Baseball for the Thinking Fan

Login | Register | Feedback

btf_logo
You are here > Home > Baseball Newsstand > Baseball Primer Newsblog > Discussion
Baseball Primer Newsblog
— The Best News Links from the Baseball Newsstand

Tuesday, June 23, 2020

Bill James: Why We Need Runs Saved Against Zero

        Absolute values, measured from zero up, create structure to our thought.  Absolute values make analysis possible.  Without reference points, it is very, very difficult to create a map, because every little error magnifies itself over space, creating an uncertain relationship between distant points.  “Surveying” is a matter of accurate measurement, yes, but it is also a system of creating multiple reference points.  Every map starts with a zero point; you don’t know what it is, but the guy who created the map does.  (For Google Earth, by the way, the zero point is in Lawrence, Kansas.  Really.)

        I understand, of course, why fielding is difficult to measure in absolute terms.  It is difficult to measure in all sports.  Offense is measured from the ground up.  Defense is measured from the sky down.  How tall is Charlie, measuring down from the sky?  It’s a hard problem.

        But the lack of a zero point, in measuring fielding, creates chaos in our ability to analyze fielding.  It creates just as much confusion in measuring fielding as it would in measuring income, or in creating a map.  And, because we have been measuring fielding in this odd way, there are lots and lots and lots of things about fielding that we SHOULD know, but we just have no way of knowing.

RoyalsRetro (AG#1F) Posted: June 23, 2020 at 01:00 PM | 202 comment(s) Login to Bookmark
  Tags: bill james

Reader Comments and Retorts

Go to end of page

Statements posted here are those of our readers and do not represent the BaseballThinkFactory. Names are provided by the poster and are not verified. We ask that posters follow our submission policy. Please report any inappropriate comments.

Page 1 of 3 pages  1 2 3 > 
   1. Rally Posted: June 23, 2020 at 01:30 PM (#5958975)
I have half followed Bill’s series of posts on this. But I just can’t get into it. He has laid out the concept that runs saved should be approximately equal to runs scored. So if .500 team scores 750 runs and allows 750, they are saving 750 runs also. So he’s setting zero point at 1500 runs, you’ll have positive runs saved as long as you stay under that level.

I am not a little bit biased towards the replacement level concept. No, I am extremely biased, so take that for what you will. Bill has had trouble with the replacement concept for years even though the roots of it can be found in his own Baseball Abstracts. As far as replacement level, I guess we’d say the team above allowing runs is about 150 runs better than a replacement level team, which would maybe give up 900. Bill seems to want a real baseline instead of a theoretical one, but while he might get something that looks like what he wants a real baseline to be, he can’t do it.

There is no zero baseline for defense. It’s true that Justin Verlander can dominate the best MLB hitters and hold them to zero runs. He could face a team of 90 year old cripples and they will score the exact same zero runs off him. But you can’t do anything similar on defense. The baseline is infinite. A team of 6 year olds could never turn a batted ball by MLB hitters into an out. A guy throwing 95 but unable to ever throw consecutive pitches in the strike zone will never record an out. Runs will just keep scoring. You have to assume some level of minimal competence. And once you do that, why that instead of replacement level?
   2. What did Billy Ripken have against ElRoy Face? Posted: June 23, 2020 at 01:37 PM (#5958976)
He has laid out the concept that runs saved should be approximately equal to runs scored. So if .500 team scores 750 runs and allows 750, they are saving 750 runs also.
Umm...why?
   3. Rally Posted: June 23, 2020 at 02:17 PM (#5958986)
If you want to know, read the article. Bill explains it better than I could.
   4. Kiko Sakata Posted: June 23, 2020 at 02:22 PM (#5958987)
Bill James is fundamentally correct here. Positive events (e.g., fielders making outs) should be grouped together and re-scaled to "wins". Negative events (e.g., fielders allowing hits or committing errors) should be grouped together and re-scaled to "losses". One should/could/can then compare "wins" against "losses" in order to calculate wins relative to average or replacement level (or what have you).

And by a happy coincidence: Here, I've done the work for you.
   5. tshipman Posted: June 23, 2020 at 02:27 PM (#5958990)
Bill's been making this argument for years, hasn't he? It's the philosophy behind win shares, and he's still steamed that people seem to prefer WAR.
   6. Kiko Sakata Posted: June 23, 2020 at 02:31 PM (#5958991)
Bill's been making this argument for years, hasn't he? It's the philosophy behind win shares, and he's still steamed that people seem to prefer WAR.


Well, one of the problems with Win Shares is that it doesn't really do this, because you need to separately track positive events - wins - and negative events - losses. You can't actually get from Win Shares to, say, Wins above Average (or Replacement level of anything else) because it's just a single number, the meaning of which is dependent on playing time, but the playing time is lost from the system - it's buried in there but you can't distinguish a guy who earned 8 Win Shares in 50 games from a guy who earned 8 Win Shares in 150 games without going to the trouble of looking up everybody's playing time. Win Shares also doesn't start from zero - mostly for the same reason; it starts from a smallish value that it treats as equivalent to zero.

But yes, much of Bill James's writing over the past few years falls very obviously into the camp of: "You damn kids and your damn WAR. Why don't you use my Win Shares instead? Damn kids these days!"
   7. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 02:38 PM (#5958992)
Im not entirely sure of what he's getting at in the article. He spends literally half the article using this analogy of a growing boy to illustrate the concept of finding an absolute reference pt. to build our map. But I dont see how this translates into a new method or perhaps some other method but maybe there's a folow up article or something. Maybe Walt or someone can explain that.

It seems he's getting at the idea of positional adjustments. Which I was raging against a few months ago. And I still hate this concept. Because I dont see the need and it also because it gives players playing key defensive position a bonus, even if they dont deserve it. Say for example Jeter playing SS getting a bonus +6 or whatever just for penciling his name at SS.

A long time ago, Bill James wrote a sentence that goes more or less like this:

"No one hits as a RF, he hits as a hitter."

Well, without any sort of context there are many ways to take that phrase, but as I recall he was trying to formulate the idea of a positional adjustment. So that instead of saying that Santo was 10 runs better at offense than the average 3Bman, there would be a way to compare him to any hitter playing any position. Well obviously we can do that, w/ batting average etc. we can measure offense vs the league average. But he wants to weight each player for his defensive contribution and then make these offensive comparisons.

Which I think is unecessary and confusing.

The most obvious response to Jame's idea, is "Of course someone hits as RF. YOu have to pencil in a lineup with one guy at each position." This seems obvious, but then most of the essay linked above is trying to make a case for positional adjustments. At least I think so...

Lets take one example: If one guy can play multiple positions and another cannot, well there's others way of quantifying the value of the versatile guy. If say Santo can play 3b/1b/CF and Dick Allen can only play 3b/1b, then obviously there's some value to Santo that's not being counted by simple def/off stats. Its his ability to fill in CF in a pinch, and presumably at a certain level where its not hurting the team. Cause obviously, well we could stick Dick ALlen out there too and that would hurt and we dont want to do that.

