it will break to the throwing-arm side, and the more it is slider-like,the more it will break to the glove-arm side.
Mariano Rivera's cutter breaks a couple inches glove-side. Roy Halladay's cutter breaks an inch throwing-arm-side.
There's no simple formula for what movement on a cutter is "good" and what movement is "bad". Presumably it's movement relative to the fastball that matters, and not so much the movement independent of the pitcher's other pitches.
Yes, I agree. To the extent that House is right, Suzuki appears to have offended #1 and #6.
I have Brent Mayne's book on catching. I should go look and see what he has to say about this.
Dan Turkenkopf looked at this back in 2008, here:
Revisiting the Run Value of Switching a Ball to a Strike
Lots of folks, including myself, have calculated the run values of various counts, including the results, not only of walks and strikeouts but also of batted balls that occur after that count. I link to Dan's work here because he specifically calculated the effect for called pitches and called pitches near the edge of the zone.
Btw, saving 20 runs in a season means getting about one extra strike call per game, out of about 75 called pitches per game. That does not seem outlandish or impossible to me.
The vast majority of the catchers are within the +/-10 runs/season range, but the very best and worst are around +/-20 runs.
Sean Smith did an excellent study of catcher ERA that was published in the Hardball Times Annual this past winter. He used catcher-pitcher pairs, and he found catcher ERA effects on a similar level. Catcher ERA would include framing, pitch calling, and any psychological/coaching effects. It would not surprise me if framing was a large portion of that.
So I think we're in the right ballpark with framing. As I said before, we're not to the point of having perfected this yet. We might be off by a factor of 2, but I doubt we're off by a factor of 5.
I wonder if this call from last year entered Cederstrom's mind:
Umpire Gary Cederstrom Admits Blow Strike Call in Tigers Loss
Cederstrom called strike three on a pitch on the outside corner against Johnny Damon to end the game. That pitch was about 7 inches farther outside than this one, but I wonder if that experience played any role last night.
Lloyd, catcher framing is something that we in the sabermetric field are just starting to be able to measure well, so I'd tread a little tentatively with our conclusions at this point.
Max Marchi did a three-part series at the Hardball Times, and he found Jose Molina, Brian McCann, and Russell Martin as the consistent leaders in runs saved by catcher framing over the last few seasons.
Evaluating Catchers Framing Pitches Part 3
I wrote some about catcher framing over the offseason.
The Real Strike Zone
I adapted a method pioneered by Dan Turkenkopf and Bill Letson, and I found Jose Molina at the top, followed by Yorvit Torrealba and Gregg Zaun.
I looked at the 2008-9 seasons, and Max looked at the 2008-10 seasons. The best catchers were saving on the order of 20 runs per season due to framing.
On a scouting level, I'm not sure which catcher has the best technique, and I don't know which techniques translate into saving runs other than from commentary from coaches such as House or other observers.
Dan Fox also had two follow-up articles to the one we resurrected a couple weeks ago. They are also worth reading:
Strike Zones, Trilobites, and a Vicious Cycle
The Moral Hazards of the Hit Batsmen
And Steve Treder wrote about it at the Hardball Times:
The HBP Explosion (That Almost Nobody Seems to Have Noticed)
Sure enough, Dave, it turns out that was a post on Plunk Everyone: http://www.plunkeveryone.com/?p=1392
I had forgotten where I had read that, and now you helped remind me. Woohoo!
Thanks, Dave. I have read Plunk Everyone before, but it was not a specific inspiration for this article. That was mainly Albert Lyu, Dan Fox, and Keith Bart.
However, there was another article online somewhere that also figured into my thinking, and I could not for the life of me remember enough specifics about it to find it again. But someone somewhere did a PITCHf/x investigation of hit-by-pitch data this year and debunked the idea that lots of batters were hanging over the plate and getting hit in the strike zone. Honestly, that came as a bit of a surprise to me when I read it there, which may be why I didn't communicate as much surprise about that fact when I wrote about it here.
I definitely believe popup rate is a skill, or to put it differently, the attributes of a pitch have a great deal of influence over the likelihood that the batter will pop it up.
But the details of the physics of why this happens are still a matter for further investigation.
Yes, it does.
He got a hit!
The HBP rate for same-handed batter and pitcher is about twice the rate for opposite-handed. That's for the time period 2007-2010.
From that you could figure out whether LOOGY/ROOGY usage is a significant part of the trend.
There is no justification, other than the assumption that he didn't dramatically lose talent between June 11 and now. He hasn't been performing well lately, but what these 45 latest at bats tell us probably pales in comparison to what the previous 1000 at bats tell us about his hitting talent today.
I say "presumably" and "probably" because I don't know that for sure. I don't have any evidence particular to Counsell's case that informs the presumption, if that's what you're asking.
Once I started watching, I couldn't avert my eyes, kinda like rubbernecking at a wreck on the other side of the freeway.
And I really started to feel for Counsell. I could see the determination and focus in his eyes every time he came to the plate, and I really started hoping that one of his batted balls would find a hole, even though I obviously knew it wouldn't.
On the one hand, the baser part of my baseball fan nature is rooting for Counsell's futility to continue such that he sets the record. On the other hand, at age 40, he comes across as such a "gritty" everyman, at least for those like myself who are approaching age 40, that seeing him continue to fail at the plate is almost as uncomfortable as watching Prince Albert stammer and stutter in The King's Speech.
What drives HBP rate far more than pitch speed is pitch type and pitcher/batter handedness.
Pitches thrown inside from a same-handed pitcher are far more likely to hit the batter in the torso--sinkers are the classic pitch type for this, but four-seam fastballs, changeups, and curveballs also occasionally qualify.
Pitches thrown inside from an opposite-handed pitcher are more likely to hit the batter in the feet or lower legs--sliders are the classic pitch type for this, and sometimes cutters and four-seam fastballs, though this doesn't happen as often as batters getting hit in the torso by a same-handed pitcher.
Hm, it strikes me I should probably get my research together and publish something on this.
This is a good idea. I think without some complicated software, though, that it's not easily possible because the camera frames from different cameras are not synchronized in time.
That is available in our sortable reports under Pitcher Season - BIP:
Great thoughts, guys. Thanks for sharing.
I also had a couple friends on a discussion board suggest that since the Posey injury, catchers have been encouraged to use more swipe tags on plays at the plate. That's an interesting facet of this that I had not considered.
Thanks for the clarification and my apologies to David. I did learn most of what I know about BaseRuns from your posts on the topic. That must have stuck in my head.
If SIERA is simply a forecasting tool, I have very little disagreement with it. I wouldn't necessarily use it as my forecaster of choice, but I would certainly be content to let it lie and don't think that those who use it that way will suffer greatly from any minor troubles it may have in that regard.
It is precisely the fact that is being advanced by Matt and others as a tool for explaining the causal effects of pitching results that bothers me, and all my objections to it are in that regard.
As I've thought about this a bit more, I think should clarify something. There has not been a purge of SIERA supporters from the BP staff. The two co-creators: Eric Seidman and Matt Swartz, have left. They are the main supporters of SIERA that I encountered within the BP staff during the period of several months in which our tenures overlapped. There are probably a not-insignificant number of the BP staff who really don't care one way or the other.
As to metrics thrown out completely when the creator left--Nate Silver's Secret Sauce got that treatment:
Nate's obviously a sharp guy, so take that for what you will, but I see/saw it as a good thing.
"It'd make me question the vetting process of everything BP produces and wonder whether metrics are used because they are a) good/proven/vetted or b) popular with whoever is in charge at BP."
It's honestly probably some of both, but I'd also point out that those two are not mutually exclusive options. In addition, it's a good thing when our less stat-heavy writers at BP tell us, "We can't use this stat in our writing, it doesn't make sense and we can't explain it or justify it to the readers." (Hypothetical example--I'm not talking about SIERA.) We are going to be human in our understanding and our choices here, for all that entails, good and bad.
Vetting is not simply a matter of Colin looking at a stat and calculating some standard errors and p-values and blessing it or consigning it to the dustbin. It's a process of filtering through various people on the team with perspectives that include statistical knowledge, philosophy of what we're trying to do at BP, knowledge of how the game works, connection with what the readers want and need, and hopefully a good dose of common sense. It is also, at times, going to be a matter of trial and error.
I don't mean "dangerous" as in someone's going to get hurt or die or anything. I mean "dangerous" in the sense of it's easy to get wrong answers and have no way to tell that they are wrong.
And I'm not sure what's ominous about that. It's not the case that people who disagree are let go from BP. We have all sorts of internal disagreements. As I've stated elsewhere, some of the former writers are people who I hold in great esteem. Matt has done good work, too. But it should be obvious that there has been a change in philosophy at BP over the last couple years, and some of that is related to a change in personnel (with causation running in both directions). One could obviously read too much into that if one assumes that every departure is due solely to that reason. In some cases it probably is and others it has nothing to do with it.
There is some good discussion of this topic at two additional places, for those who are interested:
Patriot's (developer/co-developer of BaseRuns and Pythagenpat) blog:
Tango and MGL's Book blog:
The staff who are still at BP were never in favor of SIERA from the beginning. (There might be an exception that's not coming to mind for me right now or of which I am unaware, but at least that's mostly true.) We basically don't believe that multivariate regression techniques tend to offer the best understanding of how the game of baseball works. It's very easy to get a coefficient out of a regression and read meaning back into it that may not exist. There are better ways to learn about baseball. Regression techniques have their place, but it's usually in a supporting role, not as the main feature.
The staff who are still at BP are most definitely committed to learning about how to measure and understanding pitching. I would count myself among the foremost in that desire.
SIERA is being retired because we believe that approach is dangerous in terms of thinking it's teaching you things about pitching that will end up being false and because we believe there are better ways to learn about those things.
We are not retiring from the discussion of learning about quantifying pitching skill and performance. Far from it! Matt Swartz and I are going to have very different approaches to the subject going forward. Obviously, I believe my line of inquiry will be more fruitful, or I wouldn't be pursuing it. That's not to say you shouldn't listen to both of us and decide for yourself what to think.
Chris St. John recently joined the BP staff, and one his main tasks at the moment is improving the glossary. So if you find something lacking, it would be very helpful if you would let him know about it. If for some reason you can't find his contact info, let me know, and I'll pass it on to Chris. Thanks.
I meant to say "pillars" instead of "pinnacles", but I guess you could be a pinnacle if you wanted.
I didn't need an apology from you, but I appreciate your willingness to extend one. I think what is sometimes perceived as power politics between the various saber sites, whether BP and Fangraphs or others, is often more of a difference in philosophy about how we should analyze the game of baseball. Sometimes there are personality clashes, too, for sure, but most of the lack of consensus on stats is driven by a disagreement about how we see the game of baseball or what objectives we are trying to reach. Dialogue certainly helps us move forward over time, but the more fundamental differences in perspective about what is important is why it's not simply a matter of getting our heads together to gather 'round a single resolution.
Dave, thank you very much for your thoughts. You will always remain one of the pinnacles of the sabermetric community, and of course, you have given a lot to me personally, too, for which I am grateful.
I'm completely with you on your second-to-last paragraph, and my skepticism extends to projection systems and total-win based systems, too, though there's sometimes a need to use them. For example, when it comes to my fantasy league, I love that there is a projection system I can use. In a theoretical world with infinite time, I might like to roll my own so that I understood all the pieces better, but in the real world, I'm glad that Colin and others are doing it for me.
One of my favorite sabermetric articles ever was the one that Bill James wrote in the 1987 Abstract about how to evaluate a statistic. That has been very influential on future development as a sabermetician. I want to be able to understand clearly what a statistic is telling me about a player or about the game. The mathematical pieces should simply be ways of representing other concepts with numbers or equations. That may be my physics training coming into play, too. I learned most of my higher math through physics classes. Anyhow, that's simply a different way of saying that I hear your perspective loud and clear.
I joined the BP team about nine months ago, about which time the decision was being made not to have SIERA in the annual, which ultimately led to what you see here, as Steven said.
