A FanGraphs WAR Fielding Update
4/23 Update: It has come to our attention that in the switch to Statcast Fielding Runs Prevented (RAA), outfield positional adjustments were being improperly applied. This has now been corrected by zeroing out RAA for each individual position, on a seasonal basis, specifically for use in WAR. The tables below have been updated to include this correction.
Since we launched FanGraphs WAR in 2008, we have used various components of Ultimate Zone Rating (UZR) such as Range, Outfield Arm, and Double Play Conversion to evaluate position player fielding. Today, we’re changing one of those components. Retroactive to the 2016 season, we have swapped out the Range component of UZR for the Statcast metric Fielding Runs Prevented, which is Outs Above Average (OAA) converted to runs above average. The UZR Outfield Arm and Double Play Conversion components of WAR remain unchanged.
We believe the additional data points available in Statcast, such as a fielder’s starting location, help to improve the measurement of a player’s range, especially in situations where players are shifted.
The vast majority of players’ new WAR calculations fall within a +/-1 WAR range of their previous WAR figures. Since 2016, there are 49 individual seasons that have changed by more than one win, and 14 players whose WAR has changed by more than three wins for the entire six year span starting in 2016.
On the leaderboards, we have a new custom field called “Legacy WAR (L-WAR),” which will show the results of the previous calculation for those who are interested in making comparisons.
Below is the list of players who experienced the biggest single season changes, followed by the list of the players with the largest changes since 2016.
| Name | Season | Legacy WAR | WAR | Change |
|---|---|---|---|---|
| Nick Ahmed | 2018 | 1.7 | 3.7 | 2 |
| Carlos Correa | 2018 | 1.6 | 3.5 | 1.9 |
| Ender Inciarte | 2016 | 3.1 | 4.7 | 1.6 |
| Adam Engel | 2017 | -0.8 | 0.8 | 1.6 |
| Nicky Lopez | 2021 | 4.4 | 6 | 1.6 |
| Javier Baez | 2019 | 4.3 | 5.8 | 1.5 |
| Francisco Lindor | 2021 | 2.7 | 4.2 | 1.5 |
| Jake Lamb | 2016 | 2.4 | 3.7 | 1.3 |
| Jordy Mercer | 2016 | 1.4 | 2.7 | 1.3 |
| Nick Ahmed | 2021 | 0 | 1.3 | 1.3 |
| Andrelton Simmons | 2021 | -0.5 | 0.8 | 1.3 |
| DJ LeMahieu | 2017 | 2 | 3.3 | 1.3 |
| Mallex Smith | 2019 | 0.1 | 1.4 | 1.3 |
| Fernando Tatis Jr. | 2021 | 6.1 | 7.3 | 1.2 |
| Jean Segura | 2016 | 5 | 6.2 | 1.2 |
| Travis Shaw | 2017 | 3.5 | 4.7 | 1.2 |
| Francisco Lindor | 2019 | 4.7 | 5.9 | 1.2 |
| Jonathan Schoop | 2016 | 2.3 | 3.5 | 1.2 |
| Max Kepler | 2016 | 1.3 | 2.5 | 1.2 |
| Josh Reddick | 2018 | 1.1 | 2.3 | 1.2 |
| Brian Dozier | 2017 | 5.1 | 6.2 | 1.1 |
| Tim Anderson | 2019 | 3.4 | 4.5 | 1.1 |
| Keon Broxton | 2017 | 0.7 | 1.8 | 1.1 |
| Manuel Margot | 2021 | 1.4 | 2.5 | 1.1 |
| Leonys Martin | 2016 | 2.4 | 3.5 | 1.1 |
| Addison Russell | 2018 | 1.5 | 2.6 | 1.1 |
| Nick Ahmed | 2016 | -0.5 | 0.6 | 1.1 |
| Starlin Castro | 2016 | 1.3 | 2.4 | 1.1 |
| Mike Trout | 2016 | 9.7 | 8.6 | -1.1 |
| Charlie Blackmon | 2017 | 6.6 | 5.5 | -1.1 |
| Christian Yelich | 2017 | 4.6 | 3.5 | -1.1 |
| Paul DeJong | 2017 | 3.1 | 2 | -1.1 |
| Neil Walker | 2016 | 3.6 | 2.5 | -1.1 |
| Yasmany Tomas | 2016 | 0.5 | -0.6 | -1.1 |
| Brandon Phillips | 2017 | 1.5 | 0.4 | -1.1 |
| Chase Utley | 2016 | 2.2 | 1.1 | -1.1 |
| Jon Jay | 2018 | 0.8 | -0.3 | -1.1 |
| Harold Castro | 2021 | 0.5 | -0.7 | -1.2 |
| Matt Kemp | 2016 | 0.8 | -0.5 | -1.3 |
| Brandon Phillips | 2016 | 0.9 | -0.4 | -1.3 |
| Didi Gregorius | 2019 | 0.9 | -0.4 | -1.3 |
| Dustin Pedroia | 2016 | 4.9 | 3.6 | -1.3 |
| Didi Gregorius | 2018 | 4.7 | 3.4 | -1.3 |
| Carlos Correa | 2016 | 5.2 | 3.9 | -1.3 |
| Logan Forsythe | 2017 | 1.9 | 0.5 | -1.4 |
| Zack Cozart | 2016 | 1.9 | 0.4 | -1.5 |
| Didi Gregorius | 2017 | 4.1 | 2.5 | -1.6 |
| Paul DeJong | 2018 | 3.3 | 1.7 | -1.6 |
