Andrew Miller is Scuffling, and Also Great
Here’s a sentence that I didn’t think I’d be writing in 2019: Andrew Miller has accumulated -0.2 WAR this season. By FanGraphs’ version of WAR, he’s been less valuable than a replacement-level reliever. Here’s another sentence that makes a little more sense, but is at odds with the one I just wrote: Andrew Miller might be the Cardinals’ best reliever. Now, reliever performance is volatile and all, but we’re going to need an explanation. How can those two things be true at once?
Let’s start with Miller being below replacement level, because that would have been a surprising assertion before the year. Andrew Miller has faced 139 batters this season. He’s allowed eight home runs. 6% of his plate appearances have ended with the opposing batter trotting around the bases. That surely already sounds like a ton — indeed, Miller is second in the majors in home runs allowed per nine innings, behind only Josh Osich. You don’t need a sabermetric writer to tell you that’s bad.
As bad as 2.2 home runs per nine innings sounds, though, it might be underselling how wild Miller’s season has been on the home run front. Josh Osich is a great example of the kind of pitcher who normally leads the league in home runs allowed per nine. He’s a pitch-to-contact depth reliever who works by letting opponents put the ball in play and counting on his defense to make plays behind him. Now, that strategy mostly hasn’t worked — Osich has a career ERA of 5.11, and his FIP is 5.31, so it’s not as though he’s just been getting unlucky. Still, while Osich is homer-prone, he’s mostly just contact-prone, with the home runs a cost of doing business. Strike out only 19% of the batters you face, and there will be plenty of opportunities to give up home runs.
Andrew Miller’s case of the dingers isn’t like that at all. Miller is actually one of the least contact-prone pitchers in all of baseball this year. Only 51% of the batters he has faced have put the ball in play. That severely limits the opportunities they have to hit home runs. Osich, for comparison’s sake, has let 75% of batters put the ball in play. Miller is giving up home runs at a truly alarming rate considering how few opportunities he gives batters to put a ball in play.
Look instead at home runs per batted ball, and Miller’s performance stands out. Here are the top five pitchers this year when it comes to home runs allowed per batted ball:
| Player | HR/Batted Ball | ERA | FIP |
|---|---|---|---|
| Cody Allen | 13.43% | 6.26 | 8.40 |
| Dan Straily | 12.29% | 9.82 | 9.35 |
| Josh Hader | 12.16% | 2.27 | 2.86 |
| Corbin Burnes | 11.76% | 9.00 | 6.13 |
| Andrew Miller | 11.27% | 3.86 | 5.15 |
Josh Hader is a unicorn, and also the most extreme fly ball pitcher in baseball. The rest of the list is a who’s who of pitchers on roster peripheries. Dan Straily put up 9-handles for his ERA and FIP before being sent to Triple-A by the Orioles. Cody Allen was outright cut by the pitching-needy Angels — he’s currently toiling in Triple-A Rochester with a 9.14 FIP. Corbin Burnes lost his spot in the Brewers’ starting rotation, then got demoted to the minors, then got injured. They’re a motley crew, these high-homer pitchers.
Even Hader isn’t a good comparison for Miller. When opponents put the ball in play against Hader, it’s basically always in the air. His 0.36 GB/FB ratio is the lowest in baseball. He’s not even allowing all that many home runs per fly ball (roughly 20%). He’s simply allowing fly balls for all of his contact. That leads to a high home run total, but it also lowers his BABIP — fly balls that don’t become home runs very rarely drop for hits. His .185 BABIP is among the best in the major leagues, and his xBABIP isn’t far behind at .239.
Andrew Miller isn’t much of a fly ball pitcher. His GB/FB ratio of 1.04 is barely lower than the league average of 1.21. He’s rocking a .302 BABIP, hardly exceptional in either direction. The point is, he’s not giving up his home runs through fly ball volume — literally a third of the fly balls hit against him have gone for homers, the highest rate in the majors among pitchers with 20 innings pitched. When batters put the ball in the air against Miller, they’re producing a .596 wOBA this year, nearly 100 points higher than the league average. Not exactly what you’d like to see from your relief ace.
Well, we have the “why does Andrew Miller have negative WAR?” question sorted. It’s the home runs! It’s not easy to have a 5.15 FIP and a 3.41 xFIP, but Miller has managed it. Let’s move to the second statement I made. Is Andrew Miller really the best reliever on the Cardinals, playoff hopefuls who sport the highest bullpen strikeout rate in the game? Can a team’s best reliever still have negative WAR halfway through July?
