Prospectus Feature

The AL All-Stars

1. The most obvious tactic any manager in my situation would consider is pitcher usage. It is basically universally accepted, I think, that the best starting pitchers are far better than the best relief pitchers, and if converted to relief they would put up at least elite-closer numbers. So one approach would be to eschew the tradition of rewarding the league’s top closers and set-up men and go with all starters, who would amp up to even-better-than-elite-reliever status when used in one-inning stints. Jacob deGrom, for example, averaged 98.1 with his fastball in last year’s All-Star game, while averaging 95.1 as a starter all year.

But the other approach says relievers are better at relieving, and starters wouldn’t immediately reap the transition-translation bump. That Kershaw as a reliever would certainly be better than Kenley Jansen eventually, but it might not happen in his very first outing.

I hypothesize that the first approach is truer, but I’ll take a stab at answering this empirically, if not completely rigorously. I took every All-Star Game performance since 2001 and split the pitcher performances between typically starters and typically relievers. Fascinating results:

• Typically starters: 3.93 RA/9 (183 innings)
• Typically relievers: 3.96 RA/9 (98 innings)

Fascinating because, presumably without doing a ton of research, managers have successfully intuited just the right balance. They choose and deploy the best starters and the best relievers, but not an equal number of each—the number that actually give them the best total performance. If these group numbers diverged much, we would say that the managers should be deploying far more of one group. If the starters’ RA/9 was much lower, for instance, then many more starters should be used, or vice versa. There appears to be a sort of demonstration of savant-like genius here. Congrats, managers.

Now, a few notes, because, like I said, not rigorous. For one thing, typically starters probably face tougher batters, because they face the offensive starters. For another, there’s more to life (especially in 98-inning samples) than RA/9.

• Typically starters: 21 percent K rate, 16 percent K-BB rate, 3 percent HR rate, 10.5 percent whiff rate
• Typically relievers: 24 percent K rate, 17 percent K-BB rate, 2 percent HR rate, 11.3 percent whiff rate

So, getting past RA/9, we see relievers have actually put up better numbers; but thinking contextually, we assume that starters face tougher hitters. Let’s call that a wash. Let’s call this whole thing a wash, and say that we will use a mix of starters and relievers, just like they always do, because that actually seems to be best. Starters do become elite relievers in their one-off relief performances, but not any better than that.

(I’m recently reminded that Zachary Levine wrote a real smart and more rigorous piece on this topic a couple years ago, and he found for the relievers.)

2. But how long should pitchers stay in? If a staff has Clayton Kershaw and Jon Lester, and Kershaw is obviously the better pitcher, how many innings before Kershaw stops being the better pitcher? How many before he stops being better than Lester in one inning? Should I use twice as many pitchers at one-inning per, or half as many but at two innings per?

Again, let's try empirically. I could look at every pitcher who threw two innings in an All-Star game, but that introduces a selection bias—maybe they threw two innings because they were extra good in the first inning they threw, and maybe they would have gone two but they got hit hard in their first inning. So, to consider this, I’m going to split up my Typically Starters group into two groups:

• One of the first two starters used in a game
• Not one of the first two starters used in a game
• Not one of the first two starters, but ended up going multiple innings (in extra innings, e.g.)

I’ll throw the last group out, because they just complicate things. The first group is going to be my “multiple innings” group, because relatively regularly these pitchers are asked to go two innings. Sometimes they’re not asked, and they’ll pollute my data, but alas. It’s the best I’ve got. Sometimes they are asked, but they don’t end up going two innings because they sucked; they fit right into my data. And sometimes they do go two innings. Taken together, they’re an imperfect group of Typical Starters attempting to go multiple innings.

So,

• Multi-inning cohort: 3.17 RA/9
• One-inning cohort: 4.89 RA/9

Whoa!

Actually, it’s not that big a deal. The peripheral stats are much closer:

• Multi-inning: 20.5% K, 17% K-BB, 3.2% HR, 10.6% whiff
• One-inning: 20.5% K, 15% K-BB, 2.9% HR, 11.3% whiff

I’m comfortable throwing out the RA/9 as a weirdo SSS result, and conclude that there’s little to see here—which is to say, a lot to see here. Starters don’t get worse when they’re asked to throw two innings. So, if I see a benefit to it, I will ask mine to throw two innings.

