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?
Back in the early 2000s, when I started blogging heavily on baseball and hockey, I was intrigued, then aghast at the "forecasting" systems being offered, some for a price. They all came with a pseudo-promise of some sort or other.
This is the same thing with the stock market that I used to follow back in the 1990s. I saw an article at the time about evaluating stock predictions. And wouldn't you know it: only one of the ten brokerage houses even beat the index. Basically, nobody can predict anything really. No one has any special insight. You throw thousands of people together, and Random Variation will simply start putting some folks ahead of others.
It's also when I learned how Mutual Funds would get above-average results: you'd have a fund company that has two similar types of funds. One will do better than the other. Guess what happens: one absorbs the assets of the other, but NOT the history. So, now you get survivorship bias: all the remaining mutual funds are above average! And then they create a NEW second one, to keep that cycle going.
This is also how they sell those free betting tips. You call some 1-800 number with three picks being offered for free. Well, they set up 8 different lines, each with a different combination of picks. One of them will get all of them right, and therefore 12.5% of the callers will be happy with those results, and stick with that phone line.
Anyway, back to baseball. I decided to try my hand at forecasting. I started with something simple, and just used the three most recent seasons. It worked pretty well. Then I started adding more and more. And something curious happened. It would help for 51% of the batters and hurt 49% of the batters. No matter what I tried, other than age, nothing really stuck much. A different 51% of batters helped, but no real bias. Each iteration was alot of work, for such little gain. So, I decided to take a step back and decided to have as my baseline just a Naive model: last three seasons, age, and regression.
Then, I compared that to what was being published publicly, and something interesting happened: the Naive model was as good, or better than virtually everything out there. So, instead of trying to improve the model to try to get every little gain, I decided to publish as-is, and call it Marcel The Monkey Forecasting System, aka The Marcels, as the basic most simple forecasting system anyone should expect. So, instead of trying to be the best, I'm basically saying: this is the worst (acceptable). And boy did that clear the field. If you can't beat The Marcels, then what is the value-added of your system?
And so, I published it, and kept it up for a while. In the meantime, others have implemented my model (though without me checking their code, so I can't confirm they are totally faithful, but, I'm sure they are all excellent).
On the surface, Michael Harris II had a fine batting season. His .345 wOBA was above the league average, which we can see through his wRC+ of 115. A league average wRC+ is 100, and so Harris generated 15% more runs per plate appearance (PA) than the league average batter. Since the league average batter generates about 0.12 runs per PA, then given the 539 PA of Harris, we'd expect a league average batter to generate .12 x 539 runs, or about 65 runs. Since Harris is 15% above that, he generated ~75 runs, or +10 runs above average.
Harris however saved all of his terrific batting with the bases empty, with a whopping .386 wOBA. With runners on, meaning when his bat would do the most damage, his wOBA was a very low .287. Just so you can appreciate the disparity of those numbers, among the 245 batters with at least 350 PA, .287 ranks 24th worst, while .386 ranks 13th best. In other words, he's either at the 95th percentile when the bases are empty or at the 10th percentile when the runners are on base.
Do we care? If we didn't care, then we could translate his wOBA by runners on and bases empty similar to his overall numbers, and get his splits run values. His bases empty wRC+ is 142, meaning he generates +42% runs on his 315 PA. And so, .42 x .12 x 315 = +16 runs above average. And his runners on wRC+ is 76, meaning he generates -24% runs on his 224 PA: -.24 x .12 x 224 = -6 runs. Add it up, and +16, -6 gives us +10 runs above average, back where we started.
Except, well, with bases empty, the average player does not generate .12 runs per PA, but rather 70% of that. And with runners on, the average player does not generate .12 runs per PA, but rather 1.4 times that much.
With the bases empty, we falsely gave Harris credit for +16 runs above average, when it should be 70% of that, or +11. Similarly, he wasn't -6 runs with runners on base, but 1.4X that or -9 runs. Suddenly, he is now +11, -9, or +2 runs above average, very different from where we started.
We now get into the sabermetrics world, and asking the question: how are we trying to evaluate the batting of Michael Harris? Do we care that he had a 100 point difference in his wOBA, based on whether there are runners on or bases empty? Or, do we simply evaluate him as if his performance is independent of the runner situations?
Cedric Mullins II
Now consider the flip-side with Cedric Mullins II, who had an overall wRC+ of 99, meaning virtually league average. He had an extremely low wOBA with bases empty at .256, with a 61 wRC+. While with runners on, it was an enormous .398 and a 159 wRC+. This disparity is even greater than with Harris, but in the other direction.
Let's walk through his calculations. Without considering the leverage aspect of his performance, his bases empty would be: (1 - .61) x .12 x 278 = -13 runs. And his runners on: .59 x .12 x 177 = +13 runs. So, -13 and +13 is 0, and so league average which we already knew with his 99 wRC+. But, was he really league average?
We keep saying that the average batter generates .12 runs per PA, but that's only on average. With bases empty, they only generate 70% of that, and with runners on, it's 140% of that. Mullins was not -13 runs with bases empty, but -13 x .7, or -9 runs. His performance was not as costly as that -13 would suggest. And Mullins was not +13 with runners on, but 13 x 1.4, or +18 runs. The terrific performance of Mullins was able to be leveraged, his timing of that performance to match it with the runners on base lead to +18 runs, not +13 runs. And so, we have +18 and -9, for a total of +9 runs, not 0 runs.
