Case Study: Can Pitcher Fatigue be Quantified?

1.20.20

Awhile back, I investigated the question of whether a pitcher’s fatigue can be identified by assessing their velocity as the game progressed. In that investigation, I used a LOESS regression curve to plot pitch velocities against the pitch number in each respective game for a randomly selected starting pitcher.

A few days ago, this question came back into my head and I decided to try a couple of different approaches. This time, I want to look at the average fastball velocity for each pitch number, and compare that to a handful of other pitchers to see if there were any consistencies. 

 

To start, I queried all the pitches thrown by the following pitchers, and filtered each dataframe to only fastball pitch types: 

• Charlie Morton (Tampa Bay Rays)
• Clayton Kershaw (Los Angeles Dodgers)
• Gerrit Cole (New York Yankees)
• Max Scherzer (Washington Nationals)
• Noah Syndergaard (New York Mets)

 

I then grouped the pitch numbers into discreet categories: 

• 0-24 pitches
• 25-49 pitches
• 50-74 pitches
• 75-100 pitches
• 100+ pitches

I did this so that I could run comparison of mean tests to see if there were any statistically significant differences between the average velocities of each pitch group. 

 

I decided to test the procedure with one pitcher (Noah Syndergaard). Syndergaard pitched in 32 games in 2019, posting a 4.28 ERA, 1.23 WHIP, and holding a 10-8 record. His fastball velocity is amongst the highest in the league, which is why I decided to include him in this study. 

Above is the box plot generated comparing the fastball pitch velocity against pitch count. The black line in the middle of the box represents the median velocity value, and a very small positive trend can be observed. However, after running a significance test (kruskal-wallis), the resulting value was 0.5952. This means that there is not a statistically significant difference between the average velocities of each group. 

 

This made me rethink what variables I should investigate. In my prior study, fastball velocity wasn’t a conclusive indicator either. Therefore, I shifted my focus to spin rate. Perhaps there was more of a sign that a pitcher is tiring based on the spin rate of his fastball late in the game.

To my surprise, the box plot comparing spin rate to pitch count shows more of a difference amongst medians, and the kruskal-wallis test came back with a value of 0.05. Therefore, one can state that there is a statistically significant difference between the mean spin rates of each pitch count group. 

Syndergaard’s average fastball velocity was 97.52 mph, and his average spin rate was 2105.93 RPM.

My next steps are to compare these results to the other pitchers I have queried data for and compare the results.