FREE hit counter and Internet traffic statistics from freestats.com

Friday, February 11, 2005

Looking at DIPS for 2005

One of the most interesting observations in baseball analysis in the last decade was that made by a college student from Chicago named Voros McCraken about five years ago. Essentially, McCraken argued that pitchers have little control over whether balls that are put in play (BIP) (excluding homeruns) turn into hits or outs. What follows is the realization that a pitcher's effectiveness can be directly correlated with his ability to stop hitters from making contact (strikeouts), getting on base for free (walks), and producing runs with a single swing of the bat (hitting homeruns). A corollary to this realization is that the difference in ERA between pitchers of similar ability (in terms of strikeout, walk, and homerun rates) is most influenced by luck and secondly by the defense's ability to convert batted balls into outs. In other words, the number of non-homerun hits a pitcher gives up is not indicative of his skill, as was universally assumed, but rather of luck and defense with luck being by far the larger component.

Of course, baseball fans have long known that scratch and bloop hits can beat your team just as easily as ringing line drives but the underlying assumption always was that "good pitchers" will be beat less often by those sorts of events. Well, under McCraken's view of the world that thinking is correct only if by "good pitchers" you mean those that strikeout an above average number of hitters, have good control, and stay away from the homerun.

McCraken then built a system to evaluate pitchers that he called Defense independent Pitching Statistics or DIPS that he parlayed into a consulting position for the Red Sox. For example, McCraken created a DIPS ERA that more accurately predicts a pitcher's ERA for the coming year than the previous year's ERA. McCraken created versions 1.0, 1.1, and 2.0 of this methodology but as a result of his position with the Red Sox no longer participates in the public discussion of his system.

Many in baseball's sabermetric community, notably Bill James in the New Historical Baseball Abstract, have commented that McCraken's idea is one of those that is so obvious in retrospect that it's surprising that it hadn't been explored before 1999. I liken it to the wide-spread realization from a decade before that outs are the most precious resource a team possesses and therefore they shouldn't be squandered. In recent years there has been some pushback on DIPS and several analysts have detected an ability of some pitchers, particularly knuckleballers like Phil Niekro and Tim Wakefield but also left-handers, to induce a slightly higher percentage of outs on ball put in play by reducing the number of line drives that are hit. I wrote about this work last year and discussed some of its implications for strategies that can be employed for successful pitchers.

From a practical perspective one of the conclusions that follows from DIPS is that general managers should be wary of investing long-term in young non-knuckleball pitchers with low strikeout rates. This is because these sorts of pitchers may have simply been the recipient of good luck and because their strikeouts rates will inevitably decline as they age, further reducing the wiggle room they have in terms of control and the avoidance of homeruns. The Royals Jimmy Gobble comes immediately to mind as I've written about before.

To illustrate the concept behind DIPS I used the Lahman database to take a look at the 86 pitchers who threw more than 120 innings in both 2003 and 2004. I then calculated various rate statistics including their Batting Average on Balls in Play (BABIP), Walks+Hits per inning pitched (WHIP), K/IP, BB/IP, HR/IP, ERA, Component ERA (ERAC), a quick and dirty version of DIPS ERA , and my own simple version of DIPS ERA (SDERA) which I calculated by substituting the non-homerun hit portion of the Component ERA formula with the number of hits the pitcher would have given up had his BABIP been the .288 average of the two seasons. Component ERA simply attempts to predict a pitcher's ERA given the components of his performance. I did not adjust these statistics by ballpark as called for by the full DIPS methodology.

Using this data I then calculated the correlation coefficient of each of these stats for 2003 and 2004. For Component ERA and my DIPS ERA I ran the correlation against the 2004 ERA. The assumption is that those statistics that have higher correlation can be attributed to a pitcher's ability while those that have low correlation can be attributed to other factors such as randomness. The results were:


BABIP .087
ERA .190
ERC-ERA .233
SDERA-ERA .276
HR/IP .312
DERA-ERA .322
WHIP .407
K/IP .717
BB/IP .732

A few observations of the result that illustrate the logic behind DIPS include:

  • There was virtually no correlation between BABIP in 2003 and 2004. Therefore, BABIP can attributed to randomness

  • Note too that for many of these pitchers the same defense was behind them in both seasons and yet the correlation was non-existent. This indicates that randomness is by far the larger factor in the variation of BABIP

  • Only strikeouts and walks per inning pitched had a correlation coefficient greater than .7 and could be considered strongly correlated. This indicates that pitchers have the most control of these two abilities

  • Just as McCraken discovered DIPS ERA calculated for 2003 was a better predictor of the pitcher's actual 2004 ERA than was his 2003 ERA, his 2003 Component ERA, or even my simple version using CERA as a base. The correlation for this quick and dirty version of DIPS ERA was not as strong, however, as the full version as documented by The Futility Infielder

  • I was surprised that homeruns per inning pitched did not result in a stronger correlation which may indicate that the ability to avoid homeruns is not as much of a skill as some proponents of DIPS might argue and should be factored into revisions of the formula


  • So in what other ways can DIPS be used practically?

