Defense
Independent Pitching Statistics for 2004
Here we go again... for the third year in a row, I am presenting
Defense Independent Pitching Statistics (DIPS) via this website.
DIPS was invented by an analyst named Voros McCracken, whose
studies of pitching statistics suggest
that major league pitchers do not differ greatly on their
ability to prevent hits on balls in play. Note that this
is not the same as saying that major league pitchers do not
control their ability to prevent hits on balls in play, a
statement which McCracken is often accused of making. The
rate at which a pitcher allows hits on balls in play has more
to do with defense and luck than to his own skill, and can
vary greatly from year to year.
This
controversial and somewhat counterintuitive way of looking
at pitching statistics has its advantages. The chief one is
that we can do a better job of evaluating a pitcher's future
performance by concentrating on the defense-independent things
he does strike batters out, walk them, plunk them,
and give up homers than we can by considering the effects
of the defense playing behind him.
DIPS
combines the defense-independent elements of a pitcher's record
with a league-average hit rate on balls in play. The result
is a translated line of Defense Independent Pitching Statistics,
including a DIPS ERA (or dERA); that is, an ERA based on defense-independent
pitching performance. Importantly, McCracken found that this
dERA correlates better with the following season's ERA than
the pitcher's actual ERA does, making it a useful predictive
tool.
McCracken's
work, which covered the seasons 1999-2001, has received much
acclaim in the sabermetric world, including from Bill James,
who discussed it in The New Bill James Historical Baseball
Abstract. McCracken parlayed that acclaim into a position
as consultant to baseball operations for the Boston Red Sox,
the same team for whom James is now Senior Baseball Operations
Advisor. As a result, McCracken has not published any revision
to the DIPS methodology in three years and is no longer partaking
in public dialogue regarding the system.
Which
isn't to say that our understanding of the matter is the same
as it was three years ago, when DIPS 2.0 was unveiled. Tom
Tippett of Diamond Mind Baseball published a
lengthy study on pitchers with long careers and found
evidence that those pitchers exhibited some ability to affect
the rate of hits on balls in play; however, "Their influence
over in-play hit rates is weaker than their influence over
walk and strikeout rates." Group discussions at Baseball Primer
explored the
role of luck relative to pitching, defense, and park effects
as well as other avenues.
Mitchel
Lichtman analyzed play-by-play data and found that certain
classes of pitchers appeared to have difffering abilities
to prevent hits:
Getting
back to the possibility of certain classes of pitchers having
unique hit preventing abilities, it should be clear that
fly ball pitchers, on the average, will have a different
$H than will ground ball pitchers, since a fly ball has
a higher out percentage than a ground ball. In fact, extreme
ground ball pitchers have a BABIP of .297 (1992-2003), whereas
extreme fly ball pitchers have a BABIP of .281 (extreme
= top and bottom 10% in G/F ratio for pitchers with at least
100 BIP in a season). Of course, the run value of a FB hit
is greater than that of a GB hit, such that the actual run
value of all pitchers? BABIP is almost exactly the same,
regardless of their G/F ratios.
Nate
Silver reported in Baseball
Prospectus 2004 that groundball-to-flyball ratio
has higher correlations from season to season than any other
pitching metric, slightly edging out strikeout rate (.79 to
.77 in his population of pitchers who threw at least 50 innings
in consecutive seasons since 1975), noted that GB/FB is a
better predictor of future HR rate than the HR rate itself,
and that pitchers with a higher GB/FB allow significantly
fewer HRs and significantly more base hits going forward.
Silver has incorporated the predictive powers of GB/FB into
his PECOTA
forecasting system as well as using what he calls a "weaker
version" of DIPS. More
recently, in a discussion on the BP email list, Silver elaborated:
Based
on the equations that I use for PECOTA, a difference of
20 points in GB/FB percentage (say 65% groundballs versus
45% groundballs) translates to a difference of about 1 percent
in BABIP (say 30% H/BIP versus 29%). That isn't particularly
large, and in the bigger picture, you'd rather have a guy
generating more groundballs since the positive predictive
influence on his home run rate is much greater than the
negative one on his hit rate. But it's there, and it's statistically
significant.
