We’ve Got to Get Over Our Fascination with Barrels

A friendly disclaimer because I gather this is a sensitive subject for some: truly nothing in here is a personal attack! My intention isn’t to call the folks engaged in this behavior stupid, or bad analysts. I can 100% guarantee that many of the people who’re looking–perhaps myopically–at barrels are better at quantitative analysis than I am! This isn’t so much a math issue as it is a marketing and psychology one, I think. All that said, I do think that the current fascination with “barreled balls” is a mistake, and one that speaks to the larger problem of “statistics as marketing tools” in Major League Baseball.

Naturally with that disclaimer, I’m going to ignore baseball for a minute and talk about my day job. I do analytics for a large-ish publisher. In some ways, my job could be described as “do math about screwing around on the Internet all day,” and it’s a lot of fun. There’s nothing in the following anecdote that isn’t public knowledge, but I think it’s instructive.

One of our sites is called Quartz–it’s ostensibly a business site, though you’ll occasionally find stuff on there that isn’t, to turn a phrase, strictly business. Such as an article that went up early on in its lifetime–an article about Colony Collapse Disorder in honeybees, titled “Scientists Discover What’s Killing the Bees, and It’s Worse than We Thought.

The article got a ton of traffic. Enough that in a matter of days, it was the highest-read article in Atlantic Media history (it’s no longer anywhere close to the record holder; our sites have grown quite a bit in the last several years.) But at the time, it was a Real Big Deal!

On the surface, there was nothing particularly compelling about the article. With respect to the author, it’s not as if the prose were somehow uniquely amazing. The headline used some best practices at the time (specifically the use of “worse” and “we” tended to generate slightly more traffic in the aggregate), but not to any degree that the article should’ve gone viral on headline alone. It wasn’t even the only article on the web about bees dying–there was a bunch of this at the time, and this was just Quartz’s take on it.

But, dammit, our new site’s biggest success was about honeybee death, and I’m sure you can guess what comes next. All the automated traffic reports that sort articles in descending order come out showing dead bees standing head-and-thorax above the rest of the articles, and folks are psychologically hard-wired to gravitate toward the successes.

In the next several weeks, Quartz wrote several more articles about bee death–not enough to spin up a bees channel, certainly, but enough that I could crack jokes about a bee beat. I knew none of these articles were going to do particularly well; I think on some level we all did–and indeed, they didn’t. The viral article caught a perfect storm of stuff, its subject matter being only one small part of the magic formula. Still, if one looked at the average traffic to a bees article, it remained incredibly high, the mean dragged up by the viral article with traffic in the seven figures.

To borrow some baseball parlance, though, if we looked at the batting average (meaning number of successful articles divided by number of total attempted articles about bees), it was pretty clear it was 1-for-9 or whatever. One monster home run article in a corpus of weak groundouts. By focusing on the one home run, we lost sight of all the other attempts that were screaming “bees don’t equal traffic!”

Hopefully, you see where I’m going with this. If not, I’ll do my best to spell it out.

The new hotness in MLB statistics seems to be “barrels.” Again, to MLB’s credit, the definition for a barreled ball is clearly defined on its website. To use less formal language, a barreled ball is little more than a ball that the batter squares up and hits the piss out of–balls that by definition exit the bat at nearly 100 mph, and at an angle that all but ensures the ball travels pretty far.

This makes sense! Intuitively, if a ball isn’t getting pounded into the ground (the required launch angles preclude this, by and large), a hard-hit ball should produce good results for the batter more often than not. Fielders have less time to react to hard-hit balls, and balls that are especially hard-hit find gaps, go over fielders’ heads, or leave the yard entirely–so I’d expect to see high slugging percentages to go along with the typical barreled ball.

A barreled ball is unequivocally a good thing, even if an individual barreled ball doesn’t necessarily yield a good result. Think of the last time your favorite player hit a screaming liner right at someone for an out–the ball was probably barreled, but it didn’t do your player a whole lot of good. In the aggregate, barreled balls will yield good results (indeed, as per the definition, balls with the requisite exit velocity and launch angle typically yield at least a .500 AVG and 1.500 SLG).

