clock menu more-arrow no yes mobile

Filed under:

The Devil is in the Details of WAR

ANAHEIM CA - JULY 13:  The Rays Ultimate WARrior scores the lone AL run in the All-Star game Tuesday (Photo by Stephen Dunn/Getty Images)
ANAHEIM CA - JULY 13: The Rays Ultimate WARrior scores the lone AL run in the All-Star game Tuesday (Photo by Stephen Dunn/Getty Images)
Getty Images

Erik did a great job of giving an overview on WAR in his primer this week, and in the comments there was even more great content (if you haven't read it, check out the indomitable Sky Kalkman take on all questions) pertaining to this great statistic. There's one thing that I think needs to be stressed when discussing WAR that Sky hit on in the comments: the flexibility of use it provides. If you don't agree with one aspect that Fangraphs or B-Ref uses you can toss it out and come up with your own figure. This is really great when making projections. So lets go through and see what exactly lies in the details in WAR.

Erik used Crawford's spectacular season to show how WAR breaks down to run values for each phase of the position player's game (well batting and defense are aspects of his game, but the other two-position and replacement- are more related to the nature of the stat). Let's take a look at what CC's season looks like:

Batting: +22.9 runs

Fielding: +15.2 runs

Replacement: +12.4 runs

Positional: -3.7 runs

So where do these figures come from? Well Fangraphs uses wOBA to determine the player's offensive contribution, UZR for defense, and predetermined values for replacement and position based on playing time. Let's start with those two first since they're seemingly the least likely to be fiddled with.

With any player, the replacement level adjustment remains unchanged. The theoretical framework for the replacement level adjustment is built on the finding that replacement level players are about 20 runs worse than average over a full season (or 600 PAs). As such, the replacement runs are determined by essentially using that rate over a given players actual playing time: i.e. CC has 371 PAs, so his replacement runs equal 374/600*20 or 12.4 runs. This rate holds true for all players, so there's not really any reason to ever play with this number (unless of couse you're attempting to compare two actual players with equal playing time in which case it's superflous like in the discussion about how Desmond Jennings might affect BJ's actual replacement level).

The other aspect of WAR that's more theoritcal than based on performance is the positional adjustment. The positional adjustment addresses the inequality in defensive talent across the defensive spectrum. We all know it's easier to play 1B then SS or catcher, and the positional adjustment accounts for that by attempting to put players defensive contribution in context to their position. The positional adjustments follow (each is per 162 defensive games):

Catcher: +12.5 runs
First Base: -12.5 runs
Second Base: +2.5 runs
Third Base: +2.5 runs
Shortstop: +7.5 runs
Left Field: -7.5 runs
Center Field: +2.5 runs
Right Field: -7.5 runs
Designated Hitter: -17.5 runs

As you can see, Crawford's amazing defense is somewhat mitigated by the fact that he plays a pretty easy position. When part of the "average defender" is made up of Carlos Lee, Vlad, etc. then being 20+ runs better than average isn't quite as valuable. Additionally, the theoretical spectrum of defense purports that as you move from one defensive position to another one your defensive skill should increase/decrease based on the difficulty (read positional adjustment). As we've discussed plenty of times, that means that if CC were forced were to take over for BJ in CF we'd expect his defense to take a hit equal to the difference in adjustment (10 runs). He'd still likely be a very good, above-average defender, but there wouldn't be the huge discrepancy you see in LF. Positional adjustments are generally pretty straight-forward when engineering your own WAR calculations, but sometimes they can really add some intrigue. Like you can theoretically attack where Super Utility Stud Zorilla (SUSZ pronounced SUZE!!!!) might be most valuable to the team. Assuming BenZo's offensive prowess isn't effected by his position we can  take a guess at what a reasonable projection might be for Zobrist at all the positions he's played (WARNING: SSS AHEAD!).

We'll use 20 replacement runs, 15 offensive runs (similar to career average per 162 games), and the following positional+defensive runs:

SS: -9.4 UZR/150, +7.5 pos./162; -10+7.5=-2.5 pos/def=32.5 runs above replacement or 3.25 WAR

2B: 26.6 UZR, +2.5 pos.; 28.5+2.5=31 pos/def=66 RAR=6.6 WAR (WOW)

CF: -4.6 UZR, 2.5 pos.; -5+2.5=-2.5 pos/def=32.5 RAR= 3.25 WAR

LF: -16.6 UZR, -7.5 pos.; -18-7.5=-25.5 pos/def= 9.5 RAR= .95 WAR (ugh... SSS)

RF: 26.9 UZR, -7.5 pos.; 28..5-7.5=21.5 pos/def=46.5 RAR= 4.65 WAR

So seemingly, Zobrist thrives on the right side of the field. We can all see that there are some serious SSS issues throwing this off, but if we even adjust each UZR down toward average it still seems the most efficient use of Zobrist would be at 2B. Though with the current roster construction that may not be the case.

From here, we can see why we might want to fiddle with the defensive numbers used in FG/BR's WAR calculations. It doesn't make intuitive sense that Zobrist would be other-worldy at 2B while being terrible at SS. Why would he be great in RF but terrible in LF? So one thing I like to do is take a look at +/-, UZR, and then past years to come up with a rough estimate of what the player's "true defensive value" is. This is one of the great things about WAR. If I think Zorilla is actually a guy that's +10 to +15 at 2B over the course of a season than I plug that in. It gives a better illustration of what a real projection of talent can be.

Much in the same vein, the offensive input (wOBA) can be tweaked based on intuitive issues that aren't captured on a spreadsheet. If a player's playing through injury and you want to attack what you think they'll do when healthy, you can plug a different wOBA. The formula for batting runs above average is (wOBA-league average wOBA)/1.15*plate appearnces. The 1.15 represents the scale used to make it look like OBP. Now this will not be park or league adjusted, but it's a great way to get in to a player's offensive production and toy with it. In the past, I've BABIP adjusted wOBA to try to attack luck in player's WAR.

With pitchers, you can do the same thing. Want to add leverage? Adjust for leverage. Want to use tERA rather than FIP or xFIP? Knock yourself out. Want to take an average of the three? It'll work. This is the second greatest thing about WAR. Not only is it a stat that catches all and thus makes it easier to compare players across any spectrum, but it's easy to tweak to what you believe. You can easily make the WAR stat your's. Doing this will also help you understand the big picture of this great stat.