Methodology
I decided to run a K-Nearest Neighbors search to find the most similar pitchers to Kimbrel. I looked only at Kimbrel’s performance over the last three years – his age 28-30 seasons – and compared them to the seasons of other relief pitchers over the same age range.
First, I gathered all of the statistics on FanGraphs that are available from at least the 1974 season onwards (Win Probability Added started that season and it is also about the time frame in which the era of the modern closer began so it seemed like a good breaking point). I then recorded the cumulative fWAR of each of the same pitchers from their age 31-35 seasons. My goal was to determine statistics through a relief pitcher’s age-28 to age-30 seasons had the strongest relationship with their fWAR in their next five seasons (the expected duration of Kimbrel’s next contract). I calculated the correlation matrix between all of these statistics and looked to see which metrics correlated most highly with a reliever’s fWAR between age 31 and 35. The five metrics with the strongest relationship are below. In brackets is the r-value of each metric measured against relievers’ fWAR from age 31-35.
WAR (.4913), FIP- (-.4262) , WPA/LI (.3593), WHIP (-.3081), and K/9 (.2996).
Full disclosure: ERA- and FIP actually beat out the bottom two metrics above, but I cut them on the basis that they essentially attempt to measure the exact same thing that FIP- does. I then normalized these five metrics to have a mean of zero and a standard deviation of one (this is necessary because these metrics are all in different units) and multiplied each by the absolute value of their respective r-value listed above to properly weight for importance. I looked at the Euclidean distance between Craig Kimbrel’s performance over these three seasons and all other relievers since 1974 who pitched at least 150 innings between ages 28 to 30. We will countdown his five closest matches from least to most similar.