In other words, the timing of Mullins performance, to maximize his good performance when runners are on base and minimize his bad performance with bases empty allows an overall average batter to generate in real tangible runs +9 runs above average. The Orioles obviously benefitted from those 9 extra runs: they actually scored those runs. Do we give Mullins the credit for the timing of his performance? Or, do we just consider Mullins to be an average batter in 2023, and those real runs are just put in a "timing" bucket, of which Mullins derives no benefit from?
Harris or Mullins?
This issue has perplexed sabermetrics since its foundation fifty years ago, and even before that. Really, from the dawn of baseball, we've wrestled with the idea of whether to give the batter credit for generating runs that actually scored, or whether we should just treat every plate appearance equally, regardless of its leverage aspect.
Solo HR or bases loaded HR?
It's the same thing, the batter did the same thing, and the extra benefit to the team does not translate back to the individual players.
It's ridicuously different, and of course extra benefit of the bases loaded HR has to trickle down to the players, ensuring every run is directly accounted for.
And Harris/Mullins issue comes up every single year because every single year we have dozens of players where there is a noticeable difference in how their performance are treated.
I wish I had an answer to give you, but really, I'm always left with more questions. Which is good for research purposes, but bad when trying to move forward.
The NHL released their summary reports based on player tracking. You can read about it in a few places. Here's one from The Athletic.
As this is the first version, I temper my expectations to the reality that we are dealing with production data in a production environment as a first release. Sports, unlike all other products and services, really has almost no full testing environment available. Whatever preparations we may get from putting tracking at one park (Salt River Fields) for a few weeks really is a drop in the bucket when you compare to the 30 parks x 81 games that millions are watching live, that an MLB season offers. Things that you might not even know to test for manifests itself almost immediately in a live game.
So, I look at the first release of anything sports-related as if it's batting practice, but that everyone treats as a live game. While we may all think we should be like Tim Raines on May 2, 1987, that's not reality.
Now, let's talk about player speed and MPH. First I should point out that I am in the minority here because I am dealing with Inertial Reasoning when it comes to speed. Whether MPH or KPH, speed is being represented like this. When you throw a ball and you are trying to outrace a motorcycle, sure, that seems reasonable. But life would be easier had we presented "time to plate" as the goto number. See, when you present things in MPH, there's nothing more you can do with that. It's the end-of-the-line. If you want to USE that number, that speed in MPH, the very very first thing you have to do is convert it to feet per second (or meters or yards). But the key point is that the denominator is seconds not hours.
Why is that? Because then you can actually use that number. Suppose for example I tell you a runner is rounding third at full steam, at 30 feet per second. How long until he reaches home plate? That's 90/30 or 3 seconds. And suppose an outfielder is releasing a ball at that very instant, he is 250 feet away, and the average flight speed for his throw is 100 feet per second: how long will that ball take to reach the catcher? That's 250/100 or 2.5 seconds. That's the story. The runner will get to home in 3 seconds, the ball will reach home plate area in 2.5 seconds. The catcher has 0.5 seconds to do something, whether to stand there waiting for the runner on a perfect throw, or he needs to scramble to get home on an offline throw.
An outfielder misses catching a ball by 3 feet. How much faster would be need to be to catch it? Well, if he was running at full speed for 2 seconds and 57 feet (28.5 feet per second) then he'd need to bump that up to 60 feet in 2 seconds (30 feet per second). I could go on, and have gone on. The point is simply this: make the number usable, applicable to the task at hand. And the task at hand is not to just "present a number". It's to give that number relevance, to let it resonate for the play.
Now to hockey: they are showing MPH, which of course is the default position. But suppose I tell you that Connor McDavid is skating at 30 feet per second toward the net, and in the meantime, Cale Makar is defending him by skating backwards at 20 feet per second. If I just gave you nightmares from school about two trains colliding, this is exactly correct. This nightmare for you is a dream for me. I've been waiting for this data all my life. If McDavid is going to skate for 2 seconds at this speed, he will cover 60 feet of ice. Makar in the meantime will cover 40 feet of ice in the same 2 seconds. In order for McDavid to not beat him, one on one, Makar has to have 20 feet of space between him and McDavid. (All numbers for illustration purposes only.)
There's a reason that we don't report 100m runners and 200m runners in terms of MPH. It's not relevant, and it won't resonate. What they do instead is report split times, like from 70m to 80m, they run in 0.98 seconds or something. This is something that matters, because it gives them a real target to their overall 100m run. Shaving 0.02 seconds in that split means shaving 0.02 seconds on their overall number.
It's an eventuality that the presentation of player moving speed (running, skating) will be in a form of feet or yards or metres per second. Ideally, we can set the standard from the outset, rather than needing to reset it after a long battle.
Former MLB pitcher, now MLB Network Broadcaster, Dan Plesac said something interesting: throwing all nine batters as RHH against a LHP is not a good idea. Or at least, against Montgomery specifically. Because they can get into a groove.
Which is interesting, because if there is one thing I keep pointing out is that players are human, and humans (athletes specifically) thrive on familiarity, repeatability, instincts.
Now, Montgomery faces a huge number of RHH. We're talking about Randy Johnson-level of teams avoiding throwing their LHH out there in favor of RHH. Is this smart?
How does Montgomery do in games when there are no LHH in the batting lineup? And one? And two? Glad you asked.
