Pitchers
FIP: pitcher descriptive metric; describes (part of) pitcher's season
ERA: pitcher+fielding+timing descriptive metric; describes (all of) pitcher's season, which is obfuscated by also describing team defense + random variation
Of the two, FIP better describes a pitcher's season
This shows it league-wide, for the last (up to) 10,000 pitches of each pitch type. Use this as a reference.
You can see how this could form the basis for an objective standard. The main reason it looks more splattered than we'd like is because if a pitcher says his pitch is a slider, but splatters more like a cutter, we still call it a slider. The good news is that at least we made headways in getting pitchers to split their sliders into gyro-slider and sweeper-slider. One day, the objective standards will eventually take hold, but that day is not yet upon us.
Click to embiggen
From 2016 to 2023, the pitcher among the lowest BABIP in MLB is Justin Verlander. With 3196 balls in play, his hits allowed rate is 142 fewer hits than league average. That is a substantial number, by far the highest number in that time frame. In second place is Kershaw, at 95 hits better than average. In third place is Scherzer, at 83 better than average.
This seems like a perfect refutation of Voros and DIPS, which we discussed in Part 1. If I asked you to name the three best pitchers since 2016, Verlander, Kershaw, and Scherzer could very well make up that top 3. So that potentially the three best pitchers in MLB also happens to have the best hits on balls in play is not noteworthy in the least. The next names on the list however are Julio Teheran, Cristian Javier, John Means, Tony Gonsolin, Yusmeiro Petit, and on and on it goes. deGrom is 114th out of 654 pitchers. Gerrit Cole is 89th. Aaron Nola is 466th. Wheeler is 296th. The second WORST pitcher on hits on batted balls also happens to be the 8th best in FIP-based WAR: Kevin Gausman. These 8 pitchers by the way lead in WAR on Fangraphs, a metric that ignores balls in play. Looked at it holistically, this better describes the original issue Voros found: how much attribution can we possibly give the pitcher on balls in play?
Part of the problem we have is how I even introduced it in the first paragraph. I said Verlander is among league-lows in BABIP. But more accurately, we should say that Verlander AND HIS FIELDERS are among the league-leaders. We can't just bypass his fielders. And we still have the issue that so much of what happens on batted balls goes beyond the pitcher and his fielders and their park. Random Variation weighs heavily, in ways that you don't see in other stats.
Ok, let's get into it. We can use Statcast data to directly determine the contributions of the pitcher. We can look at their launch angle and speed to determine how well effective they are. When we do that, we can see that Verlander gives up alot of soft batted balls, to the point that he ALSO happens to be the best pitcher in baseball since 2016 on launch-based hits on balls in play. Well, take that Voros! Except, well, the magnitude is not there. When we look at Verlander and his fielders, their BABIP suggests 142 fewer hits than league average. But when we look at Verlander and his allowed launch angle and speed, that suggests 76 fewer hits than league average. The breakdown for Verlander looks like this:
Here is Kevin Gausman:
Gausman is interesting in that we can explain the ENTIRETY of the poor BABIP with himself and his fielders. He's had the bad luck of having poor fielding behind him, 15 hits worse than average. But the rest of the outcomes is because of Gausman himself. Relying on FIP-only for Gausman would not be a good idea.
We can do this for every pitcher. I will show you a chart (click to embiggen), that shows, on the x-axis, how well each pitcher, and their team, do, compared to league average. You can see Verlander on the far-right and Gausman on the far-left.
On the y-axis is the direct contribution of each pitcher. While in some cases, the two correspond, like with Verlander and Scherzer and Kershaw and Gausman. In other cases, there is little overlap. Take for example Adam Wainwright:
This is a mess to resolve. Wainwright has been hit very very hard. Indeed, he's been league-worst, using launch angle and speed, at 75 more hits allowed than league average that we can directly attribute to his launch characteristics. He's had the good fortune of playing with good fielders. When he was on the mound, they made 10 extra outs than the average fielder. There was another 26 extra outs that we can't attribute to the pitcher or his fielders. Whether this is Random Variation, or it's the fielding alignment mandated by his coaches, or Wainwright somehow managing to gets more balls hit closer to his fielders, we can't really tell. All in all, Wainwright's team, with Wainwright on the mound, only gave up 39 more hits than league average.
Sometimes you get into issues like Zach Eflin, who is better than league average on how hard he is hit, and yet is worse than league average when he is on the mound with hits on balls in play. Do we really want to attribute to Eflin things that he has no control over, simply because he happens to be on the mound when those bad things happen? Why not attribute some of that to his fielders, who are equally not-complicit, but are equally present? Or maybe, stop attributing things that we don't know who to attribute to, simply because we've identified they are present? Sins of the Father and all that.
This is what it looks like if you compare the direct contribution of the pitcher, using their launch angle and speed, to the "everything else" I've been talking about. As you can see, virtually no correlation. In other words, after having identified the contribution of the pitcher directly by how hard they are being hit, whatever is left over has no association to that. Whatever is left, which is going to be mostly Random Variation, has likely very little to do with that pitcher.
When you look closely at first chart, we can come up with the general point: about half of the results we can attribute to the pitcher. In some cases more, in some cases less. In some cases, there's a reverse effect (like Eflin). But, if we simply use as our starting point that we'll count half of the outcome and give it to the pitcher, then we've taken a big step forward in better attributing outcomes to the underlying contribution.
Should we completely ignore hits on balls in play? No. The pitcher is not a pitching machine. There is some influence there.
Should we completely accepts hits on balls in play? Also no. The pitcher is not in total control here. There's alot happening that has no bearing on the pitcher.
Should we split the difference, give them half, and move on? For the pre-Statcast years: yes. Without any additional information as to their direct impact, then we have to infer their impact. And it's about half of what you see. Basically, BABIP is somewhere between Pitching Machine 4587 and that pitcher. And that's how much attribution we should give the pitcher. In Statcast years, we have more information, and so we can better attribute the impact of the pitcher to the outcome when they are present.
We can of course be a little fancier, and figure out fielder influence as well, but that's a story for another thread.