Yet…that’s not how barreled balls are being served to us. Watch an MLB broadcast and you’ll see announcers struggling to describe launch angles, exit velocity, and barrels–they’ve got access to this data, and so they think they need to use it. In the season’s first week I’ve seen Jim Palmer and A.J. Pierzynski try their damnedest to sound intelligent talking about this stuff, and fall well short. It’s not their fault.

There’s a twitter account, MLBBarrelAlert, which lets us know any time a ball has been barreled, along with its peripheral data (the player who did it, how many barrels that makes for the player on the season, the pitcher he did it to, exit velocity, launch angle, result.) Like so:

I’ve no reason to believe the person(s) operating the account are anything other than just curious fans who’ve gravitated to this new metric. After all, as pointed out previously, a barreled ball is a good thing in general. The event is good. It does not stand to reason, however, that players who hit barreled balls are good. And that’s where the disconnect lies.

In my mind, for a statistic to be worth adopting on its own (meaning I’m sure barrels have their place as peripheral data, and I’ll get to that), one of two things should be true: either it’s completely determinative of a good outcome, or it’s got great predictive value of future outcomes. Barreled balls, barreled alerts, barreled player leaderboards…none of these fit either criteria.

What do I mean by this? Think about a determinative outcome–these are typically “counting” stats. For example, I’d want to see a home run alert, because regardless of the homer’s exit velocity, launch angle, and so on, it’s a thing that 100% of the time results in a run. It probably helps that batters can pimp homers too, making for good video on mlb.com, SportsCenter, and so on. A home run, regardless of how well-struck the ball was, makes for both a good highlight and also has a meaningful impact on the outcome of a game.

Predictive value, on the other hand…these are typically “ratio” stats. Think of something like batting average (assuming we’re not in week 1 of the season, or that you’re willing to look at career average, or somesuch.) If I see a player is a .300 hitter, I might not know that the player is going to get a hit in his next at bat, but I have a general sense of his hitting ability going forward, especially relative to a .250 hitter. Sure, there are some guys that have hollow batting averages (Jon Jay, Juan Pierre, etc.), but I’ve got ISO or SLG or OPS for that. I can look at any of these ratio-based metrics and get a sense of what this hitter will do going forward.

I think we can all agree that barreled balls do not have determinative value. Look no further than this alert:

Matt Kemp hit the piss out of the ball–for an out. Nobody gets excited over line outs; it’s not gonna be at the top of anyone’s list of video highlights, either. For the casual fan, this does nothing in terms of immediate gratification–outside, perhaps, of the splendid frustration that baseball offers its fans.

For barrels to be a good standalone metric, then, they’d need to have significant predictive value–i.e., “Is a player who barrels the ball more than his counterparts a better hitter than his counterparts?” Or even “does a team that barrels the ball a lot tend to hit a bunch of homers?” The metric struggles there.

Remember, the Cardinals led the NL in homers last season (#2 in all of MLB to Baltimore). Surely they must’ve had a bunch of barrel-masters, right? Yet for all the homers the club had, you’ll never guess who the barrel leaders for the Cardinals were. Surely some great hitters, right? Well…maybe if you’re an extreme optimist. The team’s leader was Tommy Pham, a guy who didn’t even make the big league roster in 2017. Finishing second? Brandon Moss, who drew virtually no attention in free agency. It’s not until you get to #3 on the list that you reach an unequivocally good hitter: Matt Carpenter, who barreled 6.9% of all plate appearances in a nice 2016 campaign.

Thanks to the wonderful Daren Willman’s Baseball Savant, we can see this data for all players last season. So as to control for varying amounts of playing time that could inflate or deflate a given player’s cumulative barrelled balls, let’s look at the 2016 leaders in “Barreled Balls per Plate Appearance,” meaning we’re looking at a ratio rather than a counting stat.

At a glance, there’s a lot of really good hitters here! But there are also a lot of dudes on this list that aren’t good hitters. Byung-ho Park and The Ghost of Ryan Howard are perhaps the most notable bad hitters, but it’s not as if Rickie Weeks, Pedro Alvarez, or even Chris Davis would crack anyone’s list of the 50 best hitters in the game.

