May 28, 1999
Pitcher Abuse Points - One Year Later
A Look Back...and Ahead
Almost one full year ago--June of 1998, to be exact--I had become increasingly frustrated at Jim Leyland's idea of "entertainment" in what was turning into The Lost Season in south Florida. My frustration stemmed from watching as his young and promising pitchers were made to throw close to a gross of pitches on an all-too-frequent basis. I watched with dismay as Livan Hernandez twice reached the 150-pitch barrier, but when rookie Jesus Sanchez labored for 147 pitches one fine evening--accompanied by deafening silence from traditional media sources--my disappointment turned to rage and inaction no longer seemed an option. I had to do something.
So I created Pitcher Abuse Points.
The creation of any new baseball statistic is an act of arrogance. To add a new statistic to the vast landscape of numbers that crowd the baseball universe is like adding another drunk to Bourbon Street during Mardi Gras. It presumes that the new stat provides information that is not available in any other form, whether by interpreting and packaging old data in a compelling new guise, or by digging up data that simply didn't exist. And certainly, there were methods in use that attempted to measure a pitcher's workload. One, the concept of batters faced per start (BFS), had been in use since Craig Wright presented it in his book The Diamond Appraised in 1989, and could be calculated with relative ease for virtually any pitcher in major league history. The more direct approach, using pitches per start, has the historic data limitations imposed on any statistic which requires pitch-by-pitch information, but represented a huge increase in accuracy in identifying worrisome workloads.
Pitches per start has all the advantages you look for in a statistic: it directly measures the variable in question (workload); it is easy to calculate, requiring only two counting stats (pitches and starts); and the information it reveals is inaccessible via our arsenal of traditional stats. You can hype innings pitched and complete games all you want, and Greg Maddux's health record would still defy explanation--unless you look at pitch counts.
But pitch counts contain one ultimately fatal flaw: they treat all pitches the same. I couldn't accept a stat that treated the first pitch thrown as equally burdensome as the 100th or 150th pitch thrown. I couldn't accept a stat that treated two 90-pitch outings as equivalent to a 145-pitch effort followed by a 35-pitch shelling of the obviously-gassed starter. I couldn't accept a stat that rewarded a pitcher for getting bombed and taken out early--especially if the early shower was the direct result of a high workload.
Pitcher Abuse Points, better known as PAP, was an attempt to address that flaw, based on one assumption: that the body of a pitcher is able to compensate for the tiny muscle and ligament tears and associated inflammation that accompanies the unnatural motion of throwing overhand--but only up to a point. Beyond that threshold, each pitch leaves tiny but measurable damage in the pitcher's arm, and as the body becomes less and less able to compensate, each additional pitch does more and more damage.
The method for calculating PAP is familiar by now, but to briefly review: the number of pitches thrown in each start is evaluated separately. If the starter throws 100 or fewer pitches, it is assumed that he did not cross that threshold, and he is not awarded any abuse points. But starting with pitch 101, each pitch carries with it a penalty, and the number of points awarded for each pitch goes up the more pitches are thrown. Each pitch from 101 to 110 has one PAP tied to it, pitches from 111 to 120 get two PAPs apiece, pitches from 121 to 130 get three each, and so on.
Choosing The Threshold
Central to this system is the presumption that the threshold is the same for everybody: 100 pitches. Common sense dictates that all pitchers are not built alike; they have different builds, different work ethics, different mechanics and so on. And no, we don't know what the thresholds are for individual pitchers. PAP was not designed to be a universal yardstick, measuring all pitchers' workloads with disregard for context.
If your Ford Taurus gets 24 miles to the gallon, while mine gets just 18 miles per gallon, is your car more fuel-efficient? Not necessarily; what if you drive your car on empty highways across Nebraska while mine gets stuck in rush hour traffic around San Francisco Bay? Similarly, who knows for sure if Randy Johnson is more hurt by getting 45 PAPs per start than John Snyder is racking up 20 PAPs per start?
But just as we know from experience that cars get less mileage in the city than on the highway, we may use PAP to see if pitchers can stand more abuse in certain situations than in others. If older pitchers can stomach higher PAP scores than younger ones and stay healthy, we can build that into the system. If taller pitchers get injured less often than short pitchers given similar PAP scores, we can penalize short pitchers and reward tall ones. If right-handers stay healthier than left-handers...you get the idea.
And just as cars get their mileage quoted for both city and highway driving, we can adjust PAP to give a better sense of the context that each pitcher finds himself in. I made a first attempt to do that last year, by using the concept of Age-Adjusted Workload, multiplying each pitcher's PAP score by a factor depending on their age, so that given the same PAP score, a young pitcher would be penalized more than his veteran counterpart. And one of the things I want to do this year is expand on that concept, to adjust PAP scores to take other factors into account, like penalizing a pitcher more for starting on three days' rest, or after coming off a high-pitch outing in his previous start. But more on that later.
Since its unveiling last year, the future of PAP has been shaped by three events:
More Questions to Answer
What about relievers? Trying to measure the workload of relievers is a hellish task, because of all the variables involved (time off since the last appearance, the reliever's recent workload, etc.) and the fact that some information (how many times did Dusty Baker tell Julian Tavarez to warm up?) is not available and probably never will be. But the correlation between reliever workload and future ineffectiveness is certainly there; take a look at all the pitchers who have thrown 100 relief innings in a season, and watch many of their careers plummet thereafter. So we're going to try to study the issue.
And, perhaps most intriguing of all: what about the minor leagues? I don't profess to care about the usage patterns of Medicine Hat's fourth starter, but wouldn't you like to know if the Cardinals are really taking care of Rick Ankiel's arm? If the Mets are taking better care of Octavio Dotel than they did the Young Guns? The trouble is that pitch count information for the minor leagues is not readily available. At least, I haven't found the data, and I've asked colleagues of mine.
If any of you know where to find some data on minor league pitch counts--no matter how little-- please let us know. But there are techniques to estimate pitch counts from box scores, and any data is better than none.
In short, PAP is not a fixed object. Its use is changing as its value grows, and its value grows as we find more and more ways to apply the data to answer questions about baseball. And finding the answer to baseball questions is why Baseball Prospectus exists in the first place.