Twenty years ago, Voros shook the saber community with one of the most important saber discovery to that point, and still a top ten discovery of all saber-time. He called it DIPS, or Defense-Independent Pitching Statistics. My tiny contribution to that was FIP, which is merely a shortcut to the full-fledged DIPS. Had I not invented FIP, Voros would have eventually created it anyway.
The illustrations that Voros provided was extremely compelling. In 1999 and 2000, Pedro Martinez had perhaps the greatest stretch of two pitching seasons ever, in the history of baseball. It's difficult to even decide which of the two seasons was the better one. His ERAs were 2.07 and 1.74, and this is in the middle of the high scoring era. He had 313 strikeouts in one of the seasons and 284 in the other. And this is while pitching only 213 and 217 innings each season. In the season where he gave up 32 more hits, he also gave up 8 fewer HR. All in all, it's hard to decide which of the two seasons were better, and in any case, the two stood together as perhaps the best pitching seasons back to back.
What did Voros point out? If you remove the strikeouts and homeruns, and compared the non-HR hits to all remaining batted balls, what he called BABIP (batting average on balls in play), Pedro had among the league-low of .236 one season and among the league-high of .323 in the other season. This seemed ridiculous on its face. How could perhaps the greatest pitcher ever, having one of his two best pitching seasons ever, allowed hits on balls-in-play at a close to league-high rate? And how did he pair that up with a league-low rate in the other season?
This would suggest that allowing non-HR hits on balls-in-play might be pretty random. After all, Pedro would not pair a league-leading strikeout one season with a league-low strikeout another season and STILL be one of the best pitchers ever. You couldn't do that with walks either, or homeruns. It just doesn't work like that. But, non-HR hits on balls-in-play? Well, it happened. And it wasn't just Pedro either. While pitchers had a fairly stable SO, BB, HR year to year, their BABIP fluctuated greatly.
In retrospect, we should have known. Because Random Variation would have told us. But, no one ever looked, not until Voros. The key point of his discovery is that Voros created the denominator: balls in play. That was the key. Once that was done, then you could apply basic statistical principles to determine how much Random Variation could have impacted BABIP. Assuming 500 balls in play, then one standard deviation was roughly 0.46 divided by root-500 or 20 points. Two standard deviations is 40 points. So, going from 2 standard deviations worse than average to 2 standard deviations better than average is not that noteworthy from a performance standpoint. Look hard enough, and someone will do that year after year. In 1999-2000, that just happened to be Pedro. Even Pedro was subject to Random Variation.
Still, what do you do with this information, that Pedro had a .323 and .236 BABIP in back to back seasons? This is where you get into ATTRIBUTION and IDENTIFICATION. Suppose that pitching was done via pitching machines. And through Random Variation, you will end up with some games with 3 hits and other games with 13 hits. Nothing changes. It's the same machine, the same opposing batters, the same fielding alignment. Nothing changes. Except, because of Random Variation, you will get a random result of hits. We've identified the entity on the mound (Pitching Machine 4587). But do we attribute the results to that machine? Or, is the machine simply inconsequential?
Now, humans are different: they are humans. And when it comes to human behaviour and human talent, they can influence results. Now, just because they can influence SOME of the results, doesn't mean they can influence ALL the results. We can identify who the pitcher is on the mound, but do we attribute everything that happens to the pitcher? After all, we have human fielders involved, and we have the vagaries of the park and weather that day. The batters change, and heck, every ball is like a snowflake: no two balls are alike.
Just because we've identified Pedro, and we've calculated a BABIP of .323 one season and .236 another season doesn't mean we attribute all of that to Pedro. There's other entities involved here. Pedro cannot possibly absorb all those outcomes, given that he's one influence.
At the time twenty years ago, I was involved in a discussion and research called Solving DIPS, which basically determined, through basic statistical principles, that Random Variation was the large agent, while the pitcher and fielders were also significant agents, as was the park.
Next up: we'll set aside all that theory and look at things more factually.
Another excellent article from Josh, incorporating BABIP through the lens of xwOBAbip to give us xwFIP. In other words, he keeps the actual BB, SO, HBP, HR while replacing the rest of the BIP by using the xwOBA data. And it gives us a step forward.
FIP of course remains ubiquitous and offers the ideal Naive method for pitcher evaluation. The amount of effort to calculate FIP is almost non-existent, while also being completely transparent. And as we know, FIP represents a pitcher's past performance better than ERA does. Using FIP unaltered for future prediction is a happy byproduct of its construction, but not its raison d'etre. FIP was, is, and always will be an evaluation of a pitcher's past performance.
Blake Snell had a .266 wOBA. But because his BABIP was so incredibly low (.256), the rest of that wOBA, viewed through FIP gave us a good, but not great, FIP of 3.44. FIP is agnostic as to that BABIP. Hold that thought.
Spencer Strider had a .278 wOBA. But because his BABIP was so incredibly high (.316), the rest of that wOBA, viewed through FIP gave us a great, even outstanding, FIP of 2.85. FIP is agnostic as to that BABIP. Hold that thought.
So, was Snell a little bit better overall because of his wOBA, or was Strider far better because of his FIP? Ok, let's bring that thought about BABIP to the forefront.
Since each pitcher allowed about 400 batted balls in play (BIP, excludes HR), then that difference of .316 and .256, or .060 per BIP comes out to a total of 24 hits (.060 x 400 = 24). FIP says: I don't care. And it can say that in the same way that OBP says: I don't care if someone has 120 walks and 10 HR and someone else has 80 walks and 50 HR. OBP is not the last word in overall production. This is why we have wOBA. It can balance that.
Similarly, FIP is not the last word in overall production. Pitchers of course have fielders to help, or hurt them. Fortunately, we track that, at the pitcher level, right here.
Spencer Strider's fielders were actually league average. So, that .316 BABIP of his? That's all his doing. He deserves that. And so, that .278 wOBA? Yes, that's his. He deserves that. As wonderful as his FIP was, well, FIP is not the last word. And given evidence of average fielders, then wOBA is the last word for Strider.
