A different look at BABIPby Keith Isley
June 13, 2007
In January, David Gassko presented new research on DIPS theory that found a stronger relationship between pitchers’ fielding independent statistics and batting average on balls in play (BABIP) than previously thought. Though most of the information about a pitcher’s performance is still compressed into his peripherals: walk, strikeout, and home run rates, more recent research by Voros McCracken and others has led to a weaker form of DIPS theory that is now consensus. We've come a long way since McCracken’s original assertion in 2001 that “there is little if any difference among major-league pitchers in their ability to prevent hits on balls hit in the field of play."
Pitchers are thought to have “statistically significant” skill in preventing hits on balls in play, but the amount of this skill is minuscule and can be ignored in practice. David’s result bolsters the weak-form argument: He found past BABIP to be “only limitedly useful” in predicting future BABIP, and the general truth remains that DIPS metrics describe most of the variance in pitcher performance. In other words, striking out batters while avoiding walks and home runs demonstrates pitching skill, while producing outs on non-home run batted balls does not.
David concluded his article by asking whether “we can use new cutting edge data” to gain a better understanding of pitchers’ control over balls in play. In the spirit of that challenge, I’m going to investigate batted balls through a different lens, with an intensive approach using a scouting database built from video analysis of the 2004 and 2005 MLB seasons.
Hard Balls, Sweet Spots
The central idea in DIPS theory is what I’ll call the BABIP phenomenon, succinctly described by Dayn Perry in Baseball Between the Numbers: “When the bat hits the ball, and the ball stays in the park, luck, rather than the pitcher’s ability, usually holds sway.” The BABIP phenomenon is widely accepted; Curt Schilling has mentioned it in his blog. Luck implies an outcome generated by a chance process, like tossing a coin. Statistician Jim Albert reported in By the Numbers (here's the PDF file) that the hit rate on non-home run balls in play appears more random than other batting rates. (See also Dave Studeman’s comment.)
To begin to understand the BABIP phenomenon, consider what happens before and after the moment of contact. When the timing and barrel accuracy of the hitter’s swing put the sweet spot of the bat on the ball with full arm extension, solid contact produces maximum batted ball velocity. Scouting databases code these balls as well or hard hit. During the 2004-05 seasons, 59% of well-hit in-park balls were hits, while 19% of other in-park batted balls resulted in hits.
The main feature of randomness is independence from initial conditions—the outcome of an in-park batted ball cannot be fully predicted from the interaction of the pitcher and batter because neither participates in the play after contact (except in limited roles as fielder and baserunner). However, the randomness in BABIP is not itself random, and can be characterized in terms of hit probabilities: about 0.59 for a well-hit ball, about 0.19 otherwise. We can predict the likelihood that an in-park batted ball will result in a hit based upon the velocity generated at contact.
The BABIP phenomenon, then, is really a two-stage action. The first stage is the process of putting the batted ball in play; the second stage is the outcome that results from luck and defense. The dividing line is the moment of contact. The pitcher has no control over the second stage, the outcome, except as a fielder. But before contact, in the first stage, the pitcher wields substantial control over the batter’s attempt to hit the ball hard. The pitcher establishes his control through various inputs, including pitch type, spin, speed and location; pitch balance and sequence; deception and command; working the count; and the hitter’s history of previous plate appearances facing the pitcher.
In this model of BABIP, most of the luck occurs after contact in a chance process in which batted ball speed determines the probability of an outcome. BABIP is deterministic but random. By using strikeouts and home runs, which are 0% and 99.5% well hit respectively, DIPS admits pitcher control over the extremes of contact. Strikeout and home run rates serve as rough-and-ready proxies for a latent contact hardness variable that manifests as batted ball velocity. Home run is the only official batted ball stat about which we can reliably infer contact hardness.
