WAR Spreadsheet Version 1.0
by Sky Kalkman on Jan 14, 2009 5:00 AM EST
Email
PrintI've made some changes to the WAR spreadsheet in the hopes of making it more accurate and easier to use.
Changes since the beta:
- The option of using OBP and SLG instead of wOBA. Make sure they are park-adjusted, though. To use OBP/SLG instead, change the TRUE underneath "wOBA?" to FALSE to force the Hit calculation to use (1.75*OBP + SLG) / 3 to compute wOBA instead of using whatever's in the wOBA column. (The wOBA column will remain blank, though, except for the team totals row.)
- New baselines for replacement-level team wins. I started with 48 wins per team (a .300 team), added two wins for NL teams (because they are worse, but compete against each other), subtracted two wins for AL teams (similar reason), subtracted 1 win from AL teams as an estimate of added production from DHs, and subtracted 1.5 wins from all teams as a rough estimate of the average number of wins created by leveraging relievers in a bullpen. The result is a 48.5 win baseline for NL teams and a 43.5 win baseline for AL teams.
- A chart showing the probability of winning at least X number of games, given the overall talent estimate of the team, where X goes from 61 to 101 in intervals of 5 games. The first line should be read "The probability of winning at least 61 games is Y%."
- New PA estimates for a complete offense. The average PAs per position/lineup spot should be about 695, but that can range from, say, 620 to 760 depending on lineup position (18 fewer PAs for each spot lower in the order). NL teams should get another 350 PAs from pinch hitters. Do not include pitcher batting. And the higher the team's OBP (or wOBA, as a proxy for OBP), the more PAs they'll have. A rough estimate is 8 more PAs per lineup spot for each .010 points of team OBP (wOBA).
After the jump you'll find some user instructions. Feel free to ask questions in the comments.
- Change the information in the green cells, but don't touch the rest.
- Everything is measured in wins. If you prefer to do your scratch work in runs, just divide by ten when entering the info in the spreadsheet.
- If you're not up with using wOBA, you can enter park-adjusted OBP and SLG instead. You only need to enter wOBA OR OBP/SLG, not all of it.
- BR is baserunning, which probably will include both SB/CS info and non-SB/CS info, unless you're looking at Fangraphs for wOBA (which already includes SB and CS in wOBA). I'd suggest assigning most players 0 for non-SB/CS baserunning, with some at +/- .25 wins and the rare player at +/- .5 wins. Baseball Prospectus has some solid baserunning numbers, just make sure to ignore the EQSBR piece if appropriate. As an estimate, one CS cancels out two SBs and six net SBs create .1 wins.
- Fld is fielding relative to position. You could even add in OF throwing arms from THT, although 2008 data isn't available yet. The CHONEs have good fielding projections.
- ERA for pitchers is a projection of their ERA, park- and defense-adjusted. Do NOT use past seasons' ERAs as an estimate. Instead, use FIP, xFIP, tERA, or something like that. ERA is a blah stat. Projecting pitching talent on the ERA scale is safe to do, however.
- Once you have IP set for the relievers, change their leverages so that the team averages out to 1.0. Closers should stay at about 1.8 with one setup guy at 1.3 and a bunch of guys below 1.0 to average it out.
- Please, please, please don't just look at 2008 data. This is a projection. All past performance matters to varying degrees, not just last year. Even better, use projections done by people smarter than both you and me (CHONE and PECOTA are my favorites). Or get community input and take a wisdom-of-the-crowds approach. Or see how much over their heads the players on your favorite team have to perform in order to reach 90 wins.
- Be conservative. Players get injured. They don't live up to expectations. Pick rate stats and playing time estimates that account for the pessimistic possibilities, not just the optimistic possibilities. Pretend the 2009 season will be played 10 times. If a pitcher throws 200 IP 8 of those seasons, 100 IP once, and zero once, that's an expected total of 170 IP. Being conservative on the full time players also lets you give small, but significant playing time to a lot of backups and potential callups.
- WAR stands for wins above replacement. An average player over a full season is worth about 2.25 WAR. All-Stars tend to be 4+ WAR. MVPs are often at least 7 WAR. Barry Bonds' best seasons were 12+ WAR.
- FA$ is the value of each player's projected production priced at $4.5MM per win above replacement (the going rate of the free agent market), plus the $400K league minimum.
