When we use a DIPS-based pitching statistic like FIP to calculate win values, our intention is to isolate the pitcher's contributions from those of his defense. FIP, however, is only a crude approximation of the truth: it assumes that all pitchers should have a league-average BABIP, it fails to correct for double plays and pitchers who struggle out of the stretch, and as a linear run estimator, it underestimates good pitchers and overestimates bad pitchers. PZR does a much better job of isolating pitching from defense without removing the effects of timing or hit quality. PZR is UZR, but from a pitcher's perspective. The pitcher is credited the run value of each batted ball, based on its trajectory, location, and speed, regardless of whether or not the ball fell for a hit. When these run values are totaled, we obtain a measure of a pitcher's "defensive support;" we can estimate his true performance by adding his defensive support to his actual runs allowed. As a result, PZR is perfect for Win Values, since it separates defense from pitching without removing anything else.
For each pitcher-season from 2001 to 2006, MGL calculated and posted PZR here, expressed in runs above or below average. I then added this rating to each pitcher's actual runs allowed in order to isolate pitching from defense. Using each pitcher's expected runs allowed, I calculated WAR following Tom Tango's methods. First, I calculated expected RA/9 for each pitcher, and turned this number into a winning percentage by assuming a park-adjusted league average offense (using Patriot's park factors). I set replacement level at .390 for starters and .470 for relievers. In addition, I incorporated a "closer replacement level" to properly value the contributions of highly leveraged relievers. This replacement level was set at .570, and was multiplied by each pitcher's inLI - 1 and his innings pitched, and then added to his regular WAR. For example, if a relief pitcher has a .760 winning percentage, 60 IP, and a 2.0 inLI, he would have (.760-.470)*(60/9) = 1.93 regular WAR, plus (.760-.570)*(60/9)*(2-1) = 1.27 "bonus" WAR, for 3.2 total WAR. This "closer replacement level" was applied to all relievers with an inLI above 1, so that a terrible but highly leveraged reliever, like Brad Lidge, 2009, would be penalized for blowing so many games.
Now I will provide leaderboards for each year 2001-2006, with analysis of some of the more surprising leaders. I will also provide a leaderboard for the entire period. I should warn you in advance that this is a very long post; a spreadsheet of leaders and career totals can be found here (organized by year, organized by name, and 2001-2006 totals).
When I saw this leaderboard, my first thought was "What are Joe Mays and Darryl Kile doing up there?" 2001 was Mays' career year, when he pitched 233 innings with a 3.16 ERA. Traditional DIPS methods were not as kind to Mays, however, as his FIP was 4.27. PZR's view of Mays' season is significantly more favorable, with a 3.34 xERA. Part of the discrepancy is explained by Mays' 78.2% strand rate; however, his .251 BABIP was significantly below his career average. Yet the small disparity between Mays' xERA and his ERA suggests that Mays, not his defense, was largely responsible for his low BABIP. Mays may have induced weaker contact, or he may have benefited from a "lucky" batted ball distribution; either way, PZR gives Mays credit for his lowered BABIP.
Kile's case is much more clear cut. He had a fantastic season in 2001, with 227 IP and a 3.09 ERA; his FIP was 3.74, and the disparity between his FIP and his xERA can be explained mostly by his 80.9% strand rate. In fact, Kile's .308 BABIP indicates that he was hurt by his defense. Though he is mostly remembered today for his untimely death, Kile was in fact a very good pitcher in his prime.
Randy Johnson's 11.0 WAR is the highest of the 2001-2006 time period. Johnson also claims second place, with 10.8 WAR in 2001. In these two years, Johnson was both extraordinarily effective (2.27 and 2.23 xERAs, respectively) and extremely durable (249 and 260 IP, respectively).
Mark Mulder may be the most surprising name on this list--his 2.46 xERA handily beats his 3.13 ERA and 3.38 FIP, and as a result his PZR WAR of 7.5 is much higher than his Fangraphs WAR of 4.9. Mulder's BABIP was .293, very close to his career aveage of .297, so subpar fielders probably do not explain the vast difference between Mulder's ERA and his xERA. Mulder did manage to post the highest IF/FB rate of his career, at 11.3%, and he also had a 55.7% groundball rate; however, Mulder's tERA, which adjusts for batted ball types, was 3.77. Ultimately, Mulder's 2.46 xERA was probably because of a combination of allowing weaker-hit balls in play and a more favorable batted-ball distribution.
