PZR-based Win Values 2007 and 2008

MGL sent me the 2007-2008 PZR files and allowed me to share my win value calculations with you. Using the same methodology as I used for the 2001-2006 win values, I calculated PZR-based WAR for 2007 and 2008. Here are 2007-2008 win values, 2001-2008 win values ordered by name, 2001-2008 win values ordered by year, and 2001-2008 totals. I'll also summarize the 2007 and 2008 leaderboards and provide an updated 2001-2008 leaderboard.


When C.C. Sabathia won the Cy Young in 2007, most analysts were pleased that the more deserving candidate was chosen over 20-game winner Josh Beckett. But PZR rates Beckett as 1 win better than Sabathia, due to his substantial edge in xERA, 2.69 to 3.26. Beckett had a .304 BABIP in 2007 despite a 15.8% line drive rate* and a 13.6% infield fly ball rate, both of which indicate that he suffered from some poor defensive support.

*Another reminder that a pitcher's line drive rate has almost no predictive value: after Beckett's 15.8% LD season in 2007, he posted a 25.2% LD rate in 2008. His xERA went from 2.69 to 4.08 even as his FIP decreased, from 3.56 to 3.35.

Scott Kazmir is another surprising name at the top of this list, as his 2.70 xERA is well below his 3.45 FIP, and his 7.6 PZR WAR far outstrips his 5.3 Fangraphs WAR. Kazmir suffered from incredibly bad defensive support, posting a .341 BABIP despite a line drive rate of 15.6%. Why, then, is his 3.45 FIP almost equal to his 3.48 ERA? Kazmir had a LOB% of 75.3%, which is high, but not insanely so. Therefore, Kazmir must have significantly lowered his ERA through "situational pitching:" in 2007, he allowed a .587 OPS with men on (compared to a .822 OPS with the bases empty) and had a 1.0% HR/PA rate with men on (as opposed to a 3.0% HR/PA rate with the bases empty). For his career, Kazmir has allowed a .786 OPS and a 2.9% HR/PA with the bases empty but a .646 OPS and a 1.7% HR/PA with men on. This highlights another of PZR's advantages over statistics like FIP and tRA: it corrects for pitchers who are more effective out of the stretch, as well as for pitchers who are exceptional at holding runners on (and thus post a higher LOB% than would be expected).


Once again, the AL Cy Young race appeared close, with Cliff Lee amassing 7.2 Fangraphs WAR and Roy Halladay accumulating 7.4. But PZR again contradicted sabermetric conventional wisdom, declaring Lee the winner by a significant margin, 9.8 WAR to 7.5. Lee's xERA of 2.19 significantly bested his ERA (2.54), although his BABIP of .305 was only slightly below average. Thus, PZR believes that Lee allowed easier-to-field batted balls, meaning that his .305 BABIP actually indicates poor defensive support. The only batted ball statistic that really stands out is Lee's above-average IFFB rate of 11.5%; the rest of his batted ball statistics are average, indicating that Lee allowed easier-to-field ground balls, fly balls, and line drives, rather than allowing fewer line drives or more fly balls.

After Lee, the pitcher with the biggest difference between his PZR WAR and Fangraphs' WAR is Jon Lester, who had 7.3 PZR WAR but just 5.1 Fangraphs WAR. Since Lester had a .299 BABIP, almost exactly league average, we would expect him to have an xERA comparable to his ERA. Instead, Lester's xERA is significantly lower than his ERA, 2.95 to 3.21. Lester had a slightly above average ground ball rate and a slightly elevated line drive rate (20.7%, compared to league average of 19%); the only batted-ball statistic that indicates weak contact is his above average 12.9% IFFB rate. Thus, we must conclude that Lester, like Lee, allowed easier-to-field batted balls, even after controlling for batted ball type.

Chad Billingsley is another somewhat unexpected name on this list, as his 2.77 xERA outstrips his 3.14 ERA and 3.35 FIP. Billingsley's BABIP, however, was far above average at .323, and his 78.0% LOB rate pushed his FIP above his ERA. Thus PZR awards Billingsley for having an above average strand rate.

2001-2008 Totals

Seven of the ten pitchers present on the 2001-2006 leaderboard are also on this one; the three who didn't make it are Barry Zito, Roger Clemens, and Jason Schmidt, who were replaced by C.C. Sabathia, Mark Buehrle, and Brandon Webb. Roy Oswalt assumes the top spot in this ranking, passing Curt Schilling, who retired after the 2007 season, and Randy Johnson, who was injured for much of the 2007 season. From 2001 to 2008, Oswalt has been consistently very good, if not highly acclaimed; his durability and consistency allow him to overtake Roy Halladay, who was injured in 2004 and 2005, and Johan Santana, who didn't pitch a full season until 2004. Oswalt has also suffered from playing in Houston, which has resulted in a general lack of attention for the right-hander.


MGL also ran a few regressions on the PZR data that he sent to me:

When I run a correlation for years 01 on 02, 03 on 04, 05 on 06, and 07 on 08, for pitchers who have changed teams (different park and defense) from one year to the next, I get:

> 299 BIP for each year
r=.049 (-.11 to .21, 95% confidence interval)

> 399 BIP for each year
r=.104 (-.10 to .31, 95% confidence interval)

> 499 BIP for each year
r=.209 (-.05 to .45, 95% confidence interval)

So, pitchers appear to have some fairly significant control over the difference between what we expect to be the result of their BIP and what they are, at least according to the UZR methodology, which suggests that using a "PZR" to compute some defense independent kind of stat is not completely fair. It does appear that using a pitcher’s actual BIP results regressed toward their UZR, with the amount of regression based on the number of BIP, may be appropriate.

On the other hand, given the wide confidence intervals above, the sample sizes are nowhere near large enough to conclude with any certainty whether they do indeed have such control, and if they do, how much.

I have not yet incorporated any regression into my WAR calculations; the method I used here is exactly the same as the one I presented for the 2001-2006 win values. I may try to include regression later on; right now, however, I'm not entirely sure how to do so.