How about Snell? Well, his fielders were definitely above average, at +8 when he was on the mound. Indeed, most of the Padres pitchers benefitted from stellar fielding. Still, that's +8 above average, and since Strider had league average support, that difference between the two is 8 hits. So, that's 8 hits that we can bump up his wOBA by. With 742 PA, we can bump up Snell's wOBA by 8/742 or .011. And since he had an unadjusted wOBA of .266, we can have a fielding-adjusted wOBA for Snell of .277.
Well, well, well. Snell and Strider are, overall, nearly identical.
Indeed, once you park adjust (Petco is pitcher-friendly), Strider pulls slightly ahead. But, let's not talk about park factors, as you might be able to make the argument that Petco affects Snell differently, and maybe he doesn't get all the benefit. Let's call them even. Overall.
Now, Snell is famous for giving up alot of walks. Alot. Like ALOT. But not as much with runners on, and so, he can make walks less damaging. And especially that he gives up very few HR with runners on. Just 4.
Strider on the other hand gave up 10 HR with runners on. And he gives up alot more walks with runners on, making them more costly. And a huge share of his strikeouts are with no runners on. And with no runners on, a strikeout is identical to any other kind of out. Strikeouts get their extra value with a runner on 3B and fewer than two outs.
Here is how their wOBA splits look like, with bases empty (.282 v .263) and with runners on (.245 v .301). So, one pitcher drops his wOBA by 37 points when they can be more damaging, while the other increases by 38 points when they can be more damaging. I don't even have to tell you which one is Snell and which one is Strider.
Pitchers of course change their approach based on the ball-strike count, and naturally with runner-out scenarios. Batters too for that matter. Heck, even the fans have a different reaction in watching the game based on the changing conditions. But how much of that context matters? Does it matter just a little, or totally?
At a high level view, after adjusting for fielding support and park, Snell and Strider were pretty much equals. At a field-level view, Snell clearly saved more runs than Strider.
In your view: in 2023, were they in fact equals, because you see things from a high-level view? Or was Snell noticeably better because you see things from a field-level view?
One of the things that we don't measure is the finger pressure on a pitch. Alex Fast is looking to change that, and this recent research looks fascinating.
Start with the first chart. The difference between the first line and the second line is simply the effect of the hardware. So, we don't learn anything there, other than the impact of the hardware. Which as we can see, is pretty substantial: 7mph loss of speed and 8% lower spin rate. In any case, our baseline is the 2nd line, and the third line is what happens when we increase finger pressure: pitch speed goes up by 4mph and spin rate goes up by 7%. The amount of movement goes up by 14%.
Is it possible that just applying more finger pressure than usual can do that? Well, I think it's very possible and could very well explain why pitchers as relievers will throw harder than those same pitchers do as starters. They do in fact throw harder, by about 3 to 4 mph, and would explain the much better performance by those pitchers as relievers, compared to themselves as starters (aka Rule of 17). I've never looked at it beyond that (I don't know why, or maybe I have, and have forgotten), but higher spin rate and more movement would seem to be likely results as well.
The study does seem to be focused on one part of the finger to measure pressure. Eventually, we'd likely measure every part of every finger and thumb at every point in the pitch release. Exciting days and years ahead for biomechanical researchers and saberists alike.
On Savant, you can currently see the release height on a pitch by pitch basis. Here we can see it at the game level. Detmers in 2021 and in the first half of 2022 was releasing his pitches at a similar height. You can't tell by the current Savant chart, but that height for Detmers corresponds to about 42 degrees of arm angle (where 0 is a sidearmer, and 90 is over the top). In the second half, you can see his release height dropping by 3 or 4 inches. That drop is actually a drop all the way down to 30 degrees, which you can see in the inset here (click to embiggen).
Did it help? Without necessarily applying a correlation = causation, his FIP in the 2021 season was 6.36, and he followed that in the first half of 2022 with 4.81. With the drop in his release angle, his FIP was 2.57 in the second half of 2022.
The Steamer forecasts operates on identified data. And this is either an unidentified piece of data, or inconsequential, as their forecast for Detmers in 2023 is to match his career-to-date totals. It will be interesting to see how he does in 2023.
Just taking a semi-random pitcher, here's how each of Manoah's pitches looks based on his release arm angle, by pitch type. You can see that he's somewhat consistent with his pitches, except his slider. His slider is released with an arm angle of around 36-37 degrees, while his 4-seamer is at 39 degrees. Does this matter? Is consistency important? I don't know yet. We'll take a look soon enough to see how Manoah (and all pitchers) perform game by game, based on the release angle of their pitches, and we can figure out how important, or not, a consistent release angle is, across pitch types.
The image on the left is his slider released at 31 degrees. The image on the right is his 4-seamer released at 40 degrees. Both pitches were thrown in the same game. You can see that the arm angle on his slider is lower. Is this "noticeably" lower? Does it matter? We'll look into it soon.
Batters
From 2015-2019, the spread in BACON (batting average on contact) is one standard deviation of 40 points. This is for all batters with at least 100 balls hit into play (HIP). The average number of HIP was 294, and therefore we can establish that Random Variation would account for a spread of 28 points.
If we observe a 40 point spread, and we know that Random Variation accounts for a 28 point spread, then the remaining difference is 29 point spread. (That’s 29^2 = 40^2 - 28^2.) Of that 29 point spread, 13 points can be accounted for by the park. And therefore, the difference of a 29 point spread and a 13 point spread is a 26 point spread.
In other words, the TRUE TALENT spread in BACON for batters is one standard deviation = 0.026.
Thanks to a suggestion by Straight Arrow Reader GuyM, I repeated the above for xBACON. In other words, rather than taking the observed batting average on contact, I instead rely on the Statcast speed+angle quality of contact equivalency. The observed spread was much lower, at 31 points.
Random Variation accounts for 17 points.
A little sidenote. For binary results of BACON, one standard deviation for one batted ball is root of .343*(1-.343), whereby the .343 represents the average BACON. And so, one standard deviation is 0.47. However, for xBACON, it is not a binary outcome. Taking the standard deviation of every xBACON value, we get 0.29. (Which also happens to be the standard deviation of a uniform distribution.)
