A case study in hard contact and the limits of Statcast

A couple of weeks ago, Justin Upton launched an Alex Colomé fastball deep into the Seattle night. The Mariners were ahead by two and the Angels had runners on the corners with two outs. Off the bat, it looked like the ball was destined to fly over the fence. Watching at home, I immediately thought Colomé had given up the lead. To my surprise, the ball settled into Guillermo Heredia’s glove, 372 feet away from home and about 30 feet short of the fence. Crisis averted.

Here’s a GIF of that fly out for reference:

Because we live in the era of Statcast, we can find out all sorts of information about this particular batted ball. It was hit with an exit velocity of 103.1 mph with a launch angle of 36 degrees. Since 2015, a batted ball with those characteristics has gone for a home run 55% of the time. This one didn’t even really come close.

Of course, there are a number of other factors that can impact the flight of a ball: park effects, the direction it was hit, the weather on that night. Beyond illustrating the obvious limits of Statcast powered metrics like hit probability, I think this batted ball provides an interesting case study for the way the very fabric of the game has changed over the last few years. Before we get too deep, let me present another batted ball off the bat of Justin Upton with nearly the same characteristics as the previous one.

This one was hit on June 20, 2017 off an Aríel Miranda slider. It was hit with an exit velocity of 100.6 mph and a 30-degree launch angle. It traveled 411 feet, clearing the center field wall by a healthy amount. Batted balls with these characteristics become hits nearly 60% of the time and fly over the fence 46% of the time. This one was hit three miles per hour softer than the other one above—albeit with a better launch angle—and yet it traveled 40 feet farther.

It might be disingenuous to try and compare two different batted balls hit nearly a year apart because there are so many variables at play. One variable that has been examined at length and is pertinent to this discussion is the baseball itself. I’m no physicist so I’m going to call on a few others who are much more knowledgeable than I am.

Marc over at USS Mariner beat me to a version of this article last week, so you should definitely go read that one first. Rob Arthur of FiveThirtyEight and The Athletic has done most of the yeoman’s work investigating the “juiced” ball. He found that the general drag on the ball decreased significantly from 2015 to 2017. When air resistance decreases, the ball travels farther. This season, however, it appears as though air resistance has increased over the levels we saw in 2016 and 2017.

In a very practical sense, this fluctuation in the physical properties of the baseball has led to wildly different offensive environments over the last few seasons. Alex Chamberlain observed these differences in a recent article posted on FanGraphs.

Hard contact and league average exit velocity are higher than ever but the league-wide home run rate has fallen back to 2016 levels. One of the explanations could be the league-wide use of humidors to standardize the way balls are stored. But as Arthur points out, the use of a humidor would reduce the effects of drag while also suppressing the exit velocity of batted balls. Which makes the record exit velocity numbers we’re seeing this season all the more surprising. I think the data makes it clear—and Arthur and Chamberlain agree—that the physical properties of the baseball have changed again.

In the table above, that last column on the right is particularly interesting. One of the metrics that’s come out of all this new Statcast data is expected wOBA, or xwOBA. The whole premise of this metric is to use past exit velocity and launch angle data to calculate an expected outcome for any given batted ball. It’s the basis for the hit probability percentages I used in the case study above. But what happens when all of the data created for the first three years of Statcast’s existence was created while the “juiced” ball was being used? And what happens when the ball changes dramatically again, like we’ve seen this season?

In aggregate, any “expected” metric should equal the actual metric it’s estimating. We saw that, generally, for the first three years of Statcast data. But this year, xwOBA is overestimating the expected outcomes by a relatively large margin. So when we say that Justin Upton’s fly out in July 2018 had a 60% hit probability, that’s based off old, obsolete data. In the current offensive environment, it’s likely that particular batted ball had a hit probability considerably lower, hence the harmless fly out in front of the warning track. Statcast is almost four years old now and we’re still struggling to figure out how to apply all this new data to sabermetric analysis.

This is all very interesting on a macro, league-wide level. Tomorrow, I’ll examine what exactly all this means for the Mariners. Spoiler alert: it’s a very good thing.