He had 20 starts when he faced no LHH, facing 466 batters. The wOBA allowed in those games was .302. He had 58 starts when he faced one LHH in the starting lineup. The wOBA in those games was .298. So, this is the batting teams proactively applying the Randy Johnson solution, stacking RHH, and all they get is a .300 wOBA out of it.
When he faced two LHH, which to most people would be a good thing for Montgomery, but Plesac is suggesting no, well, the wOBA was .321. Plesac is right.
When he faced three LHH, the wOBA in the game was .338. Plesac is right.
Finally, there were six games when batting teams ostensibly had to suffer with FOUR LHH (and to most folks, that sounds like a gift to Montgomery), and their wOBA was .326. Plesac is right.
Here's how it looks (click to embiggen), when we split it into at most one LHH, and at least two LHH. We have 78 games with 0 or 1 LHH in the starting lineup. Those LHH hit .282 wOBA, which is certainly very very low. When he faced at least 2 LHH in the starting lineup, those LHH hit .279 wOBA. So, he eats up those LHH.
But the interesting part is the RHH. He devours RHH when there is only 0 or 1 LHH in the lineup, to the tune of .301 wOBA against those RHH. But, put in at least 2 LHH, those RHH end up hitting an astonishing .344, completely destroying Montgomery.
So, you get to the point that managers should want to feed his LHH sheep to the Montgomery wolf, and intentionally have them get devoured. This is so that his RHH sheep can turn into battering rams, as Montgomery has his head spinning between LHH and RHH.
Plesac is right.
Now: is a manager brave enough to follow through here? Would Montgomery actually WELCOME having to face 3, or 4 LHH? Or, would he rather face 0 or 1 LHH, so he can get into a rhythm against RHH?
Caveat: we need to more carefully look at the individual batters here. While it is tempting to just assume this volume of data has all the players cancelling out. Maybe there's some bias in the data, so that what makes Plesac look right is just some other underlying issue.
Aspiring Saberists, Assemble.
End-of-credits scene: As for Randy Johnson, when he faced 0 to 2 LHH, the game wOBA was .302. In the 48 games he faced 3 or more LHH, the game wOBA was .310. Indeed, the RHH hit .304 wOBA when the lineup was stacked with RHH, while they hit .329 when they weren't. Now, the caveat can simply be this: when you stack your RHH, you are bringing in a bunch of really bad hitting RHH, so that will drive down your average wOBA. But, I don't think it can explain this much of a gap. It's just far too large a gap. So, it's very possible that MLB clubs should have sent Larry Walker and John Kruk to get devoured, and they'd help out their RHH mates.
Going by the Classic Predictor in the NL, and it looks like this (with the FIP-enhanced version in parens):
Blake Snell (1)
Zac Gallen (4)
Spencer Strider (2)
Justin Steele (5)
Logan Webb (3)
Kodai Senga (7)
Zack Wheeler is 6th in FIP-enhanced.
So, Snell wins either way. And Gallen is ahead of Steele either way. And Steele is ahead of Senga either way. Strider is ahead of Webb either way. Strider is really the wildcard.
Corbin Burnes in 2021 aside, we're still not at a full paradigm shift. We are basically at a 60/40, maybe 75/25 split in terms of the Classic v FIP-enhanced Predictors.
If we were at 50/50 in terms of the weighting of the two, Strider would be just ahead of Gallen. If it were 75/25, Gallen is just ahead of Strider. If it was two-thirds/one-third, they are dead-even. Webb and Steele are dead-even if it was 80% Classic and 20% FIP-enhanced.
As you can see, we can twist ourselves into knots here trying to figure out where we are in the FIP paradigm shift.
Going by the Classic Predictor in the AL, and it looks like this (with the FIP-enhanced version in parens):
Gerrit Cole (1)
Kevin Gausman (3)
Luis Castillo (9)
Sonny Gray (2)
Kyle Bradish (5)
Zach Eflin is 4th in FIP-enhanced.
Again, Cole wins either way. Just as in 2022, we had a huge disconnect between Nola/Urias based on whether it was FIP-or-not, we see the same with Eflin and Castillo. If you go with 50/50, Eflin and Castillo are tied for 5/6.
If we go with the more reasonable 75/25, Castillo drops to 5th, while Gray and Bradish move up one slot from the Classic Predictor.
Again, we can tie ourselves into knots here. So, I'll just stick with the Classic Predictor, and let's see what we learn in 2023. If we find that what Nola/Urias taught us is that FIP is in play, then let's make a new Predictor that introduces FIP into the official, and SINGLE, forecaster.
As basestealers: Acuna has 55 more SB (worth about 10 more runs) and 10 more CS (costing 4 or 5 more runs), for a net 5 or 6 runs. As baserunners (on batted balls): they are even.
Now for defense. Betts has played 55% as RF and 45% as an infielder. This makes it a bit tricky, so let's try it anyway. For outfield range, Betts is ahead by 7 runs. For OF arm, Acuna is ahead by 4 runs. So, a net +3 for Betts in the outfield.
Betts in the infield has been 2 runs below the average infielder. Of course, the average fielding infielder is much more valuable than the average fielding outfielder. Given the amount of time Betts has played there, it's roughly a 4 run adjustment needed. So, Betts ends up being 2 runs a better fielder than a neutral fielder, when evaluated for his infield play.
Add the two together, and Betts is ahead by about 5 runs as a fielder.
