What we can glean from Statcast’s new infield defense metric

Not to say that offensive sabermetrics analysis has reached its maximum—that’s impossible for any field of study—one can justifiably say that we have better certainty about who better hitters are than who better fielders are. The reason for this is mostly historical in how we trapped ourselves into one, and then another, frame of thinking.

The first one was to tally singular events where a fielder was involved. As far back as Henry Chadwick in 1859, putouts, assists, and errors—and later fielding percentage—would become the Holy Grail for defensive analysis. The thinking was that if a player faced X opportunities, and made more of them than other players, you could directly compare, and account for things like assists where you, say, are partially involved in the play.

The problem with that, like most older statistics, was context. Certain positions faced different numbers of opportunities, and certain positions received different different difficulties of opportunities, and some failed or succeeded opportunities were more valuable than others.

That then, up until literally within the last three years, lead into the dominant mode of thinking: Zone Ratings. In this model one could put some of those contextual pieces together. Fielders are each in a zone, and each zone has a particular penalty based on how difficult/easy it is. Each piece of that zone also has a general probability of success, which carry different run values. If I am a shortstop that makes a play that only 30% of shortstops make, then I get .7 times the run value of that batted ball. You then factor in positional adjustments, and more advanced models factor in both range, a player’s arm, double plays, errors, etc.

That makes sense until modern baseball rules broke it. For one, zones only make sense if players actually stay in the zone, and with modern shifting players are constantly moving into non-traditional positions. A stringer decides if the play is considered a “shift,” which is most plays, then the data is tossed out. What that means is that the sample size is extremely small, to the point where individual plays can throw entire seasonal outlooks, and where only multi-season UZR makes any sense at all.

This was, to that point, just considered the Great Unknown of our comprehensive metrics. Because half of the WAR component is defense, and we were relying on metrics that were flawed not necessarily in their conception but in practice of modern baseball, it meant that defensive WAR has been increasingly unreliable and throws a bit of uncertainty into our projections, assessments, and the like. It has also made shifting in of itself a fault domain that’s basically uncharted, in the sense that it’s just ignored from defensive data altogether.

Statcast changes that formula. Players’ locations can be logged, so no need for pesky, generalized zones. We also have precise data on exit velocity and launch angle, which (barring the baseball itself as a large asterisk) gives us based on the pool of data a sense of Catch Probability or the probability to make that play. This was already implemented with outfielders’ Outs Above Average, which was a helpful tool recently on the Statcast leaderboard, to the point where that along with wRC+ had to be used to create a mental image of weighted value.

That was, until yesterday, missing for infielders. Tom Tango, in an excellent paper that is publicly available, breaks down both this history and also a new theory for Outs Above Average for infielders. Instead of the Zone Model with weights, we have the intercept model. This is succinctly defined as:

“[W]e have the Opportunity Distance, which is how much distance the fielder has to cover. And the Opportunity Time, which is how much time the fielder has to cover. And the two gives us the fielder’s Opportunity Space. And that point when fielder and ball cross is called the Intercept Point. In the event the ball and fielder don’t meet, we determine the Intercept Point as the point when the distance of the ball to the plate and the starting distance of the fielder to the plate are equal.”

In short, it’s the Pythagorean’s Theorem of defense, and it’s very easy to understand in the sense that we already think of defense in this way as a mental model: the ball is going to a location where the ball will land/pass the fielder, and they have to reach it in time to make that play, and that is what defines a “good defensive player.” Our thinking is off insofar as we can’t intuit how fast a ball goes, or what that intercept point is, but Statcast solves that problem.

That doesn’t mean every issue from zone rating is gone, of course. Sample size is still going to be an issue in the sense that Statcast still needs to successfully log the data, so NULLs won’t count. This has already been a problem with pitching data, where Statcast has logged more NULLs than PitchF/X did, for example.

Not only that, but exit velocity and angle matrices are somewhat foiled by the fact that we don’t know what the ball is going to be like, so it’s harder to use these opportunity times when they could vary by a couple of feet depending on the drag coefficient of the ball. It’s not the biggest factor, but it’s just a small veneer of uncertainty, a possibly unavoidable one.

It also still leaves positioning, and the coaching behind it, as that lurking Great Unknown, as I said. Though Tango defines “roles” instead of positions, which is a more intuitive understanding considering widespread shifting, someone is still ultimately choosing that role. This is no fault of the metric itself but it’s more an observation of the field writ large.

Someone is deciding that a player should be in a “role,” and they are making that decision based on spray charts, scouting, and and an understanding of where that infielder plays best. I genuinely wonder if, why, and how players excel in certain roles, and how teams balance that with, like I said, spray chart data or an understanding about the hitter when deciding a shift.

As Mike Petriello wrote for MLB yesterday, this has interesting implications when we compare this metric to UZR or DRS, though not exact when we compare outs to runs. Freddy Galvis, for example, is considered an excellent defender (at +12 OAA) despite a negative UZR/DRS. Xander Bogaerts, for example, has disagreements amongst all three, where he is -3 OAA/-20 DRS/+1.2 UZR.

It also reveals something interesting in another way. UZR and DRS really weren’t that wrong (for the obvious ones). The best advice I ever read in regards to whether a metric is good or bad is whether you get the best players for the most part at the top of the leaderboard, and we see a lot of familiar names: Javier Baez, Andrelton Simmons, Nolan Arenado, and Matt Chapman are already considered the game’s best defensive players, and that doesn’t change. It makes sense when you consider the idea that singular events can tell us a lot: even in UZR/DRS small samples, the plays that are <20% probability are probably only made in non-trivial amounts by the very best players.

While there is still a lot more testing to be done, and Kenny Kelly will discuss Galvis as an example in greater detail, the conclusion is that we now have a much better understanding of how good infielders are, and where their strengths/weaknesses truly lie. Offense isn’t set in stone, obviously, but we can at least say that for the first time, the two schools of thought are finally running closer to even with each other.