Embedded Files
Introduction
Introduction
The 2023 LSU Baseball Team dominated from a 10-0 win on Opening Day to a 18-4 win in the final game of the College World Series. They had 13 players drafted by Major League Baseball teams, the most of any school in 2023. Pitcher Paul Skenes and Outfielder Dylan Crews went 1st and 2nd overall respectively, the first time ever two teammates went back-to-back to start the draft. LSU defeated the #1 and #2 ranked teams in the country in consecutive series to win the National Championship and finish a successful season with a 54-17 record.
Their offense was especially powerful, averaging 8.81 runs per game, the 8th most in the country despite playing in the most difficult conference in college baseball. Out of 306 teams, they finished 1st in Runs, 2nd in Home Runs, 3rd in Hits, 6th in Slugging Percentage. But, to really understand their dominance, I wanted to look beyond traditional statistics and examine the underlying metrics that led to their success. This offense had a historic level of talent on it, and I wanted to look deeper into what made them so good.
To do this, I used Trackman data provided by the Rice Baseball coaches. To avoid giving away any information on our own players, I decided to focus my project on the National Champion LSU team. This Trackman dataset contains data from every pitch from every D1 baseball game from the 2023 season that took place in a stadium with a Trackman unit. Over 150 of the 306 D1 teams have a Trackman unit in their stadium, including 13 of the 14 SEC teams (not Florida). The data includes information on who was involved, the result, the game situation, and the physical movement and spin of the ball before and after being hit for each pitch. I used this data to calculate "The Big 3" Hitting Metrics for each player and team in the 2023 season, and the results are below.
Big 3 of Hitting
Big 3 of Hitting
Modern baseball research splits hitting metrics into 3 categories that are known as the Big 3 of Hitting. The DAG to the left illustrates the relationship between each category and the data that goes into each element.
Bat Speed - How Hard Can You Hit the Ball?
From pitch speed and exit velocity, we can calculate Bat Speed thanks to equations from Baseball Physicist Dr. Alan Nathan. For each player, I looked at their 90th percentile bat speed to get a sense of their maximum reachable bat speed while filtering out data errors and outlier hits.
Bat to Ball - How Often Do You Make Good Contact with the Ball?
Then, from bat speed, pitch speed, and exit velocity, we can calculate how close each hit was to perfect contact for that batter. This is called the Smash Factor of the hit. A Foul Ball is given a Smash Factor of 0.5, while a Swing and Miss has Smash Factor 0. A player's average Smash Factor is a score rating how often a hitter makes good contact on his swings.
Swing Decisions - How Well Do You Decide When to Swing?
Finally, based on the location of each pitch and the count it was thrown in, I can determine whether the optimal decision for a hitter was to swing at a pitch or not. I compare the value of the optimal result to the actual decision to calculate the Decision Quality of each decision a batter makes.
Bat Speed
Bat Speed
To Calculate Bat Speed, I used the following equation from Dr. Alan Nathan:
vs = (vb - eavp) / (1 + ea)
vs = Bat Speed
vb = Exit Speed of Ball
vp = Speed of Pitch
ea = Collision Efficiency - (0.2 is accepted maximum)
As I explained above, I took the 90th percentile bat speed for each player. Not every bat/ball collision will have perfect efficiency, so the 90th percentile of all collisions should be very close to a player's maximum level that is commonly reachable. I don't take the maximum because the dataset is not perfect. There are a few impossibly high exit speeds in the data, the wrong batter is listed for several pitches, and a player may occasionally have an exceptional hit outside of their true talent level. Taking the 90th percentile should avoid any outliers and be a good representation of a player's true talent for hitting the ball hard.
Bat Speed by Team
Bat Speed by Team
To show the importance of Bat Speed, I grouped the data by team and calculated the 90th percentile Bat Speed for each team. I compared this to the Runs Scored Per 100 Pitches for each team, and made the Pictograph of the SEC Teams to the left. (Except for Florida, who didn't have data from any of their home games)
It is clear that there's a strong correlation between bat speed and a successful offense. LSU stands out as being exceptionally good in both areas. Even in the best conference in college baseball, LSU separates itself from the rest of the pack.
Bat Speed by Player
Bat Speed by Player
This chart shows the leaders in bat speed for the Southeastern Conference. LSU had 3 players in the top 11 in the best conference in college baseball. Dylan Crews was drafted 2nd overall by the Washington Nationals this summer. Tommy White and Jared Jones were not draft eligible in 2023, but are highly touted prospects for 2024. Meanwhile, Kemp Alderman not only led the SEC in Bat Speed, but finished 1st in the entire country and was drafted in the 2nd round by Miami. LSU hitters clearly performed very well in this important area in 2023.
