Which Statcast Measures Correlate Best? 2019 Refresh
A little more than a year ago, Al Melchior had the brilliant and beautifully straightforward idea of investigating how strongly pretty much ever Statcast metric correlated with various traditional power metrics and compiling them in one post. He asked me to help out, which I was more than glad to do.
Recently, I saw folks talking about this again, and someone asked specifically about the 2019 season. I figured I could refresh the values from the original post quickly enough (certainly a lot more quickly than I did last time), and it would also help bring pertinent information to the fore for folks neck-deep in draft prep.
Spoiler alert: the results barely changed. But! I do feel more confident in this particular set of values, as I nerded out with programming instead of pulling dozens of different queries from the Baseball Savant search function and constantly getting frazzled.
OK, here’s the goods. For 2019 hitters:
| Statcast Metric | HR/oFB | ISO | wOBA | xwOBA |
|---|---|---|---|---|
| Brl/BBE % | 0.82 | 0.80 | 0.59 | 0.93 |
| Brl/PA % | 0.81 | 0.84 | 0.64 | 0.92 |
| Avg EV: FB & LD | 0.81 | 0.72 | 0.53 | 0.84 |
| Avg FB dist | 0.81 | 0.75 | 0.58 | 0.83 |
| Avg EV: all | 0.66 | 0.65 | 0.59 | 0.79 |
| Max dist | 0.66 | 0.62 | 0.48 | 0.64 |
| Max EV | 0.66 | 0.56 | 0.40 | 0.62 |
| Avg dist | 0.35 | 0.61 | 0.45 | 0.61 |
| Avg HR dist | 0.51 | 0.44 | 0.30 | 0.56 |
| 95+ mph % | 0.50 | 0.56 | 0.55 | 0.56 |
| Avg EV: GB | 0.36 | 0.35 | 0.38 | 0.37 |
(Default sort: descending by xwOBA)
Minimum 300 plate appearances
HR/oFB = home runs on outfield fly balls
Notes:
- HR/oFB = home runs per outfield fly ball. I calculated this manually using Statcast data, which provides the following two quirks: (1) Statcast codes fly balls differently from Baseball Info Solutions, which powers fly ball percentage (FB%) on FanGraphs; (2) HR/FB on FanGraphs includes pop-ups, whereas HR/oFB does not, theoretically making it a more-accurate metric as far as measuring power is concerned.
- The individual numbers are Pearson correlation coefficients, or ‘r.’ You may have heard of r-squared when folks talk about goodness of fit models. This is the same! Except not squared. And it ranges from -1 (perfectly negative relationship) to +1 (perfectly positive), with 0 indicating no relationship whatsoever. The relationship between barrels per batted ball event (Brl/BBE %) and expected weighted on-base average (xwOBA) of 0.93 is extremely strong!
And here’s how each of the metrics correlates with itself across years (year 1 to year 2, e.g., 2018 to 2019). Folks often refer to this as year-to-year “stickiness.” The sample is limited to hitter-seasons pairs that have at least 300 plate appearances in both year 1 and year 2, and spans the 2016 through 2019 seasons.
| Statcast Metric | y1 → y2 |
|---|---|
| Avg EV: FB & LD | 0.82 |
| Max EV | 0.81 |
| Brl/BBE % | 0.79 |
| Avg EV: all | 0.78 |
| Avg dist | 0.75 |
| Brl/PA % | 0.74 |
| 95+ mph % | 0.73 |
| Avg FB dist | 0.64 |
| Avg EV: GB | 0.61 |
| Max dist | 0.56 |
| Avg HR dist | 0.47 |
It’s worth noting that reducing the minimum plate appearance threshold from 300 PA to 100 PA immaterially changes the results, even though the individual player-seasons are subject to greater noise due to smaller performance samples. In fact, they’re so similar, they’re indistinguishable. It suggests to me that all of these Statcast metrics, such as maximum exit velocity (EV), average fly ball distance, etc., stabilize pretty quickly, making them good indicators of talent early in the season. This claim is presently unsubstantiated, at least by me, but, again, the evidence is there, and I’m interpreting it optimistically! A project for another day, I suppose.
Ultimately, the “best” metric to use for power purposes appears to be barrels any form. And if you’re not comfortable with using a component metric – something that requires multiple inputs (barrels requires exit velocity and launch angle) – then it appears average exit velocity on fly balls and line drives might be next-best when accounting for both strength of relationship and year-to-year stickiness.
sorry if I missed it but is that xwOBA your xwOBA or Statcast’s?
Actually, I might care more about your 3rd column, correlation to actual wOBA, more.
Hmmm. Is it bad to say I don’t know? At the very least, this xwOBA fixes the issue with “null” values. As for accounting for spray angle, I honestly can’t remember if I applied it to my whole data set yet, but I don’t think I did. Sorry that’s a terrible answer!
It’s interesting how poorly (relatively speaking) everything relates to wOBA, although, in fairness, wOBA accounts for K’s and BBs, too, which muddies everything up a bit.
How well does xwOBA correlate with wOBA within year?
~0.85
It’s funny, I was just looking at the older version of this yesterday and thought to myself “is this still up to date?” And then this hits the internet today. Good stuff
Who could possibly have imagined that how often you hit the ball hard and at what angle (i.e., the exit velocity and launch angle components of barrels) would correlate with your ability to hit for power? Stunning!
(I guess that sounds a bit snarky, but I am just teasing – this is good information.)
*every
How about correl. of each metric to future wOBA?
whats driving the gap between wOBA and xwOBA within a year? is it just too small of a sample to have a stronger relationship? does shifting matter that much?
Defensive alignment? If you hit the ball hard, the defense lines up that way?
What about statcast metrics (y1) to xwOBA (y2) across years?
I love this sort of analysis. The one I’ve also been working on is xBA which seems a tougher nut to crack. I’ve been playing around with standard deviation of launch angle and average LA, but I’m having trouble finding much strength of relationship across years.
Any ideas?
you’ll definitely need a plate discipline component