I'm not aware of what Matt or anyone else at BP did with SIERA prior to my joining. I can say that in the first few months I was here, the stats group had an extensive discussion about ERA estimators. A lot of the things that Colin and I have said here and that Matt has said in his series at Fangraphs were part of that discussion, if in more nascent forms.
I'm not sure I can easily summarize what changes have been made in terms of a list of stats added and deleted or formulas changed. Not to minimize the importance of that, because Colin and Rob and others have spent a ton of hours in those efforts.
But for me what it boils down to is mindset. I believe that Colin has been leading a change in that regard. I use Colin's name because he's in charge of the stats here, but really all of it is a group effort and there are many people to whom the positive changes should be credited. That is also not to say that this mindset was not present at BP previously either to a greater or lesser degree. I certainly hold a number of the former BP writers and founders in high regard.
What I saw that brought me back around to BP enough that I was willing to put my own name and effort to it was an emphasis on good, logical sabermetric thinking. That everything should be tested to see if it was true. I find Colin to be the clearest sabermetric thinker in the field. If there's something dodgy with the maths or the approach, he can spot it like no one else can. I need to be around someone like that to clarify my own thinking and work. I believe I have unique insights into how the game works and how to handle data, but I can also get off track in my thinking. Colin asks good questions of me that focus my ideas in a profitable direction. Dan Turkenkopf is also particularly good at that. (You guys may not see Dan's name much on bylines, but he's a big reason that everything keeps going around here stat-wise.)
For example, Colin asked a number of questions of me as I was researching my latest piece on batter plate approaches that helped to shape what the final article looked like, and he reminded me of some of Russell Carleton's previous work on the topic that is pushing me in good directions for future research.
It's that sort of thing that I believe is driving us here at BP toward a better understanding of the game. That will ultimately filter into our stats, too, and it has to some extent already, but I see the stats and their specifics as the branches of the tree, with the trunk being a solid sabermetric mindset and dialogue. The fruit, hopefully, is more enjoyment of the game for all of us, reader and writer alike.
"Wait, BP doesn't like stats to be complex?" "we want our stats to be 'close enough for government work.'"
No, absolutely not. I'm sorry if that's what you got from this.
We want our whole statistical approach to be the best in helping us and you understand and appreciate the game of baseball. There are times when that is best served by complexity, and there are times when that is best served by simplicity.
The reason we are best served by simplicity in this case is because of how little we as a sabermetric field know about how to assign credit/blame for an individual batted ball among the various participants on the field. We know fairly well how to do that once we get thousands of such events to evaluate, but not at the granular level. If we can establish some falsifiable theories about how that happens, and prove them out, then we can proceed to building more complex explanations of how to assign the credit for what a pitcher's true performance was.
Even FIP is a bit of a lie if you take it to be a true and complete record of the pitcher's performance. If you only take it to be a record of the fielding-independent portion of his performance, then you're on more solid footing. (I don't see many people take it that way in its common usage.) It's a bit harder to remember that you're looking at a lie and exactly what sort of a lie it is when you're looking at SIERA or xFIP or something other more elaborate construction that someone is using to purport to tell you a pitcher's true performance over some time period. That's why complexity is a negative here. If ERA estimators were all about telling the truth, the whole truth, and nothing but the truth, we wouldn't have this problem.
I'm as eager for a better solution to it as everyone else.
I appreciate you sharing your concerns. I definitely can tell that you have the best interests of BP at heart here as a long-time subscriber and reader who wants to be able to "trust the process" (as a long-suffering Royals fan, I think I'm entitled to use that phrase) and not have to worry whether you're just watching a pretty facade with rotting framework behind it. Your thoughts are always welcome here, and you are more than welcome to email me personally if you don't feel that you get a good enough answer in these comments.
There are, for lack of a better word, political considerations around personnel decisions, both on the part of BP management and the contributor who chooses to or is asked to leave. I know some of the details of the departure of Matt and other former BP writers, but some of them I do not know. I am glad that I am not in charge of figuring out how to manage all that kind of stuff. Steve Goldman wrote a good piece, I thought, a couple months back on that topic, and anything I would have to add to that would be worse.
I can explain further about what I mean about complexity being a negative if that is helpful. I'm not sure that it is.
Regarding ERA estimators in general, the big unresolved question for them is: What is an ERA estimator trying to do? And from that, how do you know if it is successful? If ERA estimators are not trying to predict next year's ERA, then why do we use that as the test for which ERA estimator is the best? If they are trying to tell us the pitcher's true performance in a small sample, what is the true standard for that against which we can measure? There is none, because that is an unresolved question of sabermetrics, and it will be if and until we figure that out.
We don't know how to disentangle the pitcher's performance from the performance of the batters and the fielders and the umpires in small samples. We know much better how to do that in large samples, but for that, we don't need ERA estimators.
I don't worship at the altar of FIP or kwERA any more than I do at the altar of SIERA. They are all tools that tell us something, but we don't really know what that something is, other than that it is, for FIP and kwERA at least, fielding independent, and for SIERA mostly so. So that's definitely worth more than nothing, to know what part of a pitcher's performance was independent of his fielders (other than his catcher). But it is most definitely not telling us exactly what part of his performance was due to his own effort and exactly what part was due to factors out of his control (call it luck, or what have you). We, as a sabermetric field, don't know how to separate that yet at the game level. One of my quests is to figure that out.
Let me say, simply, hell yes, that sort of process goes on at BP.
As to who exactly did what when behind the scenes, it's certainly not my place to share that, if it's even appropriate to be shared at all.
I will say for my own part that my view on SIERA has never changed. The processes at BP that led to SIERA being published and paraded with such vigor are the same processes that led me to cancel my BP subscription a couple years ago. I didn't hate SIERA then and I don't hate it now. It's not some awful stat. It's just not something I would use or believe helps our understanding of the game. I don't want to say a whole lot more than that because I don't like turning it into a clash of personalities rather than clash of approaches. But I can say that the changes that Colin has been implementing at BP are what brought me back, first as a subscriber, and now as a contributor.
I don't understand your second sentence. Would you mind restating and or explaining further? Thanks. I appreciate your comments.
FIP is on the ERA scale.
The added complexity of SIERA does not give us better info.
There are disagreements in the sabermetric community. I'm not sure how "getting our heads together" will solve that. What do you think we've been doing? This discussion is all part of getting to the truth, but it happens over the period of years. It's not the matter of one person suddenly dispensing some enlightment and everyone else bowing at the feet of their insight. Even something like Voros and DIPS took years to hash out. It's not for lack of effort and dialogue, certainly.
By the way, yes, the points I made about complexity apply to things like PECOTA and fielding metrics, too. Simpler is always better unless complexity buys you something. And when you add complexity, you better make sure you understand why and that you are testing it appropriately and thinking about whether your model is really as robust as you think it is.
I'm probably a skeptic of more baseball stats than I am an advocate, and unnecessary and/or unexamined complexity is one big reason why.
The argument here about complexity is not that "we didn't feel like putting in the extra time and research to make something like SIERA." It is that complexity brings with it several other problems.
One is that it obscures the logic of the underlying processes. Matt claims that the extra terms have real baseball meaning, but he hasn't tested that, much less proven it. I'm skeptical that any of them have much real baseball meaning, but even if some of them do, I'm quite certain that not all of them do. So, when you look at SIERA, how are you to know which terms have real baseball meaning and what they mean? If one of them is wrong, what effect does that have on the whole formula? If something in FIP is wrong, say the coefficient for the BB should be 2.5 instead of 3, it's pretty easy to see how that would affect the result.
Secondly, complexity limits applicability. SIERA is limited to the batted ball data era because of its requirement for groundball data. Who knows if it would work in the minors, or Japanese players, or elsewhere, for example? It's much easier to get the needed data for FIP and to test its applicability in other leagues and levels.
Third, the more complicated a multivariate regression, the easier it is for it to break--in other words, for the conditions that applied in sample to change in ways that cause your result to vary in ways you didn't anticipate--and the harder it tends to be to realize that this has happened.
I can't speak for Colin, but it bothers me that a few of the terms in SIERA's regression switch sign when it goes to SIRA, or from the BP to the FanGraphs version. That's a pretty good indication that those effects are not real and are just a result of overfitting to the data.
Comparing the MLBAM data to the HITf/x data that was made available for April 2009, the best and simplest division between ground balls and line drives is around a vertical launch angle of about 6 or 7 degrees, and between line drives and fly balls at a vertical launch angle of around 22 degrees, based upon how the MLBAM stringer labeled them.
Obviously you can consider launch speed, also, and that complicates the picture a bit. Here's a link to a graph I made of that a while back:
About 12 percent of home runs are scored as line drives by MLBAM stringers.
But home runs as a percentage of line drives are low: only about 2 percent of line drives become home runs (as compared to about 11 percent of outfield fly balls).
You are correct. I will get that fixed. Thanks.
Mostly because lefty hitters on average do a little bit better than righty hitters. Whether that's because of better swing decisions or different strike zones or other factors (e.g., batting more often with the platoon advantage), I don't know how to disentangle.
Thank you, Diana.
Or I should say, 12 buckets for every batter. It's only 24 buckets for the switch hitters.
I have uploaded the data for the 24 buckets for every batter from 2008-2011 to the following Google spreadsheet in case others are interested in perusing the data:
Here's the text of the rule:
6.05 A batter is out when—
(h) After hitting or bunting a fair ball, his bat hits the ball a second time in fair territory. The ball is dead and no runners may advance. If the batter-runner drops his bat and the ball rolls against the bat in fair territory and, in the umpire’s judgment, there was no intention to interfere with the course of the ball, the ball is alive and in play. If the batter is in a legal position in the batter’s box, see Rule 6.03, and, in the umpire’s judgment, there was no intention to interfere with the course of the ball, a batted ball that strikes the batter or his bat shall be ruled a foul ball;
Rule 6.05(h) Comment: If a bat breaks and part of it is in fair territory and is hit by a batted ball or
part of it hits a runner or fielder, play shall continue and no interference called. If batted ball hits part of
a broken bat in foul territory, it is a foul ball.
If a whole bat is thrown into fair territory and interferes with a defensive player attempting to make
a play, interference shall be called, whether intentional or not.
In cases where the batting helmet is accidentally hit with a batted ball on or over fair territory or a thrown ball, the ball remains in play the same as if it has not hit the helmet.
If a batted ball strikes a batting helmet or any other object foreign to the natural ground while on foul territory, it is a foul ball and the ball is dead.
If, in the umpire’s judgment, there is intent on the part of a baserunner to interfere with a batted or thrown ball by dropping the helmet or throwing it at the ball, then the runner would be out, the ball dead and runners would return to last base legally touched.
I think this is the article you were remembering:
You're right about the spin directions for an over-the-top and sidearm screwball. But the ball spinning around the east-west axis (with topspin) will drop.
Tateyama is a little bit above pure sidearm, and he may not be putting exactly the screwball spin on the pitch that you described, but he's pretty close. Anyway, he gets some drop and some movement away from a lefty hitter.
Also, his Vulcan grip seems to cut the spin rate so that he gets a bit of tumbling splitter action mixed in and not as much deflection from the spin as he would if he were spinning the ball harder with a grip like Herrera uses, for instance.
It's not a stupid question, but it's one I'll have to answer later this evening when I can have a baseball in hand. It's too easy for me to get mixed up when conceptualizing the 3-D effects of the spin without a baseball to hold while I'm doing it.
Jason Collette has written a couple articles on that topic recently:
There has been plenty of discussion on this topic among the writers behind the scenes here, some of which has shown up in other articles. (I don't remember which ones off the top of my head at the moment.)
Part of the problem with writing an article with a sweeping conclusion about run scoring is that there is no sweeping conclusion that I would believe at this point. The one thing that's a likely culprit for sweeping changes, the baseball itself, we can't really test.
One other thing that is commonly proffered as an answer is declining use of steroids. That doesn't make sense to me, but again, it's something we can't measure directly and necessarily involves a lot of speculation about how changing steroid usage would manifest itself in the game.
Some things we can test. Is the strike zone changing? Not as far as I can tell. Is cutter usage increasing? No, not really. Is the usage of other pitch types changing much? No. Are pitchers throwing more strikes? Yes, a tiny bit more, but not much. Are pitchers throwing harder? Yes, a tiny bit more, but not much.