Bat to Ball
Bat to Ball
Bat to Ball is evaluated using Smash Factor - a concept popularized by Driveline Baseball and heavily related to Bat Speed. Bat Speed allows us to calculate a player's expected maximum level of ball contact, and Smash Factor measures how often he reaches or exceeds that maximum.
SF = (1 + (vb – vs)/(vp + vs)) * (BIP?) + (0.5) * (Foul?)
vs = Bat Speed (90th percentile for that player)
vb = Exit Speed of Ball
vp = Speed of Pitch
BIP? = 1 if Ball in Play, 0 otherwise
Foul? = 1 if Foul Ball, 0 otherwise
Smash Factor by Team
Smash Factor by Team
Like I did above with Bat Speed, I aggregated Smash Factor by team and compared each team's performance with their overall run scoring success. While there is a clear relationship here, it definitely isn't as strong as Bat Speed's relationship with run scoring was.
LSU's Smash Factor was closer to the middle of the pack in the SEC, but they obviously still had the most successful offense. I attribute this to the quality of pitching in the SEC. It is very hard to consistently make solid contact against the best college pitchers in the country. Since the pitching is so good, games are low scoring and a few moments of success are typically enough to win. That is why a measure of maximum offensive potential (Bat Speed) is more closely related to offensive success than this measure of average offensive performance. LSU's offense featured many big power hitters that could score several runs at a time with homers, but would also whiff often (Jared Jones, Hayden Travinski, and Brayden Jobert all fit this archetype)
Smash Factor by Player
Smash Factor by Player
This chart shows the leaders in Smash Factor for the Southeastern Conference. Just like they did for bat speed, LSU had 3 players in the top 11 in the best conference in college baseball. However, these are 3 different players than I saw above. Tre Morgan was a 3rd round pick by the Rays, while Beloso and Nippolt both returned for 2024.
While the conference leader Luke Hancock didn't see any professional baseball interest, Enrique Bradfield and Jacob Gonzalez both appear on this leaderboard and went in the 1st round of the 2023 MLB Draft. The overall national leader in Smash Factor was Northeastern's Mike Sirota, who is a surefire 1st rounder in 2024. This skill doesn't light up the scoreboard quite like Bat Speed does, but being able to make solid contact is immensely valuable and was definitely a crucial factor in giving LSU a strong and balanced lineup in 2023.
Swing Decisions
Swing Decisions
Decision Quality Calculation
Decision Quality Calculation
To measure how well a hitter decides at what pitches to swing at, I calculated a metric called Decision Quality that works as follows. For each ball-strike count, I fit a model that calculated the expected value of a take and a swing at each location. Then, I subtracted the take value from the swing value to calculate the net value of swinging at any given pitch. If a hitter swings, his Decision Quality for that pitch equals the net swing value, while if he doesn't swing, it equals the negative net swing value.
The graph to the left shows the net swing value for each location averaged across all counts. The batter should swing at pitches in all locations with a red shading, and should take all pitches with blue shading.
Decision Quality by Count
Decision Quality by Count
This chart shows how the boundary of the optimal swing zone changes in various counts. With 2 strikes, the batter should be much more aggressive than with 0 strikes. Interestingly, this model suggests the batter should never swing on a 3-0 or 3-1 count. The value of taking a pitch right down the middle is slightly negative, but it is still higher than the value of swinging at that pitch would be in that situation (since that could lead to an out).
Decision Quality by Team
Decision Quality by Team
So as I did with the other metrics, I found the average Decision Quality for each team. Since the overall averages are quite close to 0, I multiplied the averages by 100 to calculate DQ100 - the average decision quality over 100 pitches. The relationship between the DQ100 and Runs is pretty strong for SEC teams, but not quite as strong as it was for Bat Speed.
LSU had very good decision quality scores, but only the 3rd best in the conference. South Carolina and Kentucky both had pretty mediocre offenses, but made very good swing decisions. South Carolina notably had a terrible Smash Factor, while Kentucky had very poor Bat Speed, which likely offset their strong Swing Decisions. These two outliers help show how The Big 3 metrics all play off one another to describe all aspects of a team's offensive performance.
Decision Quality by Player
Decision Quality by Player
This chart shows the leaders in Decision Quality Per 100 Pitches for the Southeastern Conference. LSU has 3 players in the top 20 for the conference. Each of these players also appeared on another leaderboard, with Crews having exceptional Bat Speed and Beloso and Nippolt having high Smash Factors.
While the Bat Speed and Smash Factor leaderboards contained entirely unique sets of players, the Swing Decision leaderboard contains several repeats from the prior two. There is especially a lot of overlap with the Smash Factor leaders, which makes sense since making good swing decisions likely makes it easier to make good contact quality.
