Friday, May 04, 2012
WAR updated on Baseball-Reference
I just saw the post, so I’ll read through it, and comment as appropriate. Sean reached out to me on a couple of things, so I’m keen on seeing what the final product looks like.
UPDATE 1: I just read through all the descriptions on Sean’s site. The explanations are tremendous, and I have no major objections. I’ll give it a second read-through, and will just make some minor, sporadic comments.
More updates below…
UPDATE 2:
I’m going to go through each of Sean’s pages of explanation, and pick out a couple of passages. You can 99% presume that if I don’t talk about something, then I either agree with it, or I consider it to be a reasonable choice.
It’s apparent that Sean took great care with the framework and calculations, and put serious thought in his choices. With Sean at Baseball-Reference and David at Fangraphs as stewards of advanced metrics, as well as having fantastically designed websites, sabermetrics has never been in a better shape to having its message put out in such an open and honest manner.
Anyway, I have several open tabs, and picking them out as I see it, here we go:
http://www.baseball-reference.com/about/war_explained_position.shtml
1. Just a note that I’m glad to see Sean holding to the WAR framework principle, that everything is about comparing to average, and then, you have the final step to compare to replacement level.
As an aside: What this does is allow people who don’t like the idea of replacement level, or, have a different idea of a replacement level, to simply substitute their own value in it. This is why the presentation that Rally had at Baseball Projection and Fangraphs has at the bottom of their player pages is so powerful. It sticks to this idea.
2. Sean switched from “Theoretical Team” BaseRuns as the basis to just linear weights. In the former, you would compute team BaseRuns with a player, and then compute it without the player, and the difference is his impact. While undoubtedly a great way to do it, and possibly even the best way, it’s not the easiest thing to program, adapt, or explain. Linear weights, which basically approximates this, is close enough, and is more flexible. You can really go either way here.
As an aside: Personally, I prefer ease and flexibility.
3. Taking out pitchers-as-hitters is always a good choice. It is more work, but the payoff here is worth it. It’s a basically minor point, but it shows great care.
As an aside: When I do my stuff, I not only always take out pitchers-as-hitters, but I also take out hitters-as-pitchers. I just pretend all those things didn’t happen. I don’t really care about many pitchers Cliff Lee has struck out, as it pertains to his value as a pitcher. I think of it as noise more than anything.
4. Sean shows the historical positional adjustments. I’m not sure if this is Forman’s or Rally’s stuff (extended to 2012). Since I myself change the values every few years by a couple of runs here or there, I’m going to always consider such positional adjustments as a work-in-progress. The numbers in the chart are certainly justifiable for the recent years, and for the most part in prior years. Like I said, we can talk about this for hours, and just take one step forward maybe.
5. Sean continues Rally’s choice of the .320 replacement level. The result is you get 875 wins every year (Sean shows 785, but that’s a typo). Fangraphs has 1100 wins, so .275 is their replacement level. You can reasonably justify anything between .250 and .350.
6. He also splits it at 59/41 between nonpitchers and pitchers, and justifies it similarly as I do, that that’s how teams pay. Now, I don’t know that we necessarily want to keep that split historically, and I know that Rally did not. But, not much of a big deal, whether you make it 40% to 45% every year, or keep it fixed at something in-between.
As an aside: I have it closer to 57/43, but, like I said, that’s no big deal. I’m not married to this number.
7. I’m not sure if those are Rally’s or Forman’s league adjustments. I personally use 5 runs as the gap, and since Forman is using 4, I have no issues. It’s cool to see it historically.
This took longer than I thought! I’ll chime in on the other pages a bit later…
UPDATE 3:
http://www.baseball-reference.com/about/war_explained_pitch.shtml
8. Sean is maintaining the runs-based version of WAR, in stark contrast to the FIP-based version that Fangraphs has. As I’ve said in the past, you can make reasonable justification for either of the two extremes, which is why I like it that they each have an extreme, so I can just take the midpoint and get on with my life.
The idea AGAINST runs-based version of WAR is as follows: A run allowed is presumed to be the responsibility of the pitcher, and adjustments are ONLY made based on the OVERALL presumed fielding performance of the team OVER THE SEASON. So, you can have Cliff Lee and Doc Halladay each performing in front of, overall, league-average fielders. But, if one guy has a .350 BABIP with men on base and the other guy has a .250 BABIP with men on base, and if we think that almost all of that is luck, then that luck will get absorbed SOLELY by the pitcher.