Allow me to convince you. First, consider this: Miller is striking out 35.3% of the batters he faces this year. That’s not the highest rate of his career, but that’s only because he was the best reliever in baseball a few years ago. It’s still tied with Aroldis Chapman for the 15th-best rate among relievers, and the second-best rate on the Cardinals behind Giovanny Gallegos. Only six of the bullpen arms ahead of him run higher groundball rates than Miller does, and mixing grounders with strikeouts is the gold standard among relievers.
As good as Miller’s season-long strikeout numbers are, though, they’re underplaying his recent form. Take a look at his strikeout and walk rates by month, and you’ll start to see what I mean:
| Month | K% | BB% | K-BB |
|---|---|---|---|
| Mar/Apr | 28.6 | 16.1 | 12.5 |
| May | 32.1 | 3.6 | 28.5 |
| June | 43.3 | 6.7 | 36.6 |
| July | 44.0 | 8.0 | 36.0 |
Early in the year, Andrew Miller had a problem. He couldn’t locate his slider for a strike. When I say couldn’t locate, I’m not kidding — he was hitting the strike zone on 35% of his sliders, by far a career low, and batters didn’t chase enough to make that work. His command was spotty overall, not just with his slider — when he got to three-ball counts, he was only getting pitches over the plate 44% of the time, a bottom-ten rate in baseball. Batters were content to take their walks, and Miller had no counter. He continued to go to his slider, even in three-ball counts, and batters took it and walked. More than a third of his three-ball pitches were ending in free passes , well above the league average of 24.6%. In short, his command was holding him back.
Since the beginning of May, Miller seems to have worked himself back into strike-throwing form. He throws his slider for a strike 47% of the time. With three balls, he’s gone from one of the worst pitchers in baseball at getting the ball over the plate to slightly better than average. He’s walking opposing hitters on 24.6% of three-ball pitches, right in line with league average. Essentially, he’s turned his biggest weakness — and make no mistake, Miller’s command was enough of a weakness that he was running a gruesome 8.16 FIP (and 5.49 xFIP) — into an unremarkable part of his game.
And when his control isn’t holding him back, he’s well-nigh unhittable. That sweeping, are-we-confident-physics-is-currently-working slider he’s known for still baffles hitters. His 14.1% swinging strike rate is stellar, and he gets misses on 32% of opposing batters’ swings, a top-20 rate in baseball. With his pesky control problems out of the way, his strikeout rate is 39.8% since May 1, his walk rate a minuscule 6%. The home run problems haven’t gone away (30.8% HR/FB), but even with those, he’s pitched to a 3.55 FIP (2.31 xFIP). By K-BB, he’s among the very best pitchers in the game:
| Player | K% | BB% | K-BB% |
|---|---|---|---|
| Josh Hader | 47.1 | 7.4 | 39.7 |
| Kirby Yates | 43.3 | 3.9 | 39.4 |
| Brad Hand | 42.9 | 5.4 | 37.5 |
| Will Smith | 42.5 | 4.8 | 36.7 |
| Chris Sale | 39.4 | 5.4 | 34.1 |
| Giovanny Gallegos | 37.8 | 3.7 | 34.1 |
| Andrew Miller | 39.8 | 6.0 | 33.7 |
| Gerrit Cole | 38.0 | 5.2 | 32.9 |
| Chris Martin | 32.6 | 0.0 | 32.6 |
| Austin Adams | 44.6 | 12.0 | 32.6 |
Remember earlier in the article when I mentioned that Miller was allowing an abysmal .596 wOBA on balls hit in the air? That was a bit of a diversion. Those might be the observed results, but Miller has been unlucky given the contact he’s allowed. His xwOBA is a still-not-amazing .519, which is a smidgen higher than the league average .508 wOBA allowed on balls hit in the air this year. That 80 point gap, though, reflects how unlucky Miller has been to have so many of his fly balls leave the park.
The point of all of this is that Miller, right now, is one of the most fearsome relievers in the game. He’ll probably strike you out, and if he doesn’t, he gets groundballs well enough. Platoon split? There’s basically none to speak of — he’s allowed a .312 wOBA to righties in his career, .298 to lefties. If you manage to hit the ball in the air, that’s good — but even there, he’s allowing a quality of contact that clocks in right around league average.
