Magnifying Class: Analyzing NLDS Decisions

I wrote last week that it’s hard to analyze playoff baseball, and that’s largely true. The playoffs are a small-sample bonanza, and we simply don’t learn enough new things over the course of five games to say much interesting about a player. Hey, did you know that Marcell Ozuna and Paul Goldschmidt are good hitters? Did you know that Jack Flaherty is a good pitcher? I thought so.

That said, it doesn’t mean there’s nothing to talk about. I built some fun tools this week over at FanGraphs to analyze intentional walks and lineup decisions, and I thought I’d turn them on the decisions Mike Shildt made in the NLDS. I’m going to highlight four important moments in the series where a manager could make a difference one way or another and see what Shildt chose, then look at whether they made sense from an analytical standpoint.

Of course, an analytical standpoint isn’t the end-all be-all, but it’s a great starting point for analysis; decisions that are close might have harder-to-quantify factors that can swing them, whereas decisions that are obvious one way or the other probably can’t be swung by looking into someone’s eyes and seeing if they have it.

That Intentional Walk

I’ve written about this one already, but here’s a quick refresher, because this is undoubtedly the most talked-about decision Shildt made. He got some criticism for it, and I understand that; walking the winning run on base and then having the guy you chose to pitch to rip a double on the first pitch doesn’t feel great.

My conclusion is that it wasn’t as bad as it looked. I still wouldn’t have done it, but I found the decision to be well within the soft-factor range; if Carlos said he didn’t feel like he had his changeup, the pitch he’d want to challenge McCann with, that would make the decision right. This one came down to platoon advantage, which we underrate, and Carlos’s platoon splits being particularly wide.

It doesn’t make sense, in our minds, that a right-hander with an elite fastball and wipeout changeup would have trouble with lefties, but Carlos invariably has over his career, and that’s enough to swing the decision to be a lot closer than it first looks. In any case, Shildt got burned by his decision, but I don’t hate this one, even though I dislike intentional walks overall.

Leaving Waino In

Because of the way baseball works, this decision gets less airtime than the intentional walk; it worked and the walk didn’t. Let’s talk about it, though, because it was also a pivotal moment. Wainwright had thrown 103 pitches in the first seven innings, and that was already curious in its own right; he’d exceeded that pitch count only four times all year, and three of those were still under 110 pitches. Leaving your league-average starter in the game for seven innings and more than one hundred pitches is hardly normal postseason managing.

In any case, that had already happened. Ignoring the even crazier decision to leave him in after allowing a baserunner, let’s talk about letting Wainwright pitch in the eighth. The Braves would be sending Brian McCann, Dansby Swanson, and a pinch hitter to the plate; anything past there would be the top of the order, and here there be dragons of the Ronald Acuna variety.

In his career as a starter, Wainwright has relatively normal times-through-the-order splits; a .291 wOBA the first time through, .294 the second time through (this increase is smaller than normal), and then .315 the third time or later. A handy rule of thumb is that if you have an estimate of a pitcher’s true talent level, that’s about how well they’ll do the second time through the order. Then, painting with a broad brush, you can make the pitcher 10% better (in ERA and FIP, 5% in wOBA) the first time through, and 10% worse the third time through.

So if you think Wainwright is a 4.38 ERA pitcher right now (FanGraphs’ projections), that works out to a 4.82 ERA guy the third time through, or something like .57 runs per inning after accounting for unearned runs. I made a rough distribution of runs allowed that work out to .57 runs per inning, and applied that to the Cardinals’ win probability. More than half of the innings are still scoreless, so that’s good, but the other ones, you know, aren’t. In the end, letting Wainwright pitch dropped the team’s win probability from 73.4% to 71.6%. That’s not a big loss, but burning 1.8% of win probability isn’t a great idea.

The general idea behind this analysis is that Wainwright the third time through lowers your odds of winning because third-time-through Waino is worse than an average pitcher. Your replacement wouldn’t be an average pitcher, though; let’s say instead that it was John Brebbia. He projects to allow .44 runs per inning. Plug in a distribution for him, and the Cardinals were 75.3% to win if he had entered in the eighth.

