Tangotiger Blog

Oh you have the perfect name for this?

I hope it’s not WAR in Pieces, since I’m about halfway done with that novel.

That is a great title!

Mine won’t be as good, but will be nerdier…

I’m surprised you would announce such a great title like that!  You should make a blog post pronto, and use that as a thread title.

Remember George Costanza: Seven?

Rally, this is like the least-important recommendation I can make: can you make your table formatting in your new pages matches those from the old page?  This is what you have (on the old pages):

BORDER=2 cellspacing=0 cellpadding=2

It’s just for the sleek/sharp presentation.  Again, sorry that I offer such little recommendation to such an overall powerful product.

I’ll check on that this evening.

I think it should be a simple fix through css.

I don’t have links to this on my homepage yet, that is coming soon. But all pages are up and you can get to them by clicking through here:

https://www.baseballprojection.com/war2/playerindex.html

Nope, it’s in the code, not css, so will have to wait. But it is worth doing.

I need to reduce the number of pages on my site, so 2 things that will probably go are:

1. WAR from 1871 to 2009, posted 15 years ago.

If I delete these I’ll replace them with a CSV download so the numbers will at least be preserved.

2. Fielder pages - some of this has moved over to the WAR2.0 pages fielding section.  What else should be prioritized and brought over?

Compare Ozzie’s defense page:

http://www.baseballprojection.com/defense/player_smito001.html

To his new WAR page:

https://www.baseballprojection.com/war2/player_smito001.html

And the new home page link is up

https://www.baseballprojection.com/

Accounting/detail questions for both of you… Is your baseline (i.e. unreconciled) GameShares number set so that each team adds up to 162 (or however many the team played)? Or so that the league adds up to 162*30 (or whatever)? I think there is a case to be made that unreconciled GS should sum to team games. But it is not hugely important issue, and, in the end, it is a choice/assumption with no right or wrong answer.

Also, I’m assuming that the reconciliation adjustment is based on playing time, and not based on fraction of team wins (which is how Bill did it).

You could call your class “The Art of WAR”, and then you’d spend the first lecture weeding out the people who thought it was going to be about something else. 😊

I make mine add up to the number of team games.  This handles extra innings, 7-inning games, etc.  W, L, G, they all add up.

***

Art of WAR: excellent!

One thing I noticed on Reference Rally: the league adjustment v the fielding scale.

From what I remember, the fielding scale was set to 0 at the… MLB level?  But then there was an issue regarding comparing AL/NL?  I don’t remember exactly I just remember it being an issue.

Also, I know we’re pushing for Statcast OAA for defence, but how does this affect how we use position modifiers going forward?

I’m fairly sure outfielders were meant to be measured relative to all outfielders, but I can’t figure out how infielders relate. Infield positions also don’t average out to 0 for any position, but the implied position modifiers are very different to the 7.5/2.5/-7.5 I’m used to. Am I missing something?

Infielders are being compared to other infielders in their “slice”. 

So, a 3B or SS or 2B that all happen to play in short-RF (back in the shift days) are compared to each other.

A 3B or SS playing in the hole are compared to each other.

I used 2016-2023, and so things don’t get to 0, but it’s close.  I probably should rebaseline things.

***

As for the position modifiers: those are still pretty good and somewhat consistent with the data of position switchers.

Game shares does not need to be reconciled, it should add up to the team games total. The reconciliation process just changes wins and losses, but games stays the same.

It is done by playing time. Players do not get docked based on a percentage of their wins. So for example on the 2001 Rockies, Juan Pierre gets the same adjustment as Todd Helton.

“From what I remember, the fielding scale was set to 0 at the… MLB level?  But then there was an issue regarding comparing AL/NL?  I don’t remember exactly I just remember it being an issue.“

Yeah, you might have NL at +200 and AL -200.

Right now I can’t remember if I fixed that for these numbers. If not, then there’s a fix for WAR 3.0 down the road.

I think I did, but will have to check. That work was 2 months back in the process.

I can happily confirm that the fielding runs are set to zero at the league level, not mlb.

#13: Ok, so we can still use position modifiers as usual, but we’d use a single modifier for OF (likely -4.17 until we have a better benchmark). That works. Thanks!

Posted something about defensive replacement level yesterday.

On BJOL.org, you can add comments. Might need to register, but it is free:

https://www.bjol.org/articles/sean-smith/533792-defensive-value-over-daniel-vogelbach

Its also on my site:

http://www.baseballprojection.com/special/defensive-replacement-level.htm

Actually, now that we have the indy W-L records, we might be able to split the offensive and defensive records.

I just hope we have enough “game space” to do this. The idea is that position players get 56% of game space, and pitchers 44%, which implies position players are 50% hitting, 6% defense. This implies about 76 PA is one “game”, and like 1300 defensive innings equals one “game”. Especially since it might need to absorb the position modifier as well.

But it might also solve the issue of DHs getting too many replacement runs, due to not contributing at all defensively, and us usually scaling to PA.

Pat: that is exactly correct. 

I wouldn’t necessarily go with the 50/6 split, but by characterizing everything as a W/L record, it makes it alot cleaner to answer Rally’s question

In other words, let’s say that Eddie Murray ends up with a career 8-12 record for his fielding (just making it up).  Well, then Rally can say “I want to compare that to a .300 player”.  Ok, go ahead and do that.

That will give you +2 WAR for his defense.

This is part of what I am (trying) to write in my (stalled) WAR book.

It’s not going to add up that well using the numbers I start with.

I have Eddie as a +30 fielder, but with a -119 position adjustment, and 17.3 defensive game shares.

http://www.baseballprojection.com/war2/player_murre001.html

So -9 wins. In that space he’d have something like -0.4 defensive wins, and 17.7 losses.

