Last week, Justin Wolfers offered an insightful analysis of Roger Clemens‘s career statistics and what those statistics imply about the likelihood that Clemens used steroids.
The latest contribution to this debate is by sabermetric legend Nate Silver of Baseball Prospectus. Using only data through 1997, Nate generates a projection of what Clemens‘s stats should have looked like from 1998-2001, the years in which the Mitchell report alleges that Clemens used performance enhancers. Silver concludes that there is nothing particularly unusual about the pitcher’s performance during this four-year period.
Does this conclusion vindicate Clemens? Not by a long shot. While statistical evidence can sometimes provide convincing evidence that something really out of the ordinary has happened (like the sumo wrestling cheating that was documented in Freakonomics), it is far from clear how to interpret the findings when things look normal. Indeed, most of the studies that have examined data on steroid use in baseball have found little evidence that steroids actually enhance performance. Stephen Stigler, a statistics professor here at the University of Chicago, came to that conclusion. So does Nate Silver in a prior analysis. The only exception I know is Ray Fair‘s study on age effects in baseball, in which he shows that almost all of the batters in his sample who far exceed his predictions late in their careers played during the 1990s.
Data are great, but in this instance analyzing dirty syringes or subjecting old blood samples to new doping tests are both more likely to provide definitive answers.
(Hat tip to Ken Kovash)