Advanced fielding metrics are often the topic of heated debate in all baseball circles. Most of the discourse surrounding these statistics has to do with the accuracy of the subjective data used. The phrase, "garbage in, garbage out" has been uttered on more than a few occasions. Though, for a second, let us assume the data's precision and discuss what these metrics tell us.
Each statistic conveys information. As fans and analysts it is our duty to choose which metric can be utilized to fit our needs.
For instance, let us look at Runs Batted In. RBI is a descriptive statistic. It clearly tells the reader the amount of times a runner has crossed home during and following a hitter's plate appearances. As you know, RBI has few uses because it lacks context. Failing to account and neutralize for the amount of RBI opportunities one has is just one problem with the old school metric.
RBI clearly is not a predictive statistic. A predictive statistic is one that correlates highly from year to year.[1]. What a player accomplishes in year X, we should expect him to accomplish in year X+1. As an example, look at xFIP. xFIP does not tell us what happened in the current year - how a pitcher was able to prevent runs - but rather tells us how one should have been able to prevent runs by looking at one's strikeouts, walks, while assuming a league average homerun rate. The assumed home run rate isn't even something that actually happened.
Given nature of predictive statistics, many look to these constructs to asses a player's true talent level.
A problem with subjective data fielding metrics is that fans and analysts alike look at them and think true talent level. Viewing fielding metrics through a true talent level lens, it is understandable one has trouble reconciling Jacoby Ellsbury's wildly fluctuating defensive statistical performance. [2].
Each opportunity for a player's On Base Percentage is marked by a plate appearance. It's safe to say that outside of the level of competition and park factors, all plate appearances are created equal. Get on base, or make an out. Simple. Clear. [3].
Often it is overlooked that the opportunities that are used for advanced fielding are not as clear. Again, forget the accuracy of the data employed and consider that two right fielders can be hit 100 fly balls, yet both have had a vast different amount of opportunities. Why? Because each fly ball - even if categorized to perfection - is different. In the given scenario, one right fielder's defensive statistical performance can be stifled by a lack of difficult opportunities while another's can be bolstered by a high volume of difficult plays.
Because the pool of fielders can have a wide array of opportunities causing fluctuating outcomes year-to-year, subjective fielding metrics are descriptive statistics, like RBI, and do not give the reader an idea of a fielder's true talent level. Conceptually, these statistics are not as useless as RBI. They tell the reader what happened using complex linear weights.[4]. But, don't fret when your perception is contradicted by these metrics. The disconnect is created, in part, by the variance in the difficultly of opportunities each fielder faces.
Colin Wyers championed the fight against the use of subjective data in fielding metrics both at The Hardball Times and Baseball Prospectus. If you consider his arguments, and you should, it only raises additional questions about the opportunities each fielder faces.
JD Sussman is full time law student and co-founder of Bullpen Banter. He can be reached at JDSussman@bullpenbanter.com or via twitter.
_____________________________________________________________________________________________
[1] If you haven't yet, check out Bill Petti's amazing Hitting Statistic Correlation Table.
[2] Obviously, Ellsbury is just one example but look at his UZR from year-to-year. Full seasons only:
|
2008 |
21.2 |
|
2009 |
-9.7 |
|
2011 |
15.8 |
[3] OBP fits the description of a descriptive statistic. However, according to Bill's table linked above, OBP correlates at .60 from year to year. So, as a secondary feature, OBP has predictive properties. That is why we love it so much!
[4] I hate putting a comment in the concluding paragraph.... But, don't forget that we're assuming the data is perfect. Clearly, since the data isn't perfect and is quite flawed, it is arguable that these statistics are useless.