- If you don't have Excel, check out OpenOffice (a downloadable, free, Excel-like application for all operating systems) and EditGrid (an online tool that simulates Excel and is way better than GoogleDocs).
57 comments
|
0 recs |
Do you like this story?
Comments
Those are a little high, appropriate for the higher level of offense that ended two years ago.
It’s more like 4.60 for relievers and 5.50 for starters these days. Plus/minus .10 runs for AL/NL for relievers and plus/minus .13 runs for AL/NL for starters.
And, to be honest, while those figures connect perfectly to the winning percentages used in the WAR spreadsheet for relievers, I used a winning percentage .010% low for starters, which I just discovered this morning. In the midst of fixing it.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Let's spitball some here.
I will list both ERA, because that’s how most of us think, and RA, because it’s how teams actually win and lose baseball games. Everything is for 1994-2008, both leagues.
The average pitcher: 4.47/4.87
Average starter: 4.58/4.98
I’ve also sorted all pitchers in that time period based upon their rotation slot, which I defined using their first starting date. This isn’t perfect, but it’ll do.
1-5: 4.43/4.82
6+: 5.01/5.42
Okay, so Win% for a replacement/6+ starter. Using Pythagenpat:
4.87+5.42=10.29
10.29^.28=1.92
There’s our exponent.
Now:
4.871.92/(4.871.92+5.42^1.92)=.448
That’s what I’m coming up with.
And yes, I realize that this suggests...
…that a replacement reliever is a league-average pitcher. I’m still hashing out some of these things.
I'm fine with the idea that rep level relievers have basically league-averaege ERAs/RAs
I’d call it a few tenths worse, but it’s close.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:41 AM EST up reply actions
MGL says
that he uses lgERA/RA is replacement level for relievers anyway, so if you like appeals to authority (and I do), then it’s cool
btw, colin, I finally got a chance to mess around with your baseruns pitching stuff in MySQL… I think I’m doing it right, might have some quesitons, but very, very cool stuff
Here’s something I should post somewhere else: do you know where I can find a set of multi-year park factors that I can easilly import into a MySQl table? they don’t have to be fancy, I’m just not willing to type all the B-R ones (yeah, I know they’re "goofy") myself for every year since 1956 or whatever…
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
by Matt Klaassen on Jan 14, 2009 11:59 AM EST up reply actions
The B-R park factors should be in the BDB teams table...
…under BPF and PPF, for batters and pitchers respectively.
d'oh. thanks
once again, I’m a lazy idiot, etc. etc.
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
by Matt Klaassen on Jan 14, 2009 12:01 PM EST up reply actions
How are you defining starter roles?
How many IPs for each role?
Let’s say each slot 1-6+ averages 170 IP for 6×160 = 960 IP per team (which is about right). Then you’re counting 1/6 of MLB IP as thrown by replacement-level relievers. Which is incorrect. You want the absolute bottom of the barrel.
I’m not trying to say that’s what you’re doing. It just MIGHT be, and if it is, I disagree.
FWIW, Tango’s 1.28xlgRA method of rep level yields a 6.23 rep level RA and 5.72 rep level ERA for starters.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:40 AM EST up reply actions
Based upon when they pitched.
The first pitcher to pitch for a team that season, is labelled number one. The second pitcher, number two. So on and so forth. R/ER is calculated only as a starter. A player’s slot is determined by what team he starts the season with. The average starter, 1-5, averages 140 IP.
How many IP from the 6+ group?
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:48 AM EST up reply actions
I can estimate it:
940 IP per team – (140×5) = 700, leaving 240 IP from the 6+ guys.
240!!!!
You can’t call more than 25% of the MLB innings from starters replacement-level.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:53 AM EST up reply actions
If we want to limit rep-level to 10% of innings pitched...
…we just use 9+ starters. We get 5.16/5.58. We’re down to very sublte differences now. The closest I can get to Tango’s rep level is something like 0.2% of all innings in sample.
Really? Huh.
Ok, let’s say rep level starters are .440 and rep level relievers are .500. What does that mean hitters have to be to make teams .300 overall. If we use the typical rep level hitter (.370? I forget), what level do you get for team rep?
