Wednesday, December 24, 2008

Nationals Wins Above Replacement

To get a handle on where the team stands now and the impact of our moves on the outlook for 2009, I will be maintaining a spreadsheet adding up the Nationals team Wins Above Replacement (WAR). I will update it periodically as the team's roster takes shape.

WAR is just what it sounds like, and although there are many ways to figure it, I'm using Tom Tango's method for hitters and the Dodger Sims method for pitchers. You can click the links if you are interested in the dirty details, but the basic idea is to take the all the component player contributions, quantify them in terms of wins and then measure the difference versus an entirely replacement-level team.

I'm using Bill James's projections from Fangraphs for offense and pitching for most players and Marcel for players James doesn't project. The Tango method is built around the composite offensive stat wOBA, which without getting into the nitty-gritty is basically a measure of overall offensive performance adjusted to look like OBP.

For defensive projections, I'm using the last three years of each player's UZR and Rate2 numbers to generate a composite runs above average metric
(this is a change from what I originally posted two days ago). Because of the still pretty slippery nature of defensive metrics, I'm just trying to get the decent sample size of a couple different stats. (I've explained UZR and Rate2 in previous posts, so you can just click control-F and search the page you're looking at for those terms.) Note that even after summing all their career innings played, however, Anderson Hernandez and Kory Casto still have so few opportunities that I just substituted average defense for them. And since Rate2 and UZR basically tell us nothing about catcher defense (it only looks at how they field balls in play--nubbers, bunts, pop-ups), I list all catchers as average defensively.

Offensive playing time is entirely my best guess--not what I think the team should do but what seems most likely based on the current roster. Innings projections are from James (or Marcel if James didn't project the player). Because of the holes in the bullpen I'm filling 120 innings with TBD replacement-level pitchers. (Although if those innings go to the likes of Jorge Sosa we'll be longing for replacement.)

So without further ado, here's where the Nationals are at right now (keep in mind an entirely replacement level team will win about 47 games, so add 47 to the Nationals total WAR for a projection of their 2009 win total):

Hitters WAR
Pos Name PT wOBA Def. RAA Pos adj WAR
C J. Flores 70% .309 0 1.25 1.38
C W. Nieves 30% .269 0 1.25 -0.10
1B N. Johnson 50% .392 7.4 -1.25 2.49
1B K. Casto 50% .317 0 -1.25 -0.03
2B R. Belliard 60% .328 0.1 0.25 1.25
2B A. Hernandez 40% .274 0 0.25 -0.42
SS C. Guzman 85% .323 -3.6 0.75 1.65
SS A. Hernandez 15% .274 0 0.75 -0.08
3B R. Zimmerman 85% .363 15.1 0.25 4.71
3B K. Casto 15% .317 0 0.25 0.22
RF E. Dukes 70% .363 6.9 -0.75 2.63
RF A. Kearns 30% .351 15.2 -0.75 1.16
CF L. Milledge 75% .347 -8.5 0.25 1.78
CF W. Harris 25% .325 11.9 0.25 0.76
LF J. Willingham 80% .373 -15.0 -0.75 1.80
LF W. Harris 20% .325 23.5 -0.75 0.63



Pitchers WAR

SP John Lannan 177 4.6 1.67
SP Scott Olsen 199 4.8 1.58
SP Daniel Cabrera 141 4.8 1.00
SP Collin Balester 120 4.6 1.19
SP Garrett Mock 50 4 0.82
SP Shawn Hill 75 3.8 1.32
SP Jason Bergmann 89 4.6 0.89
CL Joel Hanrahan 77 4.3 0.20
SU Saul Rivera 80 3.5 0.87
RP Tyler Clippard 68 4.7 -0.12
RP Jason Bergmann 50 4.6 -0.03
RP Steven Shell 66 4.6 -0.08
RP Mike Hinckley 32 3.9 0.20
RP Marco Estrada 31 4.6 -0.02
RP Shairon Martis 63 4.4 0.09
RP Other RPs 120 4.5 0.00

Total 1438

Total Team WAR
Total Offense WAR 18.66
Total Defense WAR 1.16
Total SP WAR 8.46
Total RP WAR -0.02
Total WAR 28.27
  • Update: If you are interested in seeing the full spreadsheet, it is published on Google spreadsheets here.
  • Update 2: At 11 pm 12/25, I made some corrections and revisions that affect player WARs but not the overall team WAR. I also revised the Google spreadsheet to unhide some inadvertently hidden columns and make some other calcluations easier to follow.
  • Update 3: at 10 am 12/26, I revised the defensive metrics as described above to include an average of 3 years worth of UZR and Rate2 data instead of just 2008. For this I just counted the runs saved per 150 games for each system and averaged them out after weighting them by innings sample size. The result is that the Nationals overall become out about a win better, with guys like Belliard, Zimmerman, and Kearns improving, while Willingham and a coupe others did worse.
  • Update 4: in case you're curious, adding Tex would have added a net 2.5 WAR, elevating the Nationals from a 75-win team to a hair under 78.