Park variation on xBACON accounts for 6 points.
So, with a quality of contact 31 point spread, a random 17 point spread, a park 6 point spread, that leaves us with a true 26 point spread. Which is the same number that we got from the outcome method.
Pitchers and Fielders
I repeated all this for pitchers, but this time I removed all HR. So rather than all contacted balls, it is only balls in the field of play (BIP). The true spread based on the observed outcome is 17 points, while the true spread based on the quality of contact is 13 points.
Now, why would these be so different, while the batter ones were identical? Fielders(*). While we accounted for the parks, we did not account for the fielders. And how can we account for the fielders? Well, the quality of contact establishes the spread for the pitchers. That’s 13 points. And therefore, the missing variable, what gets us from a spread of 13 points (using quality of contact) to a spread of 17 points (using observed outcomes) is the fielders. And since 11^2 = 17^2 - 13^2, then we can say that the spread in fielding is one standard deviation = 0.011.
(*) I may say Fielders here, but I actually mean Fielding. That's because the way I handle Fielding, it's a combination of Fielder skill and Fielding Alignment (which I treat as a team influence).
In other words we have this:
In other words, we can establish that each of the pitcher, fielding, and park are roughly similar in impact to the outcome of a ball in play. But more than all of them combined is good ole Random Variation.
And so, if you tried to partition responsibility based on a “left over” approach, whereby you account for two of Pitching, Fielding, Parks, and assign the remaining result, that last variable will absorb the entirety of the Random Variation. Which is almost certainly what you don’t want to do.
My preference is to just leave Random Variation unassigned. But if you must assign it to something, you may as well do it proportionally to the other three variables (Pitching, Fielding, Park).
?While I'll be using Gerrit Cole as my illustration, this isn't about Gerrit Cole. It's about any great pitcher under 30. And by great, I don't mean great, but I mean GREAT.
First I'll ask an easy question: how old is Gerrit Cole? He was born Sept 8, 1990. That makes him 29. Unless I asked you on Sept 7, in which case he's 28. Welcome to the world of truncation and math. Age is the only calculation that we make that we've all agreed to truncate. And therein lies the problem. If we ask the age question on July 1, 2019, we're going to get a different answer for players born on June 30, 1990 and July 2, 1990, even though they are only 2 days apart: we will get an answer that is one year apart. Analytically, this makes no sense. Mathematically, this makes no sense.
On July 1, 2019, the mid-point of the season, Gerrit Cole was 28.81 years old. If we are looking for pitchers at a comparable age, you can look for pitchers in the season in question who were 28.31 to 29.31 years old, or 28.81 +/- 0.5. But, we don't have to limit it that way. We can also look for comparable pitchers that are +/- 1 year from his age, or pitchers who on July 1 were 27.81 to 29.81 years old.
You can actually go even wider, but then you get into two issues. The first is that we want to make sure that Cole is at the center of the pitchers in age. You don't want to go so wide that the comparable pitchers end up being say 6 months younger on average than Cole, or 4 months older. You really want him to be right in the middle. The second issue is linearity: you want to constrain it to a point such that the pitchers older and younger than him cancel out. The wider you go, the more likely you end with some pitchers on the upward slope of the curve and other pitchers on the downward slope of the curve.
Pedro Martinez was born on October 25, 1971. In the year 2000, on July 1, he was 28.68 years old, or 0.13 years younger than Cole. In 2001, he was 29.68 years old, or 0.87 years older than Cole. If we use BOTH Pedro seasons, then two-year Pedro is 0.37 years older than Cole. Both Pedro seasons are part of the GREAT pool of pitchers at Cole's age.
I created a quick metric, weighted WAR or wWAR, that is 60% WAR (as shown on Baseball Reference) in the year in question (year T), 30% year T-1 and 10% year T-2. I selected all pitchers born since 1922, with at least 5 wWAR. And who were within 1 year of Cole's age in 2019. I also limited to pitchers whose year in question was 2010 or earlier (so that I get a full 9 years).
The most recent pitchers to qualify, if we only focus on the elite of elite, at least 7 wWAR: Tim Hudson in 2003, then two Pedros, two Madduxes, a Rijo, two Clemenses, a Viola, a Stieb, two Guidrys, a Reuschel... I guess I should stop, but I'll keep going... a Catfish, two Seavers, two Fergies, Wilbur Wood, a Marichal, two Koufaxes, a Robin Roberts, and a pear tree. Those 23 seasons, each of which was at least 7 wWAR, averaged 8.1 wWAR. This is a ridiculously high performance level.
In their first year, they averaged 5.7 WAR. Wilbur Wood and Koufax were above 10, while Dave Stieb was under 0. Such is the life of a pitcher. Another way to say this: when you observe 8.1 WAR, it's actually being generated by a 5.7 WAR pitcher, who happened to be healthy and get alot more good luck than bad luck.
In their 2nd year, they averaged 5.0 WAR, losing Koufax in the process, with a sub-zero from Catfish, his last season. But they ALSO have the other Koufax. Remember, we've got two Koufaxes, spaced one year apart. So Koufax is both the best and worst performing pitcher in year 2.
And on and on we go. In year 3, it's a 4.5 WAR. Year 4, it's 3.8. Year 5, 3.1. And years 6 through 9: 2.5, 2.7, 1.9, 1.1.
If we add up all nine years, that's 30.4 WAR. In other words, we'd expect pitchers who have a wWAR of 8.1 over their last 3 years to generate 30.4 WAR in their next 9 years, or a multiple of 3.7.
What if we expanded our pool and looked at all pitchers with at least 6 wWAR? That's 46 seasons, with an average wWAR of 7.3, and a next 9 years of WAR of 24.4. That's a multiple of 3.3.
And if we look at all pitchers with at least 5 wWAR? The most recent pitchers include two seasons of each Wainwright, Haren, CC, Johan, and Webb. You can see the future right? That's an average of 6.3, which puts Cole close to the center. And the next 9 WAR averages 22.0. That's a multiple of 3.5.
