Measuring the Impact of Pitch Location on a Player’s Performance

As technology has evolved and ideas have spurred, the game of baseball and the things that make it up have been able to be better quantified and measured. From simple stats like batting average, strikeout-rate, and earned run average, which are all calculated by easy-to-understand formulas, to metrics like exit velocity and spin rate, which are publicly available thanks to the recent technological revolution within the game. Though the complicity and detail of each metric varies, they all aim to enhance our understanding of the game.

Considering all the mainstream stats that are easily accessible, all areas of statistical explanation in baseball have been pretty well covered. You can measure a hitter’s power through slugging percentage and hard-hit rate. You can measure the physical behavior of a pitch through velocity, movement, and spin rate.

One area that hasn’t really been touched upon well is the measurement of a pitcher’s command. Strikeout-rate, walk-rate, and certain batted ball statistics have given us more than enough material to build up an idea of a pitcher’s skills, strengths, and overall game. What explanation gets excluded through all of this is the measurement of a pitcher’s command, even though it is perceived as quite an important skill to have in pitching. What should be stated before diving into any command-related research is that it has a different definition than control, a common misconception. Command is usually defined as a pitcher’s ability to successfully locate the ball with intention, whereas control is more related to a pitcher’s ability to throw strikes consistently (walk-rate being an acceptable measurement for this). Previous work related to command does exist. Listed below are three metrics that have made past attempts to quantify the skill.

• Edge%, created by Bill Petti back in 2013, measures a pitcher’s rate of locating the ball on the edge of the strike zone. This certainly can be taken as an impartial yet useful picture of a pitcher’s command. Initially, small correlations were found between the metric and overall pitcher performance, with Petti saying “,the correlation between Edge% and K% is only .08, and Edge% and BB% is -.11.”
• Commandf/x, created by Sportsvision and MLBAM, measures the average distance a pitcher locates a pitch from the catcher’s target. In a 2014 article over at Grantland, Ben Lindbergh found that Tim Lincecum had a late-career resurgence thanks to an improvement in fastball command, backed up by Commandf/x data. “In 2010, when he led qualified starters in strikeout rate for the third consecutive season, his average fastball missed its target by over an inch more than the typical pitcher’s. However, his command has steadily improved over the past several seasons, to the point where he can now claim to have average fastball command. And he needs that precision, because he no longer has the speed to survive his mistakes.”
• Command+, created by Stats LLC, measures the value of a pitcher hitting a location by assessing how similar pitches have performed in that location in the past. This data is not publicly available. Eno Sarris went over this metric in an article at The Athletic back in 2018, describing the metric as “…an overall command stat that judges outcomes by intent, separately by pitch type, and then sums it all up.”

All three of these metrics have done a fantastic job of explaining what they intend to explain, but for one reason or another, none of them have entered the mainstream world of baseball statistics. This is why I set out to conduct my own research and create my own metric for measuring a pitcher’s skill-level of locating their pitches, while also studying the impact that it may have on performance.

Process

Creating context and organizing was key in this project. My research included pitch data from seasons 2016, 2017, 2018, and 2019. Every pitch thrown across the league in that time frame was separated by pitch type and opponent batter handedness. Finally, these subsets were separated by 33 ranges of pitch location. This data was all easily exported and organized thanks to the Statcast search at Baseball Savant. For explanation purposes, the 33 pitch location ranges were extracted from the preset “attack zones” at Baseball Savant. This idea has some similarities with the aforementioned Command+, assigning a certain value to every pitch a pitcher throws in a season based on a variety of factors. Once again, that data is not publicly available.

The thought process behind this idea was pretty simple. It was to identify the performance (league-wide xwOBA) of a certain type of pitch, in a certain area of the zone, against a certain type of hitter (e.g. four-seamer, zone #4, facing right-handed hitter). The next step would be to average out the xwOBA values assessed to every pitch a pitcher threw in a season, giving out a final number, dubbed ‘Command wOBA.’