Steve Shoup said...

Just curious Steven if you looked at what the difference would be if Willie Harris got the back up time at 3B and 2B instead of Casto and Hernandez? I realize its unlikely but I feel that would be the best way to use Willie as a super utility playing almost every game. Also the arguement could be made that if Johnson only plays half the season likely a better option than Casto would be Belliard or Willingham giving Harris a further oppurtunity to get AB and show off his defense.

On a side note Happy holidays Steven thank you for doing this blog its been a great find for me this past year.

Steven said...

I just ran what I thought is most likely this year. The 10-15% PT backup allotments don't have a huge impact though. Maybe a half a win one way or another total.

Steve Shoup said...

No but you gave 40% for hernandez at 2b and 15% for Casto at 3B that is a pretty significant amount of time, esp. considering Harris's defensive value over these players. I know how you projected it is probably how it will happen, but I believe that Harris should always be the 9th man so to speak if anyone's out or needs a day off Harris should be the man. (i.e. Johnson is off Belliard or Willingham should be at 1st with Harris in their position, that suits the nats better than just going to Casto). I was just curious if you looked at that value over just going to the "designated" back-up, Casto, Hernandez, Pena ect.

Steven said...

Harris's is an excellent defensive outfielder. The metrics I've seen rate him an average defensive infielder. You could give those ABs to him and you'd only gain the benefit of his .325 wOBA over Casto's .325. To get more value out of Willie you have to put him in the OF.

Hernandez and Belliard it sounds like are set to split the starts at 2b. If it's run as a strict platoon, it'll be more like AHern 70% and Belliard 30% because Hernandez is the switch hitter. I ran it down to 40% because I just don't think Hernandez will perform at all, especially if you believe James's projection. In other wods I think it's unrealistic to think Hernandez is going to have a .274 wOBA and keep playing.

e poc said...

cool post, steven. could you please explain how you're factoring in replacement level and positional adjustment? it looks like replacement is being added to the offense and position to the defense - is that right? also, i think it would make more sense to use a projection of some sort for defense, rather than just last year's numbers (you get some wild small-sample numbers like willie's +35 in left. plus, you chop someone like ahern down to average because of the small sample, which doesn't seem right). also, what replacement level are you using for pitchers, and why do you prefer the dodger sims method to tango's? i like this feature.

Steven said...

I just published the spreadsheet on google docs so you can click through it.

Tango's post explains in detail the position adjustments and how replacement is calculated with a detailed Q&A:

The short thing is that you take league average wOBA (which always equals league-average OBP) and subtract the offensive positional adjustments to get to replacement.

There's also a positional adjustment for defensive contributions to reflect that positions on the right end of the spectrum are harder to play than those on the left.

Replacement level for a starter is 5.50 FIP and 4.50 for a reliever. This is what Dodger Sims uses, and I also think I saw the same numbers on Fangraphs but I forget just now. I didn't check the math on that though.

I prefer what Dodger Sims mainly because of simplicity. DS just takes IP and FIP and compares that to replacement. Tango factors in leverage for closers and set-up men, which is certainly important and more precise, but it's a lot more work and makes it harder for readers to understand. Since the Nationals closer, set up men, and mop-up men are all pretty similar it's not worth the work. If we had Mariano Rivera back there with a bunch of scrubs, then it would be worth factoring in the leverage for his innings because his FIP is so much better. But Hanrahan's FIP isn't all that much better than Hinckley's or Shell's so it's not worth the extra math and confusion.

Re: the defensive metrics, I agree it's all pretty slippery. I zeroed out the ones that struck me as totally off based on sample size, but left in place the ones that didn't. UZR for instance has AHern as a +38.6 2B, which seems absurdly high, and a -150.9 SS, which is definitely absurd. Harris's left field defense I think is based on a large enough sample size and matches his track record and my observations well enough to leave it in there.

A projection would be better, but no one I'm aware of does a defensive projection, and I'd rather not make this about my personal judgment. There's enough of that on this blog! I heard that Clay Davenport will have a new defensive metric in the new BP annual, so maybe that'll solve it for us.

Steven said...