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 12:15 PM EST up reply actions
Let’s use RA for a moment. 5.58 for starters and 4.87 for relief pitching. Say a 65/35 split between starters and relievers for IP, so that gives us a 5.33 RPG for a replacement-level pitching staff (presuming an average defense).
So, with a Pythagenpat exponent of 1.83, that gives us a rep-level on offense of 3.36 RPG. (I did some guessing and checking, and when I got to .300 I redid the exponent until the exponent didn’t change anymore.)
(4.87-3.36)*162=244.62
So a rep-level offense would score roughly 255 fewer runs than an average offense. That’s an absurdly low rep-level, I think -30 on offense.
If we go with a -18 rep level on offense, we get 3.98 RPG for a rep-level offense. That’d give us a team rep-level of .366, or about 59 games.
3.98 RPG seems pretty high, considering the A's, Mariners, Giants and Nationals were all right there in 2008.
(I’m ignoring the Padres because of ballpark.)
And, now I remember you’re talking 1994 to 2008, which was higher offense, overall, than just 2008. Hmm.
3.98/4.87 = .82, meaning rep level hitters are at 82% of average. That seems a bit high, but not crazy high. VORP uses 80% (mostly), and Justin’s study shows it to be somewhere between 75% and 80%. And that’s the most convincing study I’ve seen, even if it leaves some error bars.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Don't 6+ and 9+ starters include highly touted prospects?
And don’t 1-5 starters include replacement-level guys?
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 12:14 PM EST up reply actions
From time to time, sure.
I believe Rally ran a similar study where he excluded the top 100 prospects from the BA list, I just can’t find it right now.
And if the Twins want to make Livian Hernandez their #1 starter, well, not much we can do about it.
Yeah, well, good question.
I almost modified the “You can’t call more than 25% of the MLB innings from starters replacement-level” with a “Well, can you?”
I think it should be the bottom 5-10%, but that’s not based on much objectivity. And you’d have to definite that based on role, not ERA, otherwise you’ll get all the 8.00 ERA guys who are actually better than that but got unlucky.
For offense, Justin Inaz did a neat study where he looked at players based on role/playing time. The three bench players per team averaged about 225 PAs each, or 700 per team. The three fringe guys average 150, or 450 per team. The five scrubs averaged 50, or 250 per team. If you define replacement-level as bench players (which seems way high), you’re including about 10% of a team’s PAs. For fringe or scrubs, you’re more in the 3-7% range (mental math). Justin’s study shows that if we think rep level for offense is in the mid to high 70%s of average, then fringe is the category that best fits it.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 12:12 PM EST up reply actions
Link
http://jinaz-reds.blogspot.com/2007/10/what-do-we-mean-by-replacement-players.html
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 12:16 PM EST up reply actions
Meanwhile, back-calculating through Pythag...
…I come up with a 5.33/5.80 rep level for starting pitcher, per Tango’s Win%.
Actually, using a 1.9_ exponent for Pythag yields the winning percentages I was using.
And as Colin and Tom have pointed out, that’s a better exponent than 1.83.
Colin’s made some interesting points about rep level being higher for pitchers. While I like his study, I don’t love it and I don’t have enough to persuade me either way. It’s an issue under investigation.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 12:28 PM EST up reply actions
Question
FanGraphs has a different league average wOBA for CHONE and Bill James compared to Marcel. Marcel’s league average calculates out to .332 (based on reverse engineering wRAA), while CHONE and BJ appear to use the formula from The Book, which calculates out to a .345 league average wOBA. This makes a pretty big difference in the hitting wins calculation, about 0.8 wins per 700 PAs. Perhaps you should add a cell for the user to input the league average wOBA to get a more accurate calculation? You could still default it to .335 for people using OBP/SLG.
We’’re in process of trying to a guy with a trade record of working with pitches
CHONE is at .345, too?
I know Bill James’ projections are strangely/wrongly high for hitters, but I would have guessed CHONE was more appropriate, if not perfectly appropriate. Let me look into that.
A league-wOBA input wouldn’t be difficult to add at all, but I’m wondering if we want to have player data on different scales. In other words, why would someone use a Bill James projection if they knew it was “too high”? Is it really worth it to anyone to convert his projections down to more reasonable numbers? One reason some people like his projections is they “look better”, when they really just aren’t as conservative as they should be.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Sorry
I was wrong. I was working with some mixed up data that I calculated from CHONE before they were released on FanGraphs. It looks like the NL average wOBA for CHONE is around .329 (I only calculated a few Reds players), based on the corresponding wRAA. I only looked at a couple of AL players and they look to have an average of .332 or so.
