# You Can’t Fake Exit Velocity

Last week, I spent a few articles idly hunting for hitter breakouts. I centered my search on players with admirable top-end power numbers but who reached that summit rarely. I found that when those players increased their contact rate, they improved their overall line significantly. I think that finding tracks with intuition in addition to having data to back it up, so I’m overall pleased with that research.

That said, all this downloading and scraping of exit velocity data made me wonder about the opposite side of this spectrum: can hitters add power and break out from the other direction? Hitters who make a ton of contact but don’t hit the ball with much authority feel somewhat capped offensively; in my head, Luis Arraez has a 0% chance of turning in a 20-homer season. I didn’t have the numbers behind that, though, so I gathered up the same pile of data I’d used before and started hunting.

The main thing I learned from the data is something that you’ve heard over and over again: maximum exit velocity (and 95th-percentile exit velocity, which I’m using) is sticky. How hard you hit the ball in one year does a great job of determining how hard you’ll hit the ball in the next year.

More specifically, I took a sample of players with at least 100 batted balls in two consecutive seasons. I sampled from 2015 to ’22, which gave me seven year-pairs, though the ones involving 2020 were light on qualifying players thanks to the abbreviated season. From there, I asked a simple question: how much did each player’s 95th-percentile exit velocity change from one year to the next?

Obviously, players improve their exit velocity every year. Even if their true talent never changed, a laughable assumption, random variance would make some hitters get to their best efforts more frequently in games from one year to the next. The question, then, is how much a given player can improve their exit velocity by, and how likely that improvement is.

I came up with a test that I think is simple and informative. The question I’m asking, essentially, is how often a hitter moves from one tier of contact quality to another, as measured by 95th-percentile exit velocity. I measured the standard deviation of 95th-percentile exit velocity across the entire population of hitters with 100 batted balls in each year. Then I checked what percentage of the hitters who played in both that season and the next improved (or declined) by at least one standard deviation.

That might all sound like a jumble, but consider it this way: standard deviation is a good measure of how widely the population varies. If you have a statistic with huge variation from player to player, you’d expect that statistic would also be noisier for a given player on their own. If, on the other hand, there’s almost no variation across the league as a whole, an individual player changing theirs is more impressive. Indeed, exit velocity changers are rare: only 4% of hitters saw their 95th-percentile exit velocities change by at least one standard deviation from one year to the next.

“But Ben,” you might say, “I have no idea what that means.” That’s a valid point. What’s 4% compared to 6%, or 2%, or any other number I could pick out of a hat? We need to compare it to something more familiar to understand whether 4% of the population displaying a big change is a significant result.

To do so, I enlisted a little help from other statistics. I considered launch angle tightness, average exit velocity, maximum exit velocity, chase rate, contact rate, walk rate, strikeout rate, and even wRC+. I used a 150 PA minimum for the non-batted-ball statistics and stuck with my 100 batted ball minimum for the rest. This gave me a baseline number to compare the changes in 95th-percentile exit velocity against.

If you want to know why analysts focus more on top-end power than average exit velocity, here it is in a nutshell. In a given year, 15.8% of batters see their average exit velocity improve or decline by at least a standard deviation. It’s a noisy statistic, in other words; you might think that you can tell the difference between two hitters based on their average exit velocities, but there’s a decent chance that you’re being deceived by variance. Hitters change their average exit velocities by a whole standard deviation four times as frequently as they change their top-end power. One year’s data point could easily be a mirage.

What about launch angle tightness, a method for estimating consistent contact? That’s even more variable than average exit velocity: 25% of hitters saw their launch angle vary by more than a standard deviation from one year to the next. One note here: a change in the way that Statcast collects data led to a meaningful change in launch angle tightness measures from 2020 to ’21 (Hawkeye cameras capture a higher percentage of batted balls, so meaningfully fewer batted balls had imputed launch angles, which changed standard deviation significantly). Accordingly, I tossed that pair (changes from 2020 to ’21) out of the dataset.

When it comes to batted ball metrics, 95th-percentile exit velocity stands alone at the top. For the record, maximum exit velocity is too noisy; 13% of hitters changed theirs by at least a standard deviation from year to year. I think that’s a measurement issue, as it’s easier to crush one batted ball than to crush enough to move up your 95th-percentile mark. That difference matters even more over less-than-complete seasons; 95th-percentile exit velocity scales with sample size better than maximum exit velocity.

When compared to swing-based metrics, 95th-percentile exit velocity stands out again (I’m just going to call it EV95 from here on out because I’m tired of typing it). 10% of hitters change their chase rate by a standard deviation or more from one year to the next; 7.3% of hitters change their contact rate by a standard deviation or more; 12.3% change their zone contact rate by a standard deviation or more. None of those compare to the roughly 4% mark for EV95.

Naturally, PA-level statistics lag the field; they’re the noisiest, which doesn’t surprise me. Per the data, 23.3% of hitters change their walk rate by one standard deviation or more from one year to the next, and 11.4% change their strikeout rate by a standard deviation or more. wRC+ is the noisiest: a whopping 33.8% of hitters see it change by a standard deviation or more. That makes total sense to me; it’s trying to summarize how good you are at hitting in one number, so innumerable different changes all feed into the same result.