But it can't be double Dick Allen's value because of course Santo can only play one position at a time. (which rather contradicts James's statement that he doesnt hit as a 3B, he has to play a postion BIll). But one can quantify Santo's versatility with a method that doesnt rely on positional adjustments. You just have to figure how often does the avg. CF get hurt or is rested, say 15 games and then multiply by whatever Santos value above replacement would be and I guess we get some marginal value.

So we really don't need positional adjustment to account for that.

OK what about case 2: comparing two guys who dont play each other's position. What then? I think this is the gist of James's essay but whatever, I dunno. Let's take Dick Allen and Reggie Jackson. Lets assume neither can play the other's position. maybe we can put Jackson at 1b, but he's never done that before, and it would be a mistake to throw him out there in the mid season. And w/no data pts I think its mad to say he can play it at some average level. Maybe someone else can make that argument.

OK Allen we assume cannot play any OF position. And Jackson cannot play 3b/1b. Jackson is say +50 runs offensively better than the avg RF; and Allen is say 50 runs better off. than avg 3b. We'll say they are each avg at defense, so 0 runs there.

OK so its a tie. +50 vs +50 What more do we need to do?

"Aha says Bill James, its marginally harder to play RF (assuming arguendo), say 3 runs/season. So Jackson wins this comparison 53 to 50.

Its stoopid. Neither guy can play the other guys position. And they can only play one position at a time. They are each doing a fine job at holding their position called 3b or RF. Their each entrenched at that position.

"But what if say RF was ten times harder to play than 3b?"

OK, lets call it CF then. Let's say we've got Mantle in CF and we compare him to Allen at 3b. Lets say Mantle is +60 runs better at off then the avg CF and again Allen is +50. Let assume they are both 0 on defense.

So Mantle wins 60 to 50. "BUt wait he's playing a key defensive role, he should get more value.

"NO! Emphatically HELL NO."

Why not?

Cause whatever difficulty there is in playing CF is already inherently BUILT INTO THE CALCULATION because we just compared Mantle to the average CFer not the average baseball player. Presumably if CF is incredibly hard to play then we should see players with less offense ability. Right? So instead of comparing Mantle to corner OFs like Reggie Jackson or Killebrew or CLemente or whatever we are comparing Mantle to lesser offensive guys like Piersall or Dom DiMaggio or Tolan or Bill Virdon or whatever.

Presumably if CF really demands a better defensive player, then it HAS TO manifest itself in CFer's with lower offensive numbers. What else? Doing the way I suggest should already account for Mantle's value and the value of his position because we compare him to other CFers who have the burden of covering a tough position.

But no. Even this is not enuf for Bill James. He's going to make two more arguments. ONe of them is the market inefficiency:

The proposition that a thing which cannot be measured directly can be measured instead by the reaction to it, assuming that the energy of the reaction is equal to the energy of the action, is used in many places in science, and is not a bad concept. That’s what Pete was doing; he was assuming that the size of the difference in offensive value was a reaction to the difference in defensive value, so it must be of the same size as the difference in defensive value. It is not a bad concept. It is not an infallible concept, either...Shortstops and first basemen are drawn from separate markets with some crossover, but it is not clear that the talent available in one market equals the talent available in the next. There are market forces and random variables that can cause that NOT to be true in the real world.


OK first off you've got millions of kids, chasing millions of dollars to play SS and 1b. Its hard to imagine that 1b really have pretty good def skills, so that their def. numbers are somehow discounting their value and we need to assign a positional value. If they really are good, then with millions of dollars at stake presumably baseball will find a place for them.

So this would be James's argument that you cant simply compare Mantle to other CFs, cause, you know we might live in an era where there are only supermen playing CF and someone no one wanted to play 3B. So Mantle's value would be unfairly diminished by my method (cause all the great players wound up in CF), and Allen is being inflated by my method. I dont buy it. Not for a second.

Here's James second argument:

For example, although everyone agrees that right field is a more demanding defensive position than left field, right fielders also hit better than left fielders. If you follow the logic that all positions must be of the same value combining offense and defense, you arrive at a conclusion that everyone acknowledges to be false: i.e., that left fielders are better defensive players than right fielders. Tom has created a method to avoid this problem. But the REAL problem is, there is no evidence that the underlying principle is correct.


This stoopider than the first argument. The first argument you could actually make the case that their might be market inefficiencies, its possible. It cant really be proven thats impossible.

BUt here James is grasping at straws. First of all I dunno if that's even true that RFers hit better than LFer's. If it is true its some sort of aberration or some sort of illusion. Its the same thing as saying that DH's are hitting about as well as SS. OK they are but that's not a function of how hard the position is, that's a function that DHs tend to be players who are hurt, or aging or whatever. There's reasons for that not having to do with how easy it is to play.

Its like saying when metal helmets were invented (WW I) head wounds increased. Which statistically they did. CAUSE ALL THE GUYS WEARING CLOTH HATS WERE DEAD! Thats why metal helmets increase head injuries.

Its the same reason that Willie Stargell playing LF is better hitter than Willie Stargell playing 1b. It has nothing to do with how hard that position is. Its that we simply put aging players at 1b .

The reasoning in argument two just makes no sense. james has found one dumb exception that merely proves the rule. Heres the rule Its clear that as you go along the positional spectrum the better hitters are on the left side and somewhat easier def is on the left:

1b> RF> 3b > 2b > CF> SS > C > P

Oh and here comes Bill James to play his trump card:

RF > LF. "How can that be? LF is easier than RF. I win. Now you have to use positional adjustments."

Piss off James. Here:

1b > corner OF > 3b > 2b > CF > SS > C > P

Happy Now James. Its the same fukin position. One of them has a requirement for stronger arm. So do it this way:

Killebrew +30 runs off/0 def....RJackson +30 runs off/0 def

PLUS RJackson +5 runs.arm strength.

Simple. James finds one stoopid exception and now this means that all of sabermetrics has to go on a quest to find the Holy Grail of Positional Adjustments.

Gawd so sick of this...
   8. JJ1986 Posted: June 23, 2020 at 02:43 PM (#5958994)
Holy ####, the first half of this article. Is James totally nutty now?
   9. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 02:57 PM (#5958998)
its kind of a slog, yeah.
   10. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 03:00 PM (#5959000)
Bill James is fundamentally correct here. Positive events (e.g., fielders making outs) should be grouped together and re-scaled to "wins". Negative events (e.g., fielders allowing hits or committing errors) should be grouped together and re-scaled to "losses".


BUt Kiko, what's the big deal here? ("fundamentally correct")

Arent we already doing this exact same things with: weighted values of events --> Off Runs ---> WAR?

Is the only difference between Win Shares and WAR the use of baseline/zero line/ceiling of 750 runs? If so what's the big deal?
   11. Karl from NY Posted: June 23, 2020 at 03:01 PM (#5959001)
Bill's been making this argument for years, hasn't he? It's the philosophy behind win shares, and he's still steamed that people seem to prefer WAR.

People prefer WAR for good reason. One, the replacement-level baseline is set by a better method than Win Shares' arbitrary guess. Two, WAR accounts for negative numbers, while James always tried so hard to avoid either negative Win Shares or any concept of Loss Shares. Three, Win Shares' baseline is too low and overrates playing time because any marginal player accrues positive value as long as he's on the field, even if a random AAA replacement would be better. Four, WAR is denominated directly in terms of team wins, which is more intuitive than Win Shares' arbitrary multiple of 3 to make the numbers look bigger.

