Assigning Wins and Losses

Posted on September 11, 2006 by


JC Bradbury – of Sabernomics fame and the author of forthcoming book The Baseball Economist – noted in an interview with Baseball Digest Daily that he was a fan of DIPS. For those who have not heard of this, DIPS is a metric used to evaluate the productivity of a pitcher in baseball. Voros McCracken invented this measure, which he explained in a Baseball Prospectus article in 2001.

In the course of explaining his measure, McCracken made the following observation about pitchers.

“There is little correlation between what a pitcher does one year in the stat (hits per balls in play) and what he will do the next. In other words, what Eric Milton’s hits per balls in play was in 2000 tells us next to nothing about what it will be in 2001. This is not true in the other significant stats (walks, strikeouts, home runs). Walks and strikeouts correlate very well and homers correlate somewhat well.”

“This is a crucial fact. One of the more critical aspects of statistical analysis is determining how well a statistic reflects an ability. It’s the test given to clutch hitting, catcher game-calling, pitcher won/loss records, and so on. One of the first things asked when addressing this is “Does the stat correlate well with itself from year to year?” One reason clutch hitting is questioned is that the “clutch hitters” change from year to year, which indicates that it probably isn’t the hitter as much as it’s other factors. The answer to whether hits per balls in play correlates well from year to year is a fairly solid “no.””

Why are hits per balls in play not consistent across time for pitchers? It is because how many hits a pitcher allows depends upon the ability of the eight players surrounding the pitcher on defense. In other words, part of what is credited to pitchers is actually a reflection of his teammates.

The point McCracken makes about pitchers is quite similar to a point we have made about quarterbacks in football. As noted in The Wages of Wins, in a recent post to this forum, two weeks ago by Martin Schmidt in The New York Times, and this past weekend in my own New York Times Keeping Score column (okay, we told this story more than once), quarterbacks are quite inconsistent across time.

Why are quarterbacks so inconsistent, especially with respect to turnovers? Again the answer is that the statistics we credit to quarterbacks depend upon the actions of the quarterback’s teammates.

Here is the odd part of the story. In the sports of baseball and football, wins and losses are often credited to individual pitchers and quarterbacks. Basically, the two positions whose performance in their respective sports depends crucially on the actions of others, are given full credit – and blame – when their team is victorious – or not.

Given what we now know about the performances of these athletes, how long will it be before people stop evaluating these players in terms of wins and losses?

— DJ