This is how it looks: you are basically expecting some 20 to 25 wins, with a spread of 0 to 50 wins, and an outside shot of being Bob Gibson, Roger Clemens, Greg Maddux or Gaylord Perry.
?
And here's the multiple table for any number of future years for any age, smoothed out. (Note this table is based on pitchers with at least 4 wWAR. So, you should be careful in extrapolating beyond that.)
In other words, do we need to worry about DIPS (or FIP)? No.
Justin Verlander has allowed 487 batted balls, and gotten 349 outs. If we focus only on the quality of contact (launch angle+speed), we'd have expected he gets 345 outs. So, he got only 4 more outs than expected outside of his influence. If you wanted to stop reading here, you'd be fine: Verlander's results are consistent with his individual contributions.
***
What did we not control for? The fielding talent of his fielders, the team alignment of his fielders, and the spray direction of his batted balls. Naturally, all 3 of them are interlinked.
We can focus on the fielding talent of his fielders first. When he was on the mound, his fielders were a little bit above average. How much above average? +4 outs. That is, based on how much distance they had to cover, and how much time they had to get there, the Astros fielders got 4 more outs than average.
In other words, we can explain how he got his 4 extra outs. And therefore, we give Verlander credit for getting 345 outs on 487 batted balls. The league average is 65.5%, and so on 487 batted balls, a league average pitcher would have gotten 319 outs. Since he actually got 345 (after accounting for the talent of his fielders), Verlander is +26 outs.
Now, what about the fielding alignment and spray direction? So, this is an interesting question. The Astros rarely shift on RHH with Verlander pitching, while they always shift on LHH with Verlander pitching. Since Verlander is obviously well aware of the fielding alignment behind him, he is pitching to that alignment. If he can get the hitters to hit to where the fielders are, I'd contend this tells us more about Verlander than the fielders.
Now watch this. With RHH, Verlander got 181 outs, while getting almost 4 outs of support from his fielder's fielding talent. So, that's 177 outs otherwise. And based on quality of contact (speed+angle only), we expected 177 outs.
With LHH, Verlander got 168 outs, with no extra fielding support. Based on quality of contact, we expected 168 outs.
In other words, whether massively shifting all the time, or never shifting, the number of outs that Verlander got is entirely determined by the quality of contact. That is, we can safely ignore the fielding alignment, if we can also ignore the spray direction.
Verlander has a .246 wOBA and a .247 xwOBA. That he happens to have an historically low .218 BABIP is inconsequential.
Bill James invented DER, or Defensive Efficiency Record. It is put simply: outs per ball in play. You are probably more familiar with BABIP which is hits per ball in play. You would think that BABIP + DER = 1. But, there's those pesky errors. Anyway, I'm interested in outs, so I'll refer to DER.
The Whitesox had around a league average DER (.692 outs per ball in play, compared to .691 for the league). Their team UZR was somewhat below average (-18 runs). On a pitcher by pitcher level, that would work out to about -2 runs of an effect, if the Whitesox performed equally for every pitcher.
The slightly below average Whitesox fielders played slightly above average for Reynaldo Lopez, at +1.3 runs according to UZR data provided by MGL. Lopez had a wOBAcon of .369 compared to the league average of .363 (includes HR), meaning that based on the speed+angle at which balls were hit, it was slightly below average. So as best we can tell: the fielders performed around average when Lopez was on the mound, and Lopez performed around average.
Reynaldo Lopez
Reynaldo Lopez of the Whitesox had one of the highest DER in the league last year. He ended up with 20 more outs than the league average. Why does this happen? Random Variation. With over 500 balls in play, one standard deviation is just over 10 outs. We observe Lopez, or MORE ACCURATELY, the Whitesox fielders AND Lopez, at 2 standard deviations from the mean. Given that the observed distribution across all pitchers is close to what we'd expect from Random Variation, then virtually the entirety of what we are seeing with the Whitesox getting 20 more outs with Lopez on the mound is a result of Random Variation.
However. Those outs DID happen. We want to account for them. We want to ASSIGN them to... someone. Even if none of the fielders are pitchers are RESPONSIBLE for those outs, we want those outs to be ACCOUNTABLE. Now, we can decide to simply create a Random Variation bucket, and hold the accountability not to any of the players on the field, but simply the Random Variation Gods. You can certainly do that.
But suppose we want that the sum-of-players-equals-team. Then what? What do we do with those 20 outs?
Accountable
In some WAR models, like Baseball Reference, since it starts with Runs Allowed, and it strips away what it thinks is the fielding impact, whatever is left goes to the pitcher. In other words, the pitcher ABSORBS all of the Random Variation. Even if he is not directly responsible, he is being held accountable.
In other WAR models, like Fangraphs, since it starts with FIP, and tries to add in what it thinks it needs to, those outs... disappear. It is part of an implicit Random Variation Gods bucket. Since the true answer is likely between the two views, this is why I find just going 50/50 works out well enough.
We could decide to spread the Random Variation to the pitchers and the fielders. Perhaps splitting it 50/50 between the pitcher and fielders. So those 20 outs, we'd give for example 11 to the pitcher (for his pitching) and 1 to each of the 9 fielders. Or some combination thereof.
MGL
Ultimately, it brings us back to what MGL is fond of saying: all we care about is the players. And so in his view, making sure all the outs are accounted for is not a feature, but a bug, noise. It's noise to our understanding of the players. Sure, it's nice to account for everything. But assigning Random Variation to players makes it LOOK like they are responsible, even if the intent is to only make it about accountability. The Sins of the Father in essence.
So in order to reconcile all this, my suggestion is we create a bucket called Random Variation, and assign it to the pitcher. This allows folks like MGL to completely ignore it. And it allows folks like Bill James to completely include it. And he can decide even how to split up that Random Variation.
"Random Variation will be permitted, but controlled... and there will be the peace."
-- Don Barzini
Bill James has done a lengthy analysis of the Pitching component of rWAR (WAR from Baseball Reference) which you can read here:
My post here will:
First the disagreement.