So forget about comparing to .290, or any kind of recognizable and reasonable replacement level.

You have to either stick with the whole player, forget about offensive and defensive wins, or try a different approach. I decided on the latter.

Right, having 10% of his game shares being defensive shares is going to be tough here.

I have a modified approach.  When I’m further along (might be months), I’ll share that

Right. I extrapolated the 56/44 split between position players and pitchers into 50% offense, 44% pitching and 6% defense. Maybe that’s no longer accurate, since pitchers can’t contribute offensively anymore. But since this is a visualization anyway, we might not have to be as strict.

I think spread of hitting talent was 3-4 times bigger than defense, so if a full season starter gets 10 game shares, maybe 8-2 or 7-3 towards hitting makes sense. And if we can agree on a baseline of inadequate defense, maybe that guides our decision too.

It’s just DH decisions aren’t always that clean. Sometimes a team’s DH is a decent 4th OF, and sometimes a team has two terrible defenders they need to hide. Plus all the teams that just use it as a glorified bench spot.

Yeah, the problem is a long-discussed one.  The variances add, not the standard deviation.  But what we care about is the standard deviation.

In other words, if you focus on off/def, then you have:
50 offense
50 defense

Now, if you break it up into the spread in talent, it basically shakes down to:
5 shares: offense
4 shares: pitching
3 shares: fielding

So, 5 squared is 25, 4 squared is 16, and 3 squared is 9.  Therefore, 25 for offense and 16+9 = 25 for defense.  Take the square root, and it’s 5 for both.  That’s how you get 50/50.

However, if you merge offense + fielding into nonpitchers, you get root of 34 or 5.83.

So, 5.83 for nonpitchers and 4 for pitchers gives us, in this illustration, 59 to 41.

But, if you (erroneously) use the square of those numbers instead, you get 34 shares for nonpitchers and 16 for pitchers to give you a 68/32 split. THIS is the problem that Bill James never overcame in Win Shares.  Right here.

You can easily play around with the numbers to get it to 35 for pitching to match Bill (or siphon off some of the fielders into pitchers, since pitchers are fielders too).

***

But in any case, this splitting is really almost irrelevant.  It’s besides the point.  Bill James treats it as a central requirement in Win Shares.

In WAR, it’s just something you could add if you want, but it’s ultimately irrelevant.

Rally (#18): I’m wondering what the equivalent metric to DVODV would be for hitters? Maybe something like OVONP—offensive value over Neifi Perez. That is, if we define zero offensive value as the lowest practical level at which a player’s defensive value can keep him in MLB (pitchers excluded), Neifi seems like a reasonable approximation. Neifi, with a 58 wRC+, was about 35 runs below average over a full season. So you could take a player’s Rbat + Rbaser, and add 35 runs per 650 PA. So Babe Ruth had 1890 OVONP, and Barry Bonds had 1857, while Mark Belanger had 152.

I confess, though, I’m not sure what these metrics tell us in the end.

Here’s my framework:

Rule #1: Everything is “above average” as replacement value should only be used for individuals, not in aggregating them for team totals, because you have to make positional adjustments and them remove them. Instead, don’t introduce them.

#2: Express player contribution in estimated runs.

The concept is that everything a player does while his team is at bat increases or decreases his team’s runs scored, and everything done while his team is in the field (both pitching and on defense) increases or decreases his team’s runs allowed.

For example, in the 2024 NL, teams have scored an average of 4.38 runs per game, which is 709 runs per 162 games.

Therefor, expect each team to score 709 and allow 709. For each player, add their offensive and defensive runs saved to the team total. If a batter is +50 on offense (batting + baserunning) and +10 on defense (fielding + pitching) then he would turn his team’s 709/709 RS/RA to 759/699 = .541 WL% = 87.6 wins = 6.6 WAA.

Brian/27: yes, that’s how I do it.

What Bill does, trying to start from some theoretical floor is just an unnecessary requirement at best, and doomed to cause problems at worst.

Starting with the average is guaranteed to work. Not only for baseball, but for ANY sport.

I once did a WAR for Bill’s college basketball team.  This is even though I know little about basketball and zero about college basketball.

And yet when I presented Bill with the results, he said they looked reasonable for most players, and unreasonable for one or two players.

Given that I spent the number of minutes I can count on my hand, that was a major success as a proof of concept. 

Bill is just stuck in his mindset, which is odd because he always says he likes to start with a clean slate.

The book (kindle edition) is out. Print copies probably within a week.

https://www.amazon.com/dp/B0DJN8Q82T/

I posted the table of contents here, which should give the reader a great sense as to the scope:
https://x.com/tangotiger/status/1843687925701325254

On Twitter I said:

How-to book, explaining in detail all choices made; plus going through real-life players to highlight issues being tackled

If you want to create your own uber-stat; want to understand nuts and bolts of WAR; or are fascinated in how extreme players get evaluated, this is the book

That was cool.

Awesome, Rally, on the book! I did not realize you were working on this.

Other than the first post here, which I see was made in March, I didn’t mention the book much anywhere. Might have said something in th3 last few weeks on bjol.org.

Just got my copy and through the first three chapters. Loving it so far. Good history of WAR..which you could tell involved a lot of research on your end.

Although I dont think I have ever read a hard cover book that referenced an X-tweat that happened about 3 weeks before I read it (Bill James - page 6). must have been a typo

Well done Rally

Thanks Darren.

I might have the date wrong, as to when Bill posted that quote, or when it was retweeted, or responded to, or whatever to make it show up in my timeline.

Turnaround was very quick though, I saw it late in the editing process and thought it was a good fit.