Statcast’s New Defensive Metrics and Seattle’s Glove-First Outfield
I don’t want to write about yesterday’s Cactus League debacle any more than you want to read about it, and I’m loathe to look at more spring stats to see if the M’s early lead in K:BB ratio’s been frittered away the past few days. So instead, let’s look at one of the biggest stories of the offseason: the M’s remade outfield defense. You may have seen this Eno Sarris article at Fangraphs that says the 2017 M’s may be the first since the 2003 M’s/2013 Indians to field three true-talent CFs at the same time. It’s great, even if it does include video of that awful Mike Cameron injury. Importantly, it includes some quotes from Jerry Dipoto about how important defense was to their offseason plan, and how it complements the kind of pitchers they targeted – guys like extreme fly-baller Drew Smyly. Fortuitously, MLB’s Statcast group just released brand new data on outfield defense, breaking plays into 5 (with an implied 6th) buckets ranging from “nearly impossible to catch” to “nearly impossible to miss.” These new measures might help us get a handle on what the M’s baseline level of fielding was in 2016, and whether or not there’s anything about Safeco Field that will make the job easier.
The new statcast data includes 5 buckets, each given 1-5 stars. The 5 star buckets are converted into outs less than 10% of the time, and 1 stars are converted into outs about 90-95% of the time. Jeff’s got a great post on the new Statcast data today at Fangraphs, and he’s helpfully calculated the average conversion rate for the Statcast era, or 2015+2016. With that, we can compare each player’s catches to what we think a league average OF would make, given the same opportunities. I used his averages and calculated plays made above average for each player in 2015 and 2016, and then I looked at how their chances were distributed. That is, some players (Jackie Bradley, Jr.) saw proportionally more extremely difficult opportunities, while others (Kevin Kiermaier) saw fewer. The plays on the Statcast leaderboards apparently don’t include the absolutely never missed, can of corn, 100% probability balls. At least, that’s my guess just from noting that the sum of each players chances across all 5 buckets totals a fraction of their overall chances/putouts as reported elsewhere. If true, that means that the plays that really separate defenders are in a few not-terribly-large buckets in the ranges where catches are made 25-50% of the time or so. Given THAT, I’m pretty impressed with how reliable these numbers look. Kiermaier, Lorenzo Cain, Ender Inciarte, etc. look great both years despite varying playing time, opportunities, etc.
Let’s take a Mariner-centric view of this, and attempt to answer a few questions. First, how good was the M’s defense last year, and second, how much better was it than 2015’s? Third, does anything stand out about the distribution of the M’s outfield opportunities? Finally, what would replacing Nori Aoki with Jarrod Dyson and Seth Smith with Mitch Haniger do in terms of additional plays made?
Jeff mentioned it in his piece, but one of the first big takeaways is that Leonys Martin looks quite good in the Statcast-based metrics – better than he did in UZR/DRS, where he graded out as a dead-on average CF last year. Martin made 11+ plays more than expected, a bit better than the 4 runs saved by UZR, and a far sight better than DRS’s 2 runs *below* average. That edges out Lorenzo Cain, who ranked 2nd in all of baseball in 2015. On the other hand, it essentially ties him with the 11 runs saved by the M’s 2015 CF, Austin Jackson (yes, Jackson split time that year with the Cubs, but the bulk of it was in Seattle). Seth Smith and Norichika Aoki both graded out a bit below average (probably not a shock to most M’s fans), while Nelson Cruz and Guillermo Heredia kind of cancelled each other out. Interestingly, the corner OF picture wasn’t all that different in 2015. It was worse, clearly, thanks to many more innings from Nelson Cruz, the Mark Trumbo experience, and Brad Miller’s disastrous trial in the OF. That said, Seth Smith graded out as a perfectly average OF, and he played more defensive innings than Cruz, and more than Trumbo and Miller combined.
One striking thing about the distribution of chances the M’s have handled is that nearly every M’s OF has faced slightly more “midrange” opportunities, and *every* M’s OF has faced fewer 5 star opportunities. I combined the middle three Statcast buckets into one group, and compared that to the hard/easy outliers. Across the league, about 23-24% of balls are in the 5 star bucket, and 32% are in the 1 star; this was remarkably consistent in both years. Combining the rest into one larger group, you’d expect players to get about 44-45% of their chances in this maybe/maybe not zone. In general, the M’s were higher than that, with the CFs (Martin in 2016 and Jackson in 2015) at 50% or so. A few of the corner OFs had fewer opportunities in this midrange zone, but in those cases, the extra chances went to the “easy” bucket, NOT to the hardest one. Seth Smith and Mark Trumbo had a lot more “easy” chances in 2015, so every M’s player season in the list faced fewer 5 star chances than we’d expect. I’m biased, of course, but I think that adds some credence to the hypothesis in my last post on OF defense that Safeco gives OFs a higher defensive floor than most parks.
That parks could change the distribution of fly balls makes some sense, as Colorado’s massive OF would mean balls hit a hundred feet from an OF might still be in play instead of in the 8th row of seats. Boston’s huge CF might account for why, say, Jackie Bradley and Mookie Betts had so many MORE 5 star opportunities (but it wouldn’t explain why they didn’t in 2015). That said, I’m pleased to see that getting a bunch of 5 star opportunities isn’t correlated with high defensive rankings. There are many ways to succeed here, as Adam Eaton got a lot more 5 star and 1 star plays and graded out brilliantly, while Kevin Kiermaier had the opposite distribution (lower on the tails, big in the middle) and did well, too.