    First, consider the "leaders" in Batting Average on Balls in Play (BABIP) in 2003:

    2003 2004
    Glendon Rusch 0.381 0.287
    Jeff Weaver 0.343 0.293
    Rodrigo Lopez 0.340 0.277
    Shawn Estes 0.329 0.301
    Jason Jennings 0.327 0.326
    Kelvin Escobar 0.325 0.293
    Mark Hendrickson 0.325 0.296
    Josh Beckett 0.322 0.284
    Jeremy Bonderman 0.318 0.278
    Joe Kennedy 0.318 0.294

    In other words, these were the pitchers who were the unluckiest in terms of batted balls falling for hits. What you'll notice is there is little correlation between their 2003 and 2004 performance and in fact every pitcher did better in 2004. Why is that the case? DIPS says that in 2003 these pitchers were unlucky and so the probability that they would be just as unlucky in 2004 was remote and so they've regressed to the mean. For example, in 2003 opponents hit a whopping .381 on balls in play against Glendon Rusch of the Brewers resulting in his giving up 160 non-homerun hits in 123.3 innings and a painful 6.43 ERA. In 2004 his luck changed and opponents hit a more reasonable .287 (in fact .288 was the average for the two years) and with a bit better control Rusch's ERA dropped to 3.48 for the Cubs while he gave up just 127 hits in 129.7 innings. Rusch's 2003 DIPS ERA was 3.89 which was much closer to his actual 2004 ERA. Seven of the ten pitchers had an ERA in 2004 that was better than his ERA in 2003.

    On the flip side those with the lowest BABIP in 2003 were:

    2003 2004
    Barry Zito 0.239 0.291
    Ryan Franklin 0.245 0.289
    Darrell May 0.249 0.318
    Russ Ortiz 0.250 0.283
    Jason Schmidt 0.253 0.263
    Kip Wells 0.253 0.313
    Tim Hudson 0.253 0.297
    Jake Peavy 0.255 0.300
    Victor Zambrano 0.259 0.266
    Jarrod Washburn 0.259 0.284

    Opposite of the previous list all of these pitchers had 2004 seasons that were worse than in 2003. These were the lucky pitchers in 2003. A case in point is the Royals Darrell May who in 2003 gave up just 166 non-homerun hits in 210 innings while giving up 31 homeruns. Opponents, however, hit just .249 on balls in play. Had he been average in this respect he would have given up 192 non-homeruns hits which would have impacted his fine 3.77 ERA.

    Going into 2004 many of us had hoped that May had become one of the rare pitchers that can actually suppress hits on balls in play. Apparently, Royals GM Allard Baird was as well as he signed May to a 2-year $4.95M contract after the 2003 season. Alas, we were all disappointed in 2004 as his BABIP climbed to .318 and he was subsequently dealt to the Padres. His DERA in 2003 was 4.75, much closer to his actual 2004 ERA of 5.23. To be fair, he also gave up more homeruns per inning pitched and walked more batters.

    From this list nine of the ten pitchers (all accept Jake Peavy) had worse ERAs in 2004 than in 2003.

    From both of these lists that conclusion that one might come to is that pitchers with high BABIP in one season will likely have better results in the following season and vice versa. Therefore general managers should be looking for pitchers that are under-valued because of their bad luck and avoid pitchers who are over-valued because of their good luck.

    So who might be undervalued coming into 2005? Here are the top ten in highest BABIP for 2004.

    Sidney Ponson 0.327
    Derek Lowe 0.327
    Kevin Millwood 0.327
    Jason Jennings 0.326
    Kyle Lohse 0.321
    Darrell May 0.318
    Kenny Rogers 0.315
    Roy Oswalt 0.314
    Brian Anderson 0.313
    Aaron Sele 0.313

    A few observations:
  • It's interesting that both Darrell May and Brian Anderson made the top 10 and hopefully for the Royals this means that Anderson can expect better results than his awful 2004 campaign.

  • Although the Dodgers were much ridiculed for signing Derek Lowe, that deal, in conjunction with Lowe pitching half his games at pitcher-friendly Dodger Stadium, will likely have many pundits singing a different tune this season

  • The Indians seem to have made a good move in picking up Kevin Millwood, who had largest differential in DIPS ERA versus actual ERA of this group (3.79 to 4.85)


  • And who might be over-valued? Here are the leaders in BABIP for 2004.

    Al Leiter 0.240
    Johan Santana 0.250
    Kazuhisa Ishii 0.254
    Ted Lilly 0.261
    Tom Glavine 0.261
    Odalis Perez 0.263
    Jason Schmidt 0.263
    Victor Zambrano 0.266
    Jamie Moyer 0.268
    Jerome Williams 0.270

    The Marlins may be in for a surprise with Leiter as may the Twins with Santana. It is interesting that two pitchers made the top 10 in both seasons (Schmidt and Zambrano), which may indicate that these are among the few pitchers who have an ability to suppress BABIP although I'll reserve judgement since neither is a knuckle-baller or left-handed. Jamie Moyer of course may also be in this group.

    1 comment:

    Garth said...

    And who might be over-valued? Here are the leaders in BABIP for 2004.


    Al Leiter 0.240
    Johan Santana 0.250
    Kazuhisa Ishii 0.254
    Ted Lilly 0.261
    Tom Glavine 0.261
    Odalis Perez 0.263
    Jason Schmidt 0.263
    Victor Zambrano 0.266
    Jamie Moyer 0.268
    Jerome Williams 0.270


    It's funny how nearly everyone of those players has struggled mightily in '05.