Bill seems to want a real baseline instead of a theoretical one, but while he might get something that looks like what he wants a real baseline to be, he can’t do it.

Because any real baseline will have some performers beneath it, producing negatives. Bill can't or won't accept the concept of negative value. If that baseline is zero so that going under it is impossible, then playing time dominates actual performance because the vast majority of plays are routine.

Zero as a baseline just really can't exist, because there is always another player with some level of baseline ability to plug into the slot. Even I could field a grounder or bloop a single 0.1% of the time rather than 0. A player is never competing against the possibility that his position would be vacant.
   12. Kiko Sakata Posted: June 23, 2020 at 03:20 PM (#5959005)
Arent we already doing this exact same things with: weighted values of events --> Off Runs ---> WAR?


Not exactly. WAR is built up from a bunch of measures against average: +20 batting runs, -2 baserunning runs, +5 fielding runs, etc - with a final adjustment to shift from WAA to WAR. But everything's centered around zero so, as with Win Shares, there's no distinction drawn between, say, +5 fielding runs in 900 innings vs. +5 fielding runs in 300 innings. Which leaves you either beholden to the value system of the metric's designer (replacement level = .294 if you're using Fangraphs or BB-Ref) or requires a ton of extra work on the part of the user to convert from, say, a .500 baseline to a .400 baseline - 5 runs over .500 in 900 innings is not going to be the same relative to .400 or .300 or .600 as 5 runs over .500 in 300 innings.

Which is why it is preferable to be able to have wins and losses separate. So, for example, if we look at 2019 right fielders, we can see that while Bryce Harper had the most net wins, Aaron Judge was more effective per play; he just had fewer plays (because of injuries).
   13. Kiko Sakata Posted: June 23, 2020 at 03:23 PM (#5959006)
the replacement-level baseline is set by a better method than Win Shares' arbitrary guess


I like WAR and think the world of the Seans, but any specific choice of replacement level is an "arbitrary guess".

Zero as a baseline just really can't exist


Sure it can. You just need two baselines: one for wins and one for losses. You don't want to START with measures against replacement level (which WAR also doesn't start with), you want to create wins and losses and then shift to WAR (or, as I call it, WORL) as a second-level calculation.
   14. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 03:53 PM (#5959017)
Kiko what is the fundamental difference that you seem to emphasize?

Is that Win share uses two separate measures for wins/losses? Or that win shares has a baseline so low that nobody is negative? or something else ?

Im failing to see these as much different but everyone talks like it is. For example in post 6 you seem to emphasize that WS is a counting stat but so is WAR... Also you can break WAR into off/def so whats the big deal here?
   15. JJ1986 Posted: June 23, 2020 at 04:11 PM (#5959021)
I kind of understand what Bill wants. A point I've made about Andruw Jones is that his defensive numbers might be real, but they're probably measured against a weaker average defensive center fielder than a Willie Mays, so it's not fair to compare the two of them based on rfield/dWar. But he's sort of talking about a descriptive stat, while complaining about value stats and those two aren't really the same thing.
   16. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:16 PM (#5959024)
But JJ wouldnt you have to measure the league as a whole? it might have been weaker defensively in Mays time, I have no idea. But just Mays throwing the average off would not be a huge factor. Nor is it obvious that its an unfair factor.

But bringing up Mays brings up another Jame's quote. I guess it was the Historical Abstract where he defines a "level A" hall of famer as someone who's best at his position, except for the rare care of Exceptional talent doubling up at a position e.g in case of Mays and Mantle. Maybe he was thinking of that comparison when he was making the argument about market share not being a perfect proxy for averaging out talent or whatever.
   17. Kiko Sakata Posted: June 23, 2020 at 04:18 PM (#5959025)
Sunday silence,

My basic point would be that 10 wins and 7 losses is different than 16 wins and 13 losses and that converting these two records into single numbers to compare them is a secondary calculation, not the primary calculation, because the answer depends on the question you’re asking and converting to a single number will lose information.
   18. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:20 PM (#5959026)
Oh hey now that we've brought up replacement level, can I ask another elementary question:

What is replacement level for defense? Is it simply average MLB defense, or something worse?

It seems that lots of people (most?) insist that replacement level defenders are easier to find than replacement level hitters. Its almost like a fundamental aspect of WAR or something. I find that notion kind of strange cause if you could find minor league players who can field an average CF they'd probably be woeful hitters so its all going to come out in the wash anyhow. His value is based on the sum of his def and off so finding lots of MLB average defenders would seem odd because then they'd probably all be bad hitters. So you'd have to find some sort of mediocre hitters, and then in that case their defending probably goes down as well...

SO Im of the school that replacement level fielding should be below average in the same relative amounts as it is below hitting. E.g. -10 runs Off/-3 runs def... BUt most dont think like that. They think its like -10 runs off/0 runs def or something.
   19. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:24 PM (#5959029)

My basic point would be that 10 wins and 7 losses is different than 16 wins and 13 losses and that converting these two records into single numbers to compare them is a secondary calculation,


but what is the value in breaking down wins/losses? That seems so non intuitive. Everyone can easily understand breaking down off/def cause we can see those as two different abilities. But wins/losses are not like that.

Your example makes it sound like that 16/13 guy is some crazy version of Puig who hits HRs and then makes terrible blunders. But that's not what those numbers are telling us is it? Its simply a proxy for playing time? then why not convert to a rate stat e.g. wins/162 games or something?
   20. JJ1986 Posted: June 23, 2020 at 04:29 PM (#5959031)
it might have been weaker defensively in Mays time, I have no idea. But just Mays throwing the average off would not be a huge factor.
I don't mean that Mays threw off the average; I mean that Mays's contemporaries were better defensively than Jones's contemporaries.
   21. JJ1986 Posted: June 23, 2020 at 04:31 PM (#5959033)
rWar is calculated with offense, defense and replacement being 3 separate categories. The replacement component doesn't specifically go with hitting.
   22. Kiko Sakata Posted: June 23, 2020 at 04:34 PM (#5959036)
The numbers you see at BB-Ref are already calculated this way: add up the positives; add up the negatives; subtract the latter from the former. It’s the only way to do it anyway. I’m just saying those two initial numbers matter.

This is, I think, most obvious for fielding, where opportunities do not correlate perfectly with playing time.
   23. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:34 PM (#5959037)
You know I feel like I missed the boat in writing that giant essay about positional bonus or whatever. But isnt that what Bill James is getting at in at least some parts of this essay?

What percentage of a shortstop’s value is in his hitting, and what percentage is in his fielding?
You may have an opinion about that issue, and we have made efforts to measure it in the past—for example, by Win Shares. ''' In a practical sense, the largest problem RIGHT NOW is that we can only guess at the relative value of defense between positions. We’re just guessing; we have no idea.


