Madison Bumgarner’s Many Times Through the Order
If you’ve been consuming baseball analysis for any meaningful length of time, TTO comes as a particularly familiar acronym. It stands for Three True Outcomes — walks, strikeouts, and dingers — and for me, personally, it makes me think of Adam Dunn. And it makes me think of baseball writing eight or ten years ago, when at least I was starting to come into my own. But TTO can and does also stand for something else, something we’ve been talking and reading about a lot over the past several weeks — Times Through the Order. As in, the times-through-the-order penalty, that describes how starting pitchers become less effective over the course of a game. All those things you’ve read about how Starter X was left in too long by his manager? The criticisms are mostly founded upon the idea that pitchers get tired and over-exposed.
Those are tricky things to separate. It stands to reason pitchers get worse because they get more tired. It also stands to reason pitchers get worse because their opponents get more and more familiar with the pitches being thrown. You think about “looks”, and whatnot. We know there’s a TTO penalty during a game. And if any of it has to do with familiarity, then it seems like there ought also be a TTO penalty within a series.
A few weeks ago I was talking to Brandon McCarthy while I was working on something for the upcoming Hardball Times Annual. At one point he started talking about Zack Greinke, and about how he can be a difficult guy to prepare for because he’ll vary his approaches game to game. Like, one game, he might throw a ton of sliders, and then hitters will gear up for the slider in the next game, and in that game he might throw as few as zero sliders. Greinke, therefore, is thought to mix things up on the fly, even when facing a different opponent. Maybe he’s unusual in this regard, and maybe he’s not. It’s one player’s perspective.
Six days ago, Madison Bumgarner made a start against the Royals, and he went seven innings. He was good. Sunday he made another start, also against the Royals, and he went nine innings. He was good. He threw a complete-game shutout! And this despite the Royals seeing him for the second time in a week. I don’t know of any research toward this end, but I wonder if starters who make multiple starts in a series perform worse the second time around. The second time around, they’ve already recently been seen, and that strikes me as overlapping some with the TTO-penalty concerns.
Bumgarner, obviously, suffered from nothing as he cruised from start to finish. Or maybe the Royals were getting better looks and Bumgarner just got outs anyway, I don’t know. But Bumgarner, in his first start, faced 26 batters. Sunday, he faced 31. He’s had seven showdowns each against Alcides Escobar, Lorenzo Cain, Alex Gordon, and Eric Hosmer. I was interested to see if Bumgarner would do anything differently start to start. Sure enough, he approached the Royals like a different pitcher in Game 5. Oh, he still approached them like the dominant version of Madison Bumgarner, but pitch-to-pitch, at-bat-to-at-bat, Bumgarner made the Royals wonder about the significance of their recent matchup history.
Put more simply: Bumgarner did different things. However the Royals intended to learn from his Game 1 attack, he preemptively countered that by adjusting his Game 5 attack. Let us learn from our eyes, and from Brooks Baseball. A table, now, of some fastball rates:
| Start | RHB, 1st Pitch | RHB, 2 Strikes | LHB, 1st Pitch | LHB, 2 Strikes |
|---|---|---|---|---|
| Game 1 | 73% | 58% | 82% | 50% |
| Game 5 | 50% | 35% | 62% | 43% |
The last column doesn’t tell you much, but the other columns tell you more. Sunday, Bumgarner attacked fewer righties with fastballs. He attacked fewer lefties with fastballs. He tried to put righties away with fastballs less often. One of Bumgarner’s recent trends has been an increase in his four-seam fastball rate, as he’s shifted away from his slider a bit. Sunday, Bumgarner threw 44% fastballs. Last week, he threw 58%. That 44% is Bumgarner’s lowest single-game rate since August 8, when he faced…the Royals, and threw 34% fastballs. So the Royals have seen three versions of this guy in three starts in three months.
Now let’s consider overall pitch locations. I define “Inside” as inner third or beyond, and I define “Way Up” as above the upper threshold of the PITCHf/x strike zone.
| Start | RHB, Inside | RHB, Way Up | LHB, Inside | LHB, Way Up |
|---|---|---|---|---|
| Game 1 | 42% | 23% | 20% | 0% |
| Game 5 | 25% | 8% | 15% | 13% |
Bumgarner spent more time away against righties, and more time down. The high fastball didn’t disappear, though, as he started using it against lefties instead. In Game 1, Bumgarner attacked righties up and down. In Game 5, he was a lot more side-to-side. This is all very difficult to demonstrate in a few images, so let’s just consider some examples. Bumgarner’s first matchup with Lorenzo Cain, from Game 1:
And the same, from Game 5:
Different sides of the plate, different pitches. In the Game 1 example, Bumgarner started with three fastballs, then threw four of five non-fastballs. In the Game 5 example, you’ve got a fastball away, a fastball down and away, a cutter down and away, and a fastball down and away. Cain reached in both plate appearances, but in one he got hit and in the other his contact was poor.
