Batted Balls and DIPS

When Voros McCracken released his first Defense Independent Pitching Statistics (DIPS) article, a huge wave of excitement passed over the baseball statistical community. His conclusion, “that major league pitchers don’t appear to have the ability to prevent hits on balls in play,” was a breakthrough, and it led to a whole new approach to pitching analysis.

McCracken’s article was followed by Tom Tippett’s in-depth look at DIPS, which has since become the way the baseball statistics community has looked at DIPS: “pitchers do influence in-play outcomes to a significant degree, [but] there’s a lot more room for random variation in these outcomes than in the defense-independent outcomes.”

Tippett was right, but what kind of players have an impact on batting average on balls in play and why? Tippett found that knuckleballers were much better than average at preventing hits on balls in play, and in his second version of DIPS, McCracken adjusted for this and for the tendency of lefties to allow slightly higher Batting Averages on Balls in Play (BABIP). But why?

That’s where Mitchel Lichtman (also known as MGL) came in. In his article “DIPS Revisited” Lichtman uncovered the impact of infield flies, line drives and ground ball/fly ball ratio on BABIP. He concluded that “even though the overall correlation on BABIP is near zero, there appear to be several components that have some predictive value and are therefore somewhat within a pitcher’s control.”

McCracken was well aware of this; in his DIPS 2.0 article, he stated that “it has been proposed for some time that fly ball pitchers tend to have an advantage here over ground ball pitchers. In the end, I’m pretty sure this is right, but the problems currently here are tough to overcome…There are really three types of batted balls as they are counted: ground balls, fly balls and line drives…When I (or someone else for that matter) can work this out, we’ll have DIPS Version 3.0.”

This is where we stand right now. Thanks to The Hardball Times, the public now has access to fly ball, ground ball and line drive stats, as well as infield fly numbers. By regressing those, as well as walks, hit batters and strikeouts, onto 2004 season ERAs, I was able to create what I think is the next step in DIPS.

First of all, we need to contend with the problem brought up by McCracken and others: are these truly independent numbers? I think so. The idea that batters have all the impact over batted balls is bunk—Lichtman showed that pitchers have a good deal of control over ground ball and outfield fly ball rates and some control over line drive and infield fly rates. The potential of scorer bias is real and unavoidable, but if it is not considered a problem in systems like Ultimate Zone Rating (UZR), I see no reason why it should be a problem for a new version of DIPS. Moreover, these statistics fit the definition of “defense independent;” “scorer independent” does not have to be a requirement.

But this isn’t the biggest issue in devising DIPS 3.0. The greatest problem, I think, is how we define DIPS. On one hand, it seems to be a statistic that measures current season performance, one that has some more explanatory power than plain ERA. On the other hand, the whole concept of DIPS was based on year-to-year correlations, making it seem like a predictive stat.

Nevertheless, I concluded that DIPS is supposed to measure current season performance. If DIPS were trying to predict future performance, then each component would be regressed separately, since home runs, walks, strikeouts, etc. do not have the same year-to-year correlations. This is why I used same-season ERA as the dependent variable in my regression and not next-season ERA.

A third issue: Ground ball/fly ball ratios have more serious implications than just their impact on DIPS. A pitcher’s ground ball rate has a weak, but nonetheless significant, correlation with unearned runs allowed. Take a look at last year’s leaders in Unearned Runs Per Nine Innings (UER):

Player         Team     G/F    GB/9   UER/9
Lowe D.        BOS     3.07    19.7    1.38
Webb B.        ARI     3.47    17.4    1.21
Riedling J.    CIN     2.00    16.3    1.16
Biddle R.      MON     1.36    14.5    1.04
Meadows B.     PIT     1.78    15.0    1.04
Parrish J.     BAL     1.86    13.4    1.04
Anderson B.    KC      0.81    10.9    1.03
Dickey R.      TEX     1.16    14.1    1.03
Arroyo B.      BOS     1.07    11.2    0.96
Dessens E.     ARI     1.62    15.3    0.95
Minimum 75 IP

The average ground ball/fly ball ratio among these 10 pitchers is 1.82 compared to a league average of 1.17. The idea of separating earned runs and unearned runs was to separate pitching and defense. But since using batted ball type does the same, perhaps the next generation of DIPS should explain runs allowed and not ERA. I regressed the same components on Runs Allowed (RA) as well, and indeed, the only batted ball type whose value changed significantly was ground ball ratio.

Here are the coefficients I derived for each variable in my equation:

          ERA      RA
GB       +.050    +.076
OF       +.251    +.256
IF       -.041    -.038
LD       +.224    +.217
SO       -.120    -.129
BB       +.316    +.332
HBP      +.430    +.532

To calculate DIPS ERA or RA, you multiply each outcome by the appropriate coefficient, divide by innings pitched and multiply by nine. Here are the two formulas laid out in detail:

DIPS ERA:
(-0.041*IF+0.05*GB+0.251*OF+0.224*LD+0.316*BB-0.12*SO+0.43*HBP)/IP*9

DIPS RA:
(-0.038*IF+0.076*GB+0.256*OF+0.217*LD+0.332*BB-0.129*SO+0.532*HBP)/IP*9

IF = Infield flies
GB = Ground balls
OF = Outfield flies
LD = Line drives
BB = Walk
SO = Strike out
HBP = Hit by pitch

At this point, the question that will inevitably arise is “so what?” What advantage does this new version of DIPS have over the previous version of DIPS or its simpler cousin, Fielding Independent Pitching (FIP)? First of all, it correlates better with actual ERA, with a “r” of .8 as opposed to .73 for FIP.