| Corey Seager | 2016 | 6.9 | 5.1 | -1.8 |
| Name | Legacy WAR | WAR | Change |
|---|---|---|---|
| Nick Ahmed | 4.8 | 11.3 | 6.5 |
| Javier Baez | 17.8 | 22 | 4.2 |
| Jonathan Schoop | 10.7 | 14.7 | 4 |
| Rougned Odor | 4.3 | 7.7 | 3.4 |
| Adam Engel | 2.1 | 5.4 | 3.3 |
| Starlin Castro | 7.3 | 10.5 | 3.2 |
| Francisco Lindor | 28.1 | 31.3 | 3.2 |
| Tim Anderson | 14 | 17.1 | 3.1 |
| Ender Inciarte | 9.2 | 12.2 | 3 |
| Xander Bogaerts | 26.8 | 23.5 | -3.3 |
| Marcus Semien | 23.1 | 19.6 | -3.5 |
| Paul DeJong | 12.8 | 9.1 | -3.7 |
| Charlie Blackmon | 18.6 | 14.6 | -4 |
| Didi Gregorius | 14 | 7.7 | -6.3 |
David Appelman is the creator of FanGraphs.
Some nice spring cleaning! Poor Didi….
The WAR giveth and the WAR taketh away.
Are the position adjustments changing? Outfield OAA is, I believe, relative to the average outfielder rather than the average at a particular position.
We are actively exploring positional adjustment changes.
Feels like the shortstop positional adjustment needs to be lowered. Yes, it’s a difficult position to field, but there is also ample depth for many years now. So is it truly that difficult to find a good SS? Clubs would laugh at the prospect of trading a Tyler O’Neill or Arozarena for a Nicky Lopez (trying to pick guys with similar service time), but 2021 WAR values say it’s a serious question.
Also, I would think that the increase in shifts has made range at least slightly less important.
biggest knock against positional adjustments to me, is this:
it treats them as though each team has the capacity to go and always find an above average 1B hitter, an above average RF and LF. No team is comprised of a perfect array of all these players. Mike Trout has put up silly WAR totals due to being “more valuable as a CF”, but its not like his team was putting even adequate RF and LF next to him. so the argument that teams can just easily scoop up those players is wrong to me. if anything, the depth at SS right now suggests the exact opposite
It’s simple: there are fewer players who can play CF than the corners, therefore if you can play CF you are more valuable than if you play a corner.
While yeah the Angels haven’t done a great job at finding corner outfielders, imagine if Trout played LF and the Angels had to find a center fielder. That’s an even harder task.
the difference between mike trout the left fielder and mike trout the center fielder is more marginal than positional adjustments make it out to be, especially in TTO era. i think i’d be open to accept positional adjustments, i just think they should be toned down. thats my point. if anything, i think fg should offer a positional adjustment neutral WAR to accompany their standard war.
Similarly for SSs. Yeah the depth at that position is incredible! But still, the number of players that can credibly play SS is fewer than the number of players that can credibly play anywhere else on the infield. Therefore if you play SS you have more value.
Imagine if the Braves lost Swanson and Albies. Who would be harder to replace? The number of players available to replace Albies is larger than the number of players available to replace Swanson. That’s why Swanson gets more of a positional adjustment boost to his WAR than Albies.
i actually think this a perfect demonstration of my point: ozzie is better, so i would much rather have to find a replacement for swanson – who is just a league average hitter. i understand all the points your making, and i get why tango and co incorporated adjustments. i think its sound logic but it doesn’t reflect actual value as well as it is commonly perceived to.
the trevor story 2b conversion is a perfect demonstration against this, to boot. a supposedly elite player at a “scarce” position was signed to play a less valuable position
Excellent! It’s nice to see an update that should give some more precision. Thanks!