In other words, the baserunning/fielding pretty much cancel out, so you can just evaluate them as batters.
As batters: Acuna and Betts are both leadoff batters, so face a similar amount of PA with a runner on 1B and less than 2 outs. Yet Acuna has 10 more DP. That's worth about 3 runs in favor of Betts.
Acuna and Betts have almost identical rate stats. Their wOBA has Acuna ahead by 5 points (Acuna also ahead by 5 points in OBP and SLG). Truist is just slightly more conducive for batters, to the point that their wOBA, adjusted for park, are essentially even. That's why their wRC+ is 170 for Acuna and 169 for Betts. That makes Acuna ahead by 1 run.
So as a batter, Betts is ahead by 2 runs.
Playing Time: Acuna has come to bat 45 more times, which is worth about 5 or 6 runs.
Add it all up and Acuna is ahead by 3 or 4 runs, thanks to simply playing more. Put in a margin of uncertainty, and Acuna is a ahead by a smidge.
Suppose we want to tag a pitch when it's at least 3 mph above the league average. That seems too simple to do right? The league average fastball is 94 mph, so tag any fastball that is 97+. Case-closed. Or not?
Not. I did say pitch, not fastball. Ok, so then we look at league-average speed of sliders, and repeat. Rinse and repeat with cutters and curves and so on. Case-closed? No. A slider is not a slider for everyone. There is in fact an infinite number of classifications for a pitch: every pitch is as unique as a snowflake. That we've tried to compartmentalize them into a dozen groups does not mean that those groups are clear and distinct. You can have two pitches, same speed, same movement, and one would be called a slider, and another a curve. That's the reality that we live in.
Let me show you something cool. Well, cool for me, and if you are here, presumably it's going to look cool to you. This is the average pitch speed of ALL(*) pitches, based on their movement profile.
(*) See at the end where I talk about changeups.
So, a pitch that has backspin and tails arm-side, those average 94 mph. While those are typically fastballs, they could be cutters too. A pitch that has topspin and hooks glove-side, those average under 80 mph. While those are typically curves, they could be sliders too. The main point: we don't care what they are called, we just care how they move. And having established how they move, we simply figure out the average pitch speed of those pitches.
Now that we have the average speed of a pitch based on their movement (rather than some overarching pitch type), we can evaluate each single pitch, one at a time. We determine the movement profile on a pitch by pitch basis, look up the average speed for such a pitch, and compare it to the actual pitch speed thrown by that pitcher on that pitch. If the actual speed is 3+ above that baseline, great, we'll tag that pitch as being hot.
Let's look at all those pitches that were classified as four-seam fastball since 2020, and see how often each pitcher threw a hot pitch. (Click to embiggen) It should surprise no one that 100% of Jhoan Duran's fastballs were hot. You will see a bunch of familiar names. Plus Garrett Richards. Now, how does someone who averages 94-95 mph on their fastball end up with over 90% of those pitches being hot? Well, Garrett Richards may have these pitches classified as fastballs, but they could just as well be cutters. His fastball is thrown with a tremendous amount of gyro action, and so, it ends up being a cusp pitch. This is a good exception of what happens when you compartmentalize things. But in our Heater process here, it's irrelevant: all we care about is how it moves. And based on how the pitch moves, the MLB average pitch speed for that movement is 89.4 mph. And so, his fastball being 94.5 mph will end up being hot fairly often. And in this case, it's over 90%.
We can look at any other pitch type as well. Let's look at cutters. And here we can see a reverse-issue with David Robertson: while we call his pitch a cutter, and it would seem it's pretty high pitch speed at 93mph for a cutter, it's not a hot cutter, as he almost never gets above the threshold. And that's because the movement profile of his cutter is closer to a four-seamer. When we come to these cusp pitches, and we have to choose one and only one pitch type, we are basically stuck. Robertson is a cutter that moves more like a fastball (left image), Richards is a fastball that moves more like a cutter (right image). But, it's irrelevant for what we are doing here.
The top Heaters in 2023:
+10 Clase (Cutter, hot 100% of the time)
+8 Duran (4-seamer, 100% hot)
+8 Kelly (Curve, 99% hot)
+8 Ashcroft (Cutter, 99% hot)
+7 Kelly (Slider, 99% hot)
+7 Aroldis (Sinker, 99% hot)
+7 Ben Joyce (4-seamer, 100% hot)
(*) So, changeups. They are problematic. For the moment, I've accepted the classifications for changeups, splitters, forkball, and screwballs, and removed all those pitches from everything I did above. I should do something more objective, but for now, this will do. I will note that changeup speeds are not based on their movement profile. And so, how a changeup moves doesn't really matter in terms of whether a changeup is hot or not. Even the very idea of a hot changeup seems odd to begin with. In any case, just pointing out that these pitches are problematic, and I'll get back to it later.
As we know, platoon splits by handedness are a real thing. RH batters much prefer to face LH pitchers, than they do RH pitchers. And the same is true with LH batters much preferring to face RH pitchers than LH pitchers. This was most obvious when LHH Larry Walker came to bat to face his friend and former teammate LHP Randy Johnson. Johnson has a sidearm kind of delivery, and so when his slider comes in hard on the LH batter, and breaks away, it becomes an impossible pitch to hit. So, same-handed sliders are great for pitchers and terrible for batters. We'll talk about that in a minute.