For
all of these developments, nobody has come up with a revised
DIPS system or a refutation of McCracken's claims thorough
enough to merit discarding the last iteration of DIPS (version
2.0); if anything, recent work has done more to confirm than
deny his work.
So
without adding anything new to the method, I offer 2004 DIPS
2.0 data here, as well as several relevant links. McCracken's
original work (DIPS 1.x) showed how various rates (strikeout,
walk, HR, BABIP) correlated from one year to the next,
and he also did so showing how the dERA
correlated better with the following season's ERA than the
previous season's ERA did. I have taken the three years of
DIPS 2.0 that I've produced and found correlations that are
fairly consistent with his findings:
DIPS 1.x DIPS 2.0
Years 98-99 02-04
Baseline IP 162 162 100
Number of P 60 114 201
$BB .681 .699 .693
$SO .792 .736 .780
$HR .505 .317 .301
$H .153 .129 .105
Years 93-99 02-04
Baseline IP 100 162 100
Number of P 503 114 201
ERA to next ERA .407 .238 .291
dERA to next ERA .521 .332 .397
The
baseline IP is the number of innings pitched in both seasons
a pitcher needed to qualify for the study. McCracken used
separate baselines for the two comparisons, but since I had
data for both 100- and 162-inning levels, I'm running it here.
The
most striking difference in the top section appears to be
in the home run rates, none of which are park adjusted here.
Even with park adjustment, homer rates are almost certainly
more subject to park differences than other defense-independent
events, so this isn't horribly shocking. The HR data for the
last three seasons is somewhat tainted by the strange Montreal
situation of playing 22 ballgames in a Puerto Rican bandbox
and the addition of three new parks over the past two years,
one of which, the Padres' Petco, was the second-toughest park
to homer in during 2004. In the bottom section, the correlations
between dERA to following season's ERA are lower than in McCracken's
study, but it's still better a predictor than the actual ERA
is. Viva DIPS.
Moving
along to the feature pesentation, data for the post season
was obtained via ESPN.com, with Larry Mahnken of The
Hardball Times and the Replacement
Level Yankee Weblog greatly assisting me in its collection
thanks, Larry. Weighted three-year Park Home Run Factors
are used this year, a first since I've presented the data;
it's worth noting that the ERA-related correlations above
didn't budge more than .002 when I reverted to single-year
factors to insure that apples were being compared to apples.
I have not park-adjusted strikeouts or walks, though it is
possible to do so in a similar manner; Mahnken has an extremely
handy DIPS
Worksheet which, among other goodies, contains the relevant
park factors for 1969-2004 via data from Retrosheet.
Pitchers
are listed by the team with which they finished the season.
For
all pitchers, the figures for Batters Faced Pitching (BFP)
and for Batting Average on Balls in Play (BABIP, or just BIP
within the team listings) are actual; all others are DIPS.
DIPS-generated totals have been rounded to the nearest whole
number for presentation purposes.
The
calculation for Earned Runs Above Replacement (ERAR) is based
on a replacement-level ERA which 25% higher than league average
(i.e., an ERA+ of 80). For the NL in 2004, that is a 5.38
ERA against a league average of 4.30, for the AL the numbers
are 5.79 and 4.63. Pitchers who spent time in both leagues
have had their ERAR subtotals added together to determine
unique replacement levels. Unique Park Home Run Factors for
those pitchers have been calculated based on the numbers of
batters faced while pitching for each team.
In
addition to the team-by-team stats, I have presented the Top
30 (or so) in six categories: dERA, ERAR, Lower dERA than
ERA, Higher dERA than ERA, Lowest BABIP and Highest BABIP.
Those leaders are all based on a minimum of 100 DIPS Innings
Pitched (dIP) except for ERAR. An asterisk (*) denotes lefties.
Every effort has been made to ensure the accuracy of the data
here, but the possibility of human error in working with it
still exists. Caveat emptor. |