At the same time, though, he legitimately has been below replacement level this year. He’s giving out home runs like party favors, and you can’t afford to be homer-prone in 2019 if you want to be a successful pitcher. You can strike out all the batters you want, and walk as few as you want — give up home runs, and you still won’t run an elite FIP. There’s nothing fake about that.
So what do we make of Miller overall? Is he one of the best relievers in baseball, or performing about the same as a minor leaguer would in the same position? I would argue that he’s both. Put Andrew Miller on a mound, tomorrow, and there aren’t many pitchers I’d take over him. At the same time, though, sometimes you can run out a great reliever and get a 30 inning stretch of garbage results.
All of us know, I think, to look at reliever ERA’s with a healthy dose of skepticism. Edwin Diaz isn’t a 5-ERA true talent reliever, regardless of what his numbers are this year. Miller, though, shows that we should probably be careful about looking at FIP, even with more than half of a season in the books. Statistics, even the best-intentioned and most context-neutral statistics, can mislead. Relievers operate in a world of minuscule samples, a world where a single home run can be the difference between a good and bad month.
That Andrew Miller has been bad is an objective fact, and it’s also not particularly useful when you’re trying to figure out what he’ll do tomorrow. That Andrew Miller is great is something that I can only infer from a few statistics, but it’s every bit as relevant to what he’ll do tomorrow as his 2019 results. Baseball is weird and frustrating that way, and relievers particularly so.
That’s the nature of all relievers, of course. But it’s more viscerally interesting in the extremes — when a pitcher is worse than replacement level by results and excellent by other measures. Cardinals fans are likely frustrated that their team’s big offseason pitching signing has been as useful to the team as a minor leaguer on the Triple-A shuttle. It’s hard to square that in your head with the fact that in Miller’s next outing, he’s pretty likely to be great, and I get that. Doesn’t make it any less true, though.
Ben is a writer at FanGraphs. He can be found on Bluesky @benclemens.
My takeaway? Chris Martin hasn’t walked a batter since May 1st. Looked it up, his last was on April 30th That’s 25.2 IP, 31K and 0 BB. That’s pretty impressive. I mean, besides Adams none of the other guys are walking a ton – Yates has 4 over that span and Gallegos 5. But still, zero walks over 2.5+ months is pretty impressive.
Another great article from Ben!
And this time over my favorite team. Miller’s been a strange beast this year. A good headstratcher. Seems a lot like the team in general. Great in everything, but without the results.
The offense has been FAR from great, while the starting pitching has been up and down all season.
I wish I had something more constructive to say than “Good shit, per usual,” but I don’t.
Good shit, per usual.
Miller’s been a Cy Young candidate compared to some of the more recent free agent relievers we’ve signed in recent years. Holland, Cecil, Gregerson….
Here’s another explanation: FanGraphs’ Pitching WAR stat sucks! Always use Baseball-Reference for Pitching WAR (where Miller is at an even 0.0), as ERA is a much better metric than FIP for analyzing what has already occurred.
Save FIP for predicting what has yet to come, although even that’s not likely to be accurate in this case, as you are absolutely correct that Miller has pitched much better since the beginning of May than he did in April.
bWAR doesn’t use ERA. It uses RA/9. Which is important here because he’s given up 4 unearned runs and has a 4.96 RA/9. And that’s with an above average defense, a slightly pitcher-friendly park factor, and facing below average batters (all this according to bbref’s defensive, park factor, and opponent quality adjustments that go into its version of WAR). Here RA/9 says he’s been about as bad as FIP does.
Beyond that though, it’s really silly to argue that 0.0 WAR is more reflective of Miller’s performance than -0.2. They’re essentially equivalent.
OK, so bWAR Pitcher’s WAR has its own problem, since no pitching metric should be specifically giving fault for unearned runs. That means Miller’s true value to this point should be above 0.
Still, that issue and Miller’s specific case being only a small difference still doesn’t change anything about FanGraph’s Pitcher’s WAR being MUCH worse.
Lol there is no pitching metric where you are going to look good giving up 2.2 home runs per 9 innings
There are lots of issues with FIP, no doubt about it.