Now, is it necessarily true of ALL pitches thrown. Pitches break in or away, or stay up, or drop down. They come in at different speeds. Surely not every pitch has the same platoon advantage. And is it even possible that some pitches have a reverse platoon split?
Glad you asked.
Changeups and Curves
Righthanded Starting Pitchers show large reverse platoon splits with their changeups and curves. This means that RH batters perform better than LH batters against RH starting pitchers, when those pitchers are throwing changeups and curves.
RHH are around +.18 runs per 100 pitches against RH Starting Pitchers against the changeups and curves, while LHH are around -.14 runs per 100 pitches against those same pitchers. That is a reverse split of .32 runs. The normal split for a RHP is .19 runs. So, this shows a net effect of .51 runs from the average split.
RH Relief pitchers also show a deviation of this extreme.
As does LH Starting Pitchers. And LH Relief Pitchers.
While the normal platoon advantage for the pitcher is against same-handed batters, this is not true with the curves and changeups. And so, pitchers should be throwing far more curves and changeups against opposite-handed batters.
So two questions: why is this true? And do they?
Changeups and curves are the slowest of the pitches thrown. Since changeups tail away from opposite-handed batters, the likely conclusion is that the pitchers like to tail slow pitches away from the batter.
On the other hand, curveballs hooks into opposite-handed batters, but also have a big drop. And so the likely conclusion is that those batters might be more handcuffed on those dropping pitches that break-in.
Sliders
How about the other pitches? Sliders show a tremendous platoon advantage, which is basically why most LHH would sit a game against Randy Johnson. This advantage is most true for LH relief pitchers. LH batters are minus 0.88 runs per 100 pitches, while RH batters are +0.59. This is a massive 1.47 run advantage. While LH pitchers enjoy a tremendous platoon advantage to begin with (+0.70), this far exceeds that. So, it's pretty clear here: LH relief pitchers need to throw their sliders against LH batters (or more generally, there's a tremendous same-handed advantage with the slider).
Again, same two questions: why is this true? And do they?
Sliders tail away from the same-handed batter, but not as much as the curve. And it drops somewhat, but not as much as the curve. But it's thrown some 5 mph faster. So, it seems a bit strange that a pitch that follows a similar kind of path as the curve, but faster, would suddenly have a very very different platoon split. I suppose it's really about the drop of the curve, and that supercedes the horizontal movement. Sliders can be considered as a reverse-changeup. And so, just as the slider confers an advantage to the pitcher against same-handed batters, then the changeup confers an advantage to the pitcher against opposite-handed batters (aka reverse split advantage).
Since sliders and curves are in roughly a similar family of pitches, it would seem almost a necessity that a pitcher develop both a slider and a curve. And they should throw the slider predominantly to same-handed batters, and curves to opposite-handed batters.
Do they?
Max / McCullers
Max Scherzer is a good example of a pitcher who gets it. Against same-handed (RH) batters, he throws the slider 39% of the time, and the curve only 5% of the time. This is since 2020. Against opposite-handed (LH) batters, his slider is all the way down to under 1% and his curve is up 15%. Max has figured out the platoon advantage. He makes up for the gap by throwing his cutter almost exclusively to opposite-handed batters (18% of his pitches against LHH are cutters, and only 1% against RHH are cutters).
The most extreme example is Lance McCullers. Against same-handed (RH) batters, he throws 36% sliders and 6% curves. Against opposite-handed, it's 6% sliders and 48% curves. So, McCullers has also figured it out: throw curves against LHH and sliders against RHH. And where does he make up the difference? The changeup, which we also learned has reverse platoon-splits. And, McCullers must have figured it out on his own: 10% of his pitches against same-handed are changeups, while 21% against opposite-handed are changeups.
Ok, so Max/McCullers have figured it out. Have all pitchers? Surprisingly, no. There's very very few who have not, but the one who stands out is Zack Wheeler. He's at 20% slider and 9% curve against same-handed (RH) batters, and it goes UP to 29% sliders (and 14% curves). Now, how is such a successful pitcher able to pull this off? In his case, his "slider" is actually the traditional gyro-slider as well as the sweeper-slider. It's most obvious if we look at it this way (click to embiggen), with Wheeler on the left, and Scherzer on the right (2023 data).
Wheeler uses his sweeper-slider the way Scherzer uses his slider. They use their curve the same. And Wheeler's gyro-slider is more akin to Scherzer's cutter. So for all those folks asking why split up sweeper-slider from gyro-slider: we're just reflecting what MLB pitchers are doing and how they are behaving. We're not inventing anything new here. We're simply better tagging and organizing data. Without doing that, it would seem that Wheeler was clueless as to the splits of sliders. Instead, he is totally telling us that he gets it, the same way that Max gets it.
So chances are, if I were to go thru the very few pitchers who "didn't get it", they likely have some nuance to their pitches that shows they almost certainly get it.
Last year, I ran a correlation of Framing for Savant and Steamer to next-season's Savant and Steamer. The idea is similar to FIP and ERA, and correlating to next season's ERA. When FIP predicts next season's ERA better than ERA itself, this immediately tells you that ERA has alot more noise and/or that FIP has a more precise accounting of a pitcher's performance. In the case of FIP and ERA, both are true.