One reason I haven't written more on the subject is that most of the theories come up negative or with a very small impact. In that situation, I typically wait for a bigger sample so I can get a better read on the significance of these small effects in order to be more confident that they aren't just normal variations over the course of a couple months. It's possible that by the end of the season, the change in run scoring won't look quite as dramatic as it does now.
When batted balls don't fall in as much or don't go over the fence as often, which seems to be the main driver of the scoring drop this year, it's very difficult to determine why. The baseball, the batters, the pitchers, the fielders, and the weather all contribute. Big samples are the way to sort out those types of effects.
So, to sum up, I think Jason has given us a good answer based upon what we know at this point. You can expect to see more analysis of the situation as the season progresses.
I have to give credit for the headline to my editor, Ben Lindbergh, and I don't actually know what he had in mind. In any case, his headline is far better than the stupid bird song puns I was coming up with.
Colin gives more detail on this subject here:
I always appreciate getting the thoughts of an umpire. Thanks.
Thanks very much for your thoughts. That's exactly the type of insight I was hoping to get by posting these pictures.
Crawford, by being a lefty who stands at the very back of the box, increases his chances. But that in and of itself is not sufficient because there must be many other such batters. I did notice that he had a very open stance. Would that contribute?
As far as pitch locations and ball-strike counts in the photographs go, the first one (5/22/11) was down the middle on a 2-1 pitch. The next one (8/5/10) was on the outside edge, as you can see, and on an 0-1 pitch. The last one (7/26/10) was on a 1-1 pitch that was on the outside half of the plate.
Here's one with Crawford from August 5, 2010, with Drew Butera's glove about to touch his bat:
And two views of the interference by Gerald Laird against Crawford on July 26, 2010:
Here is still shot of the split second before Crawford's bat hit Castillo's glove on the catcher interference call last night.
Now here's a fun image of Crawford starting to swing (note the back foot):
And look at the lean here:
Thanks to you and thegeneral13 for your thoughts on this. It makes a lot of sense.
I do wonder what it is about Crawford's swing that makes him particularly prone to CI. I should probably watch some video and see if he's doing something different in those instances.
Generally speaking, I don't know, but Crawford doesn't seem to be a particularly egregious offender:
We have PITCHf/x data for 72 of these catcher interference calls. Nothing particularly stands out in pitch types: 71 percent fastballs, 21 percent breaking balls, 8 percent changeups. Fastballs seem overweighted a bit, but I don't know if that's significant given the sample size.
The pitch location data shows pitches at middle height across the plate, which is probably to be expected given that the batter swung at the pitch. It occurs most often with left-handed batters on pitches on the outside edge--half the sample of CI fits this criterion.
I have been thinking about this question, too. The human eye is really good at picking up on patterns (sometimes even on patterns that are really just random noise). If I could come up with a set of sound principles for identifying real speed changes, I could probably tell a computer how to do it, too. It's coming up with the set of sound principles, other than "I know it when I see it", that is difficult.
Morton chart is here:
Morton gained about 1 mph from 2008 to 2009 and another mph from 2009 to 2010. His 2011 speed is about the same as 2010.
Your other questions are good, but I don't have answers to them at this point.
Here is the chart for Bedard:
He was sitting around 91 mph in both 2008 and 2009. Of course both of those seasons were partially lost to injury. In 2011, he started slow, around 90 mph) but has picked up the pace to the 91-92 mph range in his recent starts.
We don't (yet) include fastball speed data into the PECOTA projection system. One challenge of doing that is that reliable speed data only goes back a few years, severely limiting the amount of comparable players available for comparison. Hopefully in time that will be a challenge we can overcome.
Yes, movement changes would be important to track also. IMO, it's more sensitive to pitch classification problems than is fastball speed. For example, if the distribution between two-seamers and four-seamers changes, and your classification system doesn't do a good job of picking up on that, speed isn't affected much, but movement is.
I had meant to link to Josh Kalk's 2009 study of this at an intra-game level:
I'm not sure which tables you are referring to, but Tango's win expectancy tables are available here:
We have our win expectancy reports available here:
Very interesting article, Matt.
Thank you all for the kind words.
Alan, I've taken a look at this now, and I do see about half the effect due to temperature when I look at release speed (50 ft). I see an effect of 1 mph per 70 degrees F.
I should disclose that I've worked briefly as a consultant for TrackMan, but my comments here are based entirely on what's known publicly about their system and data.
I can do a more in-depth investigation of the calibration of the camera systems with respect to velocity. That's included here in a big bucket with a number of other factors.
Correcting speed measurements is a lot more complicated than correcting position measurements. So I've debated about how to work through that and present it one piece at a time.
The article I did a couple weeks ago on how temperature and point in the season affects fastball speed was a piece of that investigation. I've done more work that I haven't published yet.
As far as TrackMan goes, it's a very promising technology, both for pitch tracking and especially for batted ball tracking. The data is very unlikely to become public. I have some reservations about the conclusions that are drawn by TrackMan's folks about curveball spin rates and release distances, as published by Verducci and others previously.
It matters because, as Sky pointed out here:
A team that already starts with a n-game losing streak is more likely to have another such losing streak in the season than a random team is two have two such streaks.
It also matters because, although a six-game sample is very small on an individual player level, it's somewhat more significant on a collective team level, and even six games affect our estimate of a team's true talent. Six games of current year performance have a lot more predictive power than six games of past performance.
Rany Jazayerli wrote about this back in 2003, and I was actually thinking about his article when I wrote yesterday about the cautionary history of previous teams in this situation. (One of them was mentioned here--the 2008 Tigers.) I couldn't find the article when I was writing yesterday, but thanks to Steve for finding the link:
I heard that last night on Twitter, too, but the PITCHf/x data shows that he was still using his sinker after the first inning. He threw about 35 sinkers and 30 four-seamers in the game.
Here's the Baltimore Sun story:
And here's the quote from Britton:
"After that first inning, I started to get a lot more comfortable out on the mound, but at the same time, I was really excited," Britton said. "We canned the sinker. We were throwing [four-seam fastballs] all day, and they hit some balls hard. But I think I was too amped up today to throw the sinker, so I wasn't able to get the ground balls that I wanted to.
I was disappointed that I didn't have the sinker, but the guys were like: 'Think about it. You can pitch here without your best pitch.' And that's something I'll take out of today."
I'll have to check which innings he used his sinker and how much. He didn't explicitly say when he canned the sinker.
I used all players who pitched in more than one game. That will bias the sample a little bit with the data from pitchers who only pitched in games in a certain small temperature range, but I don't think the effect is very significant. It will be a good idea to test that assumption, though.
Yes, thanks for the correction, Alan. I meant the air density or the drag, not the drag coefficient.
I have not done the analysis using the release speed, but that is on my list of things to do.
There is nothing inherent in the PITCHf/x measurement about pitch types. MLBAM adds their pitch type classification to the data after the fact. (By after the fact, I mean within a second or so.) PITCHf/x records the trajectory and speed, and it does that just as well for sinkers as for four-seamers.
When I say that speed is not a good way to separate four-seamers from sinkers, I mean that it's not a reliable indicator. Many pitchers throw their sinkers just as fast as their four-seamers. Other pitchers throw their sinkers a couple mph slower than their four-seamers. On average in the major leagues, the difference is about one mph.
If you are using the PITCHf/x data to classify pitches on your own (ignoring how the MLBAM algorithm decided to classify them), you would be wise not to use speed as a differentiator unless the pitcher you are evaluating actually throws his sinkers slower than his four-seamers. That has to be evaluated on a case-by-case basis. Movement, on the other hand, is a pretty reliable guide to separating sinkers from four-seamers in all cases.
I'd have to look through the PITCHf/x data on a game-by-game basis to give a complete answer, but just from looking at a few games, it seems that MLBAM's classification algorithm is keying off the spin axis angle and the speed to draw a line between sinkers and "four-seamers". (The more the spin axis is tilted, the more drop and arm-side movement a pitch will have.)
Since McClellan, as far as I can, doesn't throw a four-seamer (or if he does, it's only rarely), the MLBAM classifications ended up drawing an arbitrary line through the middle of his sinkers, dependent largely on how the PITCHf/x camera system happened to be calibrated that day rather than on any change that McClellan made.
We've also made quite a bit of progress on #3. We do a lot better these days on measuring how well catches block pitches in the dirt, and between Sean Smith's work on a WOWY approach to game calling and my proposed extension of Dan Turkenkopf and Bill Letson's work on catcher framing, I think we're starting to get a handle on that piece, too.
You're correct if you're making a general statement about classifying fastball types, but Craig was writing about McClellan in specific. McClellan does throw mostly (or maybe even exclusively) a two-seam sinking fastball and doesn't throw many (or any) four-seam fastballs. The MLBAM pitch classifications have McClellan throwing about half sinkers and half four-seamers, which is definitely not right.
Also, for many pitchers, speed is not a good way to separate four-seam fastballs from sinking fastballs.
I don't know. His cutter is normally around 88, and his four-seam fastball is around 92-93. Unfortunately there is no PITCHf/x data from Florida spring training sites, so I don't have any data to give you a better answer than what you already have.
I should clarify that when I say that the fastball speeds that I report in this article are based on the speeds measured at home plate, I mean that, unless I said otherwise, that's the speed used. In some cases, where noted, I temperature adjusted the speeds, and toward the end of the article, where I'm comparing from previous seasons to spring training, I also adjusted the plate speeds for camera calibration errors. (The camera calibration is better near home plate, but it's not perfect there, either.)
Thanks, Dave. I believe the correlation between temperature and fastball speed is partly real, occurring at the pitcher level. There is a portion of it which is a measurement/methodology artifact, probably about half of the effect that I reported here. I need to spend a bit more time figuring that out, but here's what I know right now.
The fastball speeds that I report in this article are based on the speeds measured at home plate. That helps remove a lot of the measurement error due to PITCHf/x camera calibration. I then increased the speeds by a flat 8% multiplier to get back to the standard 50-foot distance, assuming an average drag coefficient. The drag coefficient varies with temperature (and with altitude and properties of the individual baseball, etc.). A baseball going 91.8 mph when 50 feet from the plate will be going 85 mph at the plate with average temperature and altitude. A baseball going 92.3 mph when 50 feet from the plate will be going 85 mph if the temperature is 37 degrees below average because the denser air will slow it down more.
So about half the effect that I report above is probably due to air resistance changes and about half the effect is probably real.
I have some methods to correct pitch speeds for changing drag coefficients, but I haven't applied to this analysis yet.
One thing that gives me some reassurance in claiming that increased temperature helps pitchers throw harder is that the home pitcher, coming straight from warming up in the bullpen, throws harder in the first inning than does the visiting pitcher, who must go sit on the bench for a half inning before entering the game. The difference is nearly one mph in favor of the home-team pitcher, and it evaporates completely after the first inning.
Thanks, Frank. It's hard for me to get a good reading on command from just one game without having any video or record of the catcher's target. He was elevating his pitches in the first inning, but after that not as much. He was actually missing badly low with his sinker and cutter with some frequency in the 3rd and 4th innings.
Thanks, herron. The pitch classifications are my own. Pineiro's typical fastball is a sinker based upon its movement. He has also has a four-seam fastball that he throws less frequently. Something like three fourths of his fastballs are typically sinkers (roughly--I haven't spent a lot of detailed time with his data outside of the two starts I show here). In the September start shown here he didn't use the four-seamer very much, and in the March start, he didn't use the four-seamer at all.
Yes, there's a picture of his curveball grip toward the end of this article:
BABIP is dependent on a number of things. How hard the batter swings is a big one. Whether the batter pulls the ball or hits to the opposite field is another. Both the batter and the pitcher affect this, though the batter seems to have a much greater ability to affect it than the pitcher, in general. It would not surprise me if batters swing harder now than they did in the 1960s, particularly with two strikes, based upon strikeout rates, but I don't have any direct evidence of swing speed measurements to prove it one way or the other.