We’’re in process of trying to a guy with a trade record of working with pitches
I think CHONE is actually at a pretty good level.
From this comment by Tango:
I asked David to calculate the wOBA for players that are common in Marcel, James, Chone, and that Marcel forecasts for at least 300 PA. David said:
==
So, out of the 368 players that match:
wOBA avg(wOBA)
Bill James .349 .342
CHONE .340 .337
Marcel .337 .334
===
The first column is the weighted wOBA, based on however many PA each forecaster thinks the 368 players are going to get. The second column is just the simple average of those same players.
Chone is forecasting 3 wOBA points more than me, among these players.
Marcel is set for runs per 27 outs of 4.7. So, that would put Chone at 4.8. I’m fine with that.
James is 8 or 12 wOBA ahead of Marcel, which would make his run environment around 5.0 runs per game.
4.8 runs per 27 outs is 4.62 runs per 26 outs (the average number of outs per team per game). In 2008, the NL average 4.54 runs per game and the AL averaged 4.78. Not claiming it’s perfect, but quite reasonable.
Marcel seems to be a touch lower, with Bill James two to three touches higher.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Any idea how the pitchers compare?
Does James have the same run environment for them?
We’’re in process of trying to a guy with a trade record of working with pitches
I don't know.
I have a hunch he’s optimistic on them, too, meaning their wOBA-against would be something like .325. Although perhaps he’s still just using the league RPG from 2+ years ago. I think it’s the first, though.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 10:06 AM EST up reply actions
as for your second point
If you are going to compile all of these together, then you may want to set a standard for projections. I know in general you are just releasing these so people can use them for their own purposes (THANKS!), but you may want some ground rules for your compilation, if you do it.
And I don’t think it’s that big of a deal if people use wOBA from Bill James. It’s just very important that they use a league average of .345 instead of .335, or else their results will be overstated, even for the James projections. Ultimately, it shouldn’t matter though since you are comparing to a league average before calculating wins. If someone wants to use James, we just have to imagine they play in a higher scoring context. Of course, then we’ll also have to adjust the pitchers league as well.
We’’re in process of trying to a guy with a trade record of working with pitches
It doesn't matter when converting to wins, that's true.
But if we wanted to compared wOBAs, I’d have to build in an adjusted wOBA field and not just adjust it on the fly.
And I’m honestly not sure why you’d want to use a projection system that’s obviously mis-centered and has been shown to be a step worse than others out there.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 10:08 AM EST up reply actions
So... I got CHONE's lgWOBA at around .328 or .329
is it that, or higher, like .335?
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
I don't know.
The quote from Tango above makes me think it’s just fine (right at league-scoring rates overall).
When you found the average wOBA, what playing time projections did you use? Do the PAs add up on the team level? Are pitchers included?
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:33 AM EST up reply actions
to find the league wOBA from FanGraphs
I used (wRAA/PA*1.15-wOBA)*-1 for each player. It’s not precise because of rounding, but the numbers are all between .327 and .329 for the NL players that I looked at for CHONE. The few players I did in the AL come out around .332.
We’’re in process of trying to a guy with a trade record of working with pitches
All I did was take a player
and his AB+BB+HBP for PAs (CHONE in fangraphs doesn’t list PAs), then his woBA as listed on FGs, then adjusted the lgwOBA until I got close to FGs CHONE-projected wRAA, if that makes sense
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
by Matt Klaassen on Jan 14, 2009 12:00 PM EST up reply actions
That's what I got for Marcel
haven’t tried it with chone
---
Juuuust a bit outside!!
http://www.rightfieldbleachers.com
I have a question about replacement levels...
I know that you already separate it into two distinct categories for the NL and AL for the level of play in those two leagues, which means it’s a variable that changes with the competitive context, but would it make sense to break it even further down onto a divisional level? Why I ask is because the RLYW season simulations the other day showed the Rockies (88.1 wins) about equal to the Phillies (88.3) when it came to projected wins, which makes zero sense initially, and yet given the weakness of the NL West, the winner of that division whether it’s the Rockies or any other team, could well equal the Phillies or the second or third place team in the NL East or Central despite having an inferior ballclub. The Rays, Phillies and NL West winner could all wind up as +40 WAR teams under the strictest definition (88 wins minus the replacement level of 48) even though there would be three distinct levels of baseball talent represented.
