Upgrading a Deserved Barrel%

New year, new deserved barrels metric. In October, I took a crack at devising a “deserved barrels” metric in which I took the basic components of a barrel — a hitter’s exit velocity (EV) and launch angle (LA) — and determined the capacity in which the components relate to Statcast’s barrel rate metric (barrels per batted ball event, or “Brls/BBE %” on Baseball Savant). I included squared terms (EV², LA²) assuming the relationship is not linear. (A launch angle that’s too steep is detrimental, for example.)

Further offseason research led me to additional insights:

• Launch angle consistency (“tightness”) is important, correlating with higher BABIPs and more repeatable EVs.
• Launch angle grade in general is important, especially for batted ball quality, not just batted ball frequency, as indicated by batting average on balls in play (BABIP).

There exist many measures of contact quality; barrel rate captures how often a hitter produces high-quality contact. (Hard-hit rate functions similarly but ignores launch angle, to my knowledge, making barrel rate arguably superior.) It only made sense, then, that the latter finding above — that launch angle tightness matters to batted ball quality — should be incorporated into my deserved barrels work somehow.

The new deserved barrel rate incorporates the standard deviations of EV and LA — in layman’s terms, the consistency with which a hitter produces their average EV and LA. The adjusted r-squared (r²) improved from 0.68 to 0.80 — a fairly dramatic increase, dare I say.

I have been asked previously about predictiveness, so here are some important things to keep in mind:

• LA tightness bears zero correlation with barrel rate (r² = 0.00) but boasts very strong year-to-year consistency (r² = 0.53). Note that LA tightness correlates moderately with expected weighted on-base average on contact (xwOBAcon), so its incorporation here is not for naught — although there is certainly an argument to be made to exclude it from the model.
• EV tightness bears a weak relationship with barrel rate (r² = 0.09), although still stronger than that of LA tightness. However, the year-to-year consistency of EV tightness is significantly weaker than that of LA tightness (r² = 0.24).
• What this effectively implies is: LA tightness is a more repeatable skill but exerts markedly less influence on batted ball quality (and more so on batted ball frequency). On the other hand, EV tightness is a less repeatable skill but exerts more influence on batted ball quality.
  • What this implies is: EV tightness is probably more descriptive than predictive.
• It’s important to note that deserved barrel rate is a hair worse than observed/actual barrel rate in terms of predictiveness (r² = 0.64 vs. r² = 0.62). I wouldn’t disregard it because of this — the two metrics are incredibly close. Most realistically, deserved barrels will serve you best at the extremes. Who are the biggest outliers? Those hitters are likely your best buy/sell candidates (and the ones whose actual barrel rates may be less likely to predict their next-year performance).

Here is the new deserved barrel rate formula, in case the folks at home want to use it:

Deserved Barrel/BBE = 4.7656 + (0.00217 * LA) – (0.122 * EV) + (0.0000500 * LA²) + (0.000757 * EV²) – (0.00195 * stdev(LA)) + (0.0130 * stdev(EV))

Unfortunately, this equation becomes difficult to employ (unless you are crafty Savant user); player pages, leaderboards, and search queries do not have options for “standard deviation” of X, Y, or Z metric.

Here’s my best shot at an Idiot’s Guide for Calculating EV (or LA) Consistency:

1. Go to Baseball Savant’s Search page.
2. Click “Batted Ball Type” in the top row (last column) and enable all four options.
3. Click “Player Type” in the fifth row (first column) and toggle to “Hitter.”
4. Click the teal “Search” button at the bottom.
5. Once the results load, click the far right button that resembles a marshmallow with chain links in front of it. This will download the data.
   • Important! The maximum number of lines that Baseball Savant can export is 40,000. Each line will represent an individual batted ball. If you set the “Min PA” threshold (very last row, first column) to at least 400 plate appearances (which, in this case, represents 400 batted ball events), you will turn up 81 qualified hitters with a total of ~37,000 lines. A lower threshold will likely make the query time out. Just something to keep in mind. It might be easier to search specific players (“Batters” field, ninth row, last column) if you don’t care about results for everyone.
6. Open the downloaded file. Look for the columns ‘launch_speed’ (EV) and ‘launch_angle’ (LA). You’ll want to calculate standard deviations (“=stdev(range of cells)”) in these columns.

That said, if this sounds like too much work, you can use the table below for 2019 values (minimum 100 BBE). The “diff” column indicates “luck,” so to speak, where a higher positive value equates to more luck, and vice versa.

Then, once the 2020 season starts, you can refer to this Tableau I made, which will contain “xBarrel” rate (for hitters and for pitchers!). So… check back in March!

As always: click headers to sort. (Default sort from best to worst luck.)