Judging Yankees outfielders by catch probability

You know I love new metrics. I recently talked about the new incarnation of Deserved Run Average, and how that affected our perceptions of both Michael Pineda and CC Sabathia. Now we have another new stat courtesy of Statcast, and it may change the way we look at outfield defense.

The stat is “catch probability,” which was already used in a way by Baseball Info Solutions. BIS’s stat was subjective, though, so it was the responsibility of a human tracker to put batted balls into categories to determine difficulty. This stat, on the other hand, uses statcast cameras to determine the probability of a catch being made based on “opportunity time” (the time to make the play) and “distance needed” (the shortest route to the ball).

The plays are then broken up into five buckets, one through five stars, with the following probabilities attached to them:

• Five stars: 0-25%
• Four stars: 26-50%
• Three stars: 51-75%
• Two stars: 76-90%
• One star: 91-95%

Based on this, and the fact we also have 2016 data for four Yankees outfielders (I also include Matt Holliday), we can figure out just how good each player was for each type of play.

To do this, I created a metric on the same scale as wRC+, Five+ through One+, which is how much better the player is on that type of play than the league average, which I include below. So if a player has a 101 Five+, that means they’re 1% better than league average on five star plays, or plays with a 0-25% chance of being made. Here’s what I found:

The sample for Holliday is incredibly small, but it’s not great. The rest of the results are what one would expect. Hicks didn’t make any great plays but seemed to do well on others; Gardner was surprisingly good at five star plays; and Ellsbury looks a shade better than league average overall.

I find these metrics fascinating because unlike a lot of the previous Statcast data, which in my opinion was very much in the vein of look-how-cool-this-is without actual descriptive data on how this translated to value, there’s something concrete here. I could explain this to someone who understands baseball but doesn’t understand statistics and it would make sense. We all intuitively know when a player makes a great play and misses an easy one, and I feel like this maps a mathematical solution on to an already sound logical framework.

It will be interesting to see how this gets integrated in to public defense metrics, because if we know one thing, the likes of UZR and TZ are outdated and in small samples are worse than having no data at all. To have something like this, something that makes good sense and gives us an idea of a player’s range, would be useful in better quantifying the unicorn that is defensive value.