3. I do see a benefit to it. I have a 34-man roster, and I plan to use all of it—mostly on my offense. So I’m going to carry only 10 pitchers. Those 10 are: Aroldis Chapman, Wade Davis, Sean Doolittle, Cole Hamels (my only Ranger) Rich Hill, Craig Kimbrel, Corey Kluber, Andrew Miller, Chris Sale, and Steven Wright.

The relievers are simple enough. Besides being good—the AL advantage over the NL in relievers is absurd—they’re platoon dominant. Over the past three years, Doolittle has been the toughest pitcher in baseball against lefties, with 32 Ks, no walks and a .39 WHIP allowed. Davis is the equivalent, the toughest against righties. I plan to get six or seven innings out of three or four starters, then play matchups with five relievers over the final two innings.

4. Steven Wright is my starter. There are two reasons for this. One is that the knuckleball hangover effect is real, and I want Wright to face hitters who are going to stay in the game for at least one more at-bat after he exits. That means he has to face the starters, who are the only batters who usually go multiple rounds, which means he has to start. He faces nine batters, then Chris Sale comes in to face them all again, while they’re still trying to get their eyes uncrossed.

The other reason Wright is on my staff, and in a prominent role, is based on a hypothesis that the knuckleball is the great leveler—that it gets hit sometimes, that it walks batters sometimes, but that the results are far more dependent on randomness than the quality of the batter. If this hypothesis were true, then Wright would be an excellent All-Star pitcher, because while every other pitcher gets much worse than his true talent (on account of facing nothing but star hitters), Wright would basically still be close to himself.

To test, I looked at Tim Wakefield in 2001 to 2005, his best five-year run. I took the 25 best hitters in the American League during those five years—from Jason Giambi at no. 1 to Paul Konerko at no. 25—and the 25 worst hitters, minimum 1,500 PA, ranging from Eric Hinske to Dan Wilson. Then I looked at how each cohort performed against Wakefield:

• Good Hitters: .326 OBP, .432 SLG
• Lousy Hitters: .295 OBP, .386 SLG

A difference, unsurprisingly, but relatively small difference, surprisingly. To put this gap in perspective, the same hitters overall:

• Good Hitters: .373 OBP, .515 SLG
• Lousy Hitters: .324 OBP, .395 SLG

The gap between the best and worst hitters is basically halved. With the caveat that these results might be flukes, and that they might not apply to Wright anyway, I can say with absolute certainly that Steven Wright is an All-Star Killer.

5. I would probably invest some of my personal funds into a statistical analyst to do better than what I just did.

In all seriousness, I’d invest in advance scouting of previous All-Star games. Is there a different style of play in All-Star games—pitch selection, swing aggression? I’d want to know, and I’d want my team to know. (h/t Zachary Levine.)

6. So I’ve got Wright going two, maybe three innings; futzing up the other team’s starters for their second plate appearance, against Sale or Kluber; then, after maybe one or two bridge innings, going to the greatest collection of relievers (seriously; I mean this literally) that has ever been assembled. That leaves a few details to go. For one, my catchers: None of them will have ever caught these pitchers before. Does that mean framing is no concern at all? Does it mean I should bring in the White Sox catchers (uh, no) or Indians catchers (er, really no) to try to get some extra familiarity strikes? Fortunately, there’s no real point to this. Jeff Sullivan found that pitcher-catcher familiarity has virtually no effect on framing. So I can trust that even with unfamiliar pitcher-catcher tandems, my framing won’t be affected. So I’ll just do what I would normally do: Get the best catchers available to me. In this case, that would be Sal Perez and Jason Castro. Sadly, my brilliant Steven Wright strategy forces me to bring Ryan Hanigan onto the squad to start the game. Fortunately, there are basically no good AL catchers anyway, so the dropoff is slight, and there's a decent chance I'll be able to pinch-hit for him in his first AB.

7. I’ve used 13 roster spots for my catchers and my pitchers. I have 21 for the other eight positions. Ideally, I’d like to bring three infielders from the same team to play together. While the average major-league batter gets shifted 13 percent of the time, the average major-league All-Star gets shifted much, much more. My best guess at Rian’s starting lineup, in fact, has been cumulatively shifted 44 percent of the time this year. I want infielders who are used to playing with each other, because All-Star tradition be damned, I’m shifting more than a kid standing at Cruisin’ USA with a pocketful of quarters. Alas, I can’t import my infield unit from one team. Maybe if I had the NL squad and could bring the Cubs infield over, but the best I can do here would be… Lindor/Kipnis? Altuve/Correa? Eh.