And here’s Bumgarner’s first matchup with Mike Moustakas, from Game 1:
And from Game 5:
In the Game 1 example, Moustakas saw three cutters, down and away, out of the zone. In the Game 5 example, he saw a cutter high and tight, and three cutters away, in the zone. Bumgarner has talked about how he’ll add and subtract from the pitch, using it very differently depending on the circumstances, and against Moustakas it wasn’t just used as a chase pitch. Although in the seventh inning, Bumgarner moved a couple of those cutters a few inches further away, and Moustakas offered at both.
It can be tricky to discuss adjustments game to game, because we don’t know how much is deliberate, and every single plate appearance is its own unique competition. But it’s clear, at least, that Bumgarner wound up pitching differently on Sunday. I didn’t even mention that he threw a few more changeups than usual. His second time through the order Sunday, he threw 33% curveballs; no other time did he break 21%, and no other time did he break 15%. The cutter was featured more prominently Sunday, and while Bumgarner might now be done for 2014, there’s the chance he returns out of the bullpen in a Game 7. And I’m not sure what the Royals could reasonably expect.
I don’t know what the Royals were looking for throughout Game 5. All I know is that they didn’t get it. Probably had a lot to do with the guy on the mound, I’m guessing.
Jeff made Lookout Landing a thing, but he does not still write there about the Mariners. He does write here, sometimes about the Mariners, but usually not.
As a baseball fan, I admire his variety and command of several pitches; he never served that fat pitch that an opposing hitter could hammer, yet he got so many hitters to chase, what a marvel.
As a Royals fan, man, dammit. Godammit.
Excellent article. This is what I pay fo- umm, I mean, nice article!
This comment reminded me that Fangraphs+ is a thing, and the awesome game 1/5 comparison pics from Jeff got me to actually pay for it. Hat tip to both of you 🙂
Good work, Jeff. But I think if you analyzed even more closely Bumgarner’s batter-by-batter, inning-by-inning approach, it would further confirm everything you wrote. Bumgarner and Posey absolutely had an entirely different game plan for Game 5, and they executed it. Whether the decision was made before the game or as the game developed, we of course can’t be sure. But there is this:
In the first inning, Bumgarner started every hitter with a first ball fasbtall, except for Hosmer (after giving up the single to Cain). He continued that approach of establishing the fastball early in the count the first time through the order. However, the second time through the order, he began throwing more cutters and curves early in the count, and basically pitching “backwards.” Then he switched up in the fifth inning with the bottom of the order due up, and began using primarily fastballs again (and he got into a bit of a jam). He then returned to pitching backwards. And then in the ninth inning, Bumgarner changed it all up again. He threw 10 pitches, seven of them fastballs. In fact, the only hitter he threw any cutters to was Hosmer. But Bumgarner threw Hosmer more fastballs in the ninth (3) than he had thrown him in total in his three other at bats (2).
So we saw a textbook example of how to pitch against a team the second time in a short series, and how a starter needs to plan to face each hitter three or four times and mitigate being predictable. Clayton Kershaw could learn a thing or two from Madison Bumgarner.
A couple of comments:
Theoretically, a pitcher’s approach to a certain batter in a certain situation should be exactly the same in any game. There are some caveats to that. One might be how the pitcher perceives the effectiveness of his various pitches on that given day. Another might be the umpire, park, and weather. Another might be the real or perceived strengths, weakness, or expectations of the batter on that given day. And finally, it is possible that if a batter has seen you relatively recently, even if everything else remains the same, your pitching pattern SHOULD be different, although not likely VERY different.
Please do not forget that there are completely random fluctuations in the distribution of pitches thrown and their locations. For example, if in a 100 pitch outing, a pitcher is supposed to throw exactly 20% cutters, then 5% of the time, by chance alone, he will throw less than 12% or more than 28% (the binomial SD is 4%). that is a very important point – we cannot assume a change in approach based on different pitches being thrown at different percentages from game to game.
In fact, if a pitcher uses the exact same approach in very single game, it is unlikely that we will see the exact same percentages or even close on all of his pitches in any 2 randomly selected games. Not saying that he did or did not change his approach significantly in this came compared to the last one, only that we have to be careful about assuming that from changes in pitch selection,. Much of those changes are random, whether he changes his approach or not.