Outcomes of in-park balls in play appear random because no ordinary records are kept about how well they were hit. David Gassko’s study detected a predictive relationship between strikeouts per game, home runs per game and BABIP. An explanation for these correlations might be that strikeout and home run rates approximately represent a pitcher’s tendency to allow hard hit balls, a tendency that persists as a skill.
If major league pitchers are skillful at controlling whether batters make hard contact, and BABIP becomes less random in a data-rich world that includes batted ball velocities, then it follows that pitchers have more control over batted ball outcomes than previously thought. BABIP is a lucky number only to the extent that hits occur more or less frequently than predicted by the 0.59 and 0.19 hit conversion rates for well-hit and other batted balls. The pitcher controls his rate of well-hit in-park balls in the same manner as he controls strikeouts, walks, and home runs.
I’ll look for evidence of pitcher control of BABIP in the rest of this article.
It Takes Two
The study examines the performances of two pitchers in 2004 and 2005. Why only two? For one thing, several first-rate statistical studies of DIPS theory using large pitcher-samples have already been done. For another, the application of DIPS is the evaluation of individual pitchers. The BABIP phenomenon is thought to affect every pitcher in the same way, not as a generalization found only in population averages. Aggressive use of statistical analysis that fits a least squares trendline across many pitchers may, in fact, obscure the unique pitcher control that I am looking for.
In choosing the two specimens, I sought pitchers with apparently lucky BABIP stats. Starting with the 70 MLB pitchers who threw at least 100 innings for the same teams in both 2004 and 2005, I looked for guys with stable DIPS who posted the most extreme year-to-year BABIP changes. The two selected pitchers are Barry Zito, then of Oakland, and the Royals’ Zack Greinke.
Year Pitcher Tm GS IP BFP ERA DIPS BABIP 2004 Greinke KC 24 145.0 599 3.97 4.66 .270 2005 Greinke KC 33 183.0 829 5.80 4.55 .340 2004 Zito OAK 34 213.0 926 4.48 4.53 .299 2005 Zito OAK 35 228.3 953 3.86 4.51 .249Average BABIP for AL pitchers with more than 100 innings was .302 in 2004 and .296 in 2005. What I’ve attempted, of course, is to hold the other variables constant within the limitations of an observational study. Same teams, consistent pitching abilities as measured by DIPS, but large variances in batted ball outcomes. If BABIP is mostly luck, it should be evident from these two cases.
Zito Adds to His Arsenal
Barry Zito produced a near-average BABIP in 2004, and improved it by 50 points the following season. Among the 70 pitchers meeting the study’s criteria, Zito delivered the second lowest BABIP in 2005. Notice the change from 2004 to 2005 in Zito’s pitch type balance:
--2004-- --2005-- Type All LHB All LHB FB 60% 65% 51% 47% CB 23% 29% 24% 24% SL 0% 0% 9% 26% CH 17% 5% 15% 3%Between seasons Zito added a slider, a pitch that breaks down and away from a left-handed batter. In 2005, he threw two-thirds of his sliders to left-handed batters, so the most dramatic changes to Zito’s pitch mix are in the LHB columns. Ignoring the rare changeup, left-handed hitters facing Zito in 2004 could expect a 70/30 mix of fastballs/curves during a given plate appearance. Major League hitters know how to handle predictable pitchers. Sitting fastball, left-handed batters got solid contact on 48% of Zito’s in-zone fastballs that were put in play during 2004, compared to 32% in 2005. His BABIP splits are revealing:
2004 2005 LHB BABIP .368 .247 RHB BABIP .262 .233Fewer hits on left-handed hitters’ in-play balls really reduced Zito’s overall ’05 BABIP. In isolation, his slider was unremarkable, about MLB average in terms of results. As an addition to his repertoire, though, the slider enhanced Zito’s other pitches by giving hitters something else to think about—especially left-handed ones.
Anticipation is a key to hitting. In place of the predictable fastball-dominated pitch balance of 2004, a left-handed batter in 2005 got a 50/25/25 mix of 88 mph fastballs, 81 mph sliders, and 72 mph curveballs. Zito’s proportion of well-hit in-park balls in play decreased from 32% to 23% with a concomitant drop in BABIP.