So, in this sense, leaving Wainwright in cost the Cardinals 4% of a win. That might not sound like a big deal, but that’s as much of a win probability hit as Dave Martinez’s inexplicable decision to walk the winning run aboard two days earlier. This doesn’t even get into the fact that he left Wainwright in to face Ronald Acuna for a fourth time (yeah, don’t do this) and then Ozzie Albies, a switch hitter, with a runner in scoring position. This, to me, is Shildt’s worst move of the playoffs.

Letting Hudson Bat

The Cardinals had to win Game Four. Even with a lead, it was a fraught situation for them; Dakota Hudson, never the best of starters, was looking shaky. Through four innings, he’d allowed only a single run, but he had only two strikeouts (one of Dallas Keuchel on a foul bunt) and had allowed three hits and two walks already. The hits weren’t disqualifying, of course, but two walks and only two strikeouts isn’t ideal, and he wasn’t going to have any more free outs; Keuchel was done after only ten outs, having given up three solo shots.

So let’s set the situation; Hudson comes to the plate with a two-run lead, runners on first and second, and two outs. Talent-wise, he’s probably the worst pitcher the Cardinals will use as a starter this postseason, and the team knows that; they started him last, after all. It’s a must-win, empty the chamber kind of game, and there’s a day off tomorrow. What’s more, the bullpen is rested; due to the aforementioned use of Wainwright the day before, only Andrew Miller (two pitches) and Carlos Martinez (26 pitches) had even thrown in the last two days.

So let’s talk about the Cardinals’ win expectancy with Hudson at the plate. With an average batter going against an average pitcher, the Cardinals would win this game 77.2% of the time. Hudson is very much not an average batter, though; he’s a pitcher who happens to be holding a bat.

I have absolutely zero idea how good of a hitter Dakota Hudson is, so I simply used the batting line for all pitchers, though his .077/.094/.077 career line isn’t a thing of beauty. So with a generic pitcher facing Luke Jackson, after applying some statistical voodoo, the Cardinals were 75.8% likely to win. Keen mathematical minds will note that 75.8% is less than 77.2%, though not less by a lot; honestly, the Cardinals just have their two-run lead hold up enough that it’s not the highest leverage situation imaginable.

Still, let’s plug in a pinch hitter! The bench wasn’t particularly stacked that day, with Carpenter in the starting lineup, but they still had Jose Martinez. Plug him in, and they win 77.5% of the time; that makes sense; he’s a better-than-average hitter, even against righties. Okay, so Shildt cost the team 1.7% of a win by not pinch hitting for Hudson. Did he make it up by the extra innings he could expect to get out of Hudson?

Why, of course not, friends. Of course not! Dakota Hudson is worse than the Cardinals bullpen. He was about to face the top of the Braves order for a third time, which as we learned above is bad, and it’s not as though the team was in desperate need of three outs of Dakota Hudson at this point.

Let’s work out a theoretical rest of the game, leaving Carlos Martinez out (though the team didn’t) but allowing for Miles Mikolas to get his throw day in on the mound should the game go into extra innings, as actually happened. One from Brebbia, one from Gallegos, one between Miller and Webb doing their lefty voodoo, and one from Helsley; that gets us to nine. If something doesn’t work out, all of Miller, Brebbia, and Gallegos are candidates to get an extra out or two, and Martinez is available.

This probably won’t shock you, but none of those pitchers, with the possible exception of Webb facing a righty, project worse than Hudson. Hudson’s a totally acceptable starter, don’t get me wrong; he’s also a starter, as opposed to the high-octane relievers I’ve just named. Maybe his ERA was better this year, but his underlying statistics don’t tell that story, and the team placing him fourth agrees with the underlying statistics.

Naturally, Hudson gave up three runs and didn’t last another inning. The Cardinals came back to win, but Shildt burned a chunk of a win there, and he did it without good reason.

Starting Carpenter Against Keuchel

Let’s finish on a batter note. The Cardinals made the decision to get Matt Carpenter, professional hitter and unprofessional fielder, into the lineup against Dallas Keuchel. On one hand; don’t start no-glove lefties against other lefties! On the other hand, Matt Carpenter mashes, and Harrison Bader doesn’t. It has to be at least a question, right?

