Surviving 27 Outs: A Data‑Driven Look at Perfect Games

 I don’t think there’s a more celebrated feat in baseball than a perfect game. There are rarer feats, such as four home runs in one game (21 times) or 20 strikeouts in a 9-inning game (only five times), but I think what separates a perfect game is it’s less about matching a numeric record and more about accomplishing literal perfection. In a sport loaded with failure and random variation, a perfect game is not only pretty awesome by definition but also incredibly rare. There have only been 24 perfect games in history (including one in the playoffs by Don Larsen in 1956). Think about the 150+ years of baseball and hundreds of thousands of starts it took to accomplish those 24.

To gain some appreciation, I wanted to calculate just how rare it is to accomplish a perfect game, out-by-out. How rare is it to even get half-way through a perfect game? How many outs do you have to be perfect through to have accomplished better than 99% of starts? Furthermore, of those starts that were perfect through x outs, how many were ultimately converted to a perfect game? How late in a game must a pitcher be perfect through to believe they have a real shot at finishing it?

To get my dataset, I pulled all regular season plays from 1920-2025. I pulled after 1920 because some of the early play-by-play data from Retrosheet is incomplete, and 1920 is a nice “era” year to begin with for the Live Ball Era. Apologies to Lee Richmond (1880), John Montgomery Ward (1880), Cy Young (1904), and Addie Joss (1908) for filtering out their perfect games, as well as Don Larsen. That leaves us with 19 perfect games. I then had to filter my play-by-play dataset to just the pitchers who started the game, and then calculate the maximum out that they were perfect through before losing their perfect game bid. In total, we’re looking at 379,815 starts.








My favorite outs to look at here are outs 0 and 27. Of course, every single start is perfect to start the game, and the percentage of all starts to be perfect is 0.005% (19/379,815). Out number 27 should hypothetically reflect this right? If 0.005% of all starts were perfect, then how come 0.006% were perfect through 27 outs? There are two instances where a pitcher was perfect through 27 but did not throw a perfect game. In 1959, Harvey Haddix was perfect through 12 innings, before losing the perfect game when the leadoff batter of the 13th inning reached on an error and ultimately lost the game three batters later on a walk-off double. Pedro Martinez in 1995 was the next instance, when he was perfect through 9 innings before giving up a hit to leadoff the 10th inning. 19 perfect games plus these two near-perfectos leads to 0.006% (21/379,815).

So when does perfect game bid become “worth it” to tune in? Obviously that is subjective, but in order to exceed a 10% chance, you’re already looking at seven complete innings. Even then, there have only been 144 starts to go perfect that long, so don’t hold your breath for it to happen again soon.

There have been 12 instances where a pitcher was perfect through 8.2 innings before losing it. Two of those pitchers, Milt Pappas in 1972 (lost on a walk) and Max Scherzer in 2015 (lost on a controversial HBP), went on to complete no-hitters at least. Other recent instances include Yu Darvish and Yusmeiro Petit in 2013, and perhaps most famously, Andres Galarraga in 2010. Some other no-hitters that were nearly perfect include Carlos Rodon in 2021 and Jack Kralick in 1962 (perfect through 8.1).

Comments

Post a Comment

Popular posts from this blog

Analyzing Strike Zone Data From the Statcast Database

Image
 With my Statcast database in Oracle nice and handy, let’s have some more fun and look at strike zone data. Strike zone data in Statcast is measured by two variables, plate_x and plate_z, which are like respective x and y coordinates of an axis. The strike zone is the width of home plate, which is 17 inches, although the rule book says specifically part of the baseball has to hit the strike zone for it to be a strike, so truthfully the strike zone is slightly wider than 17 inches. A baseball is three inches in diameter, so really the strike zone width is a number slightly smaller than 23 inches. Statcast measures plate_x and plate_z in feet, not inches. The middle of the strike zone width is a plate_x of 0 (like an axis origin of (0,0)), with pitches to the left being a negative value of plate_x and pitches to the right being a positive value of plate_x. For example, the strike zone is 17 inches wide (1.417 feet), so the left edge has a value of -0.7083 and the right edge has a ...
Read more

Introducing the Full Statcast Database (2019-2021)

Image
Back in July, me and my school partners created a similar data warehouse for every pitch thrown in the 2021 season up until June 23. The project was a lot of fun and you can read a PDF version of it here under the Data Warehousing section. I wanted to continue that project for all of 2021 and beyond, and have finally completed it. Baseball Savant contains data from every season since 2008, and Statcast metrics were measured starting in 2015. In a perfect world, I wanted my database to include everything since 2008, and at the very least 2015 to have a "Statcast Era" database, but unfortunately my student account in Oracle has a limited table space, so I was limited to 2019, 2020, and 2021. Across those three seasons, there were a total of 1,743,857 pitches thrown.  My database is set up in a similar dimension model "warehouse" style. The "fact table" contains all the pitches that I just described, with all their details such as velocity, spin rate, pitch ...
Read more

How Predictive Are Statcast Metrics?

Image
 Intro Baseball is a hard sport to predict because of all the random variation in it, and that alone probably accounts for like half of sabermetrics. I was curious about how predictive Statcast metrics are and it got me inspired. For this assignment, I looked at the 2017-2021 player-seasons that had at least 200 plate appearances. A lot has been written about how long it takes for Statcast metrics to stabilize, and it turns out they take much, much shorter than other stats like batting average. I’ve read that Statcast metrics stabilize within 70 batted balls (some by 40 even), which a player will easily achieve by the time he’s reached 200 plate appearances. This gives me a sample size of 1548 player-seasons. Statcast’s main metrics are the batted ball ones, such as average and max exit velocity, considering those are the ones that actually need Statcast in order to exist, but any player’s Baseball Savant dashboard will also show plate discipline metrics like out-of-strike-zone swi...
Read more