But there’s also a HUGE issue with Baseball Reference’s version which I never see discussed. An example. Pitcher A leaves the game with a runner on first and two outs. Pitcher B comes in and gives up a home run to the first batter he faces. In a runs allowed model, Pitcher A is charged with a run allowed, even though he’s probably responsible for .20 runs at most. We certainly have the data and the computers to properly apportion runs allowed when there are multiple pitchers in an inning. And yet, it’s not done.
I agree completely. Or Pitcher A walks 3 batters and leaves with the bases loaded and nobody out, but Pitcher B comes in and gets 3 outs without allowing any runs. In that case Pitcher A doesn’t get any penalty for putting the team in a situation were runs will normally score.
Fair enough, but it’s still much better than Fangraph’s Pitching WAR.
Here’s an observation on the leverage adjustment for reliever WAR. The formula is “unadjusted WAR” times (1+gmLI)/2, where gmLI is the average leverage index for that reliever.
Suppose we have these three relievers on a team, each of whom has pitched to the same number of batters:
A.M Hi-Lev: unadj. WAR = 1.5, avg LI = 3.0, ==> WAR = 1.5 x (1+3.0)/2 = 3.0
A.M Med-Lev: unadj. WAR = 0.0, avg LI = 1.5, ==> WAR = 0.0 x (1+1.5)/2 = 0.0
A.M Lo-Lev: unadj. WAR = -2.0, avg LI = 0.6, ==> WAR = -2.0 x (1+0.8)/2 = -1.6
Now, suppose this was just one guy doing all this pitching instead of three. His WAR would be:
A.M Iller: unadj. WAR = -0.5, avg LI = 1.7, ==> WAR = -0.5 x (1+1.7)/2 = -0.7
But, -0.7 is not equal to 3.0+0.0+(-1.6). What am I missing?
So I’m not the progenitor of any of these stats, or even particularly privy to their inner workings, but I think I have the answer. Sabermetrics is founded on many tenets, but two of them come into play in answering your question. The first is that players can’t choose when they perform well or poorly. This is kind of folded into everything, in the background — wOBA, FIP, heck, even batting average. Nothing too crazy there — pitchers can’t choose to be good when the game is close and bad when it isn’t.
The second thing, which often goes unstated, is that different players have different levels of performance. A player can’t bring his A game and B game at will — but two players are perfectly capable of performing at different levels. The reason your leverage-adjusted thing doesn’t come out even like you’d hope is that the two scenarios are different.
In your first example, you have an elite reliever pitching a lot of extremely important innings, a replacement level reliever pitching a few important ones, and a really really garbage reliever pitching in unimportant situations. That’s the obvious division of labor you’d want if you know what performances you’ll get. In your second example, you get a bad reliever in all situations. Maybe he happens to perform well in the high-leverage one, but per tenet one, if you know his overall skill level, you should expect him to perform equally in each situation.
At its heart, the difference is that different players often do have different skill levels in predictive ways, but the same player doesn’t. I feel like I started rambling a little, but uh, did that make sense?
I think I see what you’re getting at…WAR doesn’t reflect the game-state context for each plate appearance. If it did, it would look something like WPA plus an adjustment for replacement level.
The critical point that I forgot about is that the LI adjustment for relievers is NOT about adjusting for game-state context, but is actually an approximate adjustment for the impact of chaining in bullpen roles. This adjustment is necessary to create an appropriate basis for replacement level for relievers, but the use of a multiplier leads to reliever WAR not being completely additive. This really isn’t a problem unless we’re taking a granular look at an extreme example like the one I gave here.
As a side note, the reason for my question is the crazy splits Miller has had this year by low/medium/high leverage. If you go by actual batted-ball outcomes (wOBA allowed), Miller has fared worst in high-leverage situations, resulting in a negative Clutch score. However, if you look at his FIP splits, he’s at his BEST in high- leverage spots….just happened to allow a .533 BABIP in those situations. If there was such a thing as a fielding-independent Clutch score, Miller’s would be a solid positive number instead of negative. This implies that if we folded game-state context into Miller’s fWAR, it would most like switch from negative to positive, working like a less extreme version of my hypothetical example.
Holy smokes, Josh Hader.
Relievers are a volatile example #173:
Look at Ben’s first chart on highest home run rates and who do you see: Cody Allen and Andrew Miller, the anchors of the Indians bullpen just a few years ago.