Right? Isnt this about positional bonus?
   24. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:36 PM (#5959038)

rWar is calculated with offense, defense and replacement being 3 separate categories. The replacement component doesn't specifically go with hitting.


but my question is: Is replacement fielding below MLB average? So that say an MLB average 3bman, is still putting up positive defensive WAR?
   25. What did Billy Ripken have against ElRoy Face? Posted: June 23, 2020 at 04:36 PM (#5959039)
You know I feel like I missed the boat in writing that giant essay about positional bonus or whatever.
Post 7 doesn't count?
   26. JJ1986 Posted: June 23, 2020 at 04:41 PM (#5959041)
Is replacement fielding below MLB average?
I don't think there's quite such a thing. A replacement-level player who is an average offensive player would be putting up negative rfield. A replacement level player who puts up replacement level offense would be an average fielder.
   27. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:49 PM (#5959044)
I don't mean that Mays threw off the average; I mean that Mays's contemporaries were better defensively than Jones's contemporaries.


OK, but i guess I could argue that Jones was facing better pitchers. I guess we'll never know. But lets cut to the chase....

I think what you're getting it is there no way to set an absolute standard for fielding. (correct me if Im wrong) But even if that's true then I would say to simply set the standard as whatever player is the best one at that pt. in time. So say Mays is +80 runs COMBINED off/def. And maybe Jones is +75 combined. So that number then becomes the base against which all other players are measured. Is that any better a measure?

With the reasoning that there is every reason to think that we've got the best player we can find there. There's unlikely to be some super CF kid who prefers to work in accounting at Kenner Peabody or whatever. So we can assume the market place has already found the best player to put in CF.

For the rest of the CFers in the league, perhaps in Mays time 10 were defensive genius and 5 were elite batters and 6 were some crazy extreme but net zero. And in Jones time, maybe only 6 were def genius 8 were elite batters and 3 were one dimensional/but overall mediocre. That sort of fluctuation we might chalk up to random variation and so pay less heed to it.

But we would use the very best player as some sort of standard on the basis that that guy is less prone to be an aberration, its most unlikely that we've failed to find the best CF we can.
   28. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:53 PM (#5959045)
Post 7 doesn't count?


well Im sitting there talking about positional bonus and everyone else is talking about Win Shares, so I feel like I must be not getting the pt of that Bill James essay. But then I see some of the quotes I posted and I feel he is talking about positional bonus. So are these two concepts somehow related?
   29. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 04:57 PM (#5959047)
Lets back up a bit to see what JJ is saying:



I kind of understand what Bill wants. A point I've made about Andruw Jones is that his defensive numbers might be real, but they're probably measured against a weaker average defensive center fielder than a Willie Mays,


Are you saying:

a) we have no way of measuring absolute defense, so its possible that 1990s Ofers arent as good, no way of knowing one way or the other, but just a starting assumption.. OR:

b) because the 90s were a better off. environment its reasonable to assume that managers would punt defense in order to put a better off player in CF... OR:

c)UJones def. numbers are so far off the chart something is wrong, so maybe its the level of competition..OR

d) something else?

Cause I could see any number of things you are trying to say here, and now Im not sure what..
   30. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 05:00 PM (#5959049)
A replacement-level player who is an average offensive player would be putting up negative rfield. A replacement level player who puts up replacement level offense would be an average fielder.


OK in that case, maybe my assumption is correct: A replacement level player is one who's COMBINED off/def would lead to a team with a winning pct of .292 or something.

That's kind of what I was thinking all along was the only way to do it, but people keep saying that replacement level def. is easier to find. Its confusing
   31. Kiko Sakata Posted: June 23, 2020 at 05:17 PM (#5959056)
But isnt that what Bill James is getting at in at least some parts of this essay?


Yeah, sorry, my comments were mostly a response to the title and comments that Bill's made in the past re: Win Shares vs. WAR.

To be honest, this piece was pretty hard to read; very rambly. But yes, he's trying to argue that you need a zero baseline to calculate positional adjustments correctly and I don't get what he's trying to say there at all. Look, I love measuring against a zero baseline, but it doesn't solve the issue of positional adjustments. Unless the idea is that you put all fielders into the same bucket so that you can compare the fielding numbers of a shortstop directly to those of a catcher directly to those of a left fielder. Which - I don't see it. The value of making a play is inextricably linked to how likely the play was to be made. And that's inextricably linked to where the play is made - i.e., at what position.

I guess there's a little glimmer of something where with a system that builds up from zero, you can see, for example, that shortstops made more plays in the 1920s than they did in the 2010s. But I'm not sure how far that gets you to being able to compare Rabbit Maranville to Andrelton Simmons, much less to compare Maranville to Pie Traynor or Simmons to Kevin Kiermaier, and forget completely about comparing the fielding value of Pie Traynor to Kevin Kiermaier.

I mean, James is free to try to construct his own fielding system that builds up from zero and lumps all fielders together into a single bucket. But I would be pretty surprised if the final results made any damn bit of sense.
   32. Mike Webber Posted: June 23, 2020 at 05:24 PM (#5959058)
@11
People prefer WAR for good reason.


Yes, and that reason is accessibility. When Sean Forman approached Bill about putting Win Shares on Baseball Ref, Bill should have said yes. Bill was in the lead at that point, he might have stayed there.
   33. PreservedFish Posted: June 23, 2020 at 05:33 PM (#5959061)
I recognized that James went nuts at least as early as when he claimed to have seen a feral lemur in Boston.
   34. SoSH U at work Posted: June 23, 2020 at 05:36 PM (#5959062)

I recognized that James went nuts at least as early as when he claimed to have seen a feral lemur in Boston.


I thought it was when men showering with young boys was all the rage in Kansas.
   35. Slivers of Maranville descends into chaos (SdeB) Posted: June 23, 2020 at 05:52 PM (#5959068)
Im not entirely sure of what he's getting at in the article. He spends literally half the article using this analogy of a growing boy to illustrate the concept of finding an absolute reference pt. to build our map. But I dont see how this translates into a new method or perhaps some other method but maybe there's a folow up article or something. Maybe Walt or someone can explain that.


This is just the latest in a whole series of articles (or musings) on the concept. For example, this one.
   36. Slivers of Maranville descends into chaos (SdeB) Posted: June 23, 2020 at 05:57 PM (#5959070)
Unless the idea is that you put all fielders into the same bucket so that you can compare the fielding numbers of a shortstop directly to those of a catcher directly to those of a left fielder.


I think that's exactly what James wants. We can compare a shortstop's offense directly to a catcher's but not his defense. Or we can compare Jeter's offense to Honus Wagner's, but not his defense.
   37. What did Billy Ripken have against ElRoy Face? Posted: June 23, 2020 at 06:10 PM (#5959077)
well Im sitting there talking about positional bonus and everyone else is talking about Win Shares, so I feel like I must be not getting the pt of that Bill James essay. But then I see some of the quotes I posted and I feel he is talking about positional bonus. So are these two concepts somehow related?
Oh, sorry, I misread your post in reading too quickly. I thought you had said you missed the boat in not having written a giant essay on positional bonus.
   38. Kiko Sakata Posted: June 23, 2020 at 06:15 PM (#5959080)
We can compare a shortstop's offense directly to a catcher's but not his defense.