So forget that plan. Instead, I intend to play three different units. I’m going to start with my best hitters; defense doesn’t matter to me as much, because All-Star games tend to be high-strikeout affairs. So my starting lineup is, in some order: Miguel Cabrera at first, Altuve, Machado at shortstop, Josh Donaldson, Nelson Cruz in left, Mike Trout in center, Mookie Betts, David Ortiz DHing, and Hanigan. Ortiz will play every inning of this game. He’s my best hitter, and I want him in the hitting spot all game. Otherwise, all righties, as I assume Clayton Kershaw is starting against me—and Kershaw being, after Doolittle, the next toughest pitcher against lefties over the past three years.

With Ortiz in the gamefor the whole time, I have 20 spots for seven positions; or 13, once the seven other named starters are accounted for.

After each batter hits for the second time, I’m pulling them from the game. The exceptions are Mike Trout, who will slide to LF; and Manny Machado, who will slide to third base. Each will stay in the entire game. I’ll replace each other starter on defense, with Eric Hosmer at first, Jose Iglesias at second, Andrelton Simmons at short, Kevin Pillar in center and Kevin Kiermaier in right. Each will play until its his turn to bat. There’s not a ton to be gained by squeezing in two-plus innings of the greatest defense that ever played, with strikeout pitchers Sale, Kluber and Hill on the mound, but the game is in Petco—where more batted balls stay in the yard—and there’s not much to lose, either. I still have eight roster spots after these players are accounted for.

Each of these players except Hosmer will be pinch-hit for, unless the game situation makes the gain of a pinch-hitter very, very low—two outs and nobody on, perhaps; my team leading by a ton. Most likely, pinch-hit for. Xander Bogaerts will pinch-hit for Simmons, Rob Cano for Iglesias, Jackie Bradley for Pillar and Jose Bautista for Pillar. After Sal Perez bats in the sixth or seventh, Jason Castro will take over on defense. So my closing lineup is:

• Trout (LF)
• Machado (3B)
• Ortiz (DH)
• Hosmer (1B)
• Bautista (RF)
• Cano (2B)
• Bogaerts (SS)
• Bradley (CF)
• Castro (C)

8. This sets me up for his closer, who will be, I presume, either Kenley Jansen, Hector Rondon or Noah Syndergaard. None is all that easy on left-handers, but there is at least some platoon advantage to be gained, and I’ve now got five left-handed hitters in the lineup. I also have four roster spots left, which will go to Terrance Gore, because 40 percent of All-Star games are decided by one-run or in extra innings, which means there’s a 20 percent chance my team is going to be playing for the tying or winning run in the ninth; Francisco Lindor, as my switch-hitter with speed on the bench; Carlos Beltran as my switch-hitter with power on the bench; and Joe Mauer as my lefty off the bench, and Twin. I would bring Danny Valencia as lefty-masher off the bench, but I just don’t see a lot of lefties coming into the game late, and anyway Miguel Cabrera is going to be my designated Dude Who Can Come Back Into The Game Off The Bench. Pray for Castro.

Pretty sure this has me covered, though it occurs to me now that I’ve put almost no thought into trying to figure out, and counterattack, whatever unconventional moves Rian’s got planned.

The NL All-Stars

This whole exercise is rather un-baseball-ish. That makes it rather difficult, and also possibly rather foolish. To design a system that simultaneously maximizes the odds of winning one—and only one—baseball game and comports with some recognizable form of baseball reality requires selecting from a pool of players (in this case, big-leaguers) whose talents have been heretofore understood and described in terms of their ability to contribute to wins over the rather large (but still impossibly small) sample of 162 games.

Still, I’ll give it a go.

Here’s my first assumption: that my opponent in this exercise, the canny and also-my-boss Sam Miller, is a smart man, and that he’ll therefore come to every conclusion that I come to and quite a few more besides. Thus, while I can’t control for Sam’s decisions on the basis of conclusions I don’t reach, I can design my roster as if he’s designing his on the basis of the same, or similar, underlying principles. So what are those principles?

Here’s one: that each additional time a batter and hitter face off in a single game, the batter gains an additional advantage over the hitter. Therefore, it behooves me (and would surely also behoove Sam) to design a roster which allows for each of my pitchers to face each opposing batter once, at maximum.

