Quantifying Command to Enhance Pitcher Development
This blog was authored by Kenzie Kroll (@KenzKroll), Andrew Sumner (@sumner2799), and Peter Mertka (@PeterMertka)
In recent years, many experts across the industry have begun identifying ways to quantify and evaluate pitcher command. A big reason for doing so is to provide another way to evaluate player performance and add another tool to use in player development across all levels.
Most professional organizations around baseball have been tracking pitcher command for the last several years. One specific example is the Tampa Bay Rays, who have made it public knowledge that they have their catchers set up middle-middle every pitch, looking to quantify miss distances by how far the catcher moves his glove to receive the baseball in comparison to the initial setup.
Others have a similar process to the Rays, but they do not require their catchers to set up middle-middle. Instead, they track the distance the catcher’s glove moves from the initial setup location to where the ball is actually caught. While this type of tracking is quite prevalent for professional teams, it is an area of untapped potential at the collegiate level.
However, there are some issues that arise with these tracking methods. Eno Sarris highlighted these issues in a 2018 article with The Athletic. The methodology introduced by Sarris and provided by STATS, a sports data and analytics company, is founded on the principle of using historical combinations of pitch type and location for every individual pitcher to create intended zones. The creation of defined zones allows for the further analysis of a pitcher’s ability to hit those zones, or their “command.”
Understanding how well a pitcher is able to command their pitches is a crucial element to their success. It can provide both explanatory and predictive reasons for why a certain pitcher might experience particular struggles or successes on the mound. It also give an in-depth way to analyze a pitcher’s command instead of relying on the eye test or the typical high walk rate = poor command formula. It can also provide individualized targeting methods for pitchers that can lead to better performances on the mound, and more wins for a team.
Towards the end of the Spring 2021 season, we began to explore how to quantify command in a way that better included pitcher intent and did not require hours of post-game video-watching and sign analysis. Our program recognized that command analysis provided an area for growth in the college game and that we were capable of developing an efficient quantification method.
Methodology
Before finally nailing down our current process as it stands, we went through several phases of command tracking that were made up of different strategies and styles.
Phase 1
In the Fall, our initial method was simply sitting next to our pitching coach, Robin Lund, with pen and paper to record pitch calls as they were made during team scrimmages. Attention to detail was extremely important at the start. We needed to make sure to correctly label the inning, batter, and pitcher. Above all, we had to ensure the pitches were in order so that the data aligned with what was recorded with Trackman. We then logged every pitch into an Excel file and merged that with the rest of our data during the post-game process.
Phase 2
We eventually ditched the pen and paper, figuring it would be easier to begin logging the pitches in an Excel file. However, bringing our personal laptops into the dugout became an even greater annoyance than the pen and paper. Additionally, we knew this location would not be suitable if we were to make the transition to in-game tagging.
We then made a move to the stands, but that came with a change in methodology. In order to call pitches in-game, a 4-digit number is displayed in our dugout to signify a pitch and location on a pitch calling card that the pitcher and catcher both had. Utilizing a pitch-calling card allowed us to see everything our pitching coach was calling throughout the course of the game while not being in the dugout.
At this point, we felt we were close to being ready for in-game tagging but needed a more efficient tagging interface.
Phase 3
While the Excel document was much simpler and user-friendly compared to the old pen and paper style, it still wasn’t quite cutting it. This is where our department’s specialty in R Shiny Application design came into play. We constructed an application that provided a much more user-friendly way to tag what was going on within the game. For those familiar with Trackman, it was very similar to that tagging interface with the inclusion of the intended pitch location and whether or not that pitch was executed.
This new tagging application made for easy data recording and storage option, as the pitches would appear in a final CSV in the same order they were tagged in, allowing for an easy merge with the Trackman data.
Phase 3 was fully implemented this past Fall, giving our program a finalized system of command tagging that we would use for every home game during the 2022 regular season.
Implementation
With the data collection aspect figured out, the next obstacle was manipulating our new command data to provide analysis for the coaches to help them understand how our pitchers were performing. To do this, we created a command score that would quantify our data in a concrete and usable way.
When it came to developing a command score, we considered a wide variety of factors in order to fully quantify a pitcher’s ability to command his pitches. The first data points we took into consideration were the vertical and horizontal miss distances of a pitch compared to the intended location. The miss distance is the distance (in inches) between where the pitch was thrown and the nearest edge of the intended zone.