Finally, because the situations in each game are very different with regard to inning, outs, runners, the batters at the plate in each situation, COLLECTIVE approach will appear to change from game to game more than is actually occurring. That is because “approach” is really a PA by PA phenomenon, dictated by the batter and the very specific game situation during that PA.
I don’t think you can chalk up pitch distribution – especially when broken down by counts, into a simple model of random fluctuations. While with some pitchers not named Madison Bumgarner the location may be as good as hitting a dart board, these guys are not spinning a slot machine when they decide what to throw. Bumgarner and Posey come into the game with an approach for every batter – what they’re gonna throw for the first pitch each time through, what they want to pound the guy with ahead or behind, and what to throw as an out pitch. Which I agree that the results end up a random distribution heavily influenced by the batter, it is far more influenced by that framework set ahead of time. Especially when you break it down into counts like the 1st pitch fastball%, there’s no question the approach changed. Look at how he cut the 2-strike fastball% nearly in half against RHB, that’s a fantastic stat. I’d really like to believe that’s Bumgarner trying to finish specific batters off with specific different pitches, and not Buster Posey rolling a pair of dice behind the plate.
As for the batter’s impact, he can certainly affect the count, but less so what’s called on each count. I’d guess the batter is likely to change the approach if they start fouling an excessive number of balls off, but that’ll mostly just affect 2-strike data. If a a guy starts squaring something up, it may scare them away from throwing it again – but given how that didn’t happen much and no Royal averaged more than 5 pitches per PA, the gameplan probably stuck.
“…these guys are not spinning a slot machine when they decide what to throw. Bumgarner and Posey come into the game with…”
You’re using vague words and expressions to convey an idea, but yes, every pitch is a spin of a slot machine to use your metaphor. If it weren’t, then the batter would be able to think along with the pitcher and catcher and thus have a very good idea as to the pitch selection and location. The evidence for that is that we don’t see catchers as hitters dominate. If pitchers and catchers didn’t “randomize” their pitches around some fixed distribution, then catchers as hitters and other smart batters would dominate.
Pitcher in Black: All right. Where is the pitch? The battle of wits has begun. It ends when you decide whether to swing, and find out who is right… and who is dead.
Vizzini: But it’s so simple. All I have to do is divine from what I know of you: are you the sort of man who would start the at-bat with a fastball or an offspeed ptich? Now, a clever man would start with a fastball, because he would know that only a great fool would swing at the first pitch. I am not a great fool, so I can clearly not choose the fastball. But you must have known I was not a great fool, you would have counted on it, so I can clearly not choose the offspeed pitch.
Pitcher in Black: Truly, you have a dizzying intellect.
Brad Ausmus on pitch-calling:
“[I’m thinking] what’s the score, what inning are we in, how many outs, what’s this hitter’s weakness, what’s this pitcher’s strengths, who’s on deck, who could pinch hit, who is up after the hitter on deck — and you kind of go through all of these things in an instant,” says Ausmus. “And then you make a decision and put down the next signal. [You’re also thinking] how did we get this guy out last time, what pitches did he see, what pitches did we just throw — so there’s about 10 to a dozen things that you go through in your mind before you put that signal down.”
The decisions about what pitches to call start taking place well before game time. Ausmus says he prepared graphs for his pitchers before every single series. The graphs contained information about the strengths and weaknesses of every player on the opposing team.
“We would go through this stack with their stats: how often do they walk, how often do they strike out — and this all gets condensed down into a chart,” he says. “You use that as your cheat sheet before the game before you’re going over the lineup with the pitcher that day.”
(copied from an NPR interview a few years ago)
—
If it is a slot machine, it’s at the very least a rigged one. I do agree that overall pitch selection over a whole game as appears, and from game to game can be viewed accurately as a random distribution. But when you break it down into batters and specific counts it seems clear to me each pitch is picked for a reason. No, the sequence is not 100% scripted and the catcher is making up calls as he goes along, but he’s the dice he’s rolling were loaded before the game began.
Every batter has multiple weaknesses and every pitcher has multiple strengths to attack them with. Even if you boiled it down to 3 pitches and 4 locations, there’s enough options I don’t think the smartest hitter could guess a scripted sequence. Both parties know the strengths and weaknesses and what’s been previously thrown, and maybe with a smarter batter the catcher may be obligated to call a wider distribution of pitches – but that doesn’t mean the choice -has- to be a random one for it to be unpredictable.