All 3 Together
All 3 Together
Each of the Big 3 Metrics has shown correlation with a succesful offense. By using all 3 together, we can get an overall prediction for each hitter and team in Division I Baseball.
Team Predicted Runs Per 100 Pitches
Team Predicted Runs Per 100 Pitches
First, I'll look on a team level. For each D1 team with at least 1000 Pitches of data, I used their Big 3 scores in a linear model to predict their Runs Per 100 Pitches
The predicted Runs Per 100 from the linear model is on the x-axis, and the actual Runs Per 100 is on the y-axis. The model does a good job of predicted Runs Per 100, with an R^2 value of 0.471. LSU is shown in purple, and once again we see how despite playing in the toughest conference in college baseball, their offense performed exceptionally well.
The below substantive effect plots show how each of the 3 metrics marginally affect the Runs Scored prediction in the linear model. Bat Speed has a very strong effect, while Smash Factor has a pretty strong effect too. Surprisingly, Swing Decisions has a slight negative effect, but the standard error is very high so the effect is not significant.
Bat Speed
Bat Speed
Very strong positive effect
Bat to Ball
Bat to Ball
Fairly strong positive effect
Swing Decisions
Swing Decisions
Very slight negative effect, if any
Finally, I built a similar linear model for all individual players in the SEC. Instead of Runs Scored Per 100 Pitches, I used Run Value Per 100 Pitches to quantify the overall performance of a hitter. Run Value calculates how much a hitter influences the expected number of runs scored in an inning but neutralizes the context so that each result is always worth the same amount. RV100 averages this value over 100 pitches.
Player Predicted Run Value Per 100 Pitches
Player Predicted Run Value Per 100 Pitches
Here are the results of the player model, plotted like the team results above. All SEC players with 200 Pitches were included in the linear model.
The R^2 value was 0.408, and there is clearly a moderate correlation. Several LSU players stand out in the top right corner of the graph. Also notable is that LSU had very few negative value players. In baseball, every hitter in the lineup gets opportunities to succeed or fail, so having few weak links is as (if not more) important as having strong players in a lineup.
The below substantive effect plots show how each of the 3 metrics marginally affect the Run Value predictions in the linear model for players. Like in the team model, Bat Speed has a very strong effect and Smash Factor has a relatively strong effect. The big difference is that Swing Decisions now has a slight positive effect on Run Value. It is interesting that Swing Decisions seemingly benefits players individually but not teams as a whole, but the effect is fairly insignificant so I'm not worried about the accuracy of the model.
Bat Speed
Bat Speed
Very strong positive effect
Bat to Ball
Bat to Ball
Fairly strong positive effect
Swing Decisions
Swing Decisions
Slight positive effect
Finally, the charts below compare the Actual and Predicted leaders in RV100 in the SEC. Unsurprisingly, there is a lot of overlap as well as a lot of LSU purple. Amazingly, LSU's actual run values are even higher than their projected run values based on the Big 3 metrics. LSU would have been expected to dominate offensively based purely on their metrics, but was still able to overperform those lofty expectations. This could be due to luck (from facing poor defenses or favorable umpires) or due to success in areas not covered or emphasized by the Big 3 metrics. I believe that the Big 3 metrics do a good job of covering all of the possible factors of hitting, but there is certainly a possibility that some element of hitting is missing from these metrics.
Projected RV100 Leaders
Projected RV100 Leaders
3 LSU Players in Top 18
Actual RV100 Leaders
Actual RV100 Leaders
3 LSU Players in Top 3!
and 1 more in 15th
Full LSU Ratings
Full LSU Ratings
Here are the full results for each player on the 2023 LSU Offense. I put each metric on the 20-80 scale, which is a common baseball rating system. For each metric, I calculated the mean and standard deviation for SEC hitters with > 200 pitches of data. The mean value gets a score of 50, and the values are scaled so that the standard deviation is 10. Unless any hitters are more than 3 standard deviations away from the mean in any value, all hitters will be between 20 and 80.
Conclusion
Conclusion
LSU's 2023 offense was great in nearly every area. Despite playing in the most difficult conference in college baseball, they performed extremely successfully in many different areas. LSU's hitters did significant damage at the plate, leading the SEC in 90th percentile Bat Speed, which seems to be the most important factor leading to a team's success. LSU hitters did a decent job at making high-quality contact, which allowed the lineup to be balanced with many different types of hitters. Finally, LSU hitters did a great job of choosing good pitches to swing at. Success in all of these areas, plus a bit of luck along the way, led to LSU's hitters scoring lots and lots of runs all season, all the way to the national championship.
Page updated
Report abuse