Brad Radke seems out of place on this list, but his 2004 season was legitimately outstanding. He had a 3.48 ERA and a 3.55 FIP, and his very high 13.7% IF/FB rate no doubt helped decrease his xERA. In fact, Radke was one of the majors' most underappreciated pitchers throughout his career, a result of his low strikeout rates and the Twins' national obscurity. With mediocre strikeout rates, outstanding control, and a slight flyball tendency, Radke was the archetypal Twins' pitcher--over the last few years, Nick Blackburn, Scott Baker, Kevin Slowey, Glen Perkins, and Carlos Silva have all put up similar peripherals while pitching for the Twins.
Lidge is the first reliever to make the top 10, and is in fact the highest rated reliever between 2001 and 2006. Lidge accumulated an unreal 157 strikeouts, good for a 14.93 K/9. Among all pitchers, Lidge was 15th in the league in strikeouts. Among relievers, Eric Gagne had the second-most strikeouts, with 114. Lidge's 1.32 xERA significantly surpassed his 1.97 FIP, no doubt a result of his insane 88.1% strand rate.
Note that strand rates are not completely determined by luck--an extremely good pitcher, such as Lidge in 2004, should have a much higher strand rate than normal. This is one of the problems with a linear run estimator like FIP. PZR avoids these problems by starting with actual runs allowed, removing any need for run estimation.
Washburn's FIP in 2005 was 4.35, and according to Fangraphs he earned 2.4 WAR. Nevertheless, his xERA (3.04) exceeded his actual ERA (3.20) courtesy of an 81.8% strand rate. Furthermore, Washburn's career BABIP of .280 suggests that he may have an ability to post below-average BABIPs, causing FIP to underestimate him by regressing to the wrong mean. In 2005, Washburn's .289 BABIP was below league average but above his career average, indicating that the Angel defense did him no favors.
Millwood also fares surprisingly well, but remember that he led the American League in ERA, at 2.86. Millwood's FIP , at 3.73, was quite a bit higher than his ERA; however, his 79.1% strand rate depresses his xERA.
In 2006, Barry Zito posted a 3.83 ERA, a 4.89 FIP, a 4.55 tERA, 6.15 K/9, 4.03 BB/9, and a 38.2% groundball rate. These thoroughly uninspiring peripherals should have produced average value at best, yet Zito somehow earned 6.2 WAR and a 3.38 xERA. This wasn't the only time Zito's xERA outstripped his FIP; in fact, Zito's xERA was lower than his FIP every year that I have data for. Furthermore, ZIto's ERA was always closer to his xERA than to his FIP. But FIP assumes a league-average BABIP; Zito has a career .275 BABIP, so far below league average that it very likely indicates real skill. Moreover, Zito's strand rates in this time period were consistently above league average. Among his batted ball profile, one thing stands out: Zito's ridiculously high IFFB rates, ranging from 13.3% in 2006 to 17.8% in 2002. These popups explain both his low BABIPs and his high strand rates: as near-automatic outs, they both depressed Zito's BABIP and allowed him to strand more runners, since they were the equivalent of 40 or 50 more K's per year. But Zito's story doesn't end there: after the 2006 season, he signed a much-maligned $126 million contract with the Giants. Viewed through the lens of PZR, the contract does not look nearly as bad as it seemed at the time. Nevertheless, Zito fell off a cliff in San Francisco, with his ERA rising from 3.55 with the A's to 4.56 with the Giants. Unfortunately, we don't have PZR data for Zito's years with the Giants, so we can't tell if his xERA followed suit. However, Zito's IFFB% did drop dramatically, to around 10% over the last three years, suggesting that his xERA would have risen in concert with his ERA. This sudden drop in infield fly balls suggests that Oakland Coliseum's vast foul territory helped inflate Zito's value while he was with the A's; much like Dustin Pedroia uses the Green Monster to hit 40 doubles a year, Barry Zito used Oakland Coliseum to turn himself into a very good pitcher.
Back to the WAR leaderboard...Jason Jennings' season is surprising, though largely supported by FIP (4.09) and Fangraphs' WAR (4.6). A lousy won-loss record and an ERA inflated by Coors Field largely obscured Jennings' accomplishments. Scott Kazmir's 2.59 xERA vastly exceeds both his actual ERA (3.24) and his FIP (3.36). Tampa's defense behind Kazmir was downright atrocious with a .323 BABIP; moreover, Kazmir's xERA benefits from his high 77.0% strand rate.
Schilling gets surprisingly little respect as a truly great pitcher; his 1997-2006 peak was among the best of all time. Oswalt also gets little respect, probably due to the Astros' general obscurity. With the exception of Zito, these pitchers are generally considered the best of the 2001 to 2006 period.