Moreover, it provides more, and new, information—DIPS 3.0 correlates with FIP no better than FIP does with ERA (.74), meaning that the difference between DIPS 3.0 and FIP is the same as the difference between FIP and ERA. In other words, DIPS 3.0 is providing a plethora of new information.

And while year-to-year correlations cannot be calculated until the season is over, it seems that DIPS 3.0 has better predictive power. Take a look at the players on whom DIPS 3.0 and FIP disagreed the most last year:

Player         Team     ERA     FIP    DIPS    DIFF    2005
Contreras J.   NYY     5.64    6.07    4.40    1.67    4.08
Park C.        TEX     5.46    6.40    4.83    1.57    5.66
White R.       CLE     5.29    5.84    4.55    1.29    3.75
Moyer J.       SEA     5.21    5.98    4.70    1.28    4.29
Capuano C.     MIL     4.99    5.33    4.11    1.22    3.55
Quantrill P.   NYY     4.72    3.95    5.16    1.21    3.58
Elarton S.     CLE     4.53    5.87    4.67    1.20    4.74
Fossum C.      ARI     6.65    5.80    4.65    1.15    4.44
Milton E.      PHI     4.75    5.38    4.31    1.07    6.48
Lima J.        LAD     4.07    5.11    4.07    1.04    6.47
Minimum 75 IP

FIP is a little closer to 2005 ERA than 2004 ERA if you use RMSE, vice-versa if you use mean error, but both have similar error margins. Meanwhile, DIPS is far and away better with an RMSE almost .3 points better and a mean error .4 points better. Yes, all small sample size caveats apply, but I’m willing to bet that the conclusion would remain the same if you looked at all pitchers after the season was over.

Let’s also take a look at pitchers that most under-performed their DIPS ERA last year:

Player         Team     ERA    DIPS    2005
Fossum C.      ARI     6.65    4.65    4.44
Nomo H.        LAD     8.25    5.83    7.24
Wood M.        KC      5.94    4.35    4.09
Biddle R.      MON     6.92    5.11      -
Benoit J.      TEX     5.68    4.29    3.67
Acevedo J.     CIN     5.94    4.50    6.80
Van Poppel T.  CIN     6.09    4.65      -
Contreras J.   NYY     5.64    4.40    4.08
Lowe D.        BOS     5.42    4.29    4.20
Garcia F.      CHW     4.46    3.54    3.54
Minimum 75 IP

All who are still playing except for Acevedo have improved (if you can call Nomo’s 7.24 ERA an “improvement”) on last season’s ERA, and DIPS 3.0 pretty much nailed six of the eight pitchers who are in the major leagues this year.

And what about the players who exceeded their DIPS ERA most last season? Take a look:

Player         Team     ERA    DIPS    2005
Stanton M.     NYM     3.16    4.52    7.07
Ohka T.        MON     3.40    4.89    3.88
Otsuka A.      SDP     1.75    2.65    2.58
Madson R.      PHI     2.34    3.55    3.33
Carter L.      TBD     3.47    5.27    5.16
Peavy J.       SDP     2.27    3.46    2.97
Leiter A.      NYM     3.21    4.92    5.96
Cordero C.     MON     2.94    4.58    0.98
Linebrink S.   SDP     2.14    3.50    2.45
Foulke K.      BOS     2.17    3.75    6.23
Minimum 75 IP

Nine out of the 10 pitchers saw an increase in ERA this year! DIPS absolutely nailed it with Otsuka, Madson, Carter and Peavy.

What you’re seeing here is the predictive power of DIPS, which is only improved by using batted ball information. There are a few loose ends that must be tied up:

1) Dave Studeman and I had a lengthy e-mail exchange about this, and in the end I decided that infield flies must be included. Infield flies per ball in play actually have a slight negative correlation with outfield flies per ball in play. Inducing infield flies is a skill, and while it correlates somewhat weakly year-to-year (Lichtman found an “r” of .140), a small subset of pitchers exhibits clear control over the percentage of their fly balls that are infield pop ups. I would encourage studies looking into who those pitchers are—one thing I have noticed is that extreme ground ball pitchers allow fewer than expected infield fly balls.

2) I would recommend using the DIPS version that predicts runs allowed, not earned runs. The idea of earned runs, originally, was almost somewhat ingenious; it was the first attempt to separate pitching and fielding. But once we can characterize each outcome independent of defense, the need to separate earned and unearned runs disappears.

3) You may have noticed that I did not include home runs in DIPS 3.0. I see no reason to do so since all home runs are either outfield flies or line drives; including home runs separately is thus pointless.

4) Some readers may be familiar with the concept of Pitcher Zone Rating (PZR), or UZR for pitchers. This is somewhat like that, except without knowing where each ball hit into play lands. I don’t know which is better. DIPS 3.0 should have more predictive power, but as a measure of value, PZR is surpassed only by win expectancy (and is actually how pitching and fielding would be split in a win expectancy added system). PZR is not necessarily the next step in DIPS analysis, but I’m not sure this is the last either.

5) The coefficients presented here are surely not the last word on the subject. They are based on only one year of data, and they are not adjusted for park factors. Look for more analysis into DIPS 3.0 soon.