With Framing, the conclusion was inescapable: Steamer is better, and Savant is best characterized as Steamer-lite, adding no new information. And this is as expected. After all, Savant uses larger bins, and considers less variables. The only thing that was left to determine was if Savant did anything special that Steamer did not. As of 2022, it did not. I can tell you it DID provide extra information in the past. But after I told the developer of Steamer about this insight, he incorporated a change to close off this gap. There is of course a benefit of having a lite version, or a Naive version. Chasing every last decimal place, every precision point, for extra complexity has a cost to it. That's why having a Naive version is helpful. So, if you had to use one, use Steamer. But if you had to use something simple to explain, use Savant.
FIELDING
Ok, so now I turn my attention to Fielding. Fangraphs makes life very easy, with data from UZR and DRS readily available, side by side with OAA. So, I pulled all that data from 2016-present (excluding catchers). I looked only at players with at least 300 innings for each position, year to year. Since all three methods have their own distinct approaches, one would think that each will be able to provide extra information. Did this happen?
UZR
Let's start with the clear laggard, UZR. Year to year correlation of UZR is at r=0.31. Taking OAA of year T and correlating to UZR at year T+1 gives us r=0.29. This is almost a FIP/ERA situation. OAA knows nothing at all about how UZR is constructed. And yet, using UZR in year T and OAA in year T are almost identical in helping explain UZR in year T+1. DRS is almost as helpful as OAA, with an r=0.27, in explaining next season's UZR.
And what happens if we put all three metrics together in year T, in explaining UZR in year T+1? In that case, DRS almost completely disappears. A weight of 20% UZR, 13% OAA and 2% DRS is what you need to explain UZR in year T+1. So, whatever UZR is trying to do, OAA is able to find it somewhat, and DRS add almost no new information.
DRS
Ok, how about DRS? The year to year correlation of DRS is r=0.42. Using OAA from year T to predict DRS in year T+1 gives us r=0.39. So, again, OAA is able to predict next year's DRS almost as well as current year's DRS. And if we consider all three metrics to explain next year's DRS? Well, that 28% DRS, 28% OAA, and 2% UZR. In other words, OAA and DRS are equally valuable in explaining next year's DRS.
Why would that be? Well, like with Steamer/Framing, OAA has insights that DRS simply does not have available. And those insights have value in explaining the fielding talent of the player, and so, explains next year's performances.
OAA
Finally, OAA. Year to year correlation for OAA is r=0.53, which is much higher than the others. DRS has a correlation of r=0.34 to next year's OAA, while UZR is r=0.27. That number, the 0.27, establishes basically a minimum-type of correlation level. UZR and OAA do nothing the same, though they naturally are based on the same plays. We'd get a correlation based simply on that.
Anyway, what happens when we put all three metrics together? Well, the correlation remains at r=0.53, and the weighting for DRS and UZR are each close to zero. In other words, each of DRS and UZR provide no new information in trying to explain next season's OAA.
If we set the replacement level at .300, here's how the top 3 batters look:
7.7 Mookie B
6.8 Freddie
6.8 Ronald A
That .300 win% level is somewhat arbitrary, but there's a good reason for it. But, let's say someone thinks it should be .200 win% as the zero-level. How does the race look?
8.6 Mookie B
7.7 Freddie
7.7 Ronald A
All the numbers go up by 0.9 wins, but otherwise, the disparity is the same.
How about a .400 win% level as the zero-point, the point above where value comes? If you believe that, then this is your list:
6.8 Mookie B
5.9 Freddie
5.8 Ronald A
Looks pretty similar still. How about .500, league average. Let's only count the wins above average. How does that look?
5.9 Mookie B
4.9 Freddie
4.9 Ronald A
Uhmmm... what's going on here? Why are we seeing virtually no difference regardless of what we choose as the zero-point?
At its core, WAR, or more specifically, Wins Over Readily Available Zero-Point (WORAZ), is this:
Wins = (wOBA minus Zero-Point) times PlayingTime times someConstant
Since Betts, Freeman, and Acuna all are full-time players, their Playing Time is very similar. That variable cancels out. The constant cancels out. The Zero-Point also cancels out. And so, what we end up caring about is the individual rate stats of a player. Here I am using wOBA, but it would be some all-encompassing rate stat.
So, no matter what zero-point you choose for your WAR stat, it doesn't matter. Well, at least in this particular case.
When does it matter? It matters when you want to compare players of different playing time. A .400 wOBA player with 400 PA compared to a .350 wOBA player with 500 PA compared to a .300 wOBA player with 700 PA. It's that combination of differing quality v quantity where WAR shines.
Is there a player in 2023 where it matters, some player that played well when they did, but didn't play all the time? Yes, in 2023, that player is Jose Altuve. If we set the zero-point at a .500 win%, meaning value only accrues for above-average performance, then Altuve has 2.7 wins, and puts him at #16 in MLB, just ahead of full-time players Matt Olson and Marcus Semien.
Of course, some of you out there think that's silly, there's no way Altuve in 2023 is equal to those two players. Ok, so let's say your zero-point is a .300 win%. Altuve is now 37th, just behind Christian Yelich. Some of you may think that's around correct. But others are not convinced, thinking that is STILL too high to put Altuve.
A few months back, I got interested in interference calls, as Ohtani was leading the league in getting those calls as a batter. Which of course led me to look at it by catcher. Some catchers, like Blake Sabol, were having an enormous number of interference calls, while some had little. Indeed, Salvador Perez has zero interference calls in his entire career. Which naturally led to: where are these catchers located in the catcher's box?