The coefficient of restitution of the baseball is also something which may have changed over the decades and would affect how hard balls are hit, and thus, BABIP.
The quality of fielding might also have changed, either in the speed and athleticism of fielders or in better gloves.
Improvements in hitters' backgrounds and lighting for night games might have improved the ability for hitters to make contact. Changes in ballpark sizes or altitudes might also impact BABIP, although it's not apparent to me at first glance in which direction those effects might act.
Those are few ideas off the top of my head. I'm sure there are some I've overlooked, and probably some that I proposed had only a small effect.
Also, Voros published DIPS 2.0 here:
GreggB, it's worth noting that this was the first formulation of Voros's DIPS theory to a wide audience. There have been a number of refinements, by Voros and others, in the decade since this was published.
Voros and Eric Van discuss here:
Keith Woolner examined DIPS here:
Tom Tippett looked at DIPS at the career level:
Also, you can't simply compare BABIP from different eras to the league average baseline from 1999. The league BABIP was much lower in the 1960s and 1970s, for instance.
I assume by published PITCHf/x vertical movement numbers, you are referring to the vertical component of the spin movement, often labeled as pfx_z. That is one contributor to the final vertical velocity crossing the plate, Vzf. The other contributor is the location. Location is actually as big or bigger of a contributor than the vertical spin movement.
If a pitch is released 6 feet above the ground and crosses the plate at the knees, it drops about 4 feet in .4 seconds, or around 10 feet/second. If it crosses at the letters, it drops about 2 feet in .4 seconds, or around 5 feet/second. So you can see that location makes a big difference.
The vertical spin movement can make the pitch drop a foot less than expected (fastball) or nearly a foot more than expected (curveball). For a fastball that's +1 foot/.4 seconds = +2.5 feet/sec. For a curveball that's -1 foot/.5 seconds = -2 feet/sec.
So you can see that the range of the two effects are on the same order.
PITCHf/x had his fastball averaging 89.1 mph when he pitched against Kansas City in Surprise on Tuesday.
Gregg, when I looked at changes in effectiveness due to changes in velocity on a season level, I found that pitchers who threw slower were less affected by a loss in velocity.
Whether that applies equally well within a game, I'm not sure. It's tough to separate multiple effects that are going on as pitchers pitch deeper into the games (seeing same batters multiple times, fatigue, different quality of pitchers in the sample, changing temperature, pinch hitters, etc.).
Aliendna, I'm not sure exactly what you mean by pattern analysis. Dan Brooks has done a lot of work with umpire data and signal detection theory.
I think, though, that the rounding of the corners can be explained more simply. For a pitch on the edge, an umpire only has to decide if he thinks it's left or right of the boundary. For a pitch on the corner of the zone, he has to think about left/right and up/down. If it's right on the 50/50 left/right border and the 50/50 up/down border, a simple model would say he's likely to call 25% of those pitches as strikes. So the 50/50 boundary moves in on the corners because it's affected by uncertainty in two dimensions rather than just one.
Ben, even so, that WARP value for Erickson can't be right. 251 innings of pitching at slightly better than average might give Erickson 5 wins above replacement, but there's no way that puts him at the MVP level of 11 wins.
Erickson allowed 125 runs in 1998, when an average pitcher would have allowed 141 in the same number of innings. If replacement level is around 1R/9 worse, then a replacement pitcher would have allowed 169. There are park effects and other considerations to be sure, but the simple runs tally puts Erickson in the 4-5 wins above replacement range.
Recapped game this way.
KG permits me to link.
I'll do so forthwith.
You will sadly be
out of luck unless you can
make a poem, friend.
The robot calls, "Strike."
Batter looks skyward in vain,
Missing human umps.
Sportvision does some calibration on the field prior to games. Most of this appears to take place between homestands due to issues with getting access to the field prior to games. There is some further discussion of this topic in the comments (particularly those by Alan Nathan) in the Book blog thread on Kyle Boddy's article from the Hardball Times. Alan and Kyle both suggested some ways in which the calibration might be improved. The problem remains, however, of how to calibrate the system better during the period of time during which access to the field is restricted.
The Book blog thread is here:
Kyle's article is here:
Thanks, Jon. You might have liked the alternate title I came up with after I pushed publish on this post:
"You Are the Weekend Links. Goodbye!"
No problem, Doug. I appreciate your comments.
Jeremy, you addressed the median and the mode in the text, but as I stare at the graphs, I find myself wondering if there is a simple way to draw a line that goes through both the dark cluster around the origin and the dark cluster of full playing time in the upper right. A mathematical technique, I mean. The "best fit" line is having to account for all the points along the x-axis.
No. If you're a right-handed pitcher, as Wainwright is, when you raise your arm angle, the vertical release point rises and the horizontal (left/right) release point moves toward first base. If you drop your arm angle, vertical release point drops and the horizontal release point moves toward third base.
If the vertical release point stays constant and the horizontal release point changes, then the pitcher is shifting on the rubber without changing his arm angle.
If a pitcher both changes his arm angle and moves on the rubber at the same time, the result in the horizontal dimension can be more complex (sum of the two effects).
However, if the release point doesn't change in the vertical dimension the pitcher can't be changing his arm angle, unless he's coming basically overhand to start with, which Wainwright definitely was not. You can stick your arm out at a 45-degree angle and see for yourself how the release point changes in each dimension when you raise or lower your arm angle.
Good stuff, Jeremy.
Sky, I would consider that a very open area of research.
However, one of the pieces is something that Matt Lentzner and I presented on at the 2010 PITCHf/x summit. We found that the final velocity components of the pitch as it crossed the plate, both left/right and up/down, influenced the BABIP and whiff rate of the pitch.
I believe that another important piece of the puzzle is deception. There is deception in making the spin on the pitch hard to read, and from what I know about cutters, their spin is hard/impossible to differentiate from a four-seam fastball. Then there is deception in making the trajectories of various pitch types look as much alike as possible early in the pitch's flight. Josh Kalk did a fair amount of work on this topic. However, there's one piece he didn't really look at, which is that pitches look very different from the LH batter's box than they do from the RH batter's box. It's much easier to disguise pitch trajectories to a same-handed hitter.
So, I believe there is more research to be done both in the final velocity components and in the deception to help us understand what makes pitches effective.
Doug, I definitely agree that Peterson talked about getting Axford's stride direction and spine lined up so that he was finishing lined up with the centerline.
But then he went on to show a release point graph and very clearly stated about moving Axford so that the pitch started out within the zone, and he showed on the graph what he meant by that. If you guys have watched the video and interpreted it differently, I'm willing to be corrected, but I thought he was crystal clear on the video about what he was talking about. He said it more than once and demonstrated on the chart what he meant.
I just finished reading that book last week!
It's an interesting question. I'll have to think about the best way to study that.
I don't think so, because if you interpret it that way, then it's not a reason for Axford to shift like he did.
I do think there's something to release point and how batters see the pitch moving at them or away from them early in the trajectory and then which way the pitch is moving as it crosses the plate, but you don't have to make the shift that Axford made to accomplish that, and that isn't what I saw Peterson claiming.
Peterson did offer some additional reasons, such as Axford having trouble locating to certain parts of the zone from his release point at -2 feet. Those may hold up; I haven't examined them in detail. The release point claim, though, caught my ear, because I thought I remembered that Josh's work had shown otherwise.
Everyone focuses on HR/FB and flyball BABIP as the reasons for Cain's success. IMO, it was quite a nice observation that groundball BABIP is also a driver of his success.
Another way to look at it is to simply look at the batting average on ground balls. Matt Cain's career batting average allowed on ground balls is .217. The Giants team from 2006-2010 was at .2336, and the National League was at .2374.
If we assume that Cain had an average set of Giants defenders behind him throughout that time (big assumption), then his fielders gave him slightly above average performance on ground balls, but his performance on ground balls was quite a bit better than an average set of his infield defenders would suggest.
Looking at nFRAA range runs for the Giants infielders from 2006-2010, it's a mixed bag.
Feliz +54 runs
Durham is the only awful one, and both Feliz and Vizquel were quite good. Overall it's a somewhat above average group.
Looking at data is not the same thing as doing a well-designed study of a falsifiable hypothesis.
Jeremy Greenhouse took a look at the Verducci Effect here:
David Gassko also looked at it several years ago:
Even if we wanted to lump stolen base attempts and hit-and-run plays together, the problem is that we are including the negative result of a busted hit-and-run but not the positive result of extra bases advanced when the hit-and-run succeeded. We're putting those positive runs in a different bucket because we don't have a record that they belonged with the hit-and-run plays.
PFM should be working. If you're having trouble, please email customer service with the details of the problem you are experiencing. Thanks.
He's projected to hit 12 HR in 121 PA at DH plus 129 PA at 1B, so 250 PA total. He's also, according to the weighted means spreadsheet, projected to have really poor defense at 1B (-18 runs if playing full time) that costs him all of his offensive value.
I mentioned in the article that LHB see more outside pitches than RHB. I didn't mention why this occurs, though. It's largely because LHB disproportionately face RHP, and they see a lot of changeups and fastballs on the outside half from RHP.
On the other hand, RHB face a closer to equal mix of RHP/LHP. Also, they tend to see fastballs on the inside half as much or more often than outside, and the sliders from RHP and changeups from LHP that they do see predominantly on the outside half aren't enough to tilt their location mix back to the outside to nearly the extent it does for LHB.
The work has already been done by a lot of folks before me, and I listed them in the article. The size of the called zone definitely changes by count. The question is, why? I found some effect from pitch types, but the effect remained when only looking at fastballs.
I don't think we've ruled out that (1) something is biasing the sample of pitches that I/we haven't thought to consider, and/or (2) the umpires or catchers are doing something mechanically different based upon the count. For instance, catchers may (and do) take a different stance with two strikes in order to block third strikes in the dirt.
Until we examine more of these type of possibilities, I'm not comfortable simply chalking it up to an umpire philosophy of helping the underdog and considering the case closed.
One of the things that J-Doug identified needs more exploration: that the size of the zone changes with base-out state, and in favor of the party who already has the advantage. I haven't investigated that at all, but I'm hopeful that if we can identify the cause(s) of that, it may shed some light on what is causing the zone changes with the ball-strike count.
Thanks, Dave. That means a lot coming from you.
You might be interested in these articles by Dave Allen:
They address run value on swings, which is slightly different than BABIP like you asked about. PITCHf/x definitely has data that can be used to answer your question.
It seems reasonable to me to assume that the pitch clusters generally correlate with the catcher target. That much is common sense, and it's something I've validated by charting games from video. Whether the catcher target is exactly at the centroid of the cluster is another matter. I'm sure it's not always at the center. For one thing, most of the charts I presented are an amalgamation of ball-strike counts, and pitchers and catchers do vary where they aim and set the target. But also, individual pitchers may preferentially miss in the same direction from the middle of the glove, and individual catchers may set up the glove relative to the intended aim point in different ways.
If we wanted to be as accurate as possible in assigning a value to gaining strikes by pitchers and catchers, we'd have to account for those factors as best we could. For a first pass, though, I think it would be fine to assume the catcher setting up at the cluster center and see where that leads us.
That's what John Walsh suggested. That may well be part of it, but I haven't seen any proof of that in the data. There's nothing wrong with conjecture, and it can even be helpful, but we can't say we know on that basis. I'd like to know. There's more investigation to be done.
To make your scenario more analogous to what actually happens in baseball, I'd say that they move the goalposts depending on the kicker. The officials do call field goals based on whether the ball actually goes through the goalposts, but not every kicker has the same goalposts. Where the goalposts are positioned for each kick depend upon where the kicker is kicking from and where he tends to kick to.
The St. Louis pitchers do have the lowest average pitch location, at 2.29 feet, and San Francisco is the highest at 2.52 feet.
I need to query my database again before I present all the numbers since I forgot to split it out by batter handedness when I did the team query.
I should have also mentioned that the zone that A-Rod sees is very close to that of the typical right-handed batter.
1. Batters do have an effect on the zone, and I talked about that in the article.
2. There does appear to be a home-field advantage in strike zone. I linked to Dan Turkenkopf's article on that topic, and J-Doug Mathewson wrote on that at Beyond the Boxscore and found the same size effect as Dan.