Yeah, but that's not a failing of the UZR system or whatever. That's essentially just a simple factual statement. The best you can do is compare the catcher to his catching peers and the shortstop to his shortstop peers and compare the comparisons - e.g., Jeter was 3 wins below average at SS; Piazza was 2 wins below average at catcher - numbers made up for illustration only - so Jeter was 1 win less valuable than Piazza.

I kind of, sort of, maybe could see an argument that Maranville was more valuable than Andrelton Simmons because he made more plays but you still need to also look at the other side of the ledger - Maranville made a hell of a lot more errors, too (he made 65 errors in 1914 and finished second in MVP voting!) - but that's because he played with a teeny, tiny glove on shitty fields. At which point you're right back to - you can compare Maranville to his peers and you can compare Simmons to his peers and you can then compare the comparison.

But none of that really has anything to do with whether it makes sense to measure things against a baseline of zero. It does, but only if you measure wins and losses separately, which then allows you to shift to comparisons against average or replacement level or whatever the hell you want. And you're going to need to do some kind of positional adjustment at some point somewhere in your calculations.
   39. villageidiom Posted: June 23, 2020 at 06:19 PM (#5959083)
What I got from this is that James is looking at the evolution of WAR and recognizing we built it the wrong way. We started with an absolute measurement for offense because that was easiest, then added a positional adjustment to account for an assumed defensive value. (Set aside the adjustment for replacement, for now.) Then we developed defensive measurements of each player relative to the average for their position - which makes sense given we've already accounted for the general value differences among position. But does it, really?

Ignoring offense for a minute, if we were to design a measurement of defensive value we would not develop a positional adjustment first (based on offense!) and then build a defensive metric relative to the average for the position. We'd define some baseline, and evaluate runs saved relative to that baseline.

I think James is saying that we can measure a player's value in this way, but we should also be able to do the reverse - to measure defense as an absolute, and then add offense to that - and get to the right place. Today we don't, partly because most defensive measurements aren't relative to a common reference like offensive stats are. That, in turn, is because the common reference for offense is easier. The floor of actual offensive production is 0 runs. What's the ceiling of runs saved? It's infinite. To do this approach of measuring defense, we would need to define a practical ceiling, a common reference point from which to make measurements. That's what it appears James wants to do. He'd rather not make the positional adjustment, and then measure defense relative to that. He'd rather just define a baseline, and then measure defense on that basis.

I know people will say, "The positional adjustment is a baseline, dumbass." Yes, but if you were building defensive value metrics from scratch that's not the baseline you'd use. Like, players who switch from 3B to 1B see a big drop-off in defensive performance. But nearly all 3B are right-handed, while right-handed 1B are at a disadvantage to their 1B peers. That's the sample we should use as a foundation for measurement of defensive value? No. Almost everything you can read on the positional adjustment in the last 20 years is filled with caveats that it's highly problematic but either (a) we're using it anyway or (b) we can come up with something better later. James is saying later is now.

I think there's some merit to his general proposal of starting over with defensive value. I mean, I think we should start over with offensive value, too, as BIP outcomes are the product of offense and defense, yet we credit hitters based on the outcome nonetheless. With launch angles, shifts, etc., we have the ability to measure behaviors and actions that create value; we're still assigning value to outcomes instead of actions. We are so used to measuring things in the best framework we could have used 40 years ago that it doesn't occur to people that a different framework can be used.
   40. Slivers of Maranville descends into chaos (SdeB) Posted: June 23, 2020 at 06:20 PM (#5959084)

Yeah, but that's not a failing of the UZR system or whatever. That's essentially just a simple factual statement.


It's a factual statement given the current approach to measuring defense. Whether it's a statement true for all times and places is an open question, I think.
   41. JJ1986 Posted: June 23, 2020 at 06:36 PM (#5959087)
I may just be repeating Kiko's work (sorry) and I just thought of this, but we count (in basic terms) runs allowed and outs recorded for pitchers. We kind of count runs scored and outs made for hitters. Why can't we count outs recorded and runs allowed for fielders. Shortstops and catchers would have more of each. Players in lower-strikeout eras would have more of each.
   42. John DiFool2 Posted: June 23, 2020 at 06:55 PM (#5959089)
Holy ####, the first half of this article. Is James totally nutty now?


I didn't know about the lemur story, but his dribblings on politics are enough to demonstrate that.

Two, WAR accounts for negative numbers, while James always tried so hard to avoid either negative Win Shares or any concept of Loss Shares.


Correction: he made a beeg deal about working Loss Shares into his scheme, but at some point he gave up on the notion.
   43. Walt Davis Posted: June 23, 2020 at 07:18 PM (#5959093)
Too much to respond to. I've never really tried to follow James's arguments on this, they always seemed a bit pointlessly abstract and just not worth the effort. So sorry SS, I can't tell you what he's up to and I probably disagree with him anyway.

James hits on one of the issues but then seems to argue against it. In the real world, "zero" is gnerally an arbitrary choice. Temperature is the closest we have to an every day example -- the "temperature" is the same whether you measure it in Kelvin, Celsius or Farhenheit. Celsius is Kelvin with a different zero. Fahrenheit is Kelvin with a different zero and re-scaled. Issues arise once you start doing ratios -- it's about 15 C today so it's 288 K ... tomorrow it might be 20 C which is a 33% increase or 293 K which is a <2% increase. So sure, technically speaking, you are in trouble if you say a 4 WAR player was twice as valuable or productive as a 2 WAR player but so what -- you're on solid ground if you say the difference is 2 wins (given your scale for "wins"). For C and F, there's also the issue of what to do when 0 is in the denominator. But does it really mean anything to say that 250 K is 25% "warmer" than 200 K? That probably depends on whether you're an atom out on your own or a collection of atoms taking human form.

The surveying analogy seems an odd one to use in defense of absolute zero. Yes you need a zero point but that's just so you can measure everything else relative to that zero point, it's not absolute -- it doesn't matter where you put that zero point, whether it's Lawrence, KS or Springfield, IL. And if you're trying to lay out a new road in Mittagong, NSW do you really need to know some zero point in the middle of Australia? At one point, BPro tried to stick zero somewhere around the point of "replacement-level" hitter and "replacement-level" defender -- i.e. -2 bWAR. It just led to silly-looking numbers.

I suppose you could think of a 2-dimensional space where y is offense and x is defense. The points (players) will be wherever they are (assuming we can agree on a unit of distance) ... and then you can put your (0,0) point wherever you want, it won't change the distribution of the players. You can even do a linear re-scaling of the units of distance and the "shape" of the distribution will be the same. It's more about ease of interpretability or aesthetics then -- what is the point of comparing to a team that would score zero runs and give up 1500 runs since we're never going to see anything remotely close to that. (Emerging stat nerds should look at "rotations" for some fun.)

(Another issue here I suppose is that runs and wins are not continuous but discrete (we do have true zero) yet we are forcing continuous distributions onto them but this is essentially always done whether directly or indirectly.)