Another factor that was considered in the command score calculation was batter-handedness. This is due to how our program handles intended zones for our pitchers. These zones are not the typical 1–9 that are commonly seen overlaid on a strike zone, but instead defined locations that change depending on the pitch type and batter’s handedness. By including this factor, we can quantify when a pitcher may excel or struggle with specific matchups of batter-handedness and pitch type.
Additionally, several other factors were considered but ultimately left out of the calculation because we determined they did not affect a pitcher’s command on an individual pitch. The count and game situation are two factors that do not impact our score. Although it is beneficial for a pitcher to command his pitches better in hitter’s counts or with runners in scoring position, they do not ultimately impact how we wanted to quantify command.
Considering the features discussed, we combined everything to accurately reward or punish pitches based on location relative to where it was intended to be thrown. Initially, it is quite difficult to reward a pitcher for hitting their spot beyond just equating hitting their spot to a perfect score. This is because, statistically, if the pitch is located within the defined boundaries of the intended zone, there is no way to separate it from other pitches within the same zone.
To begin quantifying command, we had to understand the boundaries of our program's intended zones. For example, a right-handed pitcher’s slider called on the outside corner to a right-handed batter is allowed to run out of the strike zone (glove-side) more than a slider on the inside corner is allowed to run over the middle of the plate. Essentially, a pitch that does not end up within the traditional strike zone may still fall within our defined boundaries. Since our pitchers have these specific targets in mind when locating their pitches, we knew it was important to quantify their command based on these definitions. Using these boundaries, we created formulas to calculate scores for pitches that miss these zones either vertically, horizontally, or in both directions.
Additionally, we differentiated between a “good” and a “bad” miss. A good miss is a pitch that misses in the direction of the intended location but lies outside of the intended zone boundaries. For example, if a pitch is called low and outside, but misses 7 inches inside, that pitch will receive a lower command score than if it missed 7 inches outside. Similarly, that pitch will receive a greater penalty for missing up in the zone 7 inches than if it missed lower in the zone 7 inches. These penalties are determined independently of each other in the x and y directions, so that the score can accurately differentiate between pitches that miss favorably in both directions, one, or neither. We also determined that some pitches miss the intended zone so badly that they deserve a score of zero. This process allows us to fairly penalize a pitch based on the intent of the pitch call.
With this half of the process, our scores can accurately reflect how the pitch missed relative to the intended zone, but that only addresses one part of our goal. The other part of our desired command score would be to reward a player who is commanding their pitches better than average.
This proved to be more complicated than simply modifying our equations. As previously mentioned, if the pitch didn’t miss the zone, then factors such as miss distance cannot be used. However, some zones (such as a high and inside fastball) may be more difficult to hit because it is a spot that can easily run away from a pitcher and can be difficult to locate when a batter is crowding the plate. Therefore, the average command score in this location will be lower than a pitch that is easier to locate, so a “hit” in this spot receives a higher score than in other zones.
Our process to consider these factors involved calculating the average command value generated by our functions for each combination of pitch type and zone. This meant that for each of our intended zones, there were three averages: one for fastballs, one for breaking balls, and one for off-speed pitches. By doing this, we were able to take the sub value of a given pitch and factor in the average for its corresponding pitch group and intended zone to better reflect a pitcher’s ability to hit that zone.
After taking the average of each zone and pitch type into consideration, we converted our command score to a Command+ metric. This is a value that we use within our program, but it is not a true plus score due to the fact that we are only calculating the value based on Iowa’s pitchers alone. Thus, the averages used are not Division I averages, but instead a localized version for our program itself. At this time, due to the nature of how different programs call their pitches, as well as how that information is typically kept private, we do not currently have a model to make a true, all-encompassing Command+ score.
By making our command score operate on a scale with an average of 100, we were able to best capture the minute details that can help us understand just how well a pitcher was commanding their pitches throughout the game relative to the rest of our pitching staff. Doing so allows us to find discrepancies with certain pitch/zone combinations and can influence pitch calling. Additionally, further analysis can be done to evaluate a player’s ability to command his pitches in various counts or situations by finding grouped averages.
Ultimately, command score is yet another statistic that we as a program can use for both player evaluation and development. It allows us to further quantify our pitcher’s performances and help them improve in all aspects of their game. In the future, we would love to bring this score to a larger scale, to encompass not only Iowa’s away games, but other team’s pitching staffs as well, allowing for a true Command+ metric for all of Division I baseball.