There are some caveats and disclaimers with this final output that should be mentioned. First, the intention of the pitcher and catcher isn’t measured in this number. Pitchers purposely and accidentally hit certain spots all the time, sometimes for better or worse. A large part of commanding the ball involves hitting a preconceived intended target. Those preconceived intentions are eliminated in this practice, as only the final results of what happened are considered. Second, the difference between the league-wide performance by attack zone and hitters that deviate away from that standard must be considered. If a certain batter consistently performs badly in a zone where the rest of the league performs well, it’s likely that the opposing pitcher will still attack that zone. Finally, variables like velocity, spin, release point, and movement are not taken into consideration. When it comes to pitch location, all these things obviously matter, but that isn’t what Command wOBA is trying to measure.

Results

Before I calculated the final Command wOBA values, I went to break every pitcher season within the aforementioned time constraints by pitch type. Breaking down each pitch type was important in this analysis, as I wanted to understand to what level on each type of pitching locating had to performance. The final command wOBA values for each player’s pitch type were compared to the xwOBA against that pitch. Here’s what the correlations looked like (minimum 1,000 pitches in a season on fastballs, 500 pitches on secondaries).

As you can see, there is a decent amount of variance when it comes to the impact of location by pitch type. Location for splitters, four-seamers, and curveballs prove to have a decent level of importance. Two-seamers, knuckle curves, and cutters level near each other in the middle, while changeups, sinkers, and sliders have a lower level of correlation between location and performance, although still existing.

Now to build up more of an idea on a player basis, here are the best individual Command wOBA seasons (minimum 500 pitches) for each of those pitch types.

Finally, looking at the overall final product, here are the top 20 Command wOBA seasons, a minimum of 2,000 pitches (Command wOBA+ = seasonally adjusted Command wOBA).

The first thing that stands out here is the match between the final product of numbers and public perception. Noted command artists like Masahiro Tanaka, Kyle Hendricks, and Mike Leake each appear in the top 20 multiple times, also indicating consistency within the numbers. For reliability purposes, it was important for the year-to-year correlations of Command wOBA to be acceptable, helping to show that the metric is more skill-driven rather than a product of randomness.

To study this, I examined the relationship between first and second-year Command wOBA out of 75 pitchers that had at least 2,000 pitches in both the 2016 and 2017 seasons. The results were promising (R = 0.87), assuring that in fact, a pitcher’s pitch location is consistent from season-to-season.

While the year-to-year consistency for Command wOBA is there, it is also worth mentioning that there is consistency in the performance aspect of pitch location. Looking at the same sample group mentioned above, the correlation between year-to-year change in Command wOBA and the year-to-year change in xwOBA against offered additional reliance on the data (R = 0.33).

Command vs Stuff

So now that we have a good measurement of a pitcher’s skill level when it comes to locating the ball and know that the measurements are reliable and stable, we are now able to draw up additional descriptions on pitchers. One area I thought this could be interesting was comparing how pitch location and the physical aspects of a pitch (velocity, movement, spin) each separately correlated with performance (in other words, confirming what matters more). For this exercise, I will be specifically looking at four-seamer performance for the sake of developing a consistent and reliable sample group. As pointed out in a table above, the correlation between Command wOBA and xwOBA against on four-seamers is somewhat strong (R = 0.432), owning the second highest correlation of any pitch type (trails only a smaller sample size of splitters).

The next thing that needed to be done was creating a measurement for which pitchers had the best overall physical movement in their four-seamer. After playing around with a few different Statcast metrics and comparing them to xwOBA against, I created a multi-regression formula, converting a handful of metrics into a predicted xwOBA figure. The metrics taken into account were pitch velocity, spin rate, horizontal and vertical movement.

Y = 0.680629–0.0000597683 X1 + 0.0101541 X2–0.0122213 X3 + 0.000163918 X4–0.00149994 X5

X1 = Spin Rate

X2 = Pitch Velocity

X3 = Effective Velocity

X4 = Horizontal Movement

X5 = Vertical Movement

Y = Predicted xwOBA

As was seen with the final Command wOBA outputs, the predicted xwOBA numbers based on the formula above (let’s call it Stuff wOBA) unsurprisingly lined up well with public perception. A top the leaderboards are names like Tyler Glasnow, Aroldis Chapman, Jacob deGrom, and Dellin Betances.

Perhaps the best way to build up an idea of how Command and Stuff wOBA interact is to build groupings of pitchers based on how they fare with each metric (once again, specifically just four-seamers). Here were the top five seasons for each grouping based on xwOBA (minimum 1,000 pitches).