The argument on "markets" strikes me as similar to Dan R's work although he gets there in different ways. He's argued that Dave Concepcion belongs in the HoF because (most definitely not Dan's words) the other SS of his era really sucked. And if you limit that comparison to SS of his era then argue that his position relative to them should be compared with SS from other eras or players at other positions relative to their position-era specific comparison groups, he's got a case. By the way, I would not believe for a second that Mays' contemporaries on average were better defenders than Andruw's in an absolute sense.

Much of the rest seems to basically be examples that, pushed to their theoretical extremes ... or maybe even just to the edges of what we observe or a bit beyond ... many systems of measurement or models break down. Yes, the Cubs' lineup will be the same regardless of what position we stick Javy Baez at (say LF) and what position we stick Kyle Schwarber at (say SS). By "dWAR theory" the defense should work out the same as well because Javy will be such an amazing LF that he will compensate for Schwarber being nearly as bad as Jeter. :-) We know (i.e. very strongly suspect) that the Cubs' defense would actually be much worse off in that alignment -- which is why no baseball manager in the entire history of baseball other than Joe Torre et al have ever gone with such an alignment. (The most obvious non-extreme (?) is DH which literally anybody can play.)

Another issue is that offensive talent and defensive talent appear to be correlated. That is, based on what I just said, we assume there is something special that guys allowed to play SS have substantially more of than other players. But there's no fundamental reason that special something makes them worse hitters (on average) -- and of course in the case of ARod, Banks, etc. they weren't -- but that's what the data very strongly suggests. One of the most constant things over time (possibly changing!) is that C and SS are the worst hitters. So we assume that there's a minimal level of defensive competence required and, for whatever reason, meeting/exceeding that level of competence usually means you have less offensive skill. What's less clear is why sacrifice offense for defense at some positions and not others -- presumably the number of plays, the number of difficult plays, that there's less return to defensive excellence at some positions than others.

I think that last point is what James is trying to get at in talking about "what proportion of a SS's values is defense vs offense?" Here we do have two extremes in actual reality -- nobody really cares how well a pitcher hits and barely cares how well he fields and (at least for that game) nobody cares how well a DH fields. (And of course nobody cares how well the DH or any of the 8 field positions can pitch.) So there is something there -- maybbe 75% of a C's role is defense while 40% of a LF's role is defense and that's where dWAR and positional adjustments break down. It probably is easier to think of separate "markets" there -- you have to reach a certain level of (very particular) defensive competence to play C so it would seem to make sense to compare them only with other Cs. The counter-argument is that every team has a pair of players that have achieved that level of competence so that value zeroes out when we compare two teams or two Cs. It may not zero out when we compare C to 1B (one of James's points I gather) but, in terms of "value to team" (which is what we claim to be measuring), what matters is that your C was 1 win better than the average C and your 1B was only average. Anyway, you hope the positional adjustment picks up the C-1B real world difference but that nobody takes dWAR so seriously as to shift Nelson Cruz to C. (Note the fact that MLB has 60 guys with at least that minimal level of competence and most have another one stashed at AAA suggests it may not be that hard to find/train players to that level so it's not really that valuable of a skill.)

Things are not so extreme at SS -- they are not doing anything radically different than what 2B and 3B are doing and many of the players at the other positions played SS in HS, college or the pros. So it's not really a different skill set, it is likely more that they tend to max out on each of range, reaction time, hands and arm instead of just 0-3 of those. But still, there's an argument that it only makes sense to comp them to other players who rate at least a 6 on all 4. (In reality, some of those other players are probably at 3B and 2B ... and some are guys who throw LH and so probably play CF or maybe never hit well enough to make the majors. If you are a LHT with Belanger's skill set, your max upside is Billy Hamilton and more likely Tony Campana.)

But James, Kiko, Sean, the nerds at BPro and fangraphs have given all of this a lot more theoretical thought and a lot more data grunt and will have a much better idea of how any of this matters than I do. But statistically speaking, put zero where you want and make any linear scale transformations you want, just make sure you know where zero is (and don't divide by it), what scale you're on, take care with ratios and comparisons.
   44. . . . . . . Posted: June 23, 2020 at 07:55 PM (#5959104)
The argument on "markets" strikes me as similar to Dan R's work although he gets there in different ways. He's argued that Dave Concepcion belongs in the HoF because (most definitely not Dan's words) the other SS of his era really sucked.


Dan R can speak for himself, but since I played a (very secondary) role in helping to flesh out his arguments and was a groomsman at his wedding, I'm going to take a run at fixing your misunderstanding of his argument.

Dan's argument is both the AVERAGE offensive value and the VARIANCE in offensive value from a position change with time (and with respect to all hitters collectively as a pool), and those changes are meaningful and not just random walk. I.e. the if there's a 15 year period where SS offensive value goes into the shitter and no one is an offensive star, that's not just because no good offensive shortstops were born for 15 years, its because there was something intrinsic to the defensive demands of the position that made it harder to excel relative to your peers AND lowered the average. So the only fair way to judge offensive value is with reference to the median and standard deviation of the position because otherwise you're assuming that, say, the defensive demands on a 3B in 1900 were the same as 3B in 1950 and 3B in 2000, which is manifestly - almost laughably - untrue.

Now, departing from Dan's arguments to my arguments:

That is, based on what I just said, we assume there is something special that guys allowed to play SS have substantially more of than other players. But there's no fundamental reason that special something makes them worse hitters (on average) -- and of course in the case of ARod, Banks, etc. they weren't -- but that's what the data very strongly suggests.


This is PROFOUNDLY untrue. There is strong - OBVIOUS - data that shows that different defensive positions impose constraints on the size of the player, and size and offensive variance are clearly correlated and probably average value as well. Put differently, big guys hit ball harder. So positions that can be played adequately by a big guy will have higher offensive value and, more importantly, more outlier offensive seasons of the sort that only an Aaron Judge (RF) or Mark McGwire (1B) sized dude could produce. Being able to play SS or 2B adequately at a size that allows the player to hit for mega power is a rare skill, and is, yes, more valuable than mega power at 1B.

Importantly, this is separate from the mere factor of selection bias - at a position where defensive ability is more leveraged, players are selected more strongly for defensive aptitude than at less leveraged positions, which means that the mean offensive performance will go down even though the variance might not (because of outliers like a young A-Rod or Ripken).

So to try to sum this up, two demonstrable factors meaningfully change the offensive performance from a position. First, whether a position selects more or less strongly for defensive ability, which skews the median but not the variance. Second, whether the demands of a position limit the physical shape of the player, which affects both variance (since the most common effect is to limit the max height/weight of a player who can cover the position) and median (since a position that requires smaller players preferentially removes the big guys who are disproportionately responsible for outlier offensive seasons out of the pool).

This was discussed 12+ years ago in connection with the HoM, it was true then and for the most part, it remains true today. And it makes calculating defensive value devilishly complex, because it is not entirely independent of offensive value.

There have been some updates though. Most of these are speculative, because I'm now a grown ass man with a grown ass job and a family and cant spend time crunching data like I used to. Think of them as educated guesses.

One thing we learned over the past few years is that there were two components in why shorter/smaller players are both worse offensively on average and have less variance. First, smaller players tend, as a rule, to generate less exit velocity. This is immutable.

