The Sabermetric Revolution: Assessing the Growth of Analytics in Baseball (28 page)

23
. Ben Baumer, “Why On-Base Percentage Is a Better Indicator of Future Performance Than Batting Average: An Algebraic Proof,”
Journal of Quantitative Analysis in Sports
4, no. 2 (2008).

24
. Indeed, this was the approach taken by Tom Tango with his Marcel projection system. Marcel is notable for being very simple, relatively easy to reproduce, and fairly accurate. It is often used as a benchmark to compare more sophisticated projection systems.

25
. Tom Tango, “Marcel 2012,”
http://www.tangotiger.net/marcel/
.

26
. In addition to the aforementioned work by Null, the appendix to
The Book
contains a discussion of what is essentially a Bayesian model for normally distributed variables with a normal prior distribution (Tango, Lichtman, and Dolphin,
The Book
). A more complete discussion is contained in B. B. McShane, A. Braunstein, J. Piette, and S. T. Jensen, “A Hierarchical Bayesian Variable Selection Approach to Major League Baseball Hitting Metrics,”
Journal of Quantitative Analysis in Sports
7, no. 4 (2011).

27
. Leon Neyfakh, “Nate Silver Signs with Penguin in Two Book Deal Worth About $700,000,”
New York Observer
, November 14, 2008.
http://observer.com
. Also confirmed via personal communication.

1
. Vörös McCracken, “How Much Control Do Hurlers Have?”
Baseball Prospectus
, January 23, 2001.
http://www.baseballprospectus.com/article.php?articleid=878l
.

2
. The Cy Young Award is given to each league’s best pitcher each year based on a Borda count vote by the Baseball Writers’ Association of America.

3
. ERA is the average number of runs a pitcher allows per nine innings that were not occasioned by an error.

4
. Tyler Kepner, “Use of Statistics Helps Greinke to AL Cy Young,”
New York Times
, November 17, 2009,
http://www.nytimes.com/2009/11/18/sports/baseball/18pitcher.html?_r=3
; Eddie Matz, “Saviormetrics,”
ESPN The Magazine
, March 5, 2012,
http://espn.go.com/mlb/story/_/id/7602264/oakland-brandon-mccarthy-writing-moneyball-next-chapter-reinventing-analytics-espn-magazine.

5
. Lewis,
Moneyball
, p. 241.

6
. That Martinez posted a .326 BABIP in 1999 (the fourteenth highest rate among the 136 pitchers with at least 500 batters faced), arguably his best season, and arguably the best season a pitcher has ever had, is truly terrifying.

7
. Tom Tippett, “Can Pitchers Prevent Hits on Balls in Play,”
Diamond Mind Baseball
, July 21, 2003.
http://207.56.97.150/articles/ipavg2.htm
. Tippett began consulting for the Red Sox around the same time, and was hired to be their director of baseball information services in 2009.

8
. This argument is severely weakened by the presence of selection bias.

9
. See, for example, Matt Swartz, “Ahead in the Count: Predicting BABIP,”
Baseball Prospectus
, March 23, 2010.
http://www.baseballprospectus.com/article.php?articleid=10333
; and, Dan Rosencheck, “Hitting ‘Em Where They are,”
Fangraphs.com
, March 13, 2013. Rosencheck found that he could explain 15 percent of the variance in a pitcher’s next year BABIP with the pitcher’s previous year’s popout rate and Z-Contact score.

10
. It may be tempting to think that these results are merely the artifact of one season’s worth of statistics being too small of a sample. One might also define reliability
in terms of the value of a statistic in one season and the cumulative average over the previous three years. While this does increase the reliability of BABIP a bit (e.g., the reliability for BABIP for pitchers improves from 0.17 to 0.21), it does not appreciably alter the conclusions.

11
. For SIERA, see
http://www.fangraphs.com/library/index.php/pitching/siera/
.

12
. It is not clear that SIERA is anything more than a multiple regression model with quadratic and interaction terms.

13
. Lewis,
Moneyball
, p. 98.

14
. See, for example, the following article from the 2009 baseball preview issue: Albert Chen, “Baseball’s Next Top Models,”
Sports Illustrated
, April 6, 2009, pp. 62–67.

