February 14, 2008
Lies, Damned Lies
The Smoking Gun?
There has been a Dueling Banjos dynamic between two conflicting analyses that attempt to address the potential impact of performance enhancing drugs on Roger Clemens's career. One, put out by Hendricks Sports Management (Clemens's agency) suggests that Clemens's late-career success is relatively normal, citing a handful of specific examples like Nolan Ryan and Curt Schilling. The other, prepared by the Wharton School of Business for last Sunday's New York Times, uses a broader set of "durable" comparable pitchers, and comes to the opposite conclusion.
There is no doubt that, as a piece of science, the Hendricks report is pretty poor. By cherry-picking from a small group of pitchers like Ryan who were known to have successful ends to their careers, the report has virtually zero statistical credibility, and instead is the sort of glossy work product meant to impress with substance rather than style. At the same time, the Wharton study may be equally flawed: the only standards it uses for selecting its comparable pitchers are based on durability, rather than how successful the pitcher was. It has long been known, for example, that power pitchers like Clemens tend to age more gracefully than finesse pitchers like Orel Hershiser-who appears to qualify for the Wharton study.
A compromise between the two approaches can be found by running Clemens through our PECOTA projection system, which considers a whole host of factors, including both durability and quality, in selecting its comparables. Specifically, we will step back in time exactly ten seasons, and analyze what we might have expected out of Clemens from 1998-2001-the period during which he's accused by the Mitchell Report of using PEDs-based on his performance through the 1997 season.
PECOTA uses as many as 100 comparable pitchers to make its forecasts, but the highest-ranking comparables receive the most weight in its forecasts. In Clemens' case, his top 20 comparables are quite favorable (comparable rank in parenthesis):
One certain future Hall-of-Famer: Randy Johnson (#13).
One pitcher who is long overdue for the Hall of Fame: Bert Blyleven (#10).
At least half of Clemens's comparables through age 34 will eventually end up in the Hall of Fame, and essentially all of them had dignified careers. Perhaps more importantly, a number of them had long careers and were successful into their late 30s or early 40s, including several of the Hendricks Group's favorite comparables, like Ryan, Schilling, and Randy Johnson.
In fact, if we analyze Clemens's performance over the four-year period, we see that the retrospective PECOTA projection comes quite close to the reality:
Projected Actual Year Team W L ERA IP BB K W L ERA IP BB K 1998 TOR 16 8 3.21 218.2 70 240 20 6 2.65 234.2 88 271 1999 NYA 15 6 3.32 197.2 71 200 14 10 4.60 187.2 90 163 2000 NYA 15 7 3.51 201 76 209 13 8 3.70 204.1 84 188 2001 NYA 12 8 3.54 177.1 56 167 20 3 3.51 220.1 72 213 4-Yr Total 58 29 3.39 794.2 272 816 67 27 3.56 847 334 835
Comparing reality to projection, Clemens pitched about 50 more innings over this period than might have been expected, and accumulated nine more wins. On the other hand, his ERA was incrementally higher than our forecast. His actual strikeout rate (8.87 K/9) was very close to PECOTA's expectations (9.24 K/9), while he walked a few more hitters than anticipated. Overall, there is nothing particularly unusual about Clemens's performance over this four-year window-pitchers of Clemens's caliber quite often do remain successful late into their thirties.
To the extent that Clemens's career has been unusual has been outside the Mitchell Commission's window: his longevity past the age of 40. But even that is not entirely without precedent. Although pitchers like Ryan, Schilling, and Johnson are not exactly typical examples, neither are they atypical, and they all showed up prominently on Clemens's comparables list. The situation is emphatically not analogous to that of Barry Bonds, who not only sustained his performance, but actually improved upon it by a couple of degrees of magnitude in his late thirties.
Naturally, I have my own opinion about whether Clemens used so-called performance enhancers, and I don't think he did himself any favors in his Congressional testimony yesterday. But where his statistical record is concerned, there is no smoking gun.