If there’s one common thread for these dudes it’s that there’s a lot of guys I’d term “home run hitters” in the mix. Which makes sense given the prerequisite batted ball velocity and launch angle for a ball to be considered barreled. But a lot of these dudes are what I’d call all-or-nothing types. “Three True Outcomes” doesn’t work here because a lot of these guys can’t draw a walk with a pen and paper.

Khris Davis, at the top of the 2016 leaderboard? A .307 OBP and a 27% strikeout rate. Mark Trumbo, last year’s home run leader? A .316 OBP and a 26% strikeout rate. Chris-with-a-C Davis? A respectable .332 OBP but a brutal 33% strikeout rate.

It’s a fair assessment that a lot of the folks on this list are frequently “selling out” in order to hit the piss out of the ball–to barrel it up. If they hit it, look out, but if they don’t barrel it up, contact is either going to be poor or nonexistent. It’s the Dave Kingman profile, or if you prefer a slightly more recent player, the Rob Deer profile.

What’s notable is that even at the very top of the leaderboard, players can’t barrel up the ball in more than about 10% of all plate appearances. (Percentages get higher when we consider batted balls, but that already removes the very real possibility of a strikeout.) The question we need to ask ourselves is “are we focusing too much on the 10% of outcomes that involve barrels that we’re ignoring the 90% of non-barreled balls?”

More specifically, if some of these players choked up (or were otherwise more conservative) a bit, would they improve the contact quality of their non-barreled balls enough to offset the handful of screamers they gain by selling out? It’s not as if guys with relatively low barrel percentages can’t hit home runs. Bryce Harper, Jose Bautista, Buster Posey, Jose Altuve, Joey Votto, and Paul Goldschmidt, amongst others, aren’t even in the top 100 of barreled ball percentage.

I strongly suspect that in many cases, hitters that are selling out are losing more than they gain, trading chances at extra base hits for chances at mere singles. Certainly, there’s no shortage of analysts, players, and coaches preaching the gospel of “swinging up,” which might explain the propensity of a hitter to “sell out” for power. Making soft contact is, not surprisingly, negatively correlated with AVG, but there’s basically no correlation at all between medium contact and batting average. It’s kind of a “duh” statement, but a player’s chances of getting a hit improves the harder he hits the ball–but even soft contact, by definition, has a greater hit probability than a strikeout. You never get a hit on a strikeout.

So…how predictive are barreled balls of whether someone’s a good hitter? It’s not terrible, certainly. (Data found here, merging the available Statcast data on Baseball Savant with that of Fangraphs). Factoring out players from 2016 with very few plate appearances, the correlation between barreled ball % and OPS is .63, which I’d characterize is “pretty good; there’s clearly a positive relationship here.” As you might expect, the correlation strengthens when we remove OBP from the equation and look purely at SLG–it jumps to .72. Makes sense, given that barreled balls result in extra bases, and that’s what feeds slugging percentage. Presumably it’d jump even further if compared to ISO, thus keeping out all the ground-ball singles that can contribute to slugging percentage.

But here’s the problem: I can do this evaluation without needing to look at barreled balls. “Hard-hit percentage” from FanGraphs, correlated with OPS, has a correlation of .60, virtually the same as when I did it with barreled %. Vs. SLG, it’s .66, again rather similar. This, too, makes sense, given that the “barreled” category is a derivative of “the launch angles and exit velocities that tend to have exceptionally high slugging percentages.”

It’s this “what’s the tradeoff for barreling a ball” question that makes me think the “barreled” stat isn’t useless, but the way it’s currently being sold is. Here’s what I’d do with it if I had the data (and seriously–if folks’re willing to share it, I’ll do it myself. The data’s already there but I lack the skills/patience to drill into each individual batter on his Statcast leaderboard, scrape the individual at-bat data, and repeat.)

