So Roger Clemens is free to go. On Monday, a jury acquitted the legendary baseball pitcher of all charges that he lied to Congress about using steroids.
But legal proceedings aside, a lot of people are still scratching their heads: how the heck did a guy pitch better, by some measures, at 42 years of age than he did at 29—or even 25?
For an answer, we suggest looking at the data. Four years ago, we analyzed Clemens’s career performance, factoring in “stable baseball pitcher performance measures” like WHIP (walks plus hits per innings pitched), and ERA (earned run average). These are better indicators of ability than wins, since they are less influenced by team quality and other factors beyond the pitcher’s control.
Using those measures, we compared Clemens with 32 other pitchers with similar longevity, and found his career trends were highly unusual. His numbers improved, declined, and improved again to a degree that was incomparable to his modern-day pitcher counterparts. In fact, Clemens is the only pitcher among the comparison set that had a WHIP statistic that got progressively worse as his career went on and then dramatically improved after he turned 40. We can’t comment on performance-enhancing drugs, since that issue is far beyond what can be concluded from a statistical analysis of his career.
Four years on, the verdict is in, but very little has changed. Our statistical analysis would remain the same: Clemens himself has thrown no more pitches in the major leagues; the comparable set of other pitchers hasn't changed; and we would use the same methods for examining trends over time. It is a glorious fact about baseball that the data is immortal.
Acquittal by a jury, as put by the prosecuting attorney, is based on a fair trial and process. Statistical analysis of his performance on the baseball field had no influence in the Roger Clemens perjury trial—but could it have? As statisticians, we can't help but wonder would there have been a different outcome if statistical evidence had been presented?
People are very influenced by the concept of extremely rare events. We teach our students, whether undergraduates, MBAs, or doctoral students, the concept of Type I error, meaning the probability of rejecting the “null hypothesis” of innocence falsely when it is actually true. In fact, statistical hypothesis testing is perfectly built for legal cases as it presumes the null (innocence) and only rejects it if there is enough evidence in the data to do so.
That Clemens was found innocent is clearly important for his legacy, but we selfishly hope that his case brings more attention to the use of statistical evidence to understand abnormal performance. Since many of us grew up loving baseball, we look for opportunities to apply statistical methods to it. Whether it is abnormal data (i.e., cheating) on a standardized test, hypervisitation to a digital-content video site (i.e., binge viewing), or lower-than-expected response to a Super Bowl advertisement (i.e., the ad made the “razzie list”), identification of unusual patterns is part of a statistician’s job.
If and when Clemens makes the Baseball Hall of Fame, we fully expect to be contacted again. Our answer though will remain the same. It is still ALL about the data.