...and takes the defensive support away from the pitcher, putting him in the position of pitching in a neutral park with an average defense behind him
It actually doesn't do this. And in fact, it does what Bill was saying it should do (or at least operate in a manner consistent with what it should do). Earlier in the article he says:
What is relevant is not what the player might have done or would have done in some other set of circumstances, but rather, the value of what he actually did in the circumstances in which he played.
I agree with that position, at least insofar we want to describe a record of what actually happened. And this is what Reference is doing. And you can see it based on the different baseline for each pitcher. That is, Reference is NOT adjusting a pitcher's Runs allowed per 9 IP. It keeps that sacrosanct. What it DOES do is ask "what would an average pitcher have done in these circumstances". This is why all the baselines are different for each pitcher.
So, what Reference is doing is adding or subtracting runs, using the team-level fielding support (or something strongly linked to that), and adding it AS PRACTICALLY A CONSTANT (the main point Bill is making) to each pitcher. This is regardless of the actual fielding support the pitcher happened to face.
If this sounds familiar to you, then let's go to the second point:
The part where I agree with him.
I wrote this article a few days ago, called Bayesian Run Support. If you read Bill's article, you see he uses the same type of analogy that I use. The basic point is what I wrote: And that is that you need to include the observations to your a priori and create a posteriori. Bill doesn't use any of those terms, but it's clear from what he wrote in his itemized list at the bottom that he's focusing on the need to have a posteriori, and not rely totally on a priori. Which is what Reference is doing. Which is what Bill's disagreement is.
And those of you who have seen what I wrote regarding Nola/Pivetta last year know I have the same issue. And I will be writing a followup called Bayesian BABIP that essentially addresses (some) of Bill's recommendations: let's establish our best estimate of the kind of fielding support they in fact did receive.
Moving forward
Having said all that, I agree with Bill's basic point that rWAR probably gets you 85% of the way there. And I've said countless times that the true Pitching WAR is somewhere between rWAR and fWAR (WAR from Fangraphs). They provide two polar opposite viewpoints. And I've recommended that the safest course of action is to just go 50/50 and get on with your day.
In the end, rWAR and fWAR are two huge services provided to the saber community. And having them is better than not having them. But having both at the same time is the best thing to do at this moment.
FIP is one of the most enduring metrics in the last 25 years. I created it in the same way that Bittersweet Symphony is the best song The Rolling Stones created in the last 25 years. FIP is really DIPS but consumable. Had someone told Voros: can you give me a quick version of DIPS, quick-DIPS, he would have created FIP. He didn't, but really, he would have, which is why I give 99% credit of FIP to Voros. I just provided the five second catchy beat that loops through Bittersweet Symphony.
Speaking of which, listen to this while you keep reading. It's going to get boring, so you'll need something to keep you awake:
FIP came to the forefront because, by itself current year FIP predicted next year's ERA (FutureERA) better than current year ERA. This was the groundbreaking part of DIPS.
Interlude
Take a step back 30 more years back: We learned that current year ERA predicted future W/L record better than current year W/L record. This is why saber-leaning folks reject a pitcher's W/L record: when the thing you are interested in is better measured by something other than itself, this is a sign that the thing you are interested in is filled with noise. And it's easily explainable: half the W/L record is actually the team's offense. Another portion is the team fielding, and the team bullpen also helps. Suddenly, the pitcher only explains a third of his own W/L record.
FIP showed similar tendencies to better predict future ERA than ERA itself. Which means we can use FIP to predict a pitcher W/L record better than either ERA or the W/L record itself.
What is FIP?
The bad part: because it did so well, people started calling FIP a metric about "what could have happened" and not "what did happen". They are wrong. All of them. FIP is 100% about WHAT DID HAPPEN, or at least a slice of what did happen, in the same way that OBP is a slice of what did happen. Neither tells the whole story of what did happen. But because FIP tells a better story of what will happen than OBP does, FIP was TREATED as "future FIP". It is not.
At its core ACTUAL FIP is:
In other words, walks have a bit more run impact than strikeouts, and HR have more than 4x the impact of walks.
Future FIP
And as a DEFACTO future-FIP, it works well. But if I were to create a FUTURE FIP for the community to use, it would look more like:
In other words, a strikeout tells you more than a walk about future runs, a reverse of "what did happen". HR only tell you 3x what walks tell you, and not the 4x "what did happen" would suggest. This particular FUTURE FIP is for illustration purposes only, but it would be along those lines conceptually.
Glenn DuPaul realized all this when he introduced Predictive FIP seven years ago:
FIP is a descriptive statistic that works fairly well as a predictive statistic, not the other way around.
And this is the core of Predictive FIP:
If you look at his Predictive FIP, all the non-K components are chopped in half. In other words, his Predictive FIP is:
***
So, what does all this have to do with Spray Angles and xWOBA? Well, the song is half-over, so, we’ll continue this in part 2.
Nola had a BABIP about 50 points lower than the Phillies BABIP, which with over 500 BIP translates to 27 fewer hits allowed than the average Phillies pitcher.
Pivetta was about 25 points higher on over 400 BIP, or about 11 more hits than the average Phillies pitcher. Indeed, Pivetta's BABIP was worst in the league among those with 162+ IP.
That is a gap of 38 hits. Did the Phillies fielders hurt them both equally, or did one pitcher receive more support than the other?
***
Well, Bayes would suggest that Nola received better fielding support. The Phillies had a negative 46 runs in fielding, which translates to about allowing 57 more hits than the average team. It seems obvious that if Nola allowed 27 fewer, and the Phillies allowed 57 more, that they probably did not play poorly for Nola.
MGL sent me his UZR by pitcher, and, as expected, the Phillies were actually league-average as fielders for Nola. In other words, those extra 57 hits their fielders allowed did not come at the expense of Nola.
And, as expected Pivetta was hurt by the Phillies fielders more than any other Phillies pitcher. Indeed, more than any other pitcher in the league! The Phillies fielders gave up 24 more hits with Pivetta on the mound.
So, adding it all up:
And so, that 38 hit gap gets reduced down to 14 hits.
