Updated: This post was updated to more accurately capture the effects increased exit velocity would have on player value. My original calculations applied the increase in xwOBA to all balls in play which over estimated the impact on performance and player value. The increase in xwOBA should only be applied to balls that are hard hit as this is where improved outcomes are observed.
As teams continue to focus their player development programs aiming to get the most value out of a player possible it will be important for them to identify which traits they want to focus on and how that will effect the player's value to the team. Working off of this idea I wondered, "What would happen to a player's value if he could add 1 mph of exit velocity to his batted balls?" This post will explain my approach to making this conversion from miles per hour to dollars and show why a hitting program that develops increased exit velocity would add tremendous value to a player or major league organization. I will also discuss some approaches to increasing exit velocity I would pursue if developing a program myself.
Overview of Methods
My jump from EV to dollars takes three steps using a combination of historical Statcast data, linear regression, and some conversion values worked out by others in the baseball community. Briefly, here is what I found:
1) +1 mph of exit velocity on well hit balls (more on that below) increases xwOBA by 0.012.
2) xwOBA relates to Batting Runs (R^2 = 0.621) using the equation delta_BattingRuns = 274.2*delta_xwOBA. That's an increase of 3.26 Batting Runs using the value from Step 1.
3) +10 Batting Runs = +1 WAR and +1 WAR = $9M
A 1 mph increase in exit velocity adds roughly 0.326 WAR and increases a player's value by $2.9M.
If this is all you need to see then you can skip the details of my analysis below. Even a fraction of this would be worth developing a training program focused around exit velocity as I am sure more and more teams will be doing.
Exit Velocity to xwOBA
Expected weighted on-base average (xwOBA) uses exit velocity (EV) and launch angle (LA) to predict the wOBA value of a ball in play. This makes xwOBA a perfect statistic to gauge how a change in EV effects a player's skill level. From the Baseball Savant website:
"xwOBA is more indicative of a player's skill than regular wOBA, as xwOBA removes defense from the equation. Hitters, and likewise pitchers, are able to influence exit velocity and launch angle but have no control over what happens to a batted ball once it is put into play."
In order to determine how much xwOBA would increase, I took a look at Statcast data for all balls in play from 2017-2019. My code allowed for this data to be broken into bins for any interval of EV which in this case was 1 mph. Using the mean difference between each bin I was able to estimate the overall value of an increase in EV. However, the image below (image source: mlb.com) shows that an increase in exit velocity doesn't effect outcomes until balls are hit hard.
MLB classifies a ball as "hard hit" if the EV is greater than 95mph. I used this cutoff to focus on balls in play that would be positively impacted by an increase in EV. Because of this cutoff it is also important to consider the MLB average hard hit rate. In recent years this has been approximately 35% of balls in play.
Using this filtered data set I calculated the average increase in xwOBA an additional 1 mph of EV would create to be 34 points. However, to add this to a player's xwOBA the value needs to be scaled down to reflect the increase only occurring for those 35% of hard hit balls. This changes the increase in xwOBA to 12 points.
For some context, in 2019 xwOBA for players with 100 PA or more ranged from 0.196 (Jeff Mathis) to 0.455 (Mike Trout) and the MLB average was 0.319 which is roughly the skill level of Albert Pujols.
xwOBA to Batting Runs
Now that an increase in EV has been converted to xwOBA I needed to convert it to something more easily related to player value. My initial attempts centered around WAR, but I was unable to find a correlation that I was happy with. However, WAR is comprised of Batting Runs, Base Running Runs, and Fielding Runs in addition to some other adjustments outside the player's control. Exit velocity only effects a player's Batting Runs and that is where I made the connection to xwOBA. This is a great conversion because Batting Runs is essentially wOBA adjusted for plate appearances, league factors, park factors, etc.
To model how an increase in xwOBA would effect Batting Runs I used a linear regression between the two on all players with more than 100 PAs in the 2017-2019 seasons. This gave me 1320 data pairs between the player's xwOBA and Batting Runs. The plot below shows the results along with the line of best fit (R^2 = 0.621).
A little bit of rearranging of variables gave me the constant needed to convert a change in xwOBA to Batting Runs:
delta_BattingRuns = 274.2*delta_xwOBA
Plugging the expected increase in xwOBA of 0.044 into this equation I project 3.26 Batting Runs added from a 1 mph increase in EV. Mike Trout led major league baseball with 61.0 Batting Runs in 2019, and Jeff Mathis had -30.6 Batting Runs. However, the nicest thing about using Batting Runs is that it converts to WAR very easily so context isn't really important.
Batting Runs to WAR and Player Value
Now that the increase in EV has been converted to Batting Runs the final two steps are fairly straight forward.
Batting Runs is one component of WAR and therefore you only need to divide the Batting Runs value by Runs per Win to get it into WAR. Using Fangraphs as reference, I divided Batting Runs by 10 to give 0.326 wins added just from increasing EV. In 2019, WAR ranged from 8.6 to -2.1 (no need to tell you who those values are from) meaning 0.326 WAR is a significant difference in performance.