DIPS Misreads Greinke
Poor Zack Greinke was nearly a first-to-worst performer in BABIP from 2004 to ’05. Of the pitchers with over 100 innings for the same teams, Greinke ranked sixth in BABIP in 2004 before collapsing to #68 out of 70 the next season. Consequently his ERA ballooned, although Greinke’s DIPS suggested he actually pitched a little better in 2005. Can this be anything other than tough luck, BABIP style?
By some objective metrics, though, Greinke did not pitch as well in 2005, and his increased BABIP can be attributed to a complex of at least three factors:
- Diminished command in and around the strike zone. In 2004 Greinke threw four types of quality pitches, meaning that he was above MLB average at locating all four pitches for strikes down and on the corners, and for "chasable" balls around the edges of the zone. His command was off in 2005, especially on off-speed pitches to left-handed batters. On curveballs his proportion of hitter’s strikes – pitches left over the heart of the plate – increased from 28% to 32%. Hitter’s strikes on changeups increased 25% to 29%. Although the fastball was Greinke’s primary pitch, 36% of his lter eft-handed batin-park batted balls in 2005 were on off-speed pitches.
LHB BABIP Type 2004 2005 All .258 .341 CB .261 .385 CH .250 .311
- Getting behind in the count more often. Probabilities for a well-hit liner or fly are highest with the pitcher behind. In both 2004 and ’05 Greinke did great jobs at getting a strike on the first pitch of a plate appearance (67% both years) and throwing strikes early in the count (90% both years). But in 2005 he had problems taking 0-1 counts to 0-2, finishing off hitters with 0-2 counts, and getting outs after falling behind. Balls were put in play when Greinke was behind 15% more often in 2005 than ’04. And on the flip side, he had fewer balls in play with two strikes or when ahead – situations when hard hits are least likely. This is one example of how pitchers affect BABIP through different rates of offensive or defensive circumstances.
- Luck. Greinke’s BABIP when behind in the count was .180 in 2004 and .391 in 2005. Part of the difference was that he allowed more hard hit balls in 2005 (34%) than in 2004 (28%), but a lot was simply luck. In general, Greinke enjoyed a lucky BABIP in ’04 followed by an unlucky ’05. He pitched better in 2004 than 2005, but the change of fortune exaggerated the magnitude of the BABIP swing.
The game of baseball is saturated with luck—round ball, round bat. Uncertain outcomes are part of the game’s appeal. The BABIP phenomenon appears random at a large scale because we typically observe it in a way that is random when part of the information is hidden from the observer. In this article I’ve exposed some of the hidden information about the performances of two pitchers whose BABIP seems especially random. Although luck was certainly a factor, a large portion of each pitcher’s BABIP variance can be explained by non-spurious baseball regularities.
Chalking up BABIP as merely the result of chance outcomes does disservice to pitchers’ skill at preventing solid contact, which is the essence of pitching. Tossing a coin is chance, not skill, because you can’t control the result by how you flip the coin. Pitchers demonstrate skill in their control over the hardness of hitter contact, which indirectly but positively affects outcomes on in-park batted balls.
BABIP inevitably includes a random element because of the many unpredictable external events involved in a putout. After a recent loss to the Giants, Tom Glavine complained that “When their guys are hitting ground balls, I’m doing my job. I’m just not getting the results. There’s nothing I can say to make people understand when you go out there, do what you want to do, make the pitches and you don’t get the results.” That’s BABIP luck, the disconnect between pitching skill and batted ball outcome.
But some of the randomness in BABIP, as shown in the Zito and Greinke examples, amounts to a missing data problem. Ordinary baseball records omit pitch and batted ball attribute values, limiting our understanding of pitcher contributions to BABIP.
Keith Isley follows baseball from Raleigh, NC. He welcomes your comments and suggestions via email.
<< Return to Article