15
. Tim Marchman, “The Problems with Defensive Stats,” SI.com, August 11, 2010.
http://sportsillustrated.cnn.com/2010/writers/tim_marchman/08/11/marchman.defen sive.stats/index.html
.

16
. Schwarz,
The Numbers Game
, p. 9.

17
. James lamented that he “feels stupid” for not coming up with DIPS thirty years prior to McCracken’s observations (Lewis,
Moneyball
, p. 240). One wonders if this comes from the fact that he invented DER to evaluate fielders, but did not think to apply the metric to pitchers.

18
. The only difference is that Reached On Errors (ROE) count as outs for the purposes of BABIP, but not for DER.

19
. Lewis,
Moneyball
, p. 76.

20
. For a more complete survey, see Ben Baumer, Andrew Galdi, and Rob Sebastian, “A Survey of Methods for the Evaluation of Defensive Ability in Major League Baseball,” JSM Proceedings, Statistics in Sports Section, 2009.

21
. In fact, SAFE does not use bins at all. Rather, it fits a continuous, piece-wise smooth surface to the data.

22
. Again, the methodology for obtaining that estimate varies. One can always take the simple arithmetic mean of the observations, but UZR, for example, makes corrections to account for differences in ballpark, among other factors.

23
. A ball is fielded successfully if it is converted into at least one out. Thus, whether an error was made on the play is irrelevant—it is only important whether an out was recorded. This removes the subjectivity of the error designation.

24
. David Appelman, “UZR Updates!” Fangraphs.com, April 21, 2010.
http://www.fangraphs.com/blogs/index.php/uzr-updates/
.

25
. What consternation this must have caused the media, which had been touting Bay’s poor UZR numbers as evidence that the Mets had been foolish in signing him to a four-year, $64 million contract in the off season. See, for example, Rob Neyer, “Upon Further Review, Bay’s D not so Bad?”
ESPN.com
, April 30, 2010.
http://espn.go.com/blog/sweetspot/post/_/id/3404/upon-further-review-bays-d-not-so-bad
.

26
. Dave Cameron, “Win Values Explained: Part Two,”
Fangraphs.com
, December 29, 2008.
http://www.fangraphs.com/blogs/index.php/win-values-explained-part-two/
.

27
. According to data downloaded from Fangraphs, 91.9 percent of the 18,847 player-position-seasons with data from 2002 to 2011 had a UZR between -5 and 5 runs.

28
. Lewis,
Moneyball
, p. 136.

29
. Hunter Atkins, “Rays’ Joe Maddon: The King of Shifts,”
New York Times
, May 7, 2012.

30
. Lewis,
Moneyball
, p. 135.

31
. For an interesting critique of UZR, though perhaps a bit intemperate at points, see Hirsch and Hirsch,
The Beauty of Short Hops
, pp. 80–86.

32
. Lewis,
Moneyball
, p. 136.

33
. In statistical terms, fielders are evaluated by the sum of the residuals between the observations and model.

34
. SAFE employs a hierarchical Bayesian model. Recall the discussion of regression to the mean in the previous section on predictive analytics.

35
. Hirsch and Hirsch,
The Beauty of Short Hops
, pp. 82–83.

36
. Baumer, Galdi, and Sebastian, “A Survey of Methods,” p. 13.

37
. One potential method for attacking this problem would be to include data from many more years, but to discount it as it is further in the past.

38
. Alan Schwarz, “Digital Eyes Will Chart Baseball’s Unseen Skills,”
New York Times
, July 9, 2009.
http://www.nytimes.com/2009/07/10/sports/baseball/10cameras.html
.

39
. This is a natural extension of UZR, SAFE, and other models. Greg Rybarczyk proposed a similar metric called True Defensive Range (TDR) at the 2010 PITCHf/x Summit. See Rob Neyer, “FIELDf/x Is Going to Change Everything,” ESPN.com, August 30, 2010.
http://espn.go.com/blog/sweetspot/post/_/id/5041/fieldfx-is-going-to-change-everything
.