***
Every year we see something like this. Back in 2011 when I first wrote about this, it was Verlander in the Nola role, and (future Cy Young winners) Scherzer and Porcello in the Pivetta role.
A continuation of a conversation from Twitter. Read that first. Please. Pretty please with a cherry on top.
***
This is what is perplexing the saber community when it comes to separating fielding from pitching: we can identify WHO is there, but we can't assign RESPONSIBILITY well enough. You start with simply ONE game. You have a perfect game, and so is 4 runs better than average and 5 runs better than replacement. But is the pitcher responsible for ALL of it? We've watched enough baseball to appreciate that there's alot of randomness. So, are perfect games usually 3 runs or 2 runs better than average for a pitcher? And are they 1 or 2 runs better than average for fielders? And how much to pure randomness? 0? 1? 4?
So that randomness, while starts to wash away over a season, doesn't completely wash away.
Jack Kralick in 1961 has this split with bases empty and runners on,respectively:
.292/.341/.429
.253/.297/.358
The OPS of those number is 14% higher than league with bases empty and 22% lower than league with runners on. And the Leverage Index with runners on is 2x that of bases empty.
So you have a pitcher that is substantially better... correction... a pitcher who has been ASSIGNED a performance record substantially better when it counts the most. And this explains why, when he's on the mound, he has among the lowest RA/9 in the league.
Do we want to credit Kralick with being on the mound getting better results with men on base, thereby limiting the impact of guys who got on base?
In other words: do we care about sequencing?
Or, do we prefer a "seasonal component" ERA, one that ASSUMES all performance is random in terms of the base-out state?
This was in effect "clutch pitching". Or "clutch results". And if we are trying to account for 101 runs allowed, and not the 110 or 120 (or whatever it is) that randomness would expect, then someone has to absorb that good result.
And you either give it to Kralick and/or his fielders and/or create a "timing-Kralick" bucket that acknowledges there was some 10 or whatever runs that were earned "on the knight's watch", but we don't know what to do with it.
Bill's methods are all about accounting for all those runs. So, we have to account for them, somewhere.
***
Fangraphs takes a polar opposite view, and assumes randomness of events, and ONLY targetting BB, SO, HR, HBP of a pitcher. The rest are essentially assigned to fielders and/or timing.
***
The true answer is somewhere in-between and since I know that we'll never come to consensus, I simply take a 50/50 approach of rWAR and fWAR and call it a day.
My Game Score v2 is in fact (a simplification of) that middle ground.
One of our findings in The Book was that a starting pitcher performed worse, the more he faced the same batter (in the same game). The other finding was that a pitcher performed better when coming in as a starting pitcher than relief pitcher (on the presumption pre-2018 that he would need to stick around for 5-8 innings as a SP). The advantage was about 8 points of wOBA for the SP for each time thru the order, back when we published in 2006 (based on 1999-2002 data). MGL has updated that over the years to something like 12 points.
I checked 2015-18,and we are now at 15 points. Here's what I did: for every batter+pitcher pairs (say Verlander+Trout), I also included whether it was the top or bottom of the inning (since there is an effect for that too, as we learned on the old book blog). I also made sure that all the players in the pool were starting batters and starting pitchers, as well as making sure it's the same bat-hand (for the odd times a switch hitter might face the same pitcher with a different hand).
The first time the players faced each other resulted in a 0.319 wOBA. The second time it was 0.335. That's a 16 point gain in wOBA.
The third time is a bit weirder to explain. Their wOBA was 0.350, but we also lose one-third of our player-pairs (usually the SP is knocked out at this point, but sometimes the batter is subbed out). So what to compare it too? If I compare it to only the 2nd time faced for the survivors, their 2nd time thru was 0.328. However, this will be biased because a SP will only be allowed to face a batter a third time if he's been doing well to begin with. Instead, I compare it to the 1st time, which was 0.323, which is a 27 point gain. Since 1st to 2nd we had previously established was 16 points, then the 2nd to 3rd is 11 points.
But waitaminute: doesn't even the 1st to 2nd ALSO have a selection bias? Yes! But in this case, we only lose 4-5% of our pairs. Whatever bias is muted. Still, it exists. If we want to instead say it's about a 15 point gain for 1st to 2nd and a 12 point gain from 2nd to 3rd, that might make more sense.
The 4th time facing each other in the same game has even bigger problems. They have a wOBA of only 0.342. But since it's only great pitchers here, we know our prior PA pool will have a much lower wOBA. And indeed the 3rd time facing is 0.305, but 1st time facing is 0.312. In other words, if the pitcher is facing the same batter the 4th time, he must have been sensational late in the game to begin with. If we use the 1st time facing as the unbiased estimator, that's a 30 point gain. And since we established that 1st to 3rd is 27 points, then 3rd to 4th is only 3 more points. Which given the highly focused sample, is likely not representative of a true 4th time thru (just like an aging chart of pitchers in their 40s might not truly be representative of all pitchers in their 40s... just those MLB pitchers who managed to survive there).
Anyway, I'd suggest the times thru the order might be something like +15, +12, +9, just for the sake of establishing something reasonable. Aspiring saberists are invited to jump in to establish a better baseline.
***
Now for relief pitchers. I did a similar thing, focused on the 1st time facing each other as a SP and as a RP. I made sure it's the same batter+pitcher pair, and everything else I had noted. And the first time facing each other, as a SP, it was a 0.345 wOBA and as a RP it was a 0.338 wOBA. In other words, there is a relief advantage of about 7 points. Of course, this is also a highly biased sample, since it's only guys that could go as a SP to begin with. Again, looking at aspiring saberists. For the sake of a baseline, I'll suggest it's more like 10 points.
***
So, our pitcher looks like this:
.300 wOBA as RP
.310 wOBA as SP, 1st time
.325 wOBA as SP, 2nd time
.337 wOBA as SP, 3rd time
.346 wOBA as SP, 4th time
The average (weighted) wOBA for a SP is about .322 using this scale, giving the RP a 22 point advantage, overall, on the SP. I've been using 27 points, but as I said, there's alot of estimating going on, so I'd love to see someone tackle this with greater care than this quick study.