You can after all instead START with the idea that the runs are the responsibility of the team, and subtract out the things we know is responsible by the pitcher, and assign the rest to the fielders.
It’s a question of how the luck is split. Runs allowed at its core is equal to the performance of the pitcher + performance of the fielders + timing. By subtracting out only the performance of the fielders (and, it’s not even their performance ON THOSE PLAYS, but a general overall performance), the pitcher absorbs the timing of all events.
The idea FOR runs-based version of WAR: over time, the timing is going to cancel out, so that if there really is a timing-based impact by the pitcher, then we’ll see it reflected in his runs allowed. Given a long-enough career, if we can adjust out all the rest, the random variation will be limited, and we’ll be left with the pitcher’s actual performance. In addition, whereas FIP focuses on HR, BB, HB, SO, a runs-based measure also includes all the little things like SB, PK, etc.
The idea that it doesn’t matter: given a long-enough career, a pitcher’s RA9 and FIP are going to converge anyway. So, it’s much ado about nothing career-wise. Season-wise, both sides have a strong case. Basically, the shorter the time frame, the more FIP makes sense, and the longer the time frame the more RA9 makes sense.
As you can see, I spend more time explaining it than simply choosing the midpoint of the two. Do yourself a favor, and split the difference and move on.
9. Good adjustment about the “opposing lineup” (whether it includes the pitcher or not batting). The natural continuation of that is to look at the ACTUAL opponents! That’s alot of work of course. Just handling the pitcher though is 90% of the payoff for 1% of the work.
10. Decent team-adjustment for fielding. If you want to make it “better”, split it by infield/outfield, and base it on the GB/FB tendency of the pitcher. Weaver/Santana get a larger OF adjustment, while Felix/Doc get a larger IF adjustment. But, it’s not going to matter much.
11. The SP / RP adjustment: if I understand it correctly, a pure starter has his replacement level at X, and a pure reliever is at .865X. So, if X is 120% of league average, then a reliever’s level is 104% of league average. That’s fairly decent. I have it a bit wider. In my case, it would be X and .833X. So, if X is 125, then a reliever is at 104. You can certainly justify it in a few ways, but it’s in the ballpark.
12. He uses the chaining explanation for using Leverage Index, so, that’s good. Interestingly, he ALSO uses it for starting pitchers, which would be a bit harder to justify. But, since SP are all so close to each other and to 1.00, it doesn’t matter much. But, you lose the chaining explanation, and instead becomes a regression explanation.
13. Sean does a good job at explaining the difference between BR and Fangraphs, and why he went his way. Like I said, you can make a convincing case either way.
UDPATE 4:
http://www.baseball-reference.com/about/war_explained_wraa.shtml
14. Alright, glad to see that Sean got that little thing I do with plate appearances (ignoring it in the rate calculation, but including it in the overall calculation). It’s one of those little things that really doesn’t do much, but it shows detail to attention.
15. He went with the additive run adjustment, rather than the multiplicative one. I think you can make a good case either way, but absent any strong reason, I’d just as well go with the easy additive one.
16. Good stuff on the SB/CS thing. He uses the same calculation I use to figure out opportunities, and then scales the SB success rate relative to his SB frequency rate.
17. I skipped over the strikeout calculation (for now).
UPDATE 5:
http://www.baseball-reference.com/about/war_explained.shtml
18. I agree about figuring out the runs per win converter based on incuding the player in the run envrionment, and in the proportion to which he played in a game. I’m pretty sure Fangraphs does the same.
19. Wonderful chart comparing Fangraphs and BR.com
UPDATE 6:
I must have closed one of the tabs, because he explained that the defense portion of WAR in the leaderboards is the fielding part plus the positional part. So, when you look at SS and RF and 1B, the SS will float to the top, even if you have a guy with “negative” fielding. It makes far more sense to include the positional adjustment to the fielding portion than not.
That’s all I got!!
As you can see, I really have no real objections to anything, which given my nature when it comes to this stuff, is a huge testament to Sean’s thoroughness.
It seems Sean was convinced: