Reviewing 2024 Arbitration Projections
Examining the results of my arbitration model.
Welcome to another installment of my series exploring Major League Baseball’s arbitration process, and projecting salaries for players who are arbitration-eligible. For the past two years, I have used a regression model to project raises awarded by the arbitration panel in order to forecast future payroll on a team level, as well as to predict the outcomes of specific arbitration cases. Given the nature of the arbitration process, as well as the restrictions on using wearable technology and Statcast data in arbitration deliberations, arbitration panels typically award raises based on more “traditional” statistics rather than the “advanced” statistics that analysts like myself use to evaluate players for our everyday analysis. I recently wrote an article for Pitcher List in which I went into further detail about the arbitration process and simulated cases for three arbitration-eligible players.
As the arbitration process for 2024 has recently concluded, this article will focus on evaluating how well my arbitration model fared at predicting the outcomes of this offseason’s arbitration cases. The process for evaluating the model is fairly straightforward. For each arbitration case, I calculated the midpoint between the team’s salary proposal and the player’s salary proposal for the 2024 season. If my projection for a given player is below the midpoint, then I predict that the team will win the arbitration case. If my projection is above the midpoint, then I predict that the player will win the arbitration case. Onto the predictions!
As shown by the table above, the arbitration model did a pretty good job at forecasting the results of this offseason’s arbitration cases, correctly predicting the outcomes of 12 out of 15 cases. In order to get a better understanding of why the model did not correctly predict the raises awarded to Tanner Scott, José Suarez, and Taylor Ward, I will conduct a deeper analysis of each player to better understand any discrepancies between my projections and the decision of the panel.
First, there is a relatively easy explanation into why the model did not correctly predict the arbitration outcome of José Suarez and that is because the model classified him a starting pitcher and not as a relief pitcher. I have two separate projection models for starters and relievers, with a pitcher being classified as a starter if they start in at least 50% of their games played. Since Suarez started 7 out of his 11 games played in 2023, he was classified as a starter when a case can be made that he should’ve been classified as a reliever. When classified as a reliever, the model projects that the team wins the arbitration case which is what occurred in reality. Perhaps the criteria for being classified as a starting pitcher will be changed for next season’s arbitration projections.
In my opinion, Taylor Ward was likely awarded a higher raise than the model suggested due to the fact that he saw limited playing time in 2023 after suffering a gruesome injury to the face in late July. The model places a value on playing time and counting stats such as number of home runs and number of hits, so it is expected that the model will produce a lower projection if a player spends a substantial amount of time on the injured list.
Lastly, I was a bit puzzled that the model did not correctly predict the arbitration outcome of Tanner Scott. Despite having an excellent season in 2023, producing 2.8 fWAR coming out of the bullpen for the Miami Marlins, I had confidence in my arbitration projection for Scott due to the value that the model places on saves for relievers, and Scott only had 12 saves last season. As evident by the result of his arbitration hearing, the panel presumably placed more value on his excellent strikeout and walk totals (33.9% SO, 7.8% BB, 26.1% K-BB) and awarded him a salary of $5,700,000 for next season. Taking a look at past salary arbitration cases for relief pitchers, I believe that 2016 David Phelps is a good comparison to use to explain the raise that Scott was awarded by the panel.
As shown by the table above, Scott and Phelps had similar strikeout rates, fWAR, and a nearly identical ERA during the seasons preceding their arbitration cases, while Scott had a lower walk rate and allowed fewer HR/9. Scott accumulated 8 more saves while Phelps pitched in 8.2 more innings, which can probably be attributed to the fact that Phelps started 5 games in 2016. Based on these factors, it is reasonable to understand why the arbitration panel decided to award Scott with a $775,000 increase over the raise that Phelps was awarded after the 2016 season.
Overall, I am quite satisfied with the results of the arbitration model from this offseason. Correctly projecting the outcomes of 12 out of the 15 cases is a success in my opinion, especially given the fact that there is a clear explanation as to why 2 of the “misses” occurred. I have already re-trained the arbitration model for next season and I am confident that I will be able to at least reproduce these results during next offseason’s round of arbitration hearings. I will be experimenting with adding new variables to the model (such as properly adjusting for inflation) throughout the season, with the intending effect of increasing the predictive accuracy of the model for relief pitchers. As we turn our attention to the upcoming season, the lessons learned from this year’s arbitration outcomes will serve as valuable guides, ensuring that our future arbitration projections capture both the essence of player value while resonating with the realities of the arbitration process.
Follow @MLBDailyStats_ on X (Twitter) and Adam Salorio on Substack for more in-depth MLB analysis. Data from FanGraphs. Photo credits to Getty Images.