In table form, you can see EV95’s pre-eminence when it comes to year-to-year stability:

Metric | Average | St. Dev | %Changes >1SD |
---|---|---|---|

EV95 | 105.2 | 2.9 | 3.9% |

Max EV | 110.3 | 3.3 | 13.5% |

Avg EV | 88.0 | 2.5 | 15.8% |

LASD | 28.4 | 3.4 | 30.6% |

Chase% | 31.1% | 6.1% | 10.0% |

Contact% | 77.4% | 6.1% | 7.3% |

Z-Contact% | 85.7% | 5.1% | 12.3% |

BB% | 8.5% | 3.1% | 21.3% |

K% | 21.9% | 6.2% | 11.4% |

wRC+ | 99 | 27 | 33.8% |

One nitpicky procedural note on the above table: the average and standard deviation numbers I displayed there are calculated across the entire population for all years from 2015 to ’22. When I did the actual calculations of which batters changed by a standard deviation or more, I calculated each year’s standard deviation individually and compared changes to the year-specific numbers.

To be clear, this doesn’t mean that players can’t change how hard they hit the ball. It’s right there in the numbers; plenty of players do it every year. My favorite example of this is Lars Nootbaar, who started clubbing the ball when he was already a major leaguer. But if you start out thinking hitters can develop patience or contact more easily than power, you’re probably coming from the right place.

There are some meaningful caveats to this analysis. This data is for major league hitters; robust minor-league batted ball data, perhaps adjusted for level of competition, might show that it’s much easier for hitters to develop power earlier in their professional careers. I also completely ignored anything other than production numbers. A more nuanced look might find different results for some body types or swing shapes. Nootbaar, for example, always *looked* like he had more power in him; he’s 6-foot-3 and strong.

“You can’t develop power” is the wrong takeaway here; “it’s hard to develop power” is closer to the truth. That holds true whether you’re hunting for fantasy breakouts or trying to sign a free agent. If you’re looking for a major league metric to trust, 95th-percentile exit velocity is a good bet. More so than any other statistic I could come up with, what you see is what you get, and what you see does a good job of telling you how dangerous a hitter can be when he’s at his best.

Ben is a writer at FanGraphs. He can be found on Twitter @_Ben_Clemens.

Where did you find EV95? I have been looking for this for the better part of a year, because I’m convinced it is a better measure of raw power than max EV (and certainly more so than average EV). Is there a way to get it on the baseball savant page? Is there a way to convince the powers that be to add it to a table here at FG?

I am assuming Ben is downloading the pbp data straight from Savant and then calculating it from there. If you are somewhat tech savvy the baseballr package can get that data for you.

In all honesty I’ve been meaning to upload the data to my github, I can do that and write a script that gets 95th percentile later this evening if you are interested and not tech savvy

Chris Clegg has found the values:

https://lookerstudio.google.com/u/0/reporting/d1c90230-cafd-4c85-b4c2-f23f6d38816e/page/p_urlxpxhwrc

This is incredible. I wonder if there is a way to collapse it to find career 95th percentile?

Now I’m itching to come up with a CTE70 on the exit velocities (that’s conditional tail expectancy, so it’s the expected value of all EVs above the 70th percentile).

That would be super cool. My level of tech-savvy is approximately “good enough to run other people’s R scripts, not good enough to write them myself”

EVAnalytics has Top 5% and “Next 20%”. Not the same, but also useful.

I’d love to see Ben regress on those two values to see if there’s any meaningful improvement vs. using EV95. Looking at the Chris Clegg spreadsheet Jeff linked above, I think that might be the best way to drop the “noise/signal” ratio in the metric using conveniently captured data – most of the intuitive ball-scorchers stand out in those percentiles, but including that “next 20%” measure should help cull out any SSS-driven outliers getting a deceptively high EV95.

(Though it’s entirely possible that EV95 is indeed the “sweet spot” among these metrics.)

It’s definitely something I’ll bring up, I agree it’d be useful but I do it myself rather than grab it from the site anywhere. To that point: Jeff linked to a page that has it down below, but here’s a very clunky shareable sheet with everyone who had 100 batted balls in a given season on it:

https://docs.google.com/spreadsheets/d/1UtyQ_m7FI1X-R7uUbsYrlFEtvOuV2SQeVcyV8-kwbr8/edit?usp=sharing

If you’re interested in making your own, I also uploaded the code I used to compile my batted ball data to GitHub. It’s in Python and requires you to have downloaded a bunch of Savant data with one tiny modification: create a batter_season line for every batted ball. It’s just the MLBAMID of the hitter plus the year the game was played in. The rest should theoretically run smoothly.

https://github.com/bclemens6/improved-waddle

Thanks much – maybe I’m misunderstanding the shareable sheet but it shows, for example, Nootbar at 115.7 last season when his max EV per his Fangraphs page was 113 (with an EV95 of 108.3 per Clegg’s sheet). Similarly, Varsho’s at 113 on the shareable sheet for 2022, actual max was 110.3, Clegg EV95 was 106.6. But in other cases, like Ohtani (119.1) and Moncada (111.5), the listed EV95 equals their max EV per Statcast. And for all players, the EV95 for 2021 and earlier on the shareable sheet corresponds with their Statcast max EV for those seasons.

Yeah uh, I pasted the Max EV sheet there haha. I updated it to have the EV95 tab instead. Sorry about that.

Edit: I’m slightly off of Clegg for Nootbaar, I have him at 107.4, I suppose I can look into it later.

Ah, that makes sense – thought I was really confused there for a minute! Yes, looks like all of your real 95 EVs are a little lower than Clegg’s, some sort of methodology difference I assume.