Given that Sam has presumably come to this same conclusion, I’m guessing that he’s going to load up his roster with a lot of the extraordinary relievers dotted throughout the American league. That means my batters have to be prepared to face the likes of Chapman, Miller, Betances, Harris, Osuna, Kimbrel, Robertson, and Gregerson. That’s a lot of good relievers, and it’ll allow Sam to ensure that my hitters, on almost a case-by-case basis, face the pitcher most-prepared to get them, individually, out. So, how do I defend against that?

Well, here’s what almost all of those pitchers have in common: they’re very good at (a) not allowing very many home runs and (b) not allowing very many baserunners. What does that mean for my hitters? Well, it could mean a lot of things, but here’s what I’m going to decide it means: that there’s no point in trying to select for hitters who do those things (hit home runs and get on base) especially well. Selecting all-or-nothing sluggers against high-strikeout, home-run-preventing relievers is rather like playing your kid cousin at chess after he’s spent the whole family vacation pestering you to play chess. He obviously wants you to play chess because he’s really good at it, and he’s gonna whoop your ass once you finally agree to take him on. So, you know, don’t play chess against him. Play basketball instead, and win.

In other words, to defeat your brilliant cousin, you must change the game.

There’s a degree to which you can overthink this, of course. If a big, powerful slugger is someone like Albert Pujols (circa 2003) or Mike Trout (circa today) then there’s no point in discounting him just because he hits home runs and gets on base. He does a lot of other things well, too, and in fact is just an all-around superb hitter. Throw that guy on the team, and stop worrying about it. The original and best competitive advantage is always selecting the player who is obviously good. But if you have a guy like Adam Dunn (circa 2004), who’s producing good numbers across a large sample size, but who really only ever walks, homers, and strikes out, you probably don’t want that guy facing a reliever with a 40 percent strikeout rate.

So who do I want? I’ve given it some thought, and I’ve decided that I want—all other things being equal—hitters who see a lot of pitches, hitters who make a lot of contact when they’ve had enough of seeing pitches, hitters who hit a lot of line drives when they make contact, and hitters who play good defense when they’re not hitting. The first three criteria are my attempt to blunt the most pernicious strength of Sam’s presumed reliever parade: their ability to strike men out, and do it fast. The fourth criteria is my attempt to support the pitching staff I intend to put together, but we’ll get to that later.

Why hitters who make contact, hit line drives, and see a lot of pitches? Well, because history tells us that the especially difficult thing to do against top relievers is to get the ball in play, at all—once you do that, though, BABIP is approximately normal, which is to say that it’s much higher than the typically BA allowed by the relievers in question, and higher still on line drives than it is on groundballs. So, selecting for guys who can routinely barrel up baseballs (if not hit home runs, directly) is probably a good idea, given the kinds of folks they’ll be facing in this game. A home run is sexier, sure, but in a game with as many good pitchers as this one will have, a few runs scraped out by hitters like this might be enough to make the difference.

And seeing a lot of pitches? I want my guys to get their looks, and I want the relievers tired out.

So. How many of these hitters do I want? Well, I’d like to ensure that every position on the defensive spectrum is covered by at least two players, and I’d simultaneously like to give my best hitters as many plate appearances as possible. The first criteria suggests I should pick lots of hitters; the second, not so many. Let’s settle on the following 13:

1. Nolan Arenado, 3b COL
2. Charlie Blackmon, of COL
3. Brandon Crawford, ss SFG
4. Bryce Harper, rf WSN
5. Jason Heyward, of CHC
6. Jonathan Lucroy, c MIL
7. Marcell Ozuna, cf MIA
8. Anthony Rizzo, 1b CHC
9. Wilson Ramos, c WSN
10. Addison Russell, if CHC
11. Corey Seager, ss LAD
12. Christian Yelich, lf MIA
13. Ben Zobrist, 2b CHC

You’ll notice that this list coincides nicely with a list of hitters who are notably good at hitting. Again, I’m trying not to overthink it. The biggest points of departure, probably, from a list of the best hitters in the league, were in choosing Christian Yelich over Kris Bryant in left field (better defense, better contact rate, better line drive rate, worse hitter) and in choosing Charlie Blackmon as a backup (he’s really good at barreling up balls). Other than that, things were pretty simple: Are there batters who are more especially good at making contact than Bryce Harper? Yeah, but only because he’s good at everything, and so especially good at nothing.

Don’t. Overthink. It.