With Statcast tracking, we can actually determine the location of every fielder at every moment of a play. Not only the fielder location, but their specific joints. Those familiar with Gameday 3D are already aware of this. This chart (click to embiggen) shows the average location of the right shoulder of each catcher, in number of inches behind the back tip of home plate, at pitch release. As you can see, the range is quite wide from 57 inches to 72 inches. The question is how interference calls correspond to the location of the catcher.
We can see that interference calls decrease the farther back the catcher is located. Which is obvious enough. At about 66+ inches is where the interference calls are minimized.
Now, why would catchers try to stay closer to home plate? Well, in a word: framing. This chart shows the framing value of each catcher, relative to where they are located. And you see a clear trend: the closer the catcher is to the plate, the better his framing numbers. The relationship is about 1 run per inch.
So you can see why catchers would try to nudge their way closer to the plate, as each inch is giving his team a run in framing. But you can also see how each inch is adding interference calls, and each interference call is worth 0.3 runs (plus potential injury). Being a catcher is a very tough job.
As we know, a solo HR has much less run value (and subsequently win value) than a grand slam. This should be obvious, but since I know there are speed-readers out there, let me slow it down. A solo HR adds one run, while the grand slam adds four. Obviously, the grand slam relies on the three runners on base, so we wouldn't credit the batter with all four runs, but a portion of those runs (it's about 2 to 3 runs we give the batter). The solo HR naturally gives the entire run to the batter.
The same applies with a bases loaded walk compared to a walk with no runner on 1B. The latter adds about 0.25 runs, while the former adds exactly 1 run. The batter gets the full run for the bases loaded walk since his starting state (bases loaded) and ending state (bases loaded, no new outs) is identical except that one extra run scored.
Indeed, a solo HR and a bases loaded walk are identical in value. Most people don't like that. While we consume baseball in a very context-specific way, most folks evaluate players in a very context-neutral way. Hence, we give every HR a run value of +1.4 runs (which is about +.13 wins). And we give every walk a win value of +.03 wins, regardless of the base situation.
But we come across some peculiarities. A strikeout with a runner on 3B and less than two outs is much different than a long fly out with a runner on 3B. Pitchers love those strikeouts, while batters enjoy those long fly outs. On the other hand, a strikeout with a runner on 1B and less than two outs doesn't have the same impact as a groundout with a runner on 1B and less than two outs.
Overall, the win value of the strikeout is .026 wins below average while the win value of a non-K out is .025 wins below average. For all intents and purposes, an out is an out is an out. When aggregated. Individual outs naturally are much different in value, depending on the context. Suddenly, we might want to consider the context for some events, and not for others. That makes things confusing.
The Standard wOBA method is to assume context-neutral, and therefore, not consider the base-out states for any event. Which leads to this excellent article by Sam Walsh, with his focus on double plays:
We can of course consider also the sac flies (or long fly outs, as we will see).
What I did was my typical process of generating Standard wOBA by the launch angle and speed. I converted that into a win value, in a fairly straightforward way.
What I also did, which I never did before was calculate the actual WPA for each of those combinations of angle and speed. Our expectation is what Sam found above: that long fly balls are going to have a different impact than short fly balls, as well as hardhit groundballs would be different from slow groundballs (the slower ones would naturally give the batter more time to reach first base, and prevent the double play).
And so we can see the effect, I am generating the difference between the two. And here are the results (click to embiggen). To read the chart, 90 means 90 to 94.99 mph, while 32 means 32 to 39.99 degrees.
Those long flyballs that lead to long outs, those in the 24-48 degrees of launch at 85+ mph, that's what is driving those extra win values of +.003 wins (or 40 points in wOBA value). And you can see similarly for those hard groundballs at the lower launch angles, going the other way. So as Sam is pointing out ( https://sam-walsh.github.io/po... ), if you base xwOBA on trying to predict Standard wOBA, as opposed to context-specific runs or wins, you will miss out on the hidden value of outs primarily. But really, any event. After all, an infield single and an outfield single have different impact as well.
So, you have to be careful and decide what it is that you are after, how much context-neutral you want your stat, and how much context-specific you might need to make it.
This is why I am a proponent of RE24, since it does indeed capture that value. But I'm also a proponent of Standard wOBA because who wants to have 24 different wOBA equations? There's also game-state wOBA and there's thousands of those. You just have to choose your solution based on your question.
As readers of The Book know, I created wOBA as the central metric for data analysis almost twenty years ago. Its basis is the run values introduced by Pete Palmer in The Hidden Game of Baseball back in the early 1980s.
The scale of wOBA is that of OBP, where naturally the floor is .000 and they share similar averages (say around .320 or .330). The run values from Pete were centered so that the average was 0, and the run value of the out was around -.25 or -.30 runs. Having average = 0 was offputting for some folks, because zero is associated to nothing, or null.
Think for example of two pitchers with a league average ERA, one with 180 IP and another with 18 IP. The 180 IP pitcher will get paid more than ten million a year, while the 18 IP pitcher might not even make the roster the next season. Both would be shown with a 0-runs-above-average, which while technically accurate is not reflective of their value if you rely one ONE number. Showing them each with a .325 wOBA, with one facing 750 batters and the other facing 75 batters is much better: it shows their value along two dimensions, and so allows for different conclusions, all depending on how you merge the two values into one.
When you are already given just the one value, be it wins above average (WAA), or wins above readily available talent (WAR), or even a player's salary, the conclusion is now forced on you. You have no way to back out of it, since the one value encapsulates everything that went into it. There's nothing left. This is why for example I would much more prefer to show the above pitchers with an Individualized Won-Loss Record, aka The Indis. If one has a 5-5 record, and the other has a 0.5-0.5 record, well, now each of us can now do something unique with that. I can compare each of those to a .300 win%. You can compare each to a .400 win%. Someone else can compare each to a .500 win%. No one is beholden to anyone. We all agree on their accomplishments, the 5-5 and the 0.5-0.5. We don't need to agree on the value of those accomplishments. That's for you to decide on your own.
WPA
I actually didn't want to talk about any of that. I wanted to talk about win probability added (WPA), or win values. Since 2010, we've had over 70,000 home runs. The win expectancy when the batter stepped up to the plate was .520. After hitting the HR, his team had a win expectancy of .653. That difference, +.133 wins, is what is called WPA. That's the win value of the HR, relative to the average (which naturally is .000). So, you can see why I had to talk about average = 0, even though that was a long-winded way to get here.
We can look at the win expectancy to make the call, if you want to be more objective. That game was in the top of the 10th, with one out, and a runner on 2B. Getting a run-scoring single adds .26 wins, while a first-and-third single adds .10 runs (over and above the first-and-second default scenario). Let's say therefore a single would add +.18 wins. An out drops it by .09 wins. So, the breakeven is getting a basehit 33% of the time (which is a wOBA of .300).
To think of it more simply: an IBB is typically win-neutral. Whether the batter gets an IBB or is allowed to be pitched normally, the end result is typically the same. So, all that Cabrera has to be thinking is if he can match his true talent wOBA. Assuming he's .400, that's what he needs to be hitting. And getting a hit 45% of the time is a .400 wOBA. So, if Cabrera thought in the moment: I can get a hit here at least 50/50, I am swinging: then tip of the cap for being faced with a situation that you likely never prepared for, and your instincts took over.
We get the unsurprising results that the faster your swing speed, the better your production. The data represents the run value (per 100 pitches) on swings, based on the speed of the swing (relative to that batter's seasonal average). It includes hit into play, fouls, and whiffs.
Now, before you go out there and just swing as hard as you can, regardless of the incoming pitch, you have to understand the selection bias at play here. In a real MLB game, a batter always has the option to slowing down his swing. So, it's very possible that those times when he is +10 mph above his seasonal average, it's because he perfectly timed his swing to exactly the pitch and location he was expecting. When he is 10 mph below his seasonal average, almost assuredly he mistimed his swing, and/or tried to (unsuccessfully) check his swing, and so on.
We measure the swing speed at the impact point and so, all those factors play a role in the chart you see here. The ideal experiment would be to have a batter who maintains his swing throughout, regardless of the incoming pitch. Of course, if he did that, he would not be in the majors too long. So, as much as we'd like to think that swing speed is an input to the entire swing model, the reality is that the swing speed, as we're capturing, has a selection bias associated to it. And therefore, is not some independent input. As one data analyst noted forty years ago: let's be careful out there.
One thing that is apparent is that swings that result in balls hit into play, and those that are foul balls or foul tip are based on similar swing speeds. This suggests it's just a matter of timing.
You can also see that the no-contact swings are often slower, especially in the case of Joey Gallo.
The speed numbers you see in the title is the average swing speed for the 90% fastest swings of each batter. This is to handle the presumed checked swings.
Bonus: Ohtani as a batter, and Ohtani as a pitcher. (click to embiggen)
We will be able to slice and dice the swing data to tell some interesting stories. To give you some snapshot of the data, we can focus on Luis Arraez and Luis Robert. (Click to embiggen)
It looks overwhelming, but let's break it down. This has all the swing data we've tracked. For Arraez, it's 709 swings and for Robert, it's 852. So, add 20% to the counts for Arraez to make an apple-to-apple comparison.
The first thing to jump out is that the number of no-contact swings (whiffs) is very low for Arraez (42 or pro-rated as 50), while it's 267 for Robert.
The second thing that jumps out to me is that Arraez has a very low swing speed on balls hit into play (68 compared to 81 from Robert), but he squares it up much higher (94% to 82%). You can also see he chokes up on the bat and/or he has a shorter swing, with the bat head 31 inches from the rotational point, while Robert is at 33.5 (around league average). You can also see it in the attack direction, where Arraez is +7 degrees, meaning the bat slices the other way, while Robert is -4 degrees, meaning he comes around more.
You will see Arraez has a missing row: he has no swings where he is tied up, while Robert has 8.
Also compare each of their hit-into-play with their strike-fouls: all their own data looks very similar, except for the Squared Up part. That is to be expected, as there's a reason the ball went foul rather than into play.
All this data will have use for two distinct groups of folks. The first is the player themselves, so they can appreciate or understand their profile. While obviously every player knows themselves, they may not know the extent of their profile, nor how much their profile may change year to year, or even how it compares to other batters. So, as a scouting tool, this data will be tremendous.
The other group is for the fans who like to think in terms of the talent of the player, trying to forecast their performance. As we get more data in the years to come, we'll be able to compare year to year changes and see how that affects their performance.
The convergence of Scouting and Performance Analysis is well upon us.
Recent comments
Older comments
Page 5 of 151 pages ‹ First < 3 4 5 6 7 > 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