Shelby Miller's been a fun story, right? Guy gets traded for a hometown favorite, guy has a habit of outpitching his peripherals, guy's somewhat a work in progress so no one's quite sure what to expect, and, oh yeah: guy has had eight starts, allowed zero runs in three of them, and his per-start ERA tops out at 3.00. Beyond the elementary joy of watching Shelby send batter after batter back to the dugout (with an added bonus that he largely goes right after guys, as compared to the nibbling we've seen from the remainder of the rotation this year), he's also been fascinating in terms of the very discrete way he's seemed to evolve through April and the first part of May.
An ugly chart's worth a thousand (well, maybe 900 because it's ugly) words, and two of them capture just how Miller's made adjustments to keep building on the success he's enjoyed.
The way I've been looking at his performance has lent itself very well to a start-to-start basis. Even though there are sample size concerns with regard to what happens in any single game (to say nothing of larger samples similarly prone to the wackiness of baseball), I think Shelby's game-to-game stats tell a compelling story.
The first is the duel between his peripherals and his runs allowed, which can be thought of as the gap between the purple and bronze lines. Miller's first three starts could be thought of as an exercise in regression to peripherals: his FIP-ERA gap went from 2.75 to 2.15 to 0.15 in the span of those starts. Starts 4 and 5 were kind of when things caught up to him a bit in different ways, leading to a ballooning FIP and a still-great-but-hey-this-is-uncharacteristic-for-him-so-far runs allowed tally. Then, over his last few starts, he's kicked it into another gear: the FIP-ERA gap may still be unsustainable in its extent, but both numbers are quite low and indicate he'll still be dominant even if his ERA inexorably increases to match his FIP.
The bottom chart above really helps to contextualize Miller's performance, His first four starts were evocative of the dearth of whiffs from opposing hitters, which had me very worried. The latter half of those (Starts 3 and 4) were somewhat of an exercise in what happens when you have good stuff but aren't gathering swinging strikes: your luck runs out eventually and some runs get on the board. (Again, that's not to say that Shelby was bad in those starts or anything like that - just worse than he's already shown he can be.)
Start 5, which came against the Reds at home and is Miller's only loss on the year, is a really interesting turning point. A switch was ostensibly flipped and took his whiff rate from the Kevin Correia Danger Zone to a ridiculously overpowering (better than any starter in baseball last year, for instance) 15.2%. But, that was only a part of the puzzle: part of the problem of going to the well with the really good fastball too many times is that major league hitters can time a fastball and hit it hard, and that's what was happening to Shelby. At the expense of that swinging strike rate came a crazy low 0.43 GB/FB and a groundball rate below 18%. Oh, and, you know, Mike Leake and something called a Tucker Barnhart taking Miller to the bleacher seats.
I think that in another world, with a Shelby Miller with a different mindset, it may have been possible that Start 5 was chalked up to good process and poor results, and that would've been that. But as Miller started the season eliciting weak contact, I think there was a clear avenue for improvement there. What Miller's done since that fifth start is to, well, put it all together: he's married a penchant for getting groundballs along with an ability to generate swinging strikes, and his performance has really taken off. His groundball rates in his last three starts aren't quite where they were before, but they're still very solid. His swinging strike rate is still very strong, and as his near no-hitter showed, he's perfectly content to let guys make outs early in the count (two Madduxes in three starts) if they don't want to hang around and get blown away by a fastball later.
Miller's recent in-start stats resemble an elite echelon of pitchers. The top 10% of pitchers over the last few seasons have, on average, been in the top quartile of generating whiffs while grabbing groundballs at an average-ish rate. Miller has been sporting a crazy-good GB% in all of his starts except Numero Cinco, and his whiff rate has been strong as well. I did a quick search on pitchers who have recently put up strong numbers akin to Miller's in those two categories and the best matches are Tyson Ross and Felix Hernandez. While Miller's got a long way to go to establish the type of dominance that King Felix has enjoyed, I wouldn't want to doubt his ceiling so long as he's combining his groundball and whiff-garnering abilities.
One last thing I found really striking on a start-to-start basis, which dovetails with his start-to-start results, has been his pitch usage. Let's go to a chart:
The above isn't anything that hasn't been covered in better qualitative and quantitative detail elsewhere, but I think it reinforces a point. In Starts 1-4, Miller was mostly working off of his two-seamer, which was responsible for heady groundball rates but a worrying lack of swinging strikes. The trend reversed itself in Start 5. The pitch that Barnhart crushed was a sinker right in the middle of the plate, a bit below the letters, while Leake took a middle-in fastball at the waist for his moonshot. Since then, Shelby's appeared to use his sinker and heater in a better combination to keep hitters well off-balance, and is primarily working off of his fastball while using the sinker in opportune times. His cutter and curveball are both important parts of his arsenal, and I think they're largely used for the element of surprise when he finds himself in a tough spot against a hitter that's getting some good swings against his four-seam/two-seam fastball combination.
Are these trends going to stick? I guess we'll wait and find out. I'm definitely curious to see whether the four-seamer usage continues to dominate the two-seamer usage going forward, or if that's an idiosyncrasy that will abate over time. Either way, I've been impressed with Miller's rapid-fire evolution from "not missing any bats" to "missing the most bats," and his starts are definitely one of the most enjoyable parts of this season.
Addendum - Don't read if you don't care about fruitless math
In writing this, I initially wanted to investigate whether there was a relationship between a pitcher's whiff rate, his groundball rate, and his overall success. Since these two things are substitutes, I was thinking about this concept as a series of indifference curves. After running some numbers, though, the data just aren't very keen on groundball percentage mattering a whole bunch as an independent contributor.
I mapped every starter over the last few years in terms of whiff rate and groundball rate, but on a percentile basis rather than a raw basis (i.e., this pitcher is in the Top X% of all starters in SwStr% or GB%), and analyzed the relationship between those two statistics and ERA/FIP/xFIP, holding the other statistic constant, just as a starting point. What I have, at this point, is something that largely accords with what we already know: whiff rate is really important, groundballs are a somewhat-preferential thing that doesn't do much to presage a pitcher's success. Specifically, this:
GB% just doesn't really seem to have much of an impact on ERA, in the end. Meanwhile, the difference between the hypothetical worst pitcher ever at getting swinging strikes, and the best, controlling for GB%, is 16 runs per 9 innings. Which is a lot. Overall, though, these two stats aren't super-explanatory of ERA (about 18 percent of the variation in ERA is explained by variation in these two things). FIP was more heartening, and since FIP is a better predictor of future ERA than ERA, I guess it makes sense that more grounders, generally as a result of limiting hard contact, matter more for a less unstable measure.
These stats are more predictive for FIP, but the key (and uninteresting) finding is that SwStr% just matters so much more than GB%. Again, on an FIP basis, the difference between the best and worst pitcher for SwStr% is 19 FIP-based runs per 9 innings, but this same difference for GB% s about half a run per 9 innings. Now, half a run of FIP is a large real-world difference for a pitcher's stats, but it's two orders of magnitude lower than the SwStr%, and that's crazy.
The xFIP results, as a coda, are pretty similar in the expected manner. Again, it's nothing we didn't suspect already, but I think it's at least marginally interesting that GB% matters less (perhaps not at all) for the stat which uses results and not peripherals, while GB% actually has a small but potentially-tangible impact on FIP, which doesn't actually directly care about the nature of balls in play directly.