clock menu more-arrow no yes

Filed under:

Thinking beyond one batted ball

Last week, I wrote about what we can learn from one batted ball. Let’s expand on that with some inferences.

MLB: Exhibition-Chicago White Sox at Chicago Cubs Kamil Krzaczynski-USA TODAY Sports

In a season of small samples, judgments need to be made rather quickly.

The Rockies might be good. The Marlins are 18.9-8.1. Mike Yazstremski is in the race for National League MVP. Of course, even in the 60-game sprint, things should stabilize, especially on the team level. But — given that we’ve all been trained to accept randomness this year — it seems to me that it’s been almost more commonplace to start making conclusions earlier, whether or not these will hold up in the long-run.

That’s why, last week, I had some fun examining what conclusions we can draw from a single batted ball. In particular, I highlighted Luis Robert, who turned on the first pitch he saw in the majors for a single with a 115.8-mph exit velocity. And though that singular exit velocity event certainly popped off the page, and did allow us to potentially adjust expectations for Robert in the isolated power department, it’s still just one datapoint. The correlation between max exit velocity and isolated power was just moderate, with an r-value of 0.55. Max exit velocity, thus, can explain roughly 30 percent of the variation in ISO. While that is a lot for one batted ball, it’s far from enough to further refine Robert’s 2020 profile.

As I mentioned near the end of the piece, thinking beyond one batted ball forces us to make some trade-offs. The minute we consider other information is the minute we turn an analysis of hard data into an analysis of hard data plus some inferences. However, while inferences may help us reduce the variability of the response variable, it also leaves some of the input variables as open-ended questions. Simply, we know Robert’s maximum exit velocity (to date), but we may not yet know more details about his batted ball profile, since they have not yet stabilized.

But we can try to estimate, and that’s what I plan to do now. One batted ball isn’t enough, of course, to make much of anything. I explained the Vladimir Guerrero Jr. phenomenon in Thursday’s piece — he’s someone with enormous power, but because he hasn’t tapped into the power on line drives and fly balls, his ISO gets suppressed.

That’s why we need to consider other variables. As long as we’re keeping ISO as the response variable, two important additional inputs are groundball rate (as previously mentioned) and hard-hit rate. When throwing max exit velocity back in the mix, we see that, while all do correlate with ISO, there is a distinct order here:

Correlations with Isolated Power

Metric R R-squared
Metric R R-squared
GB% -.466 .217
Max. EV .545 .297
Hard% .634 .402
Among players with 300+ plate appearances, 2017-19

Logically, all three variables are useful in understanding where a player’s power comes from. Maximum exit velocity effectively tells you a player’s raw power, groundball rate focuses on the activation of that power and hard-hit rate represents the repeatability of the power. It’s no good if you can hit the ball hard, but you can’t hit the ball hard often or in the air.

From there, we can develop rough models of Robert’s power potential yet again. This time, we have less information from his end, but we have more variability covered. With all three inputs considered, we can explain nearly 60% of all variability in ISO, a mark that is pretty good considering the simplicity of the inputs.

But for a player like Robert, it’s hard to know what his groundball and hard-hit rates will eventually look like. Using reliability data from FanGraphs, we can use some estimates around a 95 percent confidence interval to estimate in which range Robert’s groundball and hard-hit rates may ultimately fall. Because, into games on Sunday, Robert has only hit 37 balls into play this year, these intervals are rather wide:

Luis Robert’s bounded GB% and Hard%

Stat GB% Hard%
Stat GB% Hard%
Lower Bound 36.2% 21.2%
Robert, 2020 45.9% 32.4%
Upper Bound 54.0% 37.6%

To get a sense of how wide these ranges are, the lower bound of Robert’s groundball rate — a 36 percent mark — would put him in the 82nd percentile, where lower is better. (The lowest groundball rate, therefore, would be considered the 99th percentile.) The upper bound, meanwhile, would put him in the 2nd percentile. A similar circumstances is true when looking at the bounds of his hard-hit rate: a 21% hard-hit rate is worse than Billy Hamilton, while a 38% hard-hit rate puts him in the 40th percentile.

With these three pieces of data, you can begin to build a much more accurate case for Robert’s power potential. In what we’ll call the best case scenario, where Robert meets the 40th percentile hard-hit rate and the 82nd percentile groundball rate, we’d project a .240 ISO for this, a mark 10 points above the prediction made from his max exit velocity alone. Going from his current numbers, with the 45.9% groundball rate and the 32.4% hard-hit rate, we’d expect a .196 ISO, a mark actually below what we “learned” from one singular batted ball. In the “worst case,” a high groundball rate and a low hard-hit rate, Robert would be projected to put up a .139 ISO.

This, of course, yields a ton of uncertainty. Looking at 2019 numbers, we’re either saying he’s a Kris Bryant (.239 ISO), a Jackie Bradley Jr. (.196 ISO) or an Adam Frazier (.139 ISO). In fact, it almost feels like we’re worse off with more information, as it provides a more speculative range, with no outcomes looking particularly rosy for a player who has done nothing but impress in the first 15 games of his major league career. (For what it is worth, he has a .166 ISO into Sunday.)

Thinking outside of the reliability curves, though, we can take a closer look at the relationship between all three of these variables:

Percentile-based outcomes for GB% and Hard%

Percentile GB% Hard% Regressed ISO 2019 ISO Percentile 3-Year ISO Comp
Percentile GB% Hard% Regressed ISO 2019 ISO Percentile 3-Year ISO Comp
10th 50.3% 31.5% .181 43 Nomar Mazara (.181)
20th 47.9% 34.2% .196 53 Todd Frazier (.196)
30th 45.5% 37.0% .211 60 Max Kepler (.211)
40th 43.8% 38.0% .219 65 Rafael Devers (.219)
50th 42.0% 39.7% .229 71 Ian Happ (.230)
60th 40.3% 41.3% .239 77 Freddie Freeman (.239)
70th 38.7% 42.3% .247 81 Anthony Rendon (.247)
80th 37.2% 43.7% .255 84 Rhys Hoskins (.255)
90th 33.8% 45.7% .271 91 Nolan Arenado (.270)

What this suggests is that even if Robert was to have both a 10th percentile groundball rate and a 10th percentile hard-hit rate, he’d still be expected to put up a 43rd percentile ISO. Why? His maximum exit velocity. Raw power does indeed go a long way.

But all of this demonstrates the inherent problem with relying so heavily on one batted ball. Guerrero remains the best example of why raw power may not necessarily translate. Since he has over 550 plate appearances in the majors to date, it’s not hard to understand why the raw power hasn’t led to great results. Yes, the maximum exit velocity looks excellent, but the career groundball (51%) and hard-hit rates (36%) don’t support the same pop. The multi-linear regression implies a .213 ISO, significantly above his career mark to date (.162), but well below what the regression with maximum exit velocity alone implies (.258).

That’s naturally the problem with statistics. As much as we’d like to ooh and aah over one batted ball, we can’t. It might tell us something, but, as always, we ultimately need more information. And, in the case of Luis Robert, let’s enjoy the show while we gather it.


Devan Fink is a sophomore at Dartmouth College and a Contributor at Beyond The Box Score. Previous work of his can be found at FanGraphs and his own personal blog, Cover Those Bases. You can follow him on Twitter @DevanFink.