Notes from SaberSeminar 2018
Attending this annual baseball analytics conference for the first time was a huge step forward for me in my quest to learn more about the game of baseball and its surrounding industry. If not for my internship in Stamford about 3 hours away from Boston University where the event took place, I would not have had the opportunity to listen to such qualified speakers, meet so many like-minded students, and be exposed to so many new and innovative ideas.
Here I’ve laid out my summaries of the 10 presentations I deemed to be most relevant to the industry (in roughly decreasing order), along with a few memorable insights I gained from the rest of the talks.
Impact Presentations
The Most Likely Batter Contribution: Introducing Deserved Runs Created by Jonathan Judge
This talk unveiled Baseball Prospectus’ new publicly available metric DRS+ meant to quantify a hitter’s contributions, not their results. The metric is based on the premise that today’s statistical leaderboards all measure outcomes and aims to give hitters the appropriate amount of credit for their outcomes, teasing out some of the randomness to get a better look at the signal of the player’s individual skill. The metric uses a multinomial structure, multilevel modeling, and Bayesian-esque uncertainty model. The metric was shown to be more reliable (same over time), predictive, and descriptive than wRC+ or OPS+. BP adjusted for park by looking at players who switched teams mid-season (who should have similar DRC+ ratings with both teams if not for the park differences). DRC+ does not use batted ball data or traditional, outcome-based statistics (HR, BB, SO), so it is unclear what the inputs for this model are.
Stan is not an acronym: Bayesian solutions to Sabermetric shortcomings by Jonah Gabry and Ben Goodrich
Stan is a software that yields probability weighted outputs necessary for managerial and player decision making under uncertain conditions like those of a baseball game. Lots of our stats attempt to filter out noise, but this can alter the signal, so this method aims to quantify the noise (and therefore the uncertainty) instead of aiming to eliminate it. It seems that this software allows for many sets of simulations to be run with each set being slightly different than each other (ex: Simulating a 162-game season 10,000 times with your ideal 25-man roster, then simulating that same season 10,000 more times with a replacement level centerfielder in place of your starting centerfielder to quantify the value of your starter). This software is free and open source and can used with the R coding syntax using the rstanarm and brms functions.
Using TrackMan Spin Measurements to Characterize and Compare Pitchers by Glenn Healey
This presentation submitted that pitches should be characterized by their spin in addition to velocity and movement profiles. He pointed out that Craig Kimbrel and Dellin Betances’ fastballs have similar velocities and spin rates, but Kimbrel’s pitch has more horizontal movement because it has better spin efficiency, meaning more of its spin going in the direction of movement. (It seems to me that many people around the industry are racing to be able to calculate a pitch’s spin in three orthogonal directions, partially to better understand spin efficiency.) According to Healey, the idea of categorizing pitchers and pitches using their spin in addition to velocity and movement has immediate implications in player evaluation, player development, and health projections.
Objective Measurement for Fatigue and Concussions through eye tracking by Laura Yecies
This presenter described her company’s eye tracking technology and the device’s potential uses in baseball. Measuring the position of a person’s eyes 60 times per second while they track the path of a circle on the screen. Elite athletes are better than average at this skill, but their performance on the 2-minute test can accurately indicate concussion or physical fatigue if the results are worse than the player’s personal baseline performance. It was implied that the results of this visual test are correlated with a hitters’ pitch recognition skills. Like anything else, the skill of visual acuity appears to decline with age, but the decline is slower when it is practiced and trained with this device. This technology is a part of the wider trend of using neuroscience and biomechanics in baseball to gain an edge.
Evaluating the time it takes a hitter to check his swing by William Clark and Joe Petrich
This experiment by Diamond Kinetics aimed to identify the closest point a pitch can be to the hitter where he can still check his swing. They measured a hitter’s “trigger to impact” time on each swing. By varying the distance from the batter when he is told to check his swing (via a red light vs a green light on the pitching machine), they measured this point to be between 380 milliseconds and 240 milliseconds after pitch release. Although they only tested this on one batter, this experimental design could be reproduced to provide more universal insights. The line of thinking in this study could have important implications for the emerging study of pitch tunneling.
A Deep Dive on Speed with Statcast Data by Travis Petersen
This presentation unveiled the next evolution of baserunning metrics by MLB’s Statcast analysts. Though they had been using “Sprint Speed,” a runner’s fastest one-second window throughout a run (in feet per second), to quantify the speed of a player. But that stat doesn’t tell the story of a runner getting up the baseline after putting a ball in play, as not every player gets to, or maintains, their top speed in the same way. The new metrics introduced were the Key Step, how fast the batter was able to get three feet up the line, and Burst, how far a runner can get in the first 1.5 seconds after the Key Step. These metrics provide more context for every player’s speed profile and opens the door for more innovative definitions of seemingly discrete player characteristics like speed.
