I’ve had a number of requests for more historical data on PER, so the third installment of the PER intro will basically be a data dump of everything I have collected.
So far, I have every starting pitcher’s stat line dating back through the 1988 season (119,616 games to be exact).
Here are a variety of notes based on the entire span of data I’ve collected so far…
- In that time span PER credits 14 pitchers with PER perfect games (a score of 1.000). In addition to the 12 actual perfect games, it also gives credit to Jonathan Sanchez and Terry Mulholland for their no-hitters which were interrupted by an error.
- There have also been seven pitchers who, due to pitching into extra innings, surpassed the perfect 1.000 mark. The last pitcher to do so was Kevin Millwood, who allowed three base runners in 10 shutout innings in 1999. The others: Jose DeLeon, Dave Stewart, Greg Maddux, Kevin Appier, Erik Hanson and Andy Hawkins.
- DeLeon owns the highest PER in this time span with a score of 1.064 when he allowed one base runner in 11 shutout innings for the Cardinals against the Reds in 1989.
- The average PER during this entire time span was .737, however, it obviously varies when you break it down by era.
- The highest PER averages among pitchers with at least 100 starts in this span: Curt Schilling (.803), Greg Maddux (.796), Roger Clemens (.794) and Roy Halladay (.793).
- The lowest PER averages: Kyle Davies (.688), Kevin Ritz (.692), Casey Fossum (.694), Sean Bergman (.695) and former Indian Jason Bere (.695).
- Other former Indians near the bottom: Julian Tavarez (.698), Jaret Wright (.698), Scott Elarton (.712) and Jack Armstrong (.716).
- The highest rated Indians pitchers since ’88 (min. 50 starts in Cleveland): Tom Candiotti (.780), Greg Swindell (.778), Dennis Martinez (.773) and CC Sabathia (.768).
- Two things on Candiotti’s placement at the top.. First, keep in mind the era in which he pitched in Cleveland—late 80s/early 90s was a pitcher-friendly era. Second, I think this shows how undervalued he was during his time here. He threw over 200 innings in five straight seasons with a low walk rate. He wasn’t dominant, but as his PER shows, he consistently gave the Tribe a chance to win. The Indians were 86-90 with Candiotti on the mound, and if you exclude his only bad year (1987), they were 77-67—that’s a pretty impressive record in an Indians era where wins were tough to come by.
- The worst Indians (min. 50 starts): Jaret Wright (.691), Davd Huff (.701), Jeremy Sowers (.710), Zach McAllister (.726, through 2013 season) and Ubaldo Jimenez (.727).
- The best Individual Indians game: Billy Traber’s one-hitter against the Yankees in 2003 (.988). He fell just one baserunner shy of a perfect game.
- Bartolo Colon comes in second with a two-baserunner shutout of the Yankees in 2000 (.977). Tom Kramer comes in third at .976 with one baserunner against the Rangers in 1993, but unfortunately the one baserunner was a Julio Franco home run.
- 55 pitchers have posted a PER of .000 in this time span. This requires a pitcher to allow at least one run without recording an out. Mitch Talbot, Scott Scudder and Tom Kramer all accomplished this feat for the Indians.
I just caught up on the PER articles. Great job and way to stick with it for so long. I do have a couple of questions/concerns with the equation:
1. Did you by chance run the data with Runs as opposed to ER? I understand not wanting to penalize the pitcher for defensive miscues, but the way ER is calculated benefits pitchers in many instances. As an extreme example, say a routine GB fielding error with two outs is followed by 5 consecutive doubles. No ER are created, though pitcher effectiveness is undoubtedly in question. I’d be curious to see if the R^2 goes up if you use runs against as opposed to ER.
2. Perhaps a PER+ is the next step, as different environments will show dramatically different results. Bob Gibson’s 1968 looks terrific in raw numbers, but wouldn’t look nearly so well compared to that year in particular. The reverse can be said of 1930 and a few years the Indians benefited from recently. The percentage chance of winning a game in any particular year should be more highly correlated to the environment present in that year than to the overall counting PER statistic.
1. You’re right about Runs generating a higher R2. I was torn on how to handle this. It’s a better predictor of the team’s outcome because it then becomes a pitching + defense stat. But for individuals I stuck with ER because, while there are flaws in UER calculations as you note, most cases are legitimate and it does help (albeit imperfectly) account for just the pitcher’s performance.
2. I do plan on eventually going back to adjust for the yearly league averages. This will be easy enough to accomplish once I have all the data. However, making it a true PER+ stat would also require accounting for ballpark factors. I’m sure this is possible, but I’ll need to seek out the help of someone a little more advanced at programing than I am. Good suggestion though.