Getting to Know Exit Velocity Distributions

Although they offer really helpful snapshots of particular data points, measures of central tendency (mean, mode, etc.) sometimes leave a lot to be desired. To help with that, there are all sorts of measures for variation, which capture how data are spread around those central values. Baseball, for all its wonderful metrics, sometimes neglects these nuances, instead emphasizing mere averages (BA, OBP, OPS) or simply baking them into more complex metrics (OPS+, wRC+, which each normalize around 100). Of course, there are exceptions. Alex Chamberlain’s work on launch angle tightness is really interesting to me, for instance. Still, in some ways there seems to be a lot left undiscussed.

One particular point of interest (to me at least) are those instances when measures of central tendency diverge. When measures of central tendency do not align, it generally suggests skewed data. Thinking that any particular distribution of metrics were likely to be skewed, at least somewhat, I dug into the distribution of exit velocities (a natural place given its value and presence in today’s game).

Most often, exit velocities are presented in two forms: average and maximum. Both have proven to be of significance and value, each correlating with both power and production. Still, there seems to be a lot to address: how does any given player’s median exit velocity stack up against his average EV? Does the difference between those two metrics signify anything? Might median (as just another measure of central tendency) exit velocity in fact offer insights a simple average does not?

Data from Baseball Savant was drawn upon in order to attempt to address these questions. Very simply, event data was pulled for all batted balls struck by players with greater than 300 total events in 2021 (through August 16th). In total, 63 players qualified. Given that data, average exit velocities as well as median exit velocities for each individual were measured. It might come as no surprise to you that the two metrics in fact didn’t align (like, ever), thus indicating a skew in the distribution of player exit velocities.

Skewed distributions generally feature stretched “tails” toward those extreme values, high or low. A right skew, for instance, might feature a fairly normal looking distribution albeit with several data points trailing far rightward. As an example, personal incomes are often right-skewed as most incomes are within some general range although particularly high-paid individuals are represented by points extending far right of the most common income brackets toward the center.

As it turns out, exit velocities are (at least given this sample) exclusively left-skewed; players most often put the ball in play at roughly 90mph, yet a solid handful of times are either jammed, popped up, etc. and their corresponding exit velos are much, much slower.

As an example, below is the distribution of Nolan Arenado exit velocities in 2021, with an obvious left skew similar to that of his colleagues.

Arenado is actually an interesting example in that, while the righthand side of his exit velocity distribution is roughly normal, his distribution appears to be somewhat multimodal – meaning there are two at least semi-distinct peaks – around 90mph as well as 75mph. However, the primary point is that long-sloping tail on the lefthand side, indicating the wide array of exit velocities on balls struck sub-optimally.

For a numeric breakdown of these distributions, below is a chart of those players with 300 batted balls in 2021 with the lowest, and highest, “avg EV – median EV” figures. You’ll note that the left-skewed distribution corresponds to median EV figures being exclusively larger than the average EV measures, i.e., no average exit velocity is so great as the corresponding median exit velocity.

The distribution of Avg – Median EV is not particularly drastic, as all 63 players sit within 4.1 mph of one another. Still, there is a clear difference among those most drastic cases. To illustrate this, below is a mirror histogram featuring the EV distributions of J.P. Crawford, who tops this table, and Pavin Smith, who sits at its bottom.

Above, the majority of J.P. Crawford’s batted ball exit velocities almost make up some semblance of a (rough) uniform distribution whereas Pavin Smith’s distribution features a much longer leftward slope. Intuitively, this all checks out. Players sometimes do not hit the ball as hard as they’d like to; meanwhile, even when they really get ahold of a pitch, there are human limits to how hard they can hit it.

What does all this mean though? For performance, not much at all. Median exit velocities correlate to performance metrics like wRC+, wOBA, and xwOBA, but slightly less so than average exit velocities. Those players with a larger differential between the two metrics perform roughly the same as those with slimmer central measure differences.

Below that relationship, or lack thereof, between wOBA and Avg/Median EV differential is illustrated.

The fact that the skew of exit velocity isn’t related to performance is probably why this isn’t talked about much. Regardless, it is interesting to consider how two varied distributions of batted ball profiles can result in very similar exit velocities. Players also might do well to avoid particularly long leftward tails in those distributions as the percentage of balls his under 85mph by each player does track with performance measures like wOBA; the R-Squared figure for that percentage and wOBA is .397, while not incredibly robust, it is well worth considering.