Sorry for any vagueness. I’m no statistician and am generally out of my element talking to you guys.
Leo what you are failing to understand is that every single one of those variables simply creates a different matrix of pitch probabilities, NOT one single pitch. It can’t. If it did, the batter would be able to figure it out too and then it wouldn’t be effective.
For example a batter’s weaknesses is only in the context of how often he sees that pitch in any given situation. We all know lots of Sal Perez type hitters. Their biggest weakness is the off speed pitch low and away. They don’t seem to be able to lay off that pitch when behind in the count.
The would however lay off that pitch if it was thrown with enough frequency because a batters success on a pitch is largely dictated by whether he expects it or not. So for Perez you might throw that pitch in any given situation 60% of the time whereas with another batter it might be 40%.
But those percentages are key and they are intractable. And if you are going to throw an 0-2 curveball to Sal Perez 60% of the time on an 0-2 count (assuming many other variables like score inning etc.) that means by definition the the pitch SELECTIOB is RANDOM but the mean value for that curve ball is 60%.
I don’t mean random as in any pitch is as likely to be thrown as any other pitch.
All pitches are thrown in a random fashion around a particular mean. And that mean is usually between 0 and 1.
For those who haven’t read it yet, here’s a link to MGL’s comprehensive BP article on times through the order:
http://www.baseballprospectus.com/article.php?articleid=22156
An important issue noted by MGL here is that there appears to be a depression of offense in the 1st inning in general, which could blunt some of the possible impact of the TTO effect carrying into a second start against the same team. In principle, I’d have to think that a hitter’s ability to acclimate to a particular pitcher is a lot more powerful 30-60 minutes after the last encounter vs. 4-5 days after.
On a related note, I’d really love to see a study of how this effect might be different between starters with multiple plus pitches like Bumgartner, Greinke, Darvish, etc. and starters only one or two plus pitches. In principle, these guys would be just as likely to change up their approach within a game as they are between games (McCarthy’s comments about Greinke notwithstanding). The tough part would be finding a clear dividing line between these groups, since you couldn’t really rely on the “win values” of each pitch (because of their interdependence). The splits would probably need to be something like the % frequency of the two most-often thrown pitches to make a reasonable breakout between groups. It might shed some light into whether pitchers with a couple of great pitches and little else really do belong in the bullpen, or if their decline within a start is no worse than a guy with a large repertoire.
I had you in mind the whole time. Clearly, I can say only so much without talking to Bumgarner and getting honest responses to questions about his game plans. All I can do, really, is say here’s what happened, and you can decide how much that means and how much of it was intentional. Some of these differences seem remarkable, some not so much. I think it’s likely some was intentional, and I think it’s likely some wasn’t. I don’t think Bumgarner wanted to give the Royals exactly the same look. I also presume he didn’t go in looking to turn his whole approach upside-down.
Another factor is the different ballpark. Pitching away to righties in SF is generally a good idea as no one (like literally I don’t think but a couple if righties ever hit it out there) hits it out and there is a a whole lot of room in right center. So pitch away and let them swing as they might.
Also you wonder if they approached game 1knowing they would face him again. I would think you worry about game 1 first and go from there.
If “starters who make multiple starts in a series perform worse the second time around,” then in general we would expect Games 5, 6 and 7 of a series to be higher-scoring than Games 1, 2 and 3 of the same series. We would expect the tendency to be most pronounced in the World Series, with batters facing pitchers from the other league with whom they are less familiar the first time around. Some innings of the games are pitched by relievers, but according to this theory we would also expect the relievers to be less effective in the later games as the same batters face them multiple times.
But I find no such tendency. Over the past 40 years (1974-2013) in the World Series, Games 5, 6 and 7 have been slightly lower-scoring than Games 1, 2 and 3 of the same series (excluding extra-inning runs): There have been 16 fewer runs scored in Games 5, 6 and 7. That number is statistically insignificant of course, but the point is that there is no evidence that pitchers’ second starts were worse than their first starts.
Jeff Caveney,
Good point, and good research. However, you also would need to account for managers (especially in the past) using their best relievers more in the last few games of a series since “there is no tomorrow.” Not sure how much this matters, but it is potentially another variable.
Nice idea for an article Jeff. Was interesting to see how much his approach changed.
I did look at the issue of a pitcher facing the same team in back-to-back starts once here.
I found typically wOBA against/HR%/K%/BB%/SwStr% were not any worse the second time around. I did see that pitchers tended to throw fastballs slightly less often in the second start.
Good stuff!