Barreled balls have a strict definition, remember, dealing with launch angle and exit velocity. Rather than looking at just the barreled balls, I’d want to see what hitters do with the rest of their batted balls. Is the reason Mike Trout’s so good because even when he doesn’t barrel the ball, it’s a near-miss? Are a large portion of his batted balls just a shade off in either launch angle or exit velocity? Is the reason Byung-ho Park is distinctly not Mike Trout because when he fails to barrel a ball, he doesn’t even come in the same zip code as the requisite velocity/angle combination? Some of this is already addressed in barrel-analogous categories like “burners” and “flares,” though I’d be more interested in the individual batted ball event than in trying to create still more categories.

Basically, I’d want to see what each hitter’s variance was vs. barreled balls. I strongly suspect that the best hitters have a much tighter array of outcomes around truly barreled balls, whereas the Carters and Parks of the world have a much wider array of outcomes, most of them involving low-hit-probability angles and velocities (assuming they hit the ball at all.)

In other words, barreled balls probably aren’t useless, but neither are they good as a counting stat nor as a ratio stat. As a benchmark for the ideal batted ball against which all other batted balls can be judged, barrels could work pretty well.

Mostly, I’m taking it on faith that I’m not alone in thinking barrel ability isn’t a great standalone metric because people whose job it is to know these things seem to agree with me. Look no further than the fact that Chris Carter–high barrel percentage and all–was lucky to find a job this offseason. Same for Brandon Moss, Pedro Alvarez, and so on. Tommy Pham didn’t even find a job so that Matt Adams could patrol left field for some reason. Guys that tend to have the all-power, no-contact profile–especially when they don’t play a premium defensive position–aren’t exactly in high demand among MLB GMs. Likely related: GMs have access to far better analytics than the rest of us do.

Aside from the math of angles and velocity, consider also the fact that there’s a natural cap on what a barreled ball can do. Sure, Byung-ho Park might’ve had an average homer distance of 420 feet last season, but it doesn’t really matter. In almost every circumstance, hitting a ball 400 feet will result in a home run. You don’t get extra runs for hitting a ball so hard that it lands in the upper deck–getting a video highlight doesn’t count on the scoreboard. Once a ball is a home run, regardless of whether it was a barreled moonshot or if it was a wall-scraper, it counts the same. The surplus distance accomplishes nothing.

Perhaps if this were the grim future where Rob Manfred, seeking an improvement nobody wanted, mandated that all fields push the fences out by 50 feet in all directions, barreled balls might matter a whole lot more. The extra oomph that the barrel leaders tend to produce might mean the difference between a long fly out and a home run. But that’s not the world that we live in right now, thankfully.

All of which is a long way of saying this: barreled balls aren’t telling us much more than what we know already: hard-hit balls tend to do better than balls that aren’t. But from a player evaluation standpoint, barrels aren’t anything special. Like the Quartz bees article, by focusing myopically on a player’s ability to barrel a ball, we’re ignoring what happens with the other 90%-plus of his plate appearances. If you’re focusing on barrel leaderboards, or celebrating barreled balls as the be-all end-all of successful at bats, you’re doing precisely what Russell Carleton (one of the aforementioned people who’s better at quantitative analysis than I am) warned not to do last year.

If we’re going to make a push to market a statistic, it should have some unique merit to it. Barrels as a standalone metric are merely reinventing the wheel, yet the resulting wheel is one nobody quite knows what to do with, and when properly implemented works roughly as well as the existing wheel. It strikes me as data for the sake of data, and in a sport where “data analytics” is a sexy buzzword, it’d behoove us all to make sure we’re not just gravitating toward the shiny new metric just because it’s new.

None of this should preclude folks from enjoying it if they’re already doing so, certainly! And, while it irks me to no end that I can’t get at most Statcast data, that’s already been covered. The fact that barreled data is at least publicly accessible is great, and perhaps someone out there will eventually find a novel new use for it. As a starting point, I really hope that someone more computer savvy than I can take what’s inconvenient-but-technically-accessible on Baseball Savant right now and create a data file with which to examine barrel variance. That said, as it’s being marketed right now, both formally by MLB broadcasts and informally by well-intentioned fans, there’s nothing to recommend barrels over anything else that we have, and plenty to recommend against it.