Actually, the biggest philosophical point made by the hitters’ list is probably this: I believe that it’s best, in a one-game scenario, to carry as many pitchers as possible. Hitters generally get better the deeper they play into a game (although some of this is a reflected effect of the starter getting worse), while the opposite is generally true of pitchers, and so keeping the number of hitters to the bare minimum—basically, just enough to keep all the infield positions covered—allows me to load up on pitchers, the better to play matchup with Sam’s hitters. And, oh, the pitchers I plan to throw out there.

Here’s a few guiding principles for pitchers: I wanted pitchers who don’t allow very many home runs, pitchers who don’t allow very many baserunners, and pitchers who have one truly dominant pitch. The reasons for including the first two criteria are obvious, I think, especially in a one-game context where one home run allowed can scupper an entire ballgame.

The last, I think, is important in that it speaks to a philosophy about the relative merits of including dominant starters (Clayton Kershaw, Jake Arrieta, etc.) in an exercise like this. Are they particularly valuable in a one-game scenario? Especially: Are two of them especially valuable? There’s one school of thought that says, sure, it’d be kind of fun to watch Jake Arrieta air it out for three innings, knowing that he doesn’t have to pitch a fourth. But there’s another that says, well, does he really know how to do that effectively? So much of a starting pitcher’s game plan is about getting a hitter out within the constraints of also setting a hitter up for the next time they face him later that game; that’s manifestly not the case when you’re relieving. Clayton Kershaw’s third pitch isn’t as valuable to him as a reliever as it is as a starter. In fact, I’m not sure why he’d throw it at all, if he was only going three innings.

I’m going to split the difference here, and include the following 21 (!) pitchers on my roster—there are 34 spots to go around in total, after all, and I’ve only used 13 on hitters:

1. Jake Arrieta CHC
2. Joe Blanton LAD
3. Ryan Buchter SDP
4. Santiago Casilla SFG
5. Hunter Cervenka ATL
6. Jose Fernandez MIA
7. Daniel Hudson ARI
8. Kenley Jansen LAD
9. Shawn Kelley WSN
10. Clayton Kershaw LAD
11. Seung Oh STL
12. Mark Melancon PIT
13. Hector Neris PHI
14. Hector Rondon CHC
15. Max Scherzer WSN
16. Stephen Strasburg WSN
17. Pedro Strop CHC
18. Noah Syndergaard NYM
19. Julio Teheran ATL
20. Tyler Thornburg MIL
21. Blake Wood CIN

You might notice that the pitchers are sorted alphabetically, and not by their typical status as relievers or starters. That’s because it doesn’t matter. For my purposes, they’re all relievers. As soon as the lineup is announced for the opposing side, my manager will decide the starting pitcher for the game using a principle that is simple to explain, if admittedly complicated to execute. It is: which of my pitchers who is still available is best suited to get the next hitter out?

If, for example, the leadoff hitter has shown an inability to catch up to triple-digit stuff, perhaps Noah Syndergaard would be sent out there to face him. If he retires him, great. If he doesn’t, that’s too bad. Either way, the manager then asks a second question: Is there a pitcher on my bench who is significantly better-suited to face the next batter than Syndergaard? If so, that pitcher goes in, and the cycle starts over again. If not, Syndergaard throws to a second batter. After that batter hits, regardless of outcome, the process loops again for a third batter, but continues no further. No pitcher will ever be asked to throw to more than three hitters, regardless of whether or not they get all of them out. This is to ensure that every pitcher can air his stuff out without fear of being extended, and to ensure that no hitter gets a long look at any given pitcher.

I don’t think there’s any chance the game requires my pitchers to face 63 batters, which is the maximum number they could face in this scenario. With 21 arms available, I think we’ll be prepared to take on anything Sam throws against us. And with each pitcher only throwing to 1-3 batters, we’ll be able to play matchups to the Nth degree. That’ll require a talented manager, sure, but I’m assuming I can get the best of the best to manage my squad. It shouldn’t be a problem.

And that’s about all there is to my strategy. My defense will shift, of course, based on the data available to us, and my hitters will follow the most basic gameplan possible: You’re up, unless you’re injured, and your goal is to put the ball in play. I’ll cluster my high on-base players near the top of the lineup, try to alternate lefty-righty, and put my best hitter in the two-spot, like any good sabermetrician, but otherwise I’ll pretty much leave them alone. The offensive replacements are there in case there’s a particularly poor matchup for us late in games, but I wouldn’t be disappointed if we didn’t use them at all. The point of the exercise is to give the hitters I’ve got as many looks as I can.

May the best team win.