But also, before launch angle was measured and trained for, taller guys tended to hit with more of an uppercut. Because (unbeknownst to us all) launch angle had a strong impact on offensive value and tall guys naturally hit with more of it, shorter players were naturally worse offensively than they should've been, even accounting for the exit velocity difference. That has now changed, because short guys measure their launch angle and train away the difference with the tall guys. So with all players hitting with near optimal launch angles for their exit velocity, the difference is more purely exit velocity. So, the difference in mean and variance of offensive value between positions that demand smaller players and taller players has narrowed in that respect.

But also, teams have gotten better at optimally leveraging defensive talent. This started 40 years ago - there was never really no need for teams to limit SS to the pool they thought they did in the 1970s, even with old turf, higher GB rates, more SB and more BIP increasing the defensive demands on SS. But it continues apace. This has, IMO, compressed the defensive spectrum for positions outside of catcher, which is sort of sui generis and deserves its own rambling, long-winded post. Finally, better physical condition and younger average age has increased the ability of bigger guys to cover high-spectrum defensive positions. This also has compressed the spectrum a bit.

But at the end of the day, big guy hit ball hard, and a big guy with ++ contact skills will be the best hitter possible, and that guy almost always will end up in the OF or 1B. And if youre the exception to that rule, you're Alex Rodriguez, and you're an inner circle HOFer.

   45. Zach Posted: June 23, 2020 at 08:01 PM (#5959108)
But putting zero wherever you want means that it is impossible to say what fraction of a shortstop's value comes from defense vs offense, which is the kind of question James is interested in.

Maybe we can rephrase his point:

We all know that simply having the ability to play at shortstop and not embarrass yourself has a significant amount of value. The all glove, no hit shortstop is a cliche for a reason, and at different times in history almost every shortstop has been of that type. Even now, a utility infielder who can play short will get paid better and stick around longer than a utility infielder who can't.

But if you measure defense relative to replacement, you end up saying that a replacement level fielder has almost no value. There are lots of veterans who play an entire season and earn 0-1 WAR. If you use the Fangraphs approach and multiply WAR by WAR/dollar, you end up saying that those guys are overpaid. You'll rearrange your roster, get rid of utility infielders, and spend that money on interchangeable relievers (since everybody knows that the ability to pitch every few days and not embarrass yourself is tremendously valuable -- in part because pitching stats tend to be counting stats that measure relative to zero).
   46. Zach Posted: June 23, 2020 at 08:10 PM (#5959111)
Or consider the kind of "two cats for a dog" trade that you see every year. Most of the value in those trades isn't coming from getting guys who will excel. It comes from getting guys who can fill a role well enough to get playing time.
   47. Zach Posted: June 23, 2020 at 08:17 PM (#5959112)
With James, you always have to remember that he's as much a historian as a sabermetrician.

In the 1950s, 1960s, and 1970s there were many shortstops in baseball, like Roy McMillan, Mark Belanger, Ray Oyler, Dal Maxvill, Roger Metzger, Ed Brinkman and Bobby Wine, who were able to keep their jobs for years and years although they would often finish the season hitting .220 with zero home runs. Now, we don’t really have any players like that, but instead we have some shortstops who hit 30 homers a season.

But are the shortstops now making a defensive contribution equal to the shortstops of 1960, or have we substituted offense for defense at the position?


This is exactly the kind of question he loves writing about, but it's not directly relevant to roster construction, so it doesn't get studied as much as launch angle or spin rate.
   48. Rally Posted: June 23, 2020 at 09:14 PM (#5959120)
“I think that's exactly what James wants. We can compare a shortstop's offense directly to a catcher's but not his defense. Or we can compare Jeter's offense to Honus Wagner's, but not his defense.”

Why not? Comparing SS to C defense is really hard, different skills needed. But for 2 SS a century apart, I think comparing their offense and defense both present very similar problems. Mostly, the change in quality of competition.
   49. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 09:29 PM (#5959123)
Its a really interesting discussion and I don't have time to respond. Its almost like a Rohrschach test for primates everyone describes the model they want to see ro think they see
   50. bookbook Posted: June 23, 2020 at 09:41 PM (#5959125)
Bill James is the one who figured out players have their best year usually between ages 26-28, right? Is it offensive to suggest he’s not a replacement level sabremetrician anymore? (His politics writings have always been awful.)
   51. Slivers of Maranville descends into chaos (SdeB) Posted: June 23, 2020 at 10:00 PM (#5959128)
I think comparing their offense and defense both present very similar problems. Mostly, the change in quality of competition.


Knowing the quality of competition only matters if you want to decide who has the higher 'true talent level' (which, ironically, is generally thought of as an absolute vs. relative scale).

But to James' question -- was the value of Wagner's offense relative to his defense higher or lower than Jeter's -- I don't think you need to know that at all.


James hits on one of the issues but then seems to argue against it. In the real world, "zero" is gnerally an arbitrary choice.


I see what you're saying, but isn't 'zero' the defensive value that a DH provides?

I look at a lot of systems, and they all assign a defensive value to DHs that isn't zero. Sure, you can point out that it's not a good defensive value, but it always irks me. It's as if we decide to assume David Ortiz had 6 passing TDs and 12 interceptions in a given season 'to make the numbers work out.' It isn't really persuasive to say 'but that's really bad!" And even more so when you use it as part of an argument Ortiz was a better QB than Josh Rosen.
   52. Sunday silence: Play Guess How long season lasts Posted: June 23, 2020 at 10:55 PM (#5959140)
I think Walt really nailed it when he pointed out that James's geographical analogy is terribly misplaced. Ie its not important where you put zero on the map it won't change any thing..
   53. Kiko Sakata Posted: June 23, 2020 at 11:18 PM (#5959146)
Okay, I'm still not sure exactly what to do with this information. But, since I have a system that builds up from zero I put together some tables. Full results here.

What I did was add up all of the eWins earned by players when they're in the game as, say, shortstops - just wins, not losses - and calculate what percentage of wins were earned on offense vs. what percentage were earned while fielding. The first table includes all of my data (which go back to 1918 - I use Retrosheet's play-by-play data). Then I repeat the table by decade.

The numbers don't really work for catchers. I don't have any game-calling or pitch-framing or anything like that - just stolen bases, passed balls, and balls in play - so catchers don't really earn many fielding wins.

Here's the % of wins that are fielding over the entire history, for the 1920s (my first full decade) and the 2010s, ignoring catcher. You can see the results for catchers and the intervening decades at the link.

1918 - 2019
1B 15.4%
2B 32.0%
3B 27.6%
SS 36.3%
LF 32.1%
CF 30.7%
RF 31.4%

1920 - 1929
1B 15.6%
2B 33.1%
3B 30.5%
SS 39.9%
LF 34.4%
CF 33.1%
RF 33.1%

...

2010 - 2019
1B 15.7%
2B 29.7%
3B 26.1%
SS 31.9%
LF 31.1%
CF 30.0%
RF 31.5%

I guess it kind of works to arrange infield positions in a defensive spectrum. But it doesn't really seem to do so for outfielders and, for example, you don't see the switching of the relative importance of second base and third base that Bill James built into his Win Shares and that you can see if you calculate positional adjustments based on player offense (here's a 50-page PDF that I've written on the subject - it has some colorful graphs).