In particular, the offensive environment in the NL West seems so weak (4.25 runs per game in 2008 for the five teams despite having just one home park that favors pitchers) that I would think taking a strict NL replacement level for those teams would lead to underestimating in projections for their 2009 win totals despite giving an accurate depiction of their relative strength to the rest of the league.
Yes, I think that's a solid idea, assuming we're trying to predict actual win totals as closely as possible.
Once all the team data is put in and we can view how difficult each division is, we can talk about how best to make division-by-division adjustments.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 14, 2009 11:34 AM EST up reply actions
Cardinals
last time I ran CHONE straight through (wOBA/ERA/defense) and got 88 wins, which seemed high.
This time I didn’t enter in the exact same data, but basically the same and with this sheet I got 85.4. That seems right.
I'll be the one overrating these Faberge' eggs, thank you very much!
Future Redbirds / PAH9
Just a minor problem perhaps
that I noticed in the BETA but forgot to mention
Maybe I screwed something up but in the “totals” cells for the pitchers, the “starters” line adds up D41:D58, when it should just add up D41:D49. Doesn’t effect the projection unless you keep getting confused why it seems like your starters are always “over” the goal…
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
Hmm, I'll have to look later. That would be bad.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
probability chart
I’m having issues with the “probability of winning X games” chart. When I publish it on Google Documents, I keep getting an error message that says “Argument too large:162.” It works fine on the spreadsheet on my hard drive. My Excel skills are pretty lousy, so I’m not sure what the problem might be.
Anyone have any ideas? Thanks.
Pittsburgh Lumber Co.
http://mvn.com/pittsburghlumberco
Hm, try using 81 instead of 162 in your formula.
by R.J. Anderson on Jan 14, 2009 3:46 PM EST up reply actions
That's not the same.
It’s easier to flip 7 heads out of 10 than it is 70 heads out of 100.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Your problem is Google Docs.
Anyone having problems getting it to work with EditGrid?
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
Now there's a good question.
I’ve told people it doesn’t matter if they put PH at-bats at a position or at DH, but it obviously WILL matter, won’t it.
Let me think about it.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
I think I remember reading that the PH adjustment is -12.5
I’ll check and see if I can find that article.
---
Juuuust a bit outside!!
http://www.rightfieldbleachers.com
-17.5
Bringing you more-or-less replacement level analysis and commentary since sometime in 2008.
by Matt Klaassen on Jan 15, 2009 4:17 PM EST up reply actions
AL/NL Backwards?
Why does changing the A/N from A to N lower the win total? I’m working on the Phillies one and it seems that switching from AL to NL lowered it by two wins. That seems backwards…am I missing something?
Huh.
Just played with it, myself.
The position players are rated 5 wins worse as NL players and the pitchers are about 3 wins worse. I’m giving NL teams 5 extra wins compared to AL teams, so the net loss is 3 wins. It really should be about
Things certainly could be off.
Although, In 2008, the average NL team was about 6-9 against the AL…
Once all teams are entered, we can empirically force the numbers to average out correctly. The WAR baseline will probably be changed.
Beyond the Boxscore // Calling BJ Upton lazy is lazy.
by Sky Kalkman on Jan 19, 2009 11:21 AM EST up reply actions
Comments For This Post Are Closed
Secondary Sidebar
User Tools
FanPosts
Community blog posts and discussion.
Recent FanPosts
Follow us on Twitter!
BTB: @BtBScore
Justin: @justinbopp
Satchel: @SatchelPrice
J-Doug: @RationalPastime
Daniel: @CamdenCrazies
Peter: @CapitolAvenue
Josh: @garik16
Julian: @splashinpumpkin
Matt: @devil_fingers
Bill: @BillPetti
JD: @JDSussman
Lucas: @DBITLefty
Jeff: @jeffwzimmerman
Lewis: @LewsOnFirst
Certain photos copyright © 2012 by Associated Press or Getty Images. Any commercial use or distribution without the express written consent of Associated Press and Getty Images is strictly prohibited.