The final step in assessing an exit velocity increase's effect on player value is converting WAR to dollars. Fangraphs uses $9M per win giving a change in player value of $2.9M. That's $2.9M of added value per season per player if a training program can increase exit velocity by 1 mph. An organization who is confident in a training program can consider this into all levels of player development including signing free agents at a steep discount.
Starting Points for Increasing Exit Velocity
The first and most important part of developing a program is gathering and tracking data to monitor progress and refine programs based on what generates desired results. Systems like Rapsodo/Trackman/HitTrax are likely already in place at all major league facilities and capture hitting sessions so exit velocity is always being measured. This system also monitors launch angle which shouldn't be ignored throughout the training program and for some players may be more important than exit velocity.
As this Driveline article nicely explains, there are four components to exit velocity, and training programs can directly improve two of those: bat speed and contact quality (collision efficiency). An improvement of bat speed and/or contact quality should increase exit velocity so those will be the focus of different aspects of the training program. Contact quality is difficult to measure directly but bat speed can be monitored as often as possible using a Blast Motion bat sensor. Having data on bat speed and exit velocity will give feedback on overall progress as well as insight into where a player is improving in their swing.
There are now three areas the program is focused on and we have a way to directly measure two, giving insight into what is working and what isn't. Programs can then be adjusted for individual players and their deficiencies and new tools can be introduced and tested for efficacy. To start, here are some technologies/training methods worth exploring:
Blast Motion - monitors bat speed and other swing metrics
Axe Over/Underload Training Bats - developed and used at Driveline to improve bat speed
PlyoCare hitting balls - developed at Driveline to give feedback on and train contact quality
Vizual edge - vision training to help ball tracking and (maybe?) improve contact quality
K-Vest - swing analysis tool to improve kinetic chain timing
Motus MX - full body motion assessment tool to monitor and identify mobility deficiencies
Additionally, a Motus study showed increasing pelvis rotational energy correlated to an increase in bat speed and Driveline uses a velocity based training program in the weight room to improve player explosiveness.
Using these technologies and others, a team could build a data-driven hitting program focused on increasing exit velocity and dramatically increasing player value in the process.
To see code from my analysis please visit: https://github.com/koby-a-close/EVIncrease
Just before the 2019 MLB trade deadline Nick Castellanos was traded from the Detroit Tigers to the Chicago Cubs. With the trade, Castellanos moved from what would ultimately be the worst team in baseball in 2019 to a Cubs team making a postseason push and finishing with 37 more wins than Detroit. Castellanos' wOBA increased by 0.077 and his wRC+ increased from 105 to 154. It appears that moving to a team filled with more talent and more motivation to win caused his performance to improve but is that true for all players such that an increase in performance could be predicted? This post will use all trades from the 2017-2019 MLB trade deadlines to investigate this idea.
This project required more manual data collection than I have done in the past and I started by getting all of the players who were traded in "trade deadline deals" from 2017 to 2019. To do this I relied heavily on sports reporting and its accuracy. In total there were 49 position players and 78 batters to analyze.
Once a list was collected I needed to gather performance statistics from each player before and after they were traded. Fangraphs.com had all the information I needed and I selected two statistics for batters and two for pitchers to use in the analysis. For batters I used wOBA as a general measure of a player's hitting ability and wRC+ in an effort to control for changes in park effects. The statistics I used for pitchers were FIP- and SIERA. Both stats measure a pitcher's underlying skill level and adjust for park factors.
I also collected each team's final winning percentage in each season as a way to quantify if a player moved to a better or worse team and by how much. Using end of season winning percentages is a rough estimate since the player may effect that outcome themselves but it was significantly easier than collecting records before and after each trade for each player given I was collecting data manually.
The analysis on the data required a difference between the player's performance statistic to be compared to the difference in the winning percentage of the two teams he played for that season. My assumption was that an increase in team winning percentage would result in an increase in player performance.
Example Calculation from Nick Castellanos in 2019:
Old Team: Detroit Tigers, 0.292 Win %
New Team: Chicago Cubs, 0.519 Win %
wOBA with Tigers: 0.331
wOBA with Cubs: 0.408
Change in Win % = 0.519 - 0.292 = 0.227
Change in wOBA = 0.408 - 0.331 = 0.077
These calculations were done with each player and statistic in python (https://github.com/koby-a-close/PerformanceChanges_AfterTrade) and then plotted and checked for any trends using ordinary least squares. The following slideshow shows the results for all four statistics with the OLS line.
None of the statistics used showed any trend or statistical significance. The R^2 values were:
Based on this analysis there is no predictable change in performance when a player is traded mid-season based on the record of the new and old teams. This makes a lot of sense mainly because playing for the Cubs instead of the Tigers does not change the true skill level of Nick Castellanos and the same is true for all players. Park factors could be significant but I attempted to control for those effects in my statistic selections.
Other factors also play into the equation that are much harder to analyze like clubhouse fit, increased pressure to perform, and personal adjustments to new cities. Additionally, half a season is a very small sample size and the noise in data was significant. Practically this means that predicting a player's performance after a trade is best done using projections for the season using large amounts of historical data. Considering non-data driven factors is an art and will depend on the individual involved.
More statistics for individuals and teams could be tested but I doubt any significant would be found. Good players tend to play well and poor players tend to play poorly.