The spreadsheet is now linked on the bottom of the original post.

e poc said...

after looking at the spreadsheet, i have a few questions: 1) why do you make a positional adjustment on both offense and defense? i don't think that is necessary, and you may be overrating the offense because of it. 2) i can't tell whether you're measuring all of the starters' innings vs. replacement or only 150 of them, and then y measuring anything above 150 vs. a replacement reliever. it doesn't make a huge difference (which is why i can't tell which you're doing), but i think the former is a better way to do it. (actually, if you want to keep the 2 win difference between replacement and average, you want to make the cutoff 180 innings, i think, which kind of makes sense intuitively, too, since only 750 innings from a replacement rotation (150*5) would lead to higher replacement levels for the relievers who had to cover another 700 innings, whereas (180*5) 900 innings from your rotation is probably more accurate. it also suggests that you only want average or better starters pitching more than 180 innings in a season, which makes sense, although a lot of people who extol the virtues of "innings eaters" might disagree.) 3) for defense, it might be a pain in the ass, but career uzr numbers regressed to the mean and for age would probably be a better way to get an accurate picture of what the team will look like next year. willie's uzr in left from last year is not a very large sample at all, considering that even a whole year's worth of data isn't really a large enough sample to judge a player's defense, and even so, his uzr/150 in left from last year is about 13 runs better than his career uzr/150 in left, and he's getting older. that's a big deal and causes you to overestimate him.

Anonymous said...

Hey, thanks for the calculations! A note that the James projections tend to be pretty generous for hitters; I'd like to see those numbers tempered a little bit.

I think Kearns is being wasted a bit on that bench, as he's a great outfielder and decent bat. Maybe the Nats could trade him to a big payroll club as a 4th outfielder? They could really use the prospect/reliever they get back for him. If he rides the pine this year, he'll have very limited value from here on out.

Steven said...

epoc--thanks for the proofing. I'm the last guy who should be seen as expert on this stuff--I'm just stealing other people's work and trying to apply it to the Nationals, so don't hesitate if you find ways that I'm f-ing this up.

1. What I did was to adjust both the offensive contribution and the defensive contribution separately and then add that all together instead of just adding the defensive and offensive contributions and adjusting the whole thing. That shouldn't necessarily have been a problem, but as I was smoothing all that out to make my math easier to follow I found a different problem in how the spreadsheet pulled the PT numbers so that player WAR was allocated wrong. So the team WAR was right but the position player WARs were not. I will revise the spreadsheet to make it accurate and also easier to follow I think.

2. I am measuring all the starters innings v. replacement FIP. If you do it the other way, you are punishing pitchers for anything they throw over 150 innings. For their first 150, they are being compared to a replacement starter who puts up a 5.50 FIP, and for the next 30 you compare him to a replacement RP who gives you a 4.50 FIP. I don't see why that makes sense. You can't assume that those 30 innings will be picked up by a replacement reliever anyway.

3. For defense, I generally agree with you. The metrics are very imperfect, and there's a sample size problem. If I knew how to do regression analysis maybe I'd do what you suggest. As it stands I'll probably look to average let's say the 3 most recent years worth of data, and instead of just averaging Rate2 and UZR, also throw PMR and Dewan into the mix. For now I don't have time for that so what we have is what you get. I don't think it makes sense to factor in career-long defensive numbers, because a guy like Ron Belliard for instance just isn't the defender he used to be.

Steven said...

@redsoxtalk--agreed that Kearns in particular seems optimistic. And if he does this, he'll probably play more. And either Willingham will rot or Milledge and Dukes will lose development time. But right now Kearns can't have much if any trade value. The guy was just terrible last year, and though he's a good defender, I don't think he's good enough to carry as a defensive replacement. It may just have been the injury last year, but it's hard to know if he was bad b/c of the hurt or bad and also hurt.

e poc said...

cool stuff, steven. i got about 16.5 WAR for the position players when i ran something similar with marcel. i think your total is more likely, as there's a lot more upside than downside in the marcels (dukes, willingham, and milledge are all likely to beat their marcel projections). another thing i'd be interested in seeing is a similar projection based on an optimal use of the roster (e.g. harris/kearns instead of willingham in the outfield, etc.). or high and low end projected WAR totals (you could use chone's 80% and 20% projections, for instance). if we used our roster optimally and everything broke right, could we challenge for a playoff spot (hint: no).

the more i think about it, the more i believe that there should be a innings cutoff for judging a starter's value vs. a replacement starter. it should be roughly 180 innings, i think, if you're doing calcualtions based on dodgersim's method of fip vs. replacement ERA. this preserves the 2-win difference between replacement and average. it definitely makes sense that there must be a cutoff at some point if you're just comparing FIP to replacement ERA, since at some point the starter is taking innings away from reliever's that could pitch them. so if a below average starter is pitching innings that a replacement reliever could pitch, that is actually hurting your team, an interesting and important point to consider. (this is why i like tango's method for pitcher WAR calculations, actually, since it uses the run value of the game situation to determine a pitcher's win%. FIP v. replacement ERA begs too many questions about usage, etc.)

good work.

Steven said...

Thanks--Tango's method is definitely better in general. You need to factor in the team's ability to leverage it's best performers into more wins. I just didn't bother with it because the gap between our best reliever and worst is so narrow. But in the future I think I'll probably add that wrinkle.