Wednesday, September 30, 2020

Umpire Jordan Baker Nearly Called a Perfect Game and the World Will Never Know

On Day 1 of UEFL f/x—in which we introduced preliminary/day-of vs final/day-after plate score reporting—we discovered that umpire Jordan Baker nearly called a perfect game in Cleveland (one pitch shy), but thanks to the MLB computer's vertical strike zone blind spot, the world may never know.

During Game 1 of the Yankees-Indians AL Wild Card Series, HP Umpire Baker saw 166 callable pitches and, according to baseball's real-time pitch tracking computer, officiated 163 of those 166 pitches correctly for a plate score of 98.2%.

As we discussed in our UEFL f/x 3.0 Plate Score primer, baseball's computer quietly changes its grading system overnight, after the computer essentially double-checks itself to account for errors caused by the computer's inability to see certain vertical strike zone attributes in real-time.

Baker's game in Cleveland was no exception: While the computer scored Baker as 163/166 = 98.2% on the day of his game, MLB's computer went to work overnight re-processing and re-grading Baker's work, ultimately returning a result we hypothesized would likely occur: it increased Baker's score from 163/166 = 98.2% to 165/166 = 99.4%, an increase of +2 pitches and +1.2% accuracy.
Jordan Baker's preliminary (pre-processed) plate score for NYY-CLE WC Game 1 was 98.2%.
Perhaps a more convincing way of putting it would be the inverse of accuracy, or what Mark Williams at Boston University would call, the "Bad Call Rate" (BCR). Thus, Baker's preliminary, day-of BCR was 100-98.2= 1.8%, a strong performance in its own right.

However, after post-game processing, MLB's morning-after BCR for Baker decreased from 1.8% to just 0.6%. And, thanks to the lack of fanfare surrounding the post-game processing procedure, fans who saw Baker's game in real-time will never know what he called a better game than what ESPN K-Zone initially showed.
Baker's final (post-game processed) plate score for NYY-CLE WC Game 1 was 99.4%.

Put differently, the computer preliminarily charged Baker with three incorrect calls (163/166), but after processing changed its grade to just one incorrect call (165/166). For the computer, to decrease its red marks on Baker's game by two-of-three pitches or 66.7%, constitutes an incredibly statistically significant adjustment.

And this adjustment rate of 66.7%—which would be known as a failure rate had the computer made these calls in real-time, as would be the case with a robot umpire or computerized strike zone proposal, such as Automated Ball/Strike System (ABS)—is a blatant Achilles heel in the ABS experiment and one reason why, despite sign-off from the umpires' union and Commissioner Rob Manfred's vow to put computers in baseball games, MLB has, as of yet, been unable to assign a RoboUmp to call balls and strikes.

Video as follows:

Alternate Link: Postgame Processing - Example of MLB Computer's Fatal Flaw (CCS)


Post a Comment