Makes sense, very nice update!
Question–I thought that range for infielders was more comprehensive than for outfielders. Doesn’t infield OAA duplicate more parts of UZR than just range?
From the statcast glossary:
Is that right? If they drop the ball when they get there, that doesn’t get factored into OAA? If they don’t make the throw? There’s a % converted, is that conversion only based on running?
From the primer,
In order to get an out value on an individual play, it’s straightforward:
We add up all these partial pluses and minuses. And the total of these becomes your Outs Above Average. (This is the same process followed for Outfield Outs Above Average.)
To answer your question specifically, I’m not entirely sure who gets credited or debited for dropped balls.
Right, so my question is: Aren’t errors and throwing arm types of things being double counted for infielders? Or do we just not know?
I think the confusion might be with the ErrR UZR component? To clarify, the “range” component of UZR that we swapped included ErrR. So the ErrR or “error range” UZR component is not part of WAR where we are using OAA.
So to that end, I don’t believe there’s double counting going on?
Okay, so maybe I’m just confused about the inputs to UZR. So you swapped out range (and ErrR, which may also be included in range) for OAA, but kept DPR for infielders and ARM for outfielders?
Yep! Exactly that.
Alright, I think I got it now. Phew.
Whoa! This is fantastic.
Very cool! I seem to recall MGL himself saying that UZR would one day be surpassed by a more accurate measure. Cool change.
Great update!
This probably cost Nick Ahmed a lot of money. Oops!
I was coming here to say the same.
Would have done way better than 4/32M in 2019 if he was coming off of a two-year WAR of 7.2 instead of 4.0.
At least Baseball Reference had him at 8.3 rWAR in those two years? But yeah, this kinda sucks, especially from a site that has positioned itself as The Serious & Sensible One when it comes to fielding & framing metrics
Would you prefer FanGraphs not use higher quality, more comprehensive data just because it will change the perception of things that happened in the past? Almost every writeup or discussion about advanced defensive stats contains the caveat that defensive value is significantly harder to measure and has much wider error bars than offensive value, so the fact that it can shift by non-negligible amounts should not be that shocking to longtime readers. Incorporating Statcast data will help shrink those error bars, but the field is ever-evolving.
You’re assuming that every club uses FG or BR to come up with how they value players. I think Kevin Goldstein mentioned it somewhere previously, that lots of clubs do look at FG and BR for research purposes, and maybe for some data, but to think they don’t have their own internal valuations – especially for something as hard to quantify as defensive value – is pretty silly.
This is absolutely true…but with a pretty big caveat. A lot of teams, including the one I cover DO rely very heavily on STATCAST defensive data. And there are some possible issues with Statcast defensive data in my view. As I just mentioned above, look at SS for 2016-2022…..+544 OAA , Negative -435 OAA. That’s a problem. Also the expected hit % off of batted balls is often very inaccurate. I’m not sure, but if those hit% impact the way they calculate OAA that could be a real problem source.
Check out the gap right now between expected BA and expected Slug vs. actual. This happens every year and they they have to periodically keep updating it to bring it in line. See Tango’s response to my question on Twitter
https://twitter.com/tangotiger/status/1517152597152468992?s=20&t=gtAUIjx-sllHigr7DW3WOQ
Maybe. Teams have their own defensive metrics that are reportedly even more granular. That’s why they’re always hiring their own analysts. Not sure how much players are being paid based on fWAR and rWAR. Certainly they’re decent approximations.
Corey Seager was MVP runner-up in ’16, this 2 WAR drop for defense that year makes that seem a little shaky
Does it really matter for the runner-up? I’d only find it concerning if he had actually won the award.
He finished 3rd in the voting and is now 10th in the NL in fWAR. That isn’t too bad and is pretty much in line with a lot of historical voting. He is behind 2 catchers and Ender Inciarte who all have skillsets that are traditionally overlooked by award voters relative to their fWAR. The award going to the fWAR leader and the other two finalists both being in the top 10 fWAR finishers is way better than the historical average.
You’re right that writers often use war differences of <1 like they mean more than they do!
Does this change remove most or all credit/blame for positioning from the fielder?
Yes, Statcast considers the fielder’s initial positioning when calculating OAA
It would be nice to have a metric that credits/blames someone for positioning whether the fielder, manager or front office. It’s not clear from the metrics I see how much fielding range actually matters. For example is someone who’s 30 runs above the average range fielder actually 10 runs below that fielder because either he or someone else decided to always start him in the wrong place. Maybe hitting, base running, arm and errors are all that matter.