• +Command/+Stuff: 2017 Chris Sale, 2018 Max Scherzer, 2018 Josh Hader, 2017 Jacob deGrom, 2019 Gerrit Cole
• +Command/-Stuff: 2017 Rich Hill, 2016 CC Sabathia, 2016 Drew Smyly, 2019 Stephen Strasburg, 2017 Gio Gonzalez
• -Command/+Stuff: 2018 Vince Velasquez, 2019 Lucas Giolito, 2016 Jon Lester, 2018 Freddy Peralta, 2018 Jack Flaherty
• -Command/-Stuff: 2018 Trevor Williams, 2018 Wei-Yen Chen, 2017 Trevor Williams, 2017 Julio Teheran, 2017 Lance Lynn

Looking at a sample of 168 pitchers that had at least 1,000 four-seamers thrown in a season, the correlations showed that Command wOBA had a stronger relationship with xwOBA against (R = 0.432) than Stuff wOBA did with xwOBA against (R = 0.351), while the average of the two (R = 0.513) and the linear regression output of the two (R = 0.527) had an even higher correlation.

While these numbers would indicate that location impacts results more than physical aspects of a pitch, it should be mentioned that there is a lot more year-to-year stickiness in Stuff wOBA relative to Command wOBA.

Hitting Viewpoint

Now that Command wOBA has been described through a pitching sense, it’s time to see how the metric fares in its attempt to evaluate hitters. Calculating Command wOBA for hitters was done by using the same exact process that was used for pitchers. Conventional wisdom would suggest that pitch location is more consistent for pitchers rather than hitters, though the numbers show that there is a fair amount of year-to-year correlation with four-seamer pitch location for hitters (R = 0.394).

The numbers also indicate that a partial amount of hitter performance can be explained by pitch location, as there is a relationship between Command wOBA and xwOBA for hitters (R = 0.215). On the contrary, year-to-year change in xwOBA does not really correlate with year-to-year change in Command wOBA for hitters (R = 0.055), the opposite case with pitchers.

It does seem though that we can expect more regression-to-the-mean in Command wOBA more for hitters than pitchers. Examining the 10 highest Command wOBA seasons for hitters shows that eight of them saw a decrease in Command wOBA the next season, while the 10 lowest figures saw an increase the following season. So it does seem quite possible that a pitcher could inherently get ‘lucky’ from pitch location for a period of time, see that regress, and then consequently see a decline in performance.

Final Conclusions

As is the case with a lot of other metrics, Command wOBA is not a perfect measurement or descriptor of a pitcher’s skill at commanding the baseball. As mentioned above, it fails to consider a pitcher’s intent, the scouting reports on opposing hitters, and takes a fairly large sample of pitches to stabilize. This issue was brought up by Eno Sarris in an article over at The Athletic, which had an interesting quote from big league veteran reliever Ryan Buchter, stating that he intentionally missed the strike zone often.

“I’m not going to give in. I’m not going to throw the ball down the middle and hope it works out. It’s not like I’m wild. I’m not throwing fastballs to the backstop or in the dirt. I’m just not giving in to hitters. Even if it’s a lefty up and a righty on deck, and I fall behind, I don’t give in. That’s my game.”

But even with the flaws, the metric still proves to be worthy of use. First, it confirms a lot of what we already knew. Having perceptions backed up by data is never a bad thing. Second, once a reliable sample size has been established, it can be used as a fairly good descriptor of ‘what happened,’ along with being useable as an additional metric to analyze an aspect of a pitcher’s overall performance. Third, looking at the hitting side of things, Command wOBA can serve a purpose in analyzing hitting performance to a certain extent.

Though the importance of command may be becoming less prevalent with increasing velocity all throughout the game, command is still an important skill and making attempts to improve the measurements of the skill will benefit our understanding of it in a more exact sense.

Sources and Credits

• Eno Sarris and his amazing article at The Athletic detailing past command research and the metric Command+.
• Bill Petti and Jeff Zimmerman for their past research on Edge% and its impact over at FanGraphs.
• Ben Lindbergh for bringing Commandf/x to light and using it for his analysis of Tim Lincecum over at Grantland.
• Eno Sarris and Andrew Perpetua and their in-depth article at FanGraphs on pitch location analysis.
• Baseball Savant and FanGraphs for data