3. It still amazes me that thanks to Retrosheet, MLBAM, and Baseball-Reference, I can find the exact plate appearances you are talking about.
Nick Swisher's home run on September 8 was on a 2-0 pitch. It was in a decent location, on the black but thigh-high (-0.7, 2.4 ft), and followed a pitch that was very low and slightly more outside (-1.0, 0.9 ft).
Alex Rodriguez's home run on September 17 was on a 2-2 pitch that caught a more of the plate than Swisher's, plus it was on the inside half (-0.5, 2.3 ft). It followed a ball call on a pitch out of the rulebook zone but on the very margin of the real zone (-1.2, 2.2 ft). That pitch is called a strike to A-Rod less than 20 percent of the time.
It seems a stretch to say that the Yankees caught a break from the umpire in either of those two cases.
Matt Lentzner and I did a presentation at the PITCHf/x Summit in August 2010 about how the final velocity of the pitch in the horizontal (left-right) and vertical dimensions as it crosses the plate has a strong effect on the whiff rate and BABIP of the pitch. In the vertical dimension, we found that pitches that were moving at a 6 or 7-degree downward angle as they crossed the plate were most likely to be hit. This is close to the angle of the typical batter swing plane.
That is true in aggregate for the league, it is not true for every specific pitch type thrown by every pitcher. I suspect that the way some pitchers play one pitch off another and the associated deception, or lack thereof, is part of what makes this vary from pitcher to pitcher.
Lowe's sinkers are dropping more than the average fastball when they cross the plate, and all else being equal (is it? I don't know), should create more whiffs and lower BABIP than average.
That's an interesting thought. Dave Allen, and perhaps others I'm forgetting, have done some work on batted ball value by zone location for the league:
I've also looked some at how pitch type and location affects BABIP, but I hadn't thought of looking at in the specific way you suggested here. That might be helpful.
I don't think, though I haven't checked, that pitchers lose much precision as they age.
I would guess that changes in location for pitchers over time would show a purposeful change in approach. I did show in the strike zone article that pitchers pitch outside more the older they are. Because pitchers enter and leave the sample at different ages, that's not quite the same thing as showing that pitchers pitch outside more as they age, but I would guess that's also true. I don't know why, though.
It might be due to changing pitch types; it might be due to spotting a slower fastball more out of the hitting zone; maybe something else. It's definitely an interesting topic for research.
We know that the umpires position themselves on the inside edge of the plate, i.e., between the batter and catcher. Beyond that, though, we don't have a record. I would love to have a detailed record of umpire positions and stances.
From anecdotal evidence I agree with you. I've looked into a some of the ball calls where PITCHf/x said the pitch was down the pipe, and they were either recording errors (data assigned to the wrong pitch) or check swings.
It was around for the last two years, 2007-2008. Glavine definitely pitched on the edges and took advantage of this. I was just looking at his data two days before publication. If I'd looked earlier I might have included him in the article. He wasn't quite Livan Hernandez-like, but he was very close, on par with Mariano Rivera in hitting the edges.
Maddux didn't appear to do anything special in expanding the zone, at least at the end of his career.
Thank you all for the kind words. I'm glad you like it, and I'm glad I was finally able to spill all these thoughts out of my brain and onto the virtual page.
I know some people feel that a changing zone is anathema. I don't know how I feel, but I do think it's important to recognize that it's a collective decision by baseball to call the zone this way--players and coaches are complicit--not some arbitrary decision imposed by fiat on an unwilling game by rogue umpires.
Also, changing the way the zone is called would have a substantial impact on the game. Some would say it would make it better, but it's not clear to me whether that's true.
The middle of the plate is 0.0 in both graphs. The 0.5-ft mark is six inches toward the outside of the plate.
Most pitchers tend to pitch both lefties and righties toward the outside part of the plate, a little moreso to lefty batters.
I can take a look by team and see what I find. That's a good idea.
We appreciate your input, especially when you take the time to lay out your thoughts carefully like this. The need for a new interface is recognized, and that is in the works.
I won't begrudge anyone their opinion on Emma's article, and I appreciate the decorum with which people have discussed their feelings about it.
But as far as BP's new direction, don't think for a minute that it doesn't involve a heavy emphasis on good sabermetric analysis of the game. Jeremy Greenhouse, Dan Turkenkopf, Sky Kalkman--these guys are all heavyweight names in the saber field, and deservedly so. Hopefully, I and the other regular stats writers will also continue to produce content that will advance the thinking in the field as well as being enjoyable to read.
Yes, you have explained the basic idea well.
Do not question the Colin's powers, or you risk being ground up and fed into his nutrient bath.
The link that Juris posted above your comment is an excellent one to begin learning about what regression to the mean actually is.
I agree there's a lot of confusion about what PECOTA does. Quite a bit of it has been explained at one time or another in one place or another, some of it online, some of it in print. It might be helpful to see if some of that could be centralized or indexed in one place.
Part of the problem is that Colin's really the only one here who fully knows how PECOTA works. Nate and Clay would obviously be able to contribute on that front, but they're not around as much nowadays. Colin can't work on everything everyone thinks might be wrong with PECOTA or explain everything that everyone wants explained about PECOTA. Some high priority stuff he can and will, but other things will take longer.
So it falls to some other folks, like me, who know less about all the intricate details of PECOTA, to attend to some of the questions.
With that caveat, let me say that regression to the MLB mean does not imply that all players have the same talent level. If that were true, we'd just predict everyone to do league average next year. It's regression to the mean, not collapsing to the mean.
Tango's the king of explaining and pushing the implementation of the regression to the mean concept. The way he likes to explain it is that the longer the playing record we have for a given player, the less regression toward the mean that we need to apply. The way that Tango does regression to the mean for Marcel is to add in a half-season worth of league-average stats into the player's last three years of major-league stats.
As to the specific amounts of regression to the mean that PECOTA uses for its baseline forecast, I don't know. It's probably not terribly different than Marcel in the amount of regression for established major-league players, but for minor-league players, PECOTA has the advantage of using translated minor-league stats.
With the depth charts will come human input about likely playing time. The weighted means spreadsheet is the output that PECOTA gives without knowledge about who has already suffered an injury that will cause them to miss time next year or whose playing time may change because of change in role.
By regression to the mean, I was referring to regression toward the MLB mean. Everyone (or almost) is going to be projected to do worse than their career-best year. The human tendency is to assume that a career-best year defines a talent level, and we want to see PECOTA project them to repeat that. However, that's not the most likely outcome. Most likely, if a player did really well last year, he got a little lucky, and, conversely, if he did really poorly last year, he got a little unlucky.
You can see my specific comments about Longoria a few comments down. (Looks wrong to me, too.)
It's great to have the high level of interest in PECOTA that we do, but one of the bad side effects of 200 comments on the thread is that it's easy for ideas to get lost or crossed in cyberspace here.
For Longoria, specifically, I agree that his numbers look low. His projected batting average seems about 20 points too low to me. "Seems to low to me" is not necessarily the same as "PECOTA made a mistake here". We'll look into it and see if there's anything wrong.
The overall set of projected numbers match up pretty closely with 2010 levels of offense. That is down from previous seasons, of course, so that may be why the numbers for hitters look low across the board.
In my experience, people also tend to perceive regression to the mean, while statistically correct, as making the numbers look too low for everyone.
I'm really looking forward to that day, and being part of making that happen.
The obvious problem is that that data doesn't exist prior to 2007, so the sample size is still very small. Every additional year of data helps us solve that problem.
The other "problem" is figuring out what things to incorporate from PITCHf/x, and how. Doing that with fastball velocity isn't trivial, but it's probably doable. Doing that with other pitch data is more problematic.
If you wanted me to give a ballpark guess, I'd say that I really hope that two years from now, PECOTA includes PITCHf/x information. That's not a guarantee or even a firm estimate or something I've discussed with Colin. It's me putting my finger up into the wind and projecting into the future how well we've done at digesting the PITCHf/x information that we have.
I mentioned a couple links to studies already, below. What I am saying is nothing new. You don't seem to be completely grasping what I am saying, though. I am not saying that Marcel is identical to PECOTA in every way, nor did Colin in his article.
If all you want is a rate state projection for every player who was a regular in the major leagues last year, you will do just fine with Marcel. Don't spend your money to subscribe to BP if all you want is a PECOTA rate-stat projection for that set of established major-league players.
PECOTA offers much more than that, plus hopefully it is more accurate than Marcel even on rate stats for established players. But it's not going to be outlandishly better than Marcel on that. Nate Silver said that, Colin said that before today. I'm not making some horrible reveal of the awful truth that BP has been trying to conceal for years.
What you say on #1 is the case. BP is committed to producing the most accurate baseball projection system possible, and that includes the depth charts, Player Forecast Manager, percentiles, multi-year projections, projections of rookies and minor league players, etc., that go along with that and are not part of Marcel.
A couple of articles on projection system comparisons:
I emphatically disagree, unless by "crushed" you mean "performed marginally better". The best projection systems, by which I mean CHONE, ZiPS, PECOTA, have historically been in the same neighborhood as Marcel, and outperformance of Marcel by those systems on rate stats (e.g., OPS, ERA) has been small.
You should also note here that Colin has removed park adjustments from PECOTA and thrown out rookies in order to compare to Marcel. He mentioned that, but it bears repeating.
Colin is more of an expert than I am on this, but I can say that Marcel is far from the worst projection system out there. It is consistently among the best. It is one of the simplest, that is true, but simple does not mean bad.
The value of PECOTA over Marcel is presumably in a number of other areas, including but not limited to, forecasts of rookies and minor leaguers, forecasts of fielding, forecasting other categories that Marcel does not, depth charts, PFM. Percentiles, comparable players, breakout/collapse percents, etc., could also be on the list, though with some of those things it needs to be proven that they are accurate. (Colin published some on this in the fall.)
Also, I made the cutoff for "full-time" at 1600 fastballs across two seasons for starting pitchers and 500 fastballs for relievers. That's somewhat arbitrary, of course. If you change the thresholds, the absolute values will move up or down a little, but the relationship between LHP/RHP and relief/starting is fairly consistent.
Counting fastballs instead of pitches also dropped Wakefield from the sample. To me that seemed like a good thing, but you could do it differently if you wanted.
Top 50 RH starting pitchers averaged 92.6 mph. Top 7 LH starting pitchers averaged 92.7 mph and going to Top 8 LH moves it down to 92.5 mph. That's about 15%.
Four-seamers, sinkers, and cutters. Or for you PITCHf/x nuts: FA, FF, FT, SI, FC.
Why not just use four-seamers? (1) A number of pitchers have either sinker or a cutter as their primary or only fastball and would be excluded from the sample. (2) If you include all fastballs, you don't have to worry about the headache of properly classifying fastball types from each other.
Four-seamers and sinkers don't differ that much in speed, anyway. Cutters do, obviously, but they are a smaller piece of the pie. You could make an argument either way for cutters, and I would hear you. I opted for the bigger sample.
Tango, the fastest 23 LH starting pitchers average 91.2 mph, which is the same as the whole group of 95 RH starting pitchers. The fastest 10 LH starting pitchers average 92.2 mph.
I wanna take a nap, too. Do I get a nap?
Btw, it may be a question of semantics, too, about what wrist "break" means. But he almost has to be getting wrist supinating action to put topspin on the ball based on the pictures I've found, and he appears to be doing this in a way that's not completely different from other pitchers.
Hm. He's pretty clearly cocking his wrist for the curveball here:
So I don't know. Maybe he breaks his wrist less than the typical pitcher on a curveball, and it's certainly interesting to hear about the finger flick, but I've got to believe he's coming around the side/top of the ball with a wrist break to get topspin, too.
I abso-freaking-lutely loved this interview!
I'm looking at this picture to replicate his curveball grip:
and trying to figure out how he gets topspin with an index finger flip if he doesn't break his wrist. That's one strong index finger push!
I can't wait to see PITCHf/x data of that.