In the WAR framework that I co-developed with the Straight Arrow readers(*), we have two replacement levels for pitchers, one for relief pitchers and one for starting pitchers.(**) If this feels like you are walking into the middle of a conversation, it's because you are.(***) Since we learned of the Rule of 17(****), we know that pitchers as RP are basically 17% more effective across the board.
What that means is that essentially if the replacement level for starting pitchers is 5.5 runs allowed per 9IP (RA/9), then the RA/9 for RP is 4.5. And this has worked out spectacularly well. When used with Leverage Index, the implementation of this framework is what you see at Fangraphs (fWAR) and Baseball Reference (rWAR).(*****)
Now comes the Rays who threw their sabot(******) at Pitching WAR. With Stanek/Yarbrough being an unconvential tandem, Stanek being The Opener and Yarbrough being The Headliner (*******), the concept of Starting Pitching and Relief Pitching is flipped on its head.
We can go one way and now define what is an Opener and Headliner and use those to set the replacement level. However, RallyMonkey offered a (potentially) viable alternative: have a sliding scale of replacement level, based on number of batters faced in a particular game. Games with 8 or fewer batters faced gets the standard relief replacement level of 4.5 RA/9. Games with 18+ batters faced gets the standard starting replacement level of 5.5 RA/9. And in between, we have a sliding scale, with every batter increasing the level by 0.1 RA/9. As a programmer, this appeals to me a GREAT deal. Implementation is simple and straightforward.
There are a few drawbacks, notably pitchers who get knocked out early due to injury. But that might have some good byproduct too, since getting knocked out early means an unprepared bullpen. So those SP should suffer a bit more.
Anyway, the floor is yours. Tell me what you think about any of this.
(*) There's alot of you, and I hope I don't miss anyone, but those that come to mind: RallyMonkey, GuyM, Patriot, DavidSmyth, Ray Kerby. I apologize to others who made their contributions but I didn't single out. I'm sure for example MGL provided valuable input, but it's been 10 years. There's probably another dozen of you that brought up seemingly benign considerations that ended up being key points.
(**) This basic idea actually originated with Keith Woolner. We refined it and plugged some of the holes.
(***) Bill James said that. He noted that he can't always start from the beginning, so, you'll have to excuse me a little.
(****) Look that up.
(*****) Indeed, Straight Arrow Reader "RallyMonkey" implemented some/most of what you see at rWAR (with Sean Forman and his team adding their own flavor as well as doing a spectacular job its presentation). Dave Cameron and David Appelman implemented all of what you see at fWAR, and their presentation is equally spectacular. Two better stewards for WAR there is not.
(******) "Hence, sabotage." -- Samantha Jones
(*******) Thanks to Tess Kolp for that term.
?This is just a collection of my tweets from yesterday and today. The basic point is that Nola (a) has one of the best BABIP and (b) plays with one of the worst fielding team. And so, it boils down to: how do you adjust for his context?
And more specifically: do we treat his fielders as having the expectation to play at their typical fielding level FOR THE SEASON or ON THOSE PARTICULAR TIMES WHEN NOLA IS ON THE MOUND? It's a nuanced distinction that has a very specific implication to Nola.
Here they are:
Nov 23 14:15
Layered wOBAcon
Nov 22 22:15
Cy Young Predictor 2024
Oct 28 17:25
Layered Hit Probability breakdown
Oct 15 13:42
Binomial fun: Best-of-3-all-home is equivalent to traditional Best-of-X where X is
Oct 14 14:31
NaiveWAR and VictoryShares
Oct 02 21:23
Component Run Values: TTO and BIP
Oct 02 11:06
FRV v DRS
Sep 28 22:34
Runs Above Average
Sep 16 16:46
Skenes v Webb: Illustrating Replacement Level in WAR
Sep 16 16:43
Sacrifice Steal Attempt
Sep 09 14:47
Can Wheeler win the Cy Young in 2024?
Sep 08 13:39
Small choices, big implications, in WAR
Sep 07 09:00
Why does Baseball Reference love Erick Fedde?
Sep 03 19:42
Re-Leveraging Aaron Judge
Aug 24 14:10
Science of baseball in 1957
THREADS
October 18, 2024
Describing the season of a pitcher: ERA v FIP
May 04, 2024
Statcast Lab: Spin Axis Reports
December 28, 2023
Improving WAR - Re-solving DIPS (part 2)
December 28, 2023
Improving WAR - Resolving DIPS (part 1)
December 12, 2023
FIP and xwFIP
November 17, 2023
Blake Snell or Spencer Strider?
March 20, 2023
Measuring Finger Pressure on a Pitch
January 02, 2023
Changing the Release Angle of Reid Detmers
December 31, 2022
Arm Angles on Individual Pitches: Alek Manoah
April 05, 2020
Splitting responsibility between pitchers and fielders
December 16, 2019
How many wins can you generate over the next 9 years?
September 23, 2019
Statcast Lab: What impact does Verlander’s historically low BABIP have on our evaluation?
March 06, 2019
Sins of the Father: Who gets the random variation?
March 06, 2019
Rejoinder and/or Amicus Brief to Bill James on Pitching rWAR
January 14, 2019
Of Spray Angles, FIP, and xWOBA
Recent comments
Older comments
Page 1 of 150 pages 1 2 3 > Last ›Complete Archive – By Category
Complete Archive – By Date
FORUM TOPICS
Jul 12 15:22 MarcelsApr 16 14:31 Pitch Count Estimators
Mar 12 16:30 Appendix to THE BOOK - THE GORY DETAILS
Jan 29 09:41 NFL Overtime Idea
Jan 22 14:48 Weighting Years for NFL Player Projections
Jan 21 09:18 positional runs in pythagenpat
Oct 20 15:57 DRS: FG vs. BB-Ref
Apr 12 09:43 What if baseball was like survivor? You are eliminated ...
Nov 24 09:57 Win Attribution to offense, pitching, and fielding at the game level (prototype method)
Jul 13 10:20 How to watch great past games without spoilers