It’s a tiny sample, but Mitch Haniger looks really good by Statcast, and it’ll probably come as no shock that Jarrod Dyson does, too. I think what we’ve seen here is that the M’s OF defense wasn’t the historically awful group that, say, DRS thought they were in 2015 and they weren’t bad in 2016, either. That said, there’s still plenty of room for improvement, and in Dyson and Haniger, the M’s acquired two players who will almost certainly get to more balls in 2017. Quantifying it is difficult, but 20 additional plays seems like a very conservative number to start with, assuming 5 plays below average for Aoki/Smith and 15 above for Dyson/Haniger. Decent positioning should help the M’s get to more of the tougher “midrange” chances, and the post-2012 dimensions means the M’s don’t have to worry about as many of the nearly-impossible ones. To get the biggest bang for their trading buck, the M’s would presumably hope for more opportunities overall, and that brings us back to the weird HR effects Safeco Field saw in 2016. Regression *there* would produce big savings in runs allowed, as you’d take HRs off the board (duh) and turn more of the resulting balls in play into outs.
Other random items that come to mind after staring at a spreadsheet:
1: I left the minimum balls in play threshold at 50, the same as the default setting at Baseballsavant.com. You can play around with it, but what it does is exclude some defensive experiments that might pull down the overall average. That means that the group who gets rated was the group entrusted with fielding a few games, and not just a few innings in an emergency. Thus, the league-wide plays above expected level isn’t zero – it’s quite a bit higher. Interestingly, that was much less of an issue in 2016 than 2015, when the cutoff meant the league appeared to make 200+ plays more than expected.
2: Jeff also mentioned this (damn him), but the Angels OF defense takes a hit when comparing these new Statcast metrics to UZR/DRS. Mike Trout goes to just below average, while Kole Kalhoun goes from unheralded defensive star to solidly below average in both years. No point to this, just some Statcast schadenfreude.
3: One hypothesis I’ve had is that fly ball staffs (and the M’s have been one under Jerry Dipoto) might produce easier balls in play. That is, that Marco Estrada, for example, gives up a lot of home runs (which aren’t in play) on well-struck balls, but then a lot of pop-ups on poorly-struck balls, producing a consistently-low BABIP. These buckets exclude pop-ups, I think, so there’s no real way to tell. The M’s don’t look like they saw MORE “easy” chances, but then, we’re not seeing the easiest of the easy. It might help to compare the total opportunities in Statcast versus total chances in BBREF and see what proportion of total chances are ranked in Statcast. My assumption would be that a higher percentage of Statcasted-balls would mean a more difficult group of balls in play. Someone (not me) should do that.
4: It’s really interesting to see how bad fielders drop off precipitously around 2-3 stars. Matt Kemp can’t make any plays in the 3-4 range, and neither can Trumbo or Cruz. Like you’d expect, they’re perfectly capable of making the easiest plays, and they’re not penalized too harshly for not making highlight-reel catches. But their plays made just fall off a cliff once the expected catch rate gets to 50% or so.
Comments
9 Responses to “Statcast’s New Defensive Metrics and Seattle’s Glove-First Outfield”
Leave a Reply
You must be logged in to post a comment.
Do these new metrics take into account defensive positioning?
Also, does DRS have the same fluctuations as UZR? Specifically looking at Leonys Martin. All you have to do is watch him to know he’s not a -2 DRS center fielder.
One OF’s 3 is another’s 5. I hope no one at Statcast spent too many sleepless nights figuring that out. Also, maybe they could come up with some buckets for whether there’s a cannon arm attached to the first stat. Meh. Although I do like the image of imagining into what bucket I would have dumped Morse.
But … very grateful we seem to have a big upgrade in the OF. However it’s statted.
Great article, Marc.
I recall McLendon once saying that Cruz “caught the balls he was supposed to catch”, which is somewhat supported by this new way of looking at the data – basically “we want his bat, so as long he doesn’t completely embarrass himself in the field I’m fine with that”. It’s too bad the data isn’t available for the Morse/Ibanez years, just for the lulz.
It’s also too bad there’s not a bin for those “sure thing” catches, if only because a) then all the data is in one place and b) it might be interesting to quantify how certain parks or certain pitchers might affect that as a percentage of all balls in play.
(I’m a big fan of having all related data in one place rather than scattered around)
I also suspect this methodology might stabilize significantly faster than UZR/DRS, which take so long that it can be hard to separate ability from other factors (aging, playing through nagging injuries, etc.)
Wow, you know who else doesn’t look too bad, based on this methodology? Our favorite old man, Ichiro!
The sample size is small, of course…
Winn-Cameron-Ichiro was pretty sweet back in 2001.
that Smyly Vs. Felix was fun to watch, hopefully its a good sign of things to come.
Winn-Cameron-Ichiro was 2003.
Martin-Cameron-Ichiro was 2001.
Great piece, Marc.
I checked out the post by Eno, and while it was interesting, I thought it was odd. Kind of had an overtone suggesting the Mariners’ outfielders would be susceptible to collisions. Hard to see that being an issue.
To me the deviation in magically gauging the difficulty of outfield flies on different days in different parks is a metric-buster. But obviously the Hope meter can beep/flash with the gross statistical trend.