Stuff like this is why I like when I hear Craig Counsell talk about his “run prevention unit” or something similar. (note: he may have only said this once and it stuck with me.)
I get that it’s hard to assign everything to individual players. But there are some clearly defined buckets of data that we know that we don’t know how to assign. But we also know what their impact is.
Why not have something like “team WAR” that says the players positioning was worth ___ runs, which we know. But we don’t know who to give that credit to… so we’ll put it in the team bucket and sportswriters can argue over who should get the credit. Even something like this would, I think, give a good picture of what was happening. If Javy Baez is the only player on his team with a high team WAR at his position, that’s probably his wily-ness. And if every Cardinals player has about the same team WAR assigned to the position… good work front office nerds.
The other thing that occurred to me is the buckets of “Routine”, “Likely” etc plays. If a team has a high percentage of “Routine” plays it might be that their positioning is good and they’re better off getting guys who can hit, run and handle the “Routine” plays all 5e time.
I agree with this change, but what about a component for infield arms? While arm strength isn’t as much of a factor on the infield (save for catcher, and it’s still somewhat of a factor on te left side), throwing accuracy is even more important for infielders, otherwise fWAR is probably overrating the defense of infielders with poor throwing accuracy like Aledmys Diaz.
That’s accounted for in OAA. If you don’t complete the play, you don’t make the out therefore don’t get credit.
But isn’t fielding range considered separately from OAA?
Asked David this question above, and no one has given me a straight answer. My understanding is the same as yours: That if the play (for an infielder) isn’t completed, via error or bad throw or whatever, it doesn’t count in OAA. But then doesn’t that mean they’re double-counting arm and error rate? David responded by posting the Statcast glossary, which conspicuously does not include actually making the play.
Doesn’t infield OAA also factor in double plays? It seems like double-counting to include that and DPR for infielders
From the infield OAA Primer:
The Nick Ahmed fan club rejoices
While I will forever have serious doubts about all defensive metrics until the various systems begin to show some closely similar information the chart seems to vindicate Xander Bogaerts from the slings and arrows of all those who claim he can’t do the job at SS. Even with a slight drop in his WAR his numbers compare very favorably with those on the list.
He’s terrible by DRS and OAA and average by UZR.
Huh? I’m a BoSox fan, but if anything this adjustment just vindicates that Bogey is terrible at short. UZR’s error metric was the only thing that made him above average by that measure, and now the more legitimate hybrid hates him too
The change to Tatis’ fielding is huge. He’s now rated as a better defender than Turner or Baez, even with all of the 13 throwing errors in a partial season. Hopefully he uses his IL stint to sharpen up that part of his game and pushes into the upper tier of SS on defense
Since infield arm isn’t considered he might not be better. The errors are considered but maybe they’re a symptom of a bad arm generally
Statcast does consider infield arm. Throwing errors are accounted for in a player’s fraction of plays made (or not made)
Yes I said that but I suppose it also applies to other non outs due to weak arm
See, now this is just a tease. Trout’s 2018 WAR is up to 9.9. Please fiddle with the formula until you get it to 10. Thanks.
Can’t tell me that wasn’t the purpose. Waiting for part 2!
So there were actually two six-win shortstops in the AL last year: Carlos Correa (6.3) and Nicky Lopez (6.0).
Nicky Lopez as a six-win player has me questioning things
@MRDXol, yeah, I agree with your sentiment.
My one concern with OAA is that it is much less conservative than UZR. Not as aggressive as DRS, which regularly does things like turn 2019 Nick Ahmed and his 90 wRC+ into a 4.5 win player (this update puts Ahmed at 3.3, which seems much more plausible). But this update is definitely putting a couple of guys who you would never consider to be 4-6 win players in that territory now.
If you can buy into (a) Lopez being as good defensively last year as Lindor has been on average the last 4-5 years, (b) that level of defensive being 18 runs above average and (c) the positional adjustment for shortstop, the pieces add up for Lopez with a 6 WAR. He had a 106 wRC+ and was 8 runs above average as a baserunner, so the components add up for this particular season.
Mileage may vary on anyone’s best guess on his true talent level. Somewhere below six wins is the clear starting point
Huh? The 18 defensive runs above average (for 150 games/season) Lindor has averaged includes the positional adjustment already. I think there’s some double-counting of defensive value going on here.