40
. Hirsch and Hirsch,
The Beauty of Short Hops
, p. 85.

41
. Lewis,
Moneyball
, p. 58.

42
. Lewis,
Moneyball
, p. 137.

43
. The Rays’ pitchers’ strikeout rate was 18.6 percent in both seasons, while their walk rate declined only slightly, from 8.9 percent to 8.6 percent, and their home run rate declined from 3.1 percent to 2.7 percent.

44
. The Rays’ DER is .722 over that time period, with the Dodgers having the next best mark, at .714. The Rays’ DER is nearly 3 standard deviations above the mean.

45
. Note that this argument is independent of the question of the accuracy of defensive measurements. That is, there is a very real danger of overestimating the importance of defense because we now have the illusion of being able to measure it accurately (e.g. using UZR). However, as we have argued, there is scant evidence that the measurements made by UZR are very precise, and thus one might be tempted to argue that the importance of defense should be discounted, because the instruments for measuring it are not
precise. But the argument for the importance of defense we make above is about
team
defense, not individual fielding ability. It may very well be the case that although we are not able to measure the defensive prowess of individual fielders accurately, the effect of better team defense still plays a large role in determining the outcomes of games. Thus, it should remain an important consideration for general managers.

46
. This estimate is based on Bill James’s formula for expected winning percentage, which we discussed previously. The idea is that the value of the partial derivative of the function f(rs, ra) = 162/(1 + (ra/rs)
²
) with respect to runs scored (runs allowed) is about 0.1 (−0.1) around the point (750, 750). Thus, an additional 10 runs scored (allowed) would add about 1 (−1) win to an average team’s season total. While this is a convenient rule of thumb, a more careful analysis reveals that the value of these derivatives change considerably based on the run-scoring environment. For example, an additional 10 runs in the environment (600, 900) would translate to only 0.725 wins.

47
. Fangraphs David Wright page,
http://www.fangraphs.com/statss.aspx?playerid=3787&position=3B#value
.

48
. The player is 4A in the sense that he is better than a AAA player but worse than a major league player.

49
. It is, however, important to remember that this is an abstraction. While such players may be plentiful, they may have significantly differing skill levels. In this sense as well, then, WAR is not a precise measurement.

50
. It is useful to note that the replacement level player is significantly worse than the league-average major league player. In fact, replacement level production has no standard definition, but is usually in the range of 75–80 percent of the league average. It is also important to understand that this represents a much greater level of production than a randomly selected farmhand from A ball would likely produce, let alone your average person off the street. Some players also produce a negative value of WAR, because they perform worse than the threshold for replacement level.

51
. For example, the Keeping Score column in the
New York Times
contains frequent references to WAR.

52
. For a catalog of the differences between the three metrics, see
http://www.base ball-reference.com/about/war_explained_comparison.shtml
.

53
. Dave Cameron, “Win Values Explained: Part Seven,” Fangraphs.com, January 5, 2009.
http://www.fangraphs.com/blogs/index.php/win-values-explained-part-seven/
.

54
. Among the 95,547 player-team-seasons from 1871 to 2011, nearly 82 percent had a bWAR between −1 and +1.

55
. Retrosheet data is “free as in freedom.” MLBAM data is only “free as in beer.”

56
. This would include any R, SQL, Python, or bash scripts necessary to perform the computations. It would not include any Excel procedures. This may seem inflexible, but this level of transparency is necessary for a model as complex as WAR. Also, as we have
argued previously, the sabermetric community has a long history of parallels with the open-source community that should not be discarded.

57
. J. C. Bradbury,
Hot Stove Economics
(New York: Copernicus Books, 2011), pp. 92–96.

58
. Of course, in practice it is considerably more complicated for a GM to sign a free agent. Such salaries are invariably guaranteed over multiyear contracts, so the issue becomes not only evaluating what the player was worth last year, but what he will be worth the coming year, and the year after next, and the year after, and so on.

59
. Tango, Lichtman, and Dolphin,
The Book
.

60
. Bill James, “A History of Platooning,” in
The Complete Armchair Book of Baseball
, edited by John Thorn (New York: Galahad, 2004).