A quick look at some of the numbers suggests to me that the sample sizes in some of the boxes are very small. For example, .636 is 7/11, .556 is 5/9, .375 is 3/8, etc.
If you choose to use the numbers, you have to use them with that understanding, or acquire a much bigger sample if you want all the numbers in all the boxes to be accurate.
I'm on board with the basic idea (though, I'm not on staff, so implementation is not strictly my purview), but I hate the example.
I personally was very disappointed when BP published Bradbury's work on aging after its deficiencies had been identified and thoroughly discussed by the rest of the sabermetric community. To me that was a low point for BP. I know some people felt otherwise, but for my part, I would much rather see examples of the best work from around the sabermetric community published here rather than the poorest quality work.
Lest it be unclear, I'm not advocating that everything published at BP must have the same mindset. Far from it. I think different viewpoints are great, and ideas that challenge the mainstream thought are valuable. My own ideas about how baseball works and what is important for sabermetrics don't fit very well with a lot of mainstream sabermetric thinking these days. But there is no sense in getting controversial viewpoints while abandoning the principles of sound logic. Or even a step back from that, publishing outside viewpoints that may have some support in the sabermetric community but which the bulk of the leadership at BP believe are fundamentally flawed.
1. Agreed. I'm happy to see studes here and always love reading his stuff.
2. I know Tommy used to do links roundups both back when he was at Beyond the Box Score and since he came here to BP. Is there a desire to see more of that?
3. A lot of the sharing of links and ideas among saber folk from different sites happens on Twitter these days.
4. The different leading sites tend to have different philosophies about things which leads to certain ideas/topics being popular on one site and relatively ignored on another.
Btw, of the pitchers shown on the zoomed chart, I consider that Meek, Soria, Rodriguez, and Dotel throw cutters as their primary pitch (Bailey also, with the caveat listed in the article).
Bell, Betancourt, Clippard, Rauch, Broxton, Axford, and Storen throw normal four-seam fastballs but get minimal arm-side movement because their arm angles are close to being over the top.
Axford is at 94.9 mph and -3.1 inches horizontal spin deflection.
I created a zoomed-in image of the portion of the graph where the right-handed pitchers were too densely packed to place names on the full chart:
Ditto. I always enjoy reading your stuff, Craig.
I hope that conformity of thought is not expected of BP writers. In my opinion, a diversity of thought and an understanding and usage of ideas developed outside of BP is a big advantage, not a detriment.
I personally prefer Kevin's prospect lists, and I'm not wild about any of the ERA estimators, but that's my opinion, and I'm not going to argue too much with someone who chooses differently.
But I do feel strongly that BP should not be a closed shop in terms of ideas and statistics. While I believe that very good work has been and is being done here (or I wouldn't have joined), much good work has also been done outside the walls of BP. It doesn't serve the progress of thought within BP, and thus doesn't serve our readers, to ignore or disdain ideas from other sources.
If you read the initial article (linked in the first sentence of this post), I talked about your question toward the end.
Dan Turkenkopf found an advantage in ball-strike calls to the home team worth 0.12 runs/game.
In addition to being a hard thrower, Jeffress has a very nice curveball. Of course, if he can't throw it for strikes more often than he does right now, hitters can ignore it.
Nice work, Chase. I do believe some, but not nearly all, of Harden's velocity drop in 2010 was due to PITCHf/x calibration issues in Texas.
Lucas Apostoleris had a nice post on Harden at Beyond the Box Score:
Thanks. I appreciate the constructive feedback.
Thanks. Yes, those are the HBPs.
I've added some lines to the batter silhouette and uploaded the image here:
Do people the additional lines on the silhouette helpful or distracting?
What I probably should do is to draw a couple lines inside the silhouette to clarify which arm and leg is which like I did with the legs on the pitcher silhouette here:
I understand that you perceive the pictures as being backwards. That's how they look to you and I can't change that, and I'm not trying to tell you that you don't see what you see.
However, the images are in fact flat. There's no way from the image itself to tell which is the batter's left or right arm. Your brain is adding depth information to the image that isn't there in order to tell you what you think is the right or left arm.
If I simply posted the batter silhouette by itself with no labels, how would you be able to tell whether you were seeing the front or back of the batter? Since there's only an outline, which would be the exact same from the front or the back, there would be no way for you to tell.
As far as which way to present the data, I personally like to see the data from the pitcher perspective, but I've had far more people tell me it makes sense to see the data from the catcher perspective, so that's what I've adopted. I suppose if I get a lot more input from people who feel like you and I do, I would consider changing it back. However, most of the PITCHf/x data in the world gets presented from the catcher perspective, so I'm loathe to buck the trend without a good reason to do so.
I've added some information on Pettitte's pitching repertoire in an Unfiltered blog post here:
I'm no fan of considering FIP the be-all-end-all of pitching talent measurement, but if you look at Pettitte's FIP over the last three years, it's been 3.71, 4.15, 3.85. That's probably a decent first estimate of his pitching talent.
I would consider injury a bigger impediment to him pitching effectively in the next year or two, either by putting him on the DL or into retirement or by causing him to lose speed off his fastball or his ability to command one or more of his pitches effectively.
It's called a sinking fastball, or sinker, because it drops more than the regular four-seam fastball.
Taldan was responding, appropriately, in my view, to a previous version of this article. I would appreciate if people would not rate his comment negatively as a result.
Taldan, I agree with your comment, and hopefully this version of the article expresses that.
Those are just drawings I did of a generic batter silhouette. It's not necessarily supposed to be from the front or the back. It's just there to indicate which side the batter is standing on and give a general impression of the size and stance of a typical batter. I think the mind tends to fill in details which aren't there, like whether you're seeing the front or back of the hitter's head, an arm or leg in front or behind, etc. If I were a better artist with MS Paint perhaps I could draw a better batter silhouette with more detail.
The current version of the article is slightly different than the one that was originally on the site. Sorry for any confusion.
Thanks for Matt Lentzner for communicating an important correction to me about how pitchers are getting topspin on the forkball.
It does not occur due to supination of the arm/wrist, as it does with the curveball.
With a fastball, the fingers start on top of the ball and roll across the back of it on release, creating backspin. The same thing happens on the changeup, and to a lesser degree, with the splitter.
In the case of the forkball, the pitcher gets topspin by applying force behind the ball but contacting it below the center of gravity of the ball. With a wrist flip/snap on release, the ball rolls out forward over the fingers, creating topspin. Of course, you can't get very much spin on the ball this way.
Matt's explanation is more consistent with the observed evidence, and many thanks to him for figuring it out and communicating to me.
I highly recommend The Neyer/James Guide to Pitchers. It's one of my favorite books and continues to be a good reference. It's amazing how much they got right without the benefit of PITCHf/x data and how little they got wrong. If you love baseball history, it's a great read from that perspective, too.
Correct. Because gravity is a constant force with a known magnitude and fixed direction (straight down), all one needs to know is the time the ball was in flight in order to know the effect of gravity to that point.
The PITCHf/x data contains all that one needs in order to calculate the full trajectory of each pitch. It contains the three-dimensional position and velocity of the pitch at 50 feet from home plate and the acceleration in each dimension.
According to the PITCHf/x data we have from 2007, Clemens threw a classic splitter that ran in the upper eighties, about 6 mph slower than his fastball, and with about eight inches of arm-side movement.
The spin deflection is relative to the path the baseball was on when it left the pitcher's hand (and relative to gravity as well, in the vertical dimension only, of course). So yes, for a right-handed pitcher, the ball is moving toward first base in order to cross the plate, and the spin deflection is relative to that motion. The spin deflection shows whether the path is bending back toward third base, i.e., arm-side for a RHP, or bending even more out toward first base, i.e., glove-side for a RHP.
Yes, a screwball should have even more arm side deflection. The only two pitchers in MLB that I know throw the screwball are both lefties, Danny Ray Herrera (often) and Dallas Braden (rarely).
Last I heard, El Duque's still a free agent, ready to bring his eephus back to a major league city.
Tango, you quote me, but I wouldn't say to Christina what I said to the reader in that thread, for at least two reasons.
One, Christina expanded quite a bit on what she meant us not being better off in measuring defense, and I tend to agree with the broad strokes of what she said, if not every detail. The reader in the previous thread may have meant similar things (I don't know), but Christina actually specified quite well what she meant.
Two, my opinion of the quality of UZR and its input data has gone significantly downhill in the last four and half months as we've done more investigation into the data. The flaws have been identified more concretely and I'm left wondering what value, if any, that UZR and similar frameworks bring to the table. Based on the evidence now available, I'm far more critical and skeptical of those systems today than I was in July.
Those pitch classifications from BIS aren't the greatest for Benoit. They're missing his cutter completely in 2007, which makes his overall fastball speed in 2007 look slower than it really was. They have a similar problem with their 2008 data for Benoit, although his pitch types changed a bit from 2007 to 2008.
That is an excellent point about variation based upon intended location. In fact, as I was thinking about this article after publication, I realized that the in-game variation might be very dependent on whether a pitcher threw his fastball mainly to one side of the plate or whether he was confident enough in his command of it to throw it to both sides. A pitcher throwing it to both sides would have to vary his release point left to right to change his aim. In fact, I've looked for and found that effect in the PITCHf/x data before. So it might turn out that a pitcher with good command would have a higher in-game standard deviation for release point. That may be why the in-game numbers showed no correlation with walks. I'll have to think about how to study that further.
As far as the classifying between four-seamers and two-seamers, I tend to use my own pitch classifications for that (though they're all lumped together for this article) rather than the ones that MLBAM produces from the PITCHf/x data.
The reason I focused on fastballs only for this article is that curveballs in particular tend to have a higher release point because the pitcher's hand goes around and under the ball rather than staying behind the ball like it does for a fastball. I wanted to include the largest set of pitches that were basically released in a similar fashion regarding hand position relative to the ball. I could have probably included changeups and splitters, too, on that account, but I did not. Since fastballs make up the bulk of most pitchers' repertoires; however, it shouldn't make a big difference which other pitches I excluded.
There are definitely ways to improve this analysis, and I think your suggestions are on the right track.
Dave, whatever the quality of the work of MGL and Dewan and others and however dedicated and commendable their efforts, it's not science if it's not evaluated objectively in a way that other people can reproduce. That should be the standard in getting new approaches validated, and for whatever reason large portions of the baseball analysis community abandoned that standard in the case of fielding metrics. We have to get back to that standard if we want to make real progress in evaluating fielding.
Thanks, Jeff and JD.
May I suggest singing it to the tune of "Nobody doesn't like Sara Lee" instead?
Thanks for the link, Jay. I was going off the list at Cot's Contracts, but I guess it's not up to date or entirely accurate.
In the unlikely event the Yankees decided to let Jeter walk, wouldn't they sign a free-agent shortstop like Hardy, Peralta, etc., rather than turning to an internal option like Nunez?
I love the interview, as always, David.
I was particularly struck by this response: "You have to analyze hitters’ swings—what they’re trying to do as hitters—and you start to recognize swing paths. You’ll see a hitter and you’ll find out which pitches he struggles with by analyzing how they swing the bat."
I've seen other pitchers mention similar things and would love to know specifically what they are looking for from the hitters in terms of leans, steps, swing paths, etc., and what that means about the pitches they need to throw them.
I just ran across the article by Patrick Newman on Iwakuma and recommend it for anyone who is interested:
RedsManRick, I definitely agree with the gist of what you're saying.
I wonder, though, to what extent the umpires know what areas they struggle with and what causes the struggles. I imagine they know a lot, and I'd love to pick their brains on the subject. Nonetheless, having specific, quantitative records of one's performance can often be more helpful than just a general idea of where one is good or bad. Moreover, we could evaluate across the population of umpires whether certain umpiring techniques are good or bad.
Analyzing and making that sort of information available to the umpires, combined with something like you suggest with an LED indicator for feedback could be a very powerful tool for improvement. I forget the name of the gentleman who advocated for this approach with the Questec data, but his story is covered in Weber's book. Unfortunately, his approach didn't win out at that time.
I frankly don't see any difference between "how much better than human umpires" and "how close to perfection." The answer should be exactly the same either way.