Extending Advanced Defensive Analytics to the Minor Leagues and Japan by Brian Reiff
The goal of the metric mentioned in this talk is to mimic major league defensive metrics at levels of baseball that don’t have Statcast. Although Baseball Info Solutions has Defensive Runs Saved for MiLB, this metric throws out plays on shifts, which are increasingly common. The PART system includes the contribution of defensive runs saved from positioning, airballs, range, and throwing separately. This number was applied to MiLB back to 2013 retroactively, but it was unclear if this was by watching video or by some other method. This metric is limited to infielders, and range is where players generally separate themselves from the competition. This may even be applicable to college teams which is largely void of defensive metrics.
Pitch Tunneling, Pitch Calling and Expected Outcome: A Former Pitcher’s Perspective by Dan Blewett
This talk brought the topics of pitch tunneling and pitch selection out of their usual theoretical realms and directly onto the mound. Blewett encouraged pitchers to evaluate their strengths and weaknesses, the characteristics of the hitter, and what outcome they wanted to achieve. He mentioned that tunneling has flaws, including that it is not observable in-game and tunneling doesn’t account for intentional changes in location (because an outside curveball won’t tunnel with an inside fastball). Also, he suggested that most pitchers can only control their release trajectory and that tunneling is hard to teach. With regards to pitch calling, he implied that pitchers make suboptimal decisions fairly often as a result of conflicting scouting reports or a lack of foresight/focus.
Explaining the Home Run Surge: The Commissioner’s Report by Alan Nathan, Peko Hosoi, and Reed MacPhail
In this hour, members of the Commissioner’s committee to investigate the post-2015 All Star break home run increase. The group detailed their logical process to approaching the problem and reiterated the findings stated in their 80-page report; namely that the baseball itself is the cause of balls traveling farther as a result of a lower drag coefficient. They could not conclude why newer balls had less drag, but an earlier presentation by Dr. Meredith Wills suggested that it may be the wider seams on the new ball causing the difference in drag and therefore the spike in home runs. It was also revealed that there is a large variation between the drag coefficients of individual baseballs in the sample regardless of when they were used in game, but that Rawlings’ manufacturing procedures are not to blame for this variance. The MLB did not intentionally change the ball in any way that contributed to the rise in home runs.
Rescaling the 20–80 scale with Rap Score by Rohan Gupta
Rapsodo introduced their new pitching metric aimed at grading pitches more objectively in this presentation. Called Rap Score, there are five components to this score. Pitchers get credit for more difference between the velocities of their pitches and the movements of their pitchers, their consistency of release point, their pitch’s proximity to a corner of the strike zone, and the company’s existing “Pitch Score” for every pitch. How these five characteristics of a pitch were weighted in the calculation of Rap Score was unaddressed. Although the Rapsodo software will show pitchers their Rap Score (which changes by game) for pitches thrown on a Rapsodo device, it appears the metric can also be calculated using TrackMan data. Though it is a valiant attempt, there are some logical questions about the methodology of this metric’s calculation
Quick Hits
Why Japanese (NPB) pitchers rely heavily on breaking and off-speed pitchers compared to their stateside counterparts (MLB) by Kazuto Yamazaki
Japanese baseball values a well-located fastball whereas American baseball values whiffs above all else. For this reason, Japanese pitchers throw their fastballs in pitchers counts more than Americans.
Changing the College Summer Baseball Landscape: player Management and Winning by Ryan Mossman
In an environment where there is a conflict between winning games and developing players, it is important to provide players with access to their data so that they can do some of their own development.
Exploring Minor League Strike Zones by Nate Freiman
In his sample of 20,000 minor league pitches, between 40–47% of all pitches were “borderline.”
Can Adjusting the Strike Zone Change the Time of Game? by Andy Andres
Andres built the case that time of game is strongly correlated with pitches per plate appearance and pitches per game, so widening the strike zone would decrease time of game in a way that a pitch clock or any other rule change couldn’t.
Contributions to Ball Movement from Arm Slot to Spin Rates by Cyrus Shirazi
This experiment looked for correlations between arm movement (measured by a Motus sleeve) and various pitch characteristics (captured by Rapsodo) for pitchers at a training facility in his area. He found that arm slot only correlates with vertical and horizontal movement.
Sports Medicine in 2018 by Chris Geary
Hamstring injuries are the most common injury in baseball with 0.7 injuries per 1000 player games. Most occur when running to first, they usually don’t require surgery, and these injuries average 24 days missed.
Is an all-or-nothing hitter the most valuable offensive asset? by Cameron Rogers
Players in the top 50th percentile in BIP% and OBP above .340 had higher offensive stats in 2017 than players in the bottom 50th percentile in BIP% and ISO above .200 implying that players with a more balanced approach are more valuable and more consistent than pure power hitters.
Until next time, Boston!
