Syndicate this site: (RSS)

A DIPS Study

If you aren't already familiar with DIPS, Futility Infielder has a pretty good collection of links on the subject.

The motivation of this study is one done by Tom Tippett. What he found was a definite trend: "The pitchers with longer careers were progressively better than their shorter-career counterparts in every respect."

The "every respect" bit bothered me. If there's any sort of correllation at all among the different skills, you would expect that those pitchers better at every other aspect would also be better at preventing hits. Can we factor that out?

Sifting through the Lahman database, I pulled out each pitching stint from 1969 to 2003. Note that a pitcher might have more than one stint in a season, if he is traded to another team. For each of these stints, I calculated the DIPS values for $BB (using BB and HBP in the numerator), $K, $HR, balls in play, hits, and an approximation of the adjustment for context, based on the team score for $H compared to the league.

I also aggregated career totals (not exactly, since some of the pitchers in the study debuted before 1969), and used those to calculate the DIPS numbers as above.

Then I walked through all of the stints, computing a distance between each set of dips variables. Then, sorting the stints by proximity to a career, I created a composite pitching career with roughly the same number of balls in play (slightly more, as I simply added to the totals until it exceeded my target).

For example, in creating Composite Clemens, the following seasons were used:

Todd Worrell 1987
John Wetteland 1989
Dave Veres 1998
Rick Aguilera 1990
Scott Elarton 1999
Jim Brewer 1972
Mike Scott 1987
Chuck Finley 2002
Mike Timlin 1996
Tom Henke 1988
Dan Miceli 1998
Fernando Valenzuela 1981
Kevin Appier 1996
Bob Walk 1984
Greg McMichael 1997
Scott Strickland 2000
Kevin Brown 2001
Enrique Romo 1977
Darren Dreifort 1998
Tom Seaver 1972
Steve Carlton 1981
Rob Murphy 1987
Francisco Cordero 2002
Doug Brocail 1999
Jose Rijo 1994
Bob Moose 1969
Jeff Fassero 1993
Trevor Hoffman 1993
Randy Wolf 2001
T.J. Mathews 1997
John Wetteland 1997
Danny Cox 1993
Dave Righetti 1984
Sid Fernandez 1989
David Cone 1988
Paul Assenmacher 1990
David Cone 1998

These all feature unadjusted DIPS numbers rather similar to Roger Clemens. For comparison


BFP BB+HBP K HR H-HR
Roger Clemens 17653 1520 4099 321 3356
Composite 18351 1587 4245 336 3402
(scaled) 17653 1526 4083 323 3272

So Clemens is a bit better than his matchup, except in hits, where he trails by about 84. However, Clemens probably gave up about 90 extra hits because of his environment (approximated to first order), where the composite saved 37. So we are looking at a discrepency of 1-2%.

Are we on to something? Sadly, no. Clemens just happens to be an example where the numbers come out. The career players who stood out on Tippett's chart still are way, way out line with what you would expect if the pitchers skill were wrapped up in those three numbers.

June 3, 2004 1:09 AM | TrackBack

Comments
Post a comment




Who are you?