I'm not arguing in favor of human umpires. I'm arguing that everything needs to be compared against the same baseline, which is how they perform in practice. People seem to want to castigate human umps because they're not 100% perfect. The minute a technological solution, e.g., PITCHf/x, is shown to be less than 100% perfect, people are looking for replacements. I'm trying to remind everyone that the question is which reasonably applicable system is best over time, not which one never makes a mistake.
I've done quite a bit of investigation into how well both the umpires and the PITCHf/x operators do at the top and bottom lines. I have more investigation to do before I'm ready to present those results. However, my sense is that the umpires generally do better than the PITCHf/x operators. Umpires definitely call zones that take into account the vertical size of hitter's zone based on his stance. I understand skepticism of my claims on this until I present my evidence.
I don't believe calibration of the PITCHf/x system is an intractable problem. Sportvision has a procedure for doing it that involves something similar to what you suggest--putting markers on the field. For practical reasons, that's not something they currently do before every game. I imagine if their system was being used to replace the umpire, the money and personnel and unfettered access to the field would be available to make that happen on a daily basis. Right now that happens at best once per homestand and usually not that often.
Fantasyking, that's an excellent question, probably worth an article in itself.
The quick answer is that for PITCHf/x, the operator watches the center field TV camera view and sets two lines as the batter settles into his stance. The first line is at the hollow of the back knee, and this is used for the bottom of the zone. The second line is set at the belt, and four inches are added to this to get the top of the zone.
How much judgment/error is involved in the process? A lot. The top/bottom limits produced by this method are not very reliable. There's a better answer to this question that takes more explanation. I'll try to get to it some time.
The other thing to notice is that the top line is not being set according the rulebook. It's set closer to where the umpires actually call the top of the zone, around 3.4 feet for a typical batter. The rulebook top line is more like 3.7-3.8 feet for a typical batter.
We've not yet done anywhere close to all we can do with the data we have for evaluating umpires.
Dan Brooks has great data available with his strike zone map tool at BrooksBaseball.net. Unfortunately, the way I see that data used 99% of the time, though, is simply to count up the ball and strike calls inside and outside of the drawn box.
I suppose that's a mostly legitimate way to evaluate umpires in one sense, but it strikes me as being very divorced from the reality of how the game is played.
Are you thinking about something akin to the garage door sensors, where if the ball broke the beams, it would register a strike?
I can see two big obstacles/challenges with that approach. One would be where to put the emitters and sensors. The other would be how to prevent false strikes from registering due to things other than a pitched ball passing through the detection grid.
Tom, good questions. My response ended up going into a bit more detail than would fit well here, so I added a new blog post for it:
Danny Ray Herrera throws a screwball pretty regularly. Braden throws one on rare occasions.
I mentioned the Neyer/James Guide to Pitchers, but if you have the book, it's worth checking out the sections on the cut fastball and the sailer on pp. 12-13 and p. 18. Neyer argues that the cut fastball has been around for a long time under other names. He has a good account of the possible history of the pitch. However, I found the following quote from Tom Seaver's 1984 book on pitching to be quite poignant:
"You hear a lot of TV talk these days about the 'cut fastball.'"
What's new is old. What's old is new.
Wouldn't any technology you use be subject to problems, just like the PITCHf/x cameras? They do a pretty good job, but there's nothing in this world that's going to be perfect and reliable 100% of the time.
I've notified Sportvision of the problem so they can look into what went wrong. They're the ones with access to the raw data from the cameras that could provide a definitive answer to what happened to the camera. I could speculate what might have happened, but that wouldn't be productive.
The standard deviation on my method for one game is about one inch. So ~68% of the games will be estimated within one inch, and ~95% of the games will be estimated within two inches. It's possible that this game was an outlier at four standard deviations and the umpire was off by five inches on his zone. However, it seems reasonable to me that since the estimate from my method and the zone called by the umpire agree, the PITCHf/x system really was off by 5-6 inches.
I'm always happy with more views. However, if PITCHf/x is well calibrated, the PitchTrax graphics should do a very good job of reproducing the pitch trajectories and locations.
In terms of horizontal offset, the PITCHf/x strike zone bias is within half an inch 50% of the time, within one inch 80% of the time, and within two inches 98% of the time.
So to see it this far out of calibration is really surprising.
Btw, I didn't mean that to be as combative as perhaps it sounds. I appreciated your comment, and it actually makes a ton of sense to me. However, when I looked, expecting to see it, I couldn't find the data to support the impact.
It does make some difference, but perhaps not as much as you think. For one thing, the career win totals at a given age for pitchers today are not much lower than they were in 1996. One win lower, on average. So you could say that quantifies how much difference it makes, I suppose.
Secondly, Glavine had six seasons where he got a couple starts more than pitchers today. That's twelve extra starts. Glavine won about 50% of his games in those seasons, so that's six extra wins. Yes, it makes some difference, but it's not a huge difference.
Moreover, durability is very important in getting to 300, and it's possible, though I don't know for sure, that pitchers now are less likely to have career-derailing injuries. Certainly you can see the impact of injuries when you look at the post-WWII leaderboard.
I just don't see any evidence that career win totals are going down by very much, for whatever reason. It plausible that declining seasonal win totals should affect career win totals, and that's basically what Eric argued in his column that touched on this, but if it's not showing up in career win totals to date, how much difference can it be making?
This comment is a winner. He's definitely one of my favorite pitchers.
That's definitely true, and it's nigh impossible to predict over the time frames needed to get from 100 or 150 wins to 300. Even someone on a long-term contract probably doesn't get half way with one team. There are exceptions of course--Maddux and Glavine made it substantially on the basis of their time with those excellent Braves teams--but it's tough to predict ahead of time (1) which good pitchers will stay with the good teams for most of their career and (2) which teams will be good over the next decade or more.
Ryan started fairly strong but not at a world-beating pace. You can see in the second graph that he was right behind the Seaver-Sutton-Carlton group throughout his twenties.
His peak relative to the Glavine line came in 1977 when he put up a 19-win season for the Angels. He struck out 341 that year with 204 walks (!) and 21 wild pitches. That season put him two wins ahead of the Glavine line. Mostly though, he was between two and 15 wins behind that pace.
His thirties were mostly spent with the Astros, and that put a big damper on his win pace. For one thing, they didn't score a lot of runs for him, and his record was only 106-94 in nine seasons with the Astros despite an ERA+ of 110. However, he also pitched a lot fewer innings with the Astros than he did in his Angels days, an average of 206 innings per year with the Astros versus 273 innings per year with the Angels. He averaged 20 complete games per year for the Angels and only four CG/year for the Astros.
Altogether, he finished his time in Houston with 273 career wins at age 41. Mike Mussina had almost that many wins when he retired at age 39. Of course, we know that Ryan wasn't done at that point. He went on to put up three good seasons for the Rangers plus parts of two more, crossing the 300-win mark at the end of his second season in Texas at age 43.
While Nolan Ryan would always have been in the discussion for 300 wins, I don't know that he would have ever seemed an outstanding candidate based on his win totals or performance at any point before he started getting really close. Of course, his durability ultimately made him a much stronger candidate than his other numbers.
Wouldn't it be TNSTAATHWPP? Because we know TISATAATHWP.
In any case, it's a fun acronym to pronounce.
Fun stuff, Eric, and I like Bill's ideas, too.
One of the particular inconsistencies with a PITCHf/x strike zone at this point is that a human operator is required to set the top and bottom of the zone from video. This human operator can be and has been badly off at times.
It's probably true that the camera calibrations for the left/right edges could also be improved, but that's not a trivial task, and it's much easier to accomplish post hoc, which does no good for calling strikes accurately during the game. If systematic problems could be identified and if these were amenable to correction ahead of time, the accuracy of the systems probably could be improved. Keeping 30 separate systems in good health, though, would be a big challenge.
NYYanks826, I'm curious, what reference was used to identify that their recorded pitch locations were off?
Bill, I've not published my method. However, the basic idea is that I compare a pitcher's pitch locations from one park to another. I do this for all the pitchers in the league for every game, and calculate the average shift.
For a single game, the error in the method is about +/- 0.10 feet. With more games in the sample, the error goes down. In this specific case, it looks like the PITCHf/x calibration at Target Field shifted between the September 21 and 22 games, giving us seven games of data to calculate the current plate location offset for the system. A seven-game sample gives an error of about +/- 0.04 feet.
Actually, I should probably exclude Ichiro's air balls from those numbers, in which case he hit 23 balls in the gaps and 9 balls near the infielders. That comes in at about 2.5 standard deviations from the mean. (The previous one, with air balls included, was actually closer to 2 std dev, I reported that wrong in the comment above.)
I looked at Ryan Howard, too, and excluding his air balls, he hit 13 balls in the gaps and 11 balls near the infielders, not accounting for any abnormal shifting by the infielders beyond what they normally do for LHB.
If I divide the 90 degrees between the foul lines into 45 degrees nearest the infielders and 45 degrees "in the gaps", Ichiro hit 28 balls in the gaps and 14 balls near the infielders.
Whether that is a repeatable skill or not, and if it is, to what extent, I don't know. However it is over three standard deviations from the mean in the binomial distribution, under the assumption that the spray angle is simply random.
Here's a graph showing the HITf/x data for Ichiro that I was talking about.
There are two lines on the graph. The blue one is the horizontal angle at which Suzuki hit his batted balls in April 2009, grouped in bins five degrees wide. The red one is the BABIP for left-handed batters on batted balls with a vertical launch angle of less than 8 degrees (basically ground balls and borderline line drives). The idea is to find the positions of the infielders. Where the BABIP is lowest, that's the likely position of the infielder, and where the BABIP is the highest, that's the gap between the fielders.
You can see that Ichiro did a pretty impressive job of hitting the gaps. I included all of Suzuki's 42 batted balls, but the pattern doesn't change much if his 10 air balls are excluded.
When guys like Keith Woolner, Dan Fox, Russell Carleton, etc., are hired by teams, yes, it's a talent drain for BP, but I'm not sure that's something that BP needs to explain or elaborate on or consider a bad development. Similarly with Nate Silver moving into the political field--I'm not sure there's much that BP could offer Nate to keep him from doing that. So it's hardly fair to write a list of all the great writers of BP past and act like BP hasn't done what it could to keep them.
I don't begrudge anyone the right to say that Joe Sheehan or Will Carroll were the reason they subscribed and that they don't see the value any more. However, Joe and Will obviously are a different case than Woolner et al. I don't see that they fit in a larger or long-standing pattern at BP, unless two makes a pattern (maybe it does) or you're simply looking at the fact that in life everyone comes and goes eventually.
IIRC, Will had access to some actuarial/insurance books from MLB, though I don't remember if he specified the exact source.
The problem with that is that, as far as I know, the accuracy of the projection percentiles was never tested. Tango has, in fact, offered some good reasons to believe they were problematic. I don't know if Tango's right or Nate's right, but without evidence I don't have a lot of confidence in the percentile forecasts.
Presumably this is something that Colin is going to address at some point.
It shouldn't take a few years worth. When we're relying on cameras rather than subjective observers to tell us where the ball went, we can be a lot more confident in the data. But I agree that I'd like to have a sample bigger than 40 batted balls before I can make a statement with confidence about Ichiro's placehitting skills.
The HITf/x data that was made available for April 2009 implied to me that Ichiro did have some ability to hit 'em where they ain't. Other analysts were skeptical that that was anything more than a fluke.
I haven't seen anything in the currently available batted ball location data sets that suggests we have enough precision and accuracy in the data to tell one way or the other.
Steven, thanks for sharing the backstory on that one. Some folks may have gotten too exercised about it, and I was one of them, but at the time it appeared that Baseball Prospectus thought they were better than the rest of us and their excrement didn't stink. Not simply from that line on the book cover but also from other things that were happening. By now, it's clear that BP is serious about addressing and fixing errors and playing on the same field with everyone else, and that's very good. It wasn't so clear back then that it was made in humor and not a claim to be better than the rest of the sabermetric community.
I think it's understandable that it tweaked a few people and good to know that it wasn't meant that way.
Craig, I really like what you did with the superimposed images there.