I don't know. I'm still not buying that shifting to a zero baseline gets you anywhere in terms of trying to compare players across different defensive positions.
   54. greenback used to say live and let live Posted: June 24, 2020 at 12:17 AM (#5959159)
Kiko, is the reduction in defensive value roughly aligned with changes to K-rates and BABIP? It's actually smaller than I would have guessed based on the change in K-rates (went from 8% to 23%, I think).
   55. Sunday silence: Play Guess How long season lasts Posted: June 24, 2020 at 08:48 AM (#5959191)
I find it hilarious that in any discussion such as this inevitably the guy trying to create the defensive adjustment just has to admit that pitchers don't count in the defensive spectrum. Cause obviously he cant compare the defensive ability of the pitcher to anyone else.

But the pitcher is playing the same game as the SS or Catcher, he's out there on defense just like they are. If you can't provide a defensive adjustment for the pitcher you should just admit the entire exercise is doomed.
   56. Mefisto Posted: June 24, 2020 at 09:44 AM (#5959201)
"you don't see the switching of the relative importance of second base and third base"

That really jumped out at me. If 1B and 3B don't change as you go back from 1918, that will be another big hit to "conventional wisdom" (which I never bought).

More generally, I can think of 2 ways to approach the idea of a "zero point", one theoretical and one based on modeling:

1. Assume that a defense turns all BIP into outs, that is, its Defensive Efficiency is 1.000. Then allocate the actual runs allowed to individual positions.

2. For the StatCast era, create a model such that an average pitcher takes the field with the best defenders at each position. These probably won't be the all-time greatest, but the best for which we have StatCast data. Now run a simulation which tracks the relevant location of all BIP (velocity, direction, location, etc.) and create the probability that the good defenders make a putout. That would establish a "for practical purposes" zero point: all runs allowed which they can't prevent are treated as intrinsic to the game.
   57. Kiko Sakata Posted: June 24, 2020 at 09:54 AM (#5959205)
Kiko, is the reduction in defensive value roughly aligned with changes to K-rates and BABIP? It's actually smaller than I would have guessed based on the change in K-rates (went from 8% to 23%, I think).


Mostly. I also allow the pitcher/fielder split to vary over time although I'm not sure there's an obvious trend there - but there could be a subtle underlying trend in either the distribution of balls in play or the types (more infield pop-ups at some points, which are mostly a pitcher thing).
   58. Kiko Sakata Posted: June 24, 2020 at 09:56 AM (#5959206)
I find it hilarious that in any discussion such as this inevitably the guy trying to create the defensive adjustment just has to admit that pitchers don't count in the defensive spectrum. Cause obviously he cant compare the defensive ability of the pitcher to anyone else.

But the pitcher is playing the same game as the SS or Catcher, he's out there on defense just like they are. If you can't provide a defensive adjustment for the pitcher you should just admit the entire exercise is doomed.


I don't understand what you're saying here.
   59. villageidiom Posted: June 24, 2020 at 10:57 AM (#5959220)
Let me pose it in a different way. If the average runs scored by the average team in the average game is 4.50 in a season, we've been essentially calibrating all offensive stats to reflect that, on average, teams are creating 4.50 runs per game on offense. Easy.

Now let's talk defense. If on average 4.50 runs are scored per game per team, how many runs have been prevented?

James is suggesting that we can't, as a practical matter, say infinite runs have been prevented. (i.e. measuring height by starting from the sky.) Infinite runs have infinite value; and if that's the case on defense then any finite number of runs created on offense has relatively no value. We have to define a reference point and measure from there, but the same issue exists: the choice of reference point could inflate or deflate the value of defense compared to offense.

I think he's saying we should be able to build a start-with-defense framework with an absolute reference point that will get us to the same overall value as we see today in a start-with-offense framework. And then once we have that reference point, then we can measure offense as an absolute, and defense as an absolute, and throw out all the other stuff like positional adjustments.
   60. villageidiom Posted: June 24, 2020 at 11:12 AM (#5959222)
I find it hilarious that in any discussion such as this inevitably the guy trying to create the defensive adjustment just has to admit that pitchers don't count in the defensive spectrum. Cause obviously he cant compare the defensive ability of the pitcher to anyone else.
The defensive adjustment arose to bring positional value into a framework built on offense. Nobody cares about a pitcher's ability on offense in discussions of value.

Pitchers exist on the defensive spectrum. People generally don't bother trying to pinpoint where pitchers are relative to other positions on that spectrum because there's little practical value in doing so.

On that note... If a 3B isn't playing the position well, and the team wanted to shift him to a different position, would you recommend pitcher?
   61. Kiko Sakata Posted: June 24, 2020 at 11:20 AM (#5959223)
I think he's saying we should be able to build a start-with-defense framework with an absolute reference point that will get us to the same overall value as we see today in a start-with-offense framework. And then once we have that reference point, then we can measure offense as an absolute, and defense as an absolute, and throw out all the other stuff like positional adjustments.


For offense, you can build up offense, since we're talking Bill James, let's call that "Runs Created". But to evaluate that number you need an offset. In his New Historical Abstract, James would frequently write about a guy who, say, "created 98 runs while making 367 outs". The problem is those two things are different scales - runs vs. outs. So, step one is to put them on the same scale. You can convert outs to runs via linear weights or something or, as I do, you can calculate both sides directly as wins and losses.

Having done that, the defensive side falls out naturally as a pure accounting problem. If Pedro Strop allows a double to Yasiel Puig, whatever the runs created by Puig for that are the same as the runs allowed by Strop (and his fielders - this is the tricky bit on the defensive side).

So, in the same way "Roy White created 95 runs while making 414 outs" leads to a way to measure Roy White's offensive contribution, so does "Rawley Eastwick allowed 65 runs while generating 235 outs". Again, with the non-trivial caveat that some of the debit for some of the runs allowed and some of the credit for some of the outs made while Eastwick were pitching really belong to the Reds' fielders rather than Eastwick. (note: I made up all of the numbers in this paragraph; don't go looking to see what specific seasons I'm comparing.)

Personally, I think all of this is easier if you calculate everything play by play and convert to wins and losses as you go. But in theory, one could calculate linear weights on a pitcher's "batting against" line in the same way that one calculates linear weights on a batter's batting line at the seasonal or even career level.

But I still don't see what any of that has to do with positional adjustments. Shifting to a zero baseline doesn't automatically put all of your fielders into a single bucket and I don't really see why you couldn't put all of your fielders into a single bucket while keeping your zero point at average as most fielding systems currently do. (And perhaps more importantly, I personally don't think you should put all of your fielders into the same bucket. I think that positional adjustments are something that you want to be able to measure explicitly, not bury them in the fielding numbers and hope they work out okay.)
   62. Slivers of Maranville descends into chaos (SdeB) Posted: June 24, 2020 at 11:27 AM (#5959224)
I think Walt really nailed it when he pointed out that James's geographical analogy is terribly misplaced. Ie its not important where you put zero on the map it won't change any thing..


I think James is well aware of the fact that you can put 'zero' in more than one place -- that's why he's thinking about measuring defense