Again, let’s go to the math. In this game, we know upfront Keuchel won’t be throwing deep into the game as he’s going on short rest, so I’m going to give Carpenter two plate appearances against Keuchel and two against a generic, platoon-ignoring opponent. I won’t apply a rest penalty to Keuchel, both because I don’t know what it should be and because I’m also calculating Bader’s projected offense against the same pitching.

Carpenter projects as a .320 wOBA hitter against lefties, while Bader checks in at .315 against lefties, which means that yes, Carpenter is better in the generic case without the handedness advantage than Bader is with it. That’s not surprising, because Matt Carpenter is a phenomenal hitter. When you add in odds ratio math to account for Keuchel’s contribution to the whole mess, it looks worse for Carp, because Keuchel displays strong platoon splits.

In all, that makes Carpenter a .286 wOBA guy against Keuchel and Bader a .300 wOBA guy. Why is this the case? Well, basically Keuchel kills lefties, even for a lefty. His projections work out to a .275 wOBA allowed against left-handers, miles lower than the league average lefty-lefty result of a .308 wOBA. Carpenter, meanwhile, is only slightly better than average, which means that Keuchel’s contributions matter much more in determining the outcome. At the same time, Keuchel is a tiny bit better than average as a lefty facing righties, while Bader is actually below average when it comes to righties hitting with the handedness edge, so there we are.

Of course, the two plate appearances that don’t feature Keuchel are massively the other way. Matt Carpenter is a much better hitter than Harrison Bader overall, to the tune of a projected 39-point edge in wOBA. One of wOBA’s big benefits is that it can be converted into runs pretty easily, and the math works out to an edge of roughly .05 runs per game from starting Carpenter.

Can that make up for the cascading defensive downgrade the team gets from putting Carpenter in the lineup? In a word: not-even-close-and-so-not-even-close-that-I-needed-a-compound-word-to-describe-it. Let’s do some lazy chaining here: Carpenter is 4 runs per 150 games worse than Edman at third, Edman is a wash in right field with Fowler (and this is being generous, I think, given some of Edman’s routes, though he’s fast enough that it’s not crazy), and Fowler is, conservatively, 15 runs per 150 worse than Bader. That works out to .13 runs per game, and even if you want to devalue outfield defense with Hudson on the mound based on his groundball rate, I can’t find a way to lower that below .09 runs per game.

So in essence, Mike Shildt traded .04 runs of offensive gain for .09 runs of defensive loss. What does that cost? Per a lazy model I have built, the generic Cardinals are a .514 true-talent team with Hudson pitching in a neutral site. They’re .508 with the altered lineup. Against the Keuchel-led Braves, that dropped the team’s odds of winning a home game from 49.4% to 48.7%. That’s not a huge dropoff, but it’s .007 wins burned, and that is probably more generous to Shildt than I needed to be; I could easily redo the math to have the team get Carpenter an at-bat or two without starting, and to pull Bader if they fall behind, which would make the offensive gap even smaller.

In summary, I didn’t love Mike Shildt’s managing decisions this series. None of them were absolute howlers (except letting Wainwright pitch, which was oof not great), but he consistently erred on the side of more innings for starters in situations where that didn’t necessarily make sense. He also gave up the platoon advantage willingly, though it ended up being a close call because Matt Carpenter is, indeed, a great hitter, even in his current diminished state.

Overall, these four decisions worked out to about .07 wins of value that Mike Shildt donated to the Braves. That doesn’t sound big, but it’s absolutely not nothing; that’s about as much value as Marcell Ozuna contributed with a solo home run in Game Four that took the lead from 1-0 to 2-0. Managers have tough decisions to make all the time, of course, but Shildt essentially un-hit a home run with these choices throughout the series.

Another way to think about this is that the Cardinals, by construction, had a total of .5 wins above average for the whole series, 2.5 wins being the average amount you’d get out of five games. The entire contribution of the whole Cardinals team was .5 wins, and Mike Shildt, on his own, was worth -.07. That doesn’t sound particularly good for him, and it shouldn’t; he doesn’t *have* to get better, because the playoffs are a tiny sample and anything can happen, but his decisions so far have been rough.

Note: the section on Wainwright has been updated to reflect a calculation error on times through the orders effect on ERA. The updated win probability is 1% higher for the Cardinals.