Personally, I’ve always thought WAR overrated defense (and baserunning) in general.
This isn’t a fault of the defensive metrics themselves, but I feel the metrics’ data shouldn’t be weighed as heavily for Baserunning Runs and Defensive Runs as they are (or indirectly that Batting Runs should be weighed more heavily than they are compared to baserunning and defense) when calculating WAR.
This update also dropped Trout out of the exclusive five 9 WAR season club. So, boo.
For now.
Can UZR be renamed Penultimate Zone Rating (PZR)?
I wonder how many millions this update just cost Bogaerts
Teams certainly already have their own internal WAR equivalents or similar stats. I seriously doubt they are basing free agent contract offers solely on Fangraphs.
Was being slightly sarcastically. However, with that said, the pro argument for Bogaerts’ defense was that UZR doesn’t hate him. So now that UZR is being fazed out it does feel like his defensive reputation (at least to the public eyes) took a hit. Hard to look at his OOA or DRS and feel he can handle the position much longer
Really excited about this change. Curious to see whether Statcast can help down the road in evaluating baserunning, too.
A belated congratulations to Tatis Jr. for leading MLB in WAR in 2021!
Until the next tweak!
Trading Didi instead of Nick Ahmed is probably the only time being opposed to any kind of advanced metrics worked out for Dave Stewart
Is Fielding Runs Prevented the same as Defensive Runs Saved (DRS)?
Interesting to see Carlos Correa at the top and bottom of the list.
Imagine if MLB used FanGraphs WAR for the bonus pool. Would these changes have been prevented from being pushed through?
Is there any sense as to whether this makes fWAR align better, worse, or no differently to historical wins?
I suppose anyone could yank the data themselves, but am wondering whether the Fangraphs crew tested this before implementing it.
While WAR has many uses, and OAA seems at least more intuitive and sensible than either UZR or DRS in terms of how it works and the specific measurements it takes and converts into run values, I’m still wondering whether the elimination of “positioning as a skill” from the team WAR calculus has made fWAR more predictive of team wins, less predictive, or neither.
I went ahead and pulled the current WAR and L-WAR to check. Using 47.7 for replacement wins and assuming my spreadsheet work is correct, I’m getting an average absolute error of the new WAR of 4.38 wins and 4.25 wins for L-WAR. There might be a better way to judge this than average error though.
I used the 2016-2019 and 2021 seasons.
Thanks, I appreciate your work. I think in a lot of cases, RMSE is “preferred” over MAE because of the undesirability of “large” errors which RMSE penalizes for more heavily, but with a small sample in each year of just 30 team-seasons I’m not sure it matters too much.
I think this highlights an interesting tension in how WAR is a Swiss army knife of a stat — for comparing player “skill” it is probably more appropriate to use OAA, which is positioning-neutral. But if positioning is a player or team skill, removing it altogether by switching to OAA from something that wasn’t positioning-neutral makes it worse at explaining team wins. There are probably lots of ways to get around this, but none of them are currently implemented, so we have the option between L-WAR which includes some of it but tells you less useful things about player defensive skill, and current WAR which is somewhat less useful for explaining overall team context-neutral quality but has the edge on defensive information about the players involved.
Is Fielding supposed to equal RAA+DPR for infielders and RAA+ARM for outfielders? Looking at random players these seem to not add up correctly. All of the discrepancies are less than a run, but I am curious what I’m missing here.
I noticed that in this current season – Luis Arraez went from being around a 0.3 WAR (solid for 10 games) to a negative – 0.3 with this adjustment. His DEF rating is -5.5 worst in all of baseball so far. Is there a mistake in his numbers somewhere? I know he has been a liability at third especially so far for the Twins but not seemingly enough to drag him down 0.6 WAR over 10 games??
What about the issue that positions don’t zero out. For example from 2016 to 2022 Shortstop totals +544 runs but only negative -435 runs when adding up the teams.
That seems quite problematic to me.
LINK to Statcast page
Sorry, that’s the OAA number, but it’s the same issue with Runs instead of OAA…they don’t zero out.
We’ve made an update that zeros out RAA by season/position to mitigate this issue. It will not show up in the OAA / RAA numbers on the site, but it is now part of the Fielding metric in WAR.
Can we please get more data earlier than 2016 to see how much lower Ryan Dogmit’s war will go based on this new change. Please find the data back to 2009
If you’re embracing alt-WAR displays, can you guys please provide Total Zone data for post 2001 years? I’d love to play around with it to see how dWAR rankings look when using the one uniform defensive stat that goes back to before MLB’s pitch/play tracking era.