J.C. Bradbury back in 2005 found that strikeout rate did correlate with lower allowed BABIP:
I'm not aware that he looked for a difference between the impact of swinging strikes and called strikes on BABIP. That would be an interesting research question.
Ricky Zanker looked briefly at the changes in O-Swing% earlier this year:
He found that PITCHf/x did not show an increase in O-Swing% from 2008 to 2010.
Ross Paul tweaks his neural net classification algorithm for MLBAM from year to year and sometimes tunes it for specific pitcher repertoires. Neural nets can be very sensitive to noise on the inputs, so a small change in speed or movement of a pitch can result in different classification by the neural net.
That in itself shouldn't be sensitive to parks, but the classifications are run on un-corrected PITCHf/x data, so if there is a calibration problem with the PITCHf/x cameras in a given park, it can affect the accuracy of the classifications from that park. I don't see that that's what happened with Vazquez, necessarily, but I haven't checked closely.
I talked some about pitch speed calibration errors in the comments to Kevin's article a week or so ago:
I wrote about that topic more thoroughly at THT Live later that day (Sept. 3rd), but I don't know if it's kosher to post a link to that here.
Those numbers line up with the MLBAM classifications of the PITCHf/x data, which you can find at pitchfx.texasleaguers.com.
From my quick look at the data, I think MLBAM is over-eager on classifying the curveball this year, lumping in some sliders with the curves, whereas last year they were lumping in some curves into their slider classification. Whatever the classifications, though, as Craig pointed out, the whiff rate on everything is down from decent rates last year on the offspeed stuff and good rates on the fastball to well below league average this year on offspeed whiffs and merely average on fastball whiffs.
Justin, I'm still trying to wrap my mind around this. When you say "it's definitely performing a heck of a lot better than any of the traditional offense-based position adjustments", what do you mean specifically? How can you tell?
I do it by comparing how pitchers who pitched in those games pitched in other games during the season. It's a little tough to get a read on a single series, but a homestand that lasts longer, say 10 games or so, gives you 70 or 80 data points. That is enough to get a pretty good read.
I don't see how the Royals can be at the bottom of the saber-friendly list unless people are judging by results or by the sarcastic commentary of the bevy of sabermetricians who are also frustrated Royals fans.
The Royals have three sabermetricians on staff, which is a lot more than the majority of clubs, certainly enough that they shouldn't be at the bottom of the list. I suppose it's an open question how much the sabermetricians have meaningful input into the organization's decision-making process, but I don't see how they can be ranked below teams like the Orioles, Phillies, Astros, etc.
The effect of wind on the pitched ball speed seems to be very small compared to other effects we observe.
The drag on the ball is proportional to the air density. Air density is dependent on altitude, temperature, humidity, etc. However, the drag is not going to have much effect at the points near the release of the ball where the speed is being measured. It's going to affect how much speed the ball loses from release to plate crossing.
There is also, however, an additional effect that I have observed from temperature on pitch speeds. Pitchers tend to throw about 1 mph harder for every 40 degrees that the temperature increases. I don't know the cause of this, but possibly it's because it's easier for them to get loose and warmed up.
I have Toronto at +0.5 mph in the last homestand and Chicago Cellular Field at +1.5 mph.
I show Kansas City's PITCHf/x system as recording speeds about +1.6 mph too fast for the year as a whole and +1.9 mph for the most recent homestand.
Cincinnati, on the other hand, is around -0.2 mph for the whole year and very close to correct for the most recent homestand.
So you can take Chapman's numbers at face value, but subtract a couple mph from Feliz in KC.
As far as I know, Josh Kalk and I are the only ones who run a comprehensive set of adjustments to the PITCHf/x data. Josh's haven't been public since he joined the Rays, and I don't publish mine. Someone else may be doing it, too, but not that I know of. None of the major public sources for PITCHf/x data (TexasLeaguers, Fangraphs, BrooksBaseball, etc.) make any adjustments to the data.
Chapman's velocities in Cincinnati are being recorded by PITCHf/x, which uses cameras, not radar, for its measurements. Cincinnati's PITCHf/x system has been pretty accurate for speed measurements this year, and this week's games don't appear to be an exception.
Kansas City's PITCHf/x system, on the other hand, has been recording speeds that are around 1 mph too fast this year, as compared to other parks. I'll have to check Wednesday's game to see if it was even faster than that, but I've noticed that people tend to see the pitchers and games where velocities are 2-4 mph faster and ignore the pitchers/games where velocities are the same or slower, even if they altogether average out to around +1 mph.
Jeremy Greenhouse did a study on fastball speed aging using the BIS velocity data:
Josh Kalk did a study using PITCHf/x data, but a much smaller data set:
SIERA may not apply very well to Rivera, but more advanced stats do tell us why he is able to limit hits so well. He has incredible command of his pitches. Reference Dave Allen's article:
Colin, I don't think I understand what you did in the part where you "regressed" offense and defense to make the error symmetrical.
Where did the 9.8 and 5.9 come from?
Is the +21.2 estimate for offense of the first player saying that he's equally likely to have been a +10 hitter as +32.4 hitter? That doesn't seem right to me when our best guess is that he was a +30 hitter. Is the error distribution really skewed that much? Or am I misunderstanding the meaning of what you are doing there?
Thanks for the explanation.
Hi, Ken. I enjoy your writing and agree with a number of your listed beliefs. However, on #32, hitter-vs-pitcher matchups, have you read what Tango, et al, published on that topic in the Book? They seem to have studied it pretty thoroughly, including addressing your point about outliers. I'm curious if you hold your belief in spite of what they found, and if so, what your thought process is.
It's one standard deviation, which is about 68% of potential outcomes, right?
Dave, you might enjoy the thread that Tango started in response to Colin's article at the Book blog where similar questions are being discussed:
Colin, as I mentioned on Twitter, can you use these numbers to estimate the magnitude of range bias for various advanced fielding systems (and at various positions)? Over a large sample of players, the park-scorer bias should become much less important.
If the ~70 run difference for Ozzie Smith is due to range bias, and 1 play = 0.8 runs, and Ozzie played about the equivalent of 17 seasons, then 70 / 0.8 / 17 = about 5 runs per season due to range bias.
If we apply the same method to a large group of players, we ought to be able to estimate the range bias.
Then, I wonder--thinking out loud here--if the range bias for large samples of players is known, could you then turn around and estimate the park bias?
Excellent work, Colin!
I'm trying to digest this and see if there is any way to squeeze any more adjustments into it without adding too much bias.
Particularly, I'm wondering if there's a way to adjust for pitcher tendencies without looking to the results of the plays made. The only ones I can think of where we have historical data are pitcher handedness and the groundball-airball split (as you defined it) by pitcher.
As Colin mentioned, the margin of error is proportional the square root of the number of balls in play. The margin of error as a percentage, then, is proportional to the inverse of the square root of the number of balls in play.
So if you multiply your balls in play by three (i.e., three seasons instead of one), your percentage error will drop by 1.7x. (The square root of 3 is about 1.7.) If you multiply balls in play by 10, your percentage error will drop by about 3x.
Small correction: the graph is for the deflection by spin and drag. He's subtracted out the effect of gravity.
What you see in his graph is mostly spin-related since the drag force mainly operates in the direction into the screen (i.e., toward the pitcher's mound). The drag deflection is only an inch or two in that graph.
The effect of gravity, however, is huge. It looks like about a two-foot drop for Wright's slider. You can see a similar chart for Wright, but including all three forces on the baseball here:
That graph is for spin-induced movement only. It's fairly typical for a slider to have only a few inches of spin-induced deflection, or even zero. Of course, gravity still acts on the slider, which it makes it drop in an accelerating fashion by a foot or more toward the end of its flight.
Tango does much of this with the Fans Scouting Report that he gathers every year.
Most of the advanced fielding systems also separate out an arm rating for outfielders.
Unless FIELDf/x or similar data is ever released, it will be tough (though not necessarily impossible) to separate positioning, range, and hands in what we have from the batted ball data.
Colin said he wasn't sure and wasn't sure how to tell. You seemed to wipe away the uncertainty and conclude that we have made no progress.
There is reason to believe we have made progress, certainly on the theory side. But until we can test, we won't know for sure. That's historically been the standard in sabermetrics.
However, people who have developed fielding metrics will take your criticism very differently if you say, "I don't see how to tell how accurate your metric is" versus "Your metric is worthless." The former is a statement of fact that can be contested and explained, though it may raise some emotions. The second is a very value judgment that comes across as very dismissive and not focused on the examination of facts.
I don't think it's fair to say we haven't made much progress in 20 years with fielding metrics. We would seem to have made substantial progress. The methodology outlined by MGL in UZR, for instance, which built on previous work by others, is very thoughtful theory that's far beyond the days of simple range factor.
What we don't have is any quantification of which pieces of that work are most helpful and which don't add anything or perhaps are troublesome in their application.
Our fielding data collection is also much more precise than it was 20 years, but we have no idea when it's better, and how much better it is in those cases, or when it's worse.
If we can sort that out, I have hope that we will find that much of the work that's been done by MGL, Dial, Dewan, Tango, Sean Smith, and others can be made more useful, not less.
Nice work, Ken. I enjoyed the article. I always like when someone gets his hands dirty by watching game video and shares the results.
Nice work, Ben. Very interesting.
When Kevin Jepsen debuted his slider (some call it a cutter) on July 3rd, it made a world of difference in his performance. It's allowed him to cut back drastically on the use of his curveball, which wasn't a very good pitch. The slider has become his strikeout pitch against right-handed batters, something he didn't have before.
His numbers since adding the slider: 38.1 innings, 30 hits, 11 walks, 36 strikeouts, 1 HR, and a 1.88 ERA.
Prior to adding the slider, he was getting 10% swinging strikes on the fastball to RHB and 7% swinging strikes on the curveball to RHB. Post-slider-addition, he's getting 8% swinging strikes on the fastball and 26% on the slider.
Eric, I wonder if something went wrong in your calculations. By looking at still images, I find that most pitchers, Chris Young included, release the ball roughly in line with their plant foot. A pitcher's stride length is typically about his height or a little less.
In the case of Chris Young, his stride is possibly a little longer than his height, but not much. Take 60.5 feet and subtract 7 feet, and you get a release point around y=53 feet for Chris Young and more like y=55 feet for the typical pitcher. I don't think it's physically possible for Young to be releasing the ball inside of y=50 feet.
Why not? What is the best?
Everything I\'ve read indicates that PECOTA is right up there in the group with the best. Some fare a little better by one measure and some a little better by another, but all the systems that use at least three years of historical data and some regression to the mean end up in basically the same accuracy basket.
Yes, Posada should definitely be in there. CHONE says .266/.363/.434 for Posada and Oliver says .288/.380/.480.
I\'m not sure I buy the assertion that PECOTA is the most accurate projection system in this business. If you\'d said that it\'s as accurate as any of the others, then sure.
Sean Smith\'s CHONE, which has been shown to be just as good at projecting hitters as PECOTA, has Wieters projected at .274/.352/.439, compared to Mauer at .314/.410/.452, McCann at .297/.368/.503, and Sota at .279/.363/.474.
The new promising kid on the block in the projections world, Brian Cartwright\'s Oliver, has Wieters projected at .294/.373/.487, Mauer at .308/.377/.435, McCann at .285/.339/.491, and Soto at .270/.338/.443.
So Wieters is clearly considered a bright star in the sabermetric world, but CHONE is not as high on him as PECOTA and Oliver.
David, I always love coming to BP and seeing a new interview from you. You are great at what you do; keep up the good work!
Will, Dan Brooks has a great tool at his website for answering questions about PITCHf/x data like yours about Pedro Martinez.
You can look at the data for his last start here:
You can see, particularly if you look at the \"Horizontal Movement vs. Speed\" and \"All Game Pitch Plot\" graphs, that Pedro\'s changeup was running in the 72-77 mph range. His fastball started at 80-84 mph, but he cranked it up to 88-90 mph by the 4th inning.
You can see a similar pattern is his previous start from August 31st:
In his August 26 start, he was mostly 86-92 mph with his fastball: