Here are the updated projections for game three!
When a dominant pitcher like Tyler Glasnow faces a high-flying offense like the Dodgers who has the advantage? When an analytically-minded team like the Rays rolls with an opener like John Curtiss, what is gained and what is lost? How long should the starter pitch? What are the most dangerous moments he’ll face? And why ever is Cody Bellinger batting sixth?
One of the better way to answer questions with these is with handedness split-based matchup projections, and for the World Series I’ve changed the belts and put a fresh coat of paint on an old tool.
Here are the answers.
Let’s hit some of the details.
How does the projection work?
If you already know the concept, or if you just want to look at a lineup with different shades of colors laid over it, you can skip this part.
The Handedness Split
In general, batters perform better against opposite-handed pitching than they do against same-handed pitching. This is called their “handedness platoon split,” “handedness split, or “platoon split.” It is not the only kind of split. It is not the only type of split that matters in baseball, but it is the easiest to study and the most commonly discussed.
Almost all batters display some type of handedness split, but they don’t all struggle with same-handers or succeed with opposite-handers at the same rate. As with everything else in baseball, it’s not good enough to just take a look at the last several plate appearances, or the past month, or the past season and simply take what we observe as the true talent of a player and the expected performance going forward. Instead we have to interrogate the sample and separate signal from noise. Luckily, when it comes to handedness platoon splits, smart people have already done that work.
I first saw this analyzed in the sabermetrics classic, The Book, and then saw that research updated by Bojan Koprivica in 2013. This implementation uses the numbers from Koprivica’s work. To create a matchup projection, I’ve calculated the observed split a player has shown in his career and then regressed that split a set amount toward the league average split as per Koprivica’s work (note that lefties have larger splits than righties, on average). I’ve then redistributed that regressed split in a ration based off of career plate appearances against each handedness around the Rest of Season (RoS) projection from FanGraphs Depth Charts, which is a combination of ZiPS and Steamer.
Put simply, this method asks and answers three questions:
- How good are the involved players overall? Check the Depth Charts projection.
- How wide is their split? - Check the career split and regress it to the mean.
Then it combines those two answers into a projection for each player vs an average lefty or an average righty. Then it combines those two projections (using the log5 method) into a single projection for the matchup, which it then sets in the correct park (but note that for this season I’m calling Globe Life Field a neutral park).
The Times Through the Order Penalty
To understand what’s going on in modern baseball pitching management (of which both the Rays and the Dodgers are pioneers and adherents), it’s important to understand the “Times Through the Order Penalty” (TTO) If you never did when it was new, stop and read MGL’s articles about it here, here, and here, then follow that up with his work here, here, and here on whether it’s smart to draw conclusions about future pitcher performance in a game based on how a pitcher has performed so far in that same game (short answer: it’s not).
To better help follow in-game managerial decisions, I’ve built the TTO penalty into these projections. For relievers, their first time facing a batter will be their mean projection, after which each TTO will add an additional 2.5% of wOBA to their projection. For starters, their second time facing a batter will be their mean projection, while the first TTO will be 97.5% of the mean, and the third and fourth TTO would add an additional 2.5% to the projection.
This is a good implementation of good research, and I’m generally happy with it. Nevertheless, there are some aspects of the implementation that I know are incorrect, but I’m running with anyway:
- It’s not actually correct to spread the platoon split for a player around the projection based on the ratio of career PAs against opposing left-handed or right-handed players. What we want is to use the same ratio that the projection itself was based on. But I don’t know what ratio Steamer and ZiPS assumed, so career numbers will have to do.
- In a similar vein, the reason I have to subtract some wOBA projection from starting pitchers is that the Depth Charts projections know that they’re starters (or bulk guys), and so they’re already accounting for a TTO penalty in their overall number. But this, like the handedness split is not uniform. Some pitchers are expected to go one and a half or twice through the order, and some are expected to go three times or into the fourth. The projections may or may not know who’s who (I’m really not sure), but I definitely don’t, so the TTO penalty is a blunt aggregate. I’m sure Dan Szymborski could do this part better. Maybe he already does.
There are some worthwhile things that these projections by their nature are going to miss. They are:
- Other types of splits. There are lots of splits in baseball! Groundball vs flyball. High velocity vs junkballer. Pitch type groupings. Release points. All of these matters, and they’re things that teams use. They’re harder to look at though, so right now we won’t.
- Changes in the true talent level of the split. Because I’m using the career split to get a useful sample size, these projections will miss on any player who either significantly narrows or widens their split.
- Dramatic changes in overall true talent level or projections of young players, both of which that the projection systems struggle with. Is Randy Arozarena a slightly above average hitter right now (.327 wOBA)? Probably not, but that’s what Depth Charts has for him, because the magnitude of his current breakout is insane and unexpected. What is Shane McClanahan? Who knows.
Okay, here’s the fun part. Let’s start with comparing the two lineups against average pitchers.
The shadings are distributed around average, with magnitude of shading determined by the total set of batters. For reference, league average wOBA is 0.320, league average ERA is 4.45.
First, here’s each projected lineup against an average righty.
The Rays are a decent lineup. If they were to sacrifice some defense to get Yoshitomo Tsutsugo in there from the start they’d look better. But the Dodges are, frankly speaking, insane. Cody Bellinger and Corey Seager far outstrip the Rays best anti-righty hitter, Brandon Lowe. They have three other hitters (Mookie Betts, Max Muncy, and Joc Pederson) in Lowe’s neighborhood.
Now here are the two lineups (or close enough), against an average lefty.
The imbalance is the same. The Rays are still okay (and can get a little bit better if they get Mike Brosseau in there), but the Dodgers are on another planet (and also can get better depending on what they do to get Enrique Hernandez in the lineup).
The main difference between the lefty and righty lineups is that versus lefties, the biggest danger man is leadoff man Betts. Against lefties he’s the “slumping” Bellinger, who has been dropped to six, where he is more likely to face a Rays lefty reliever the third time he comes to the plate.
Quite candidly, it’s game time, and I’m out of time. Will come back and update this through the night, but for now, here are the bullpens:
The relievers are at the top, the starters are below. Red on the name means they’re probably unavailable, yellow means that if they pitch it will probably make them unavailable tomorrow.
And finally, here are the starting lineups vs today’s pitchers.
Clayton Kershaw is pretty good. Tyler Glasnow will get two times through the lineup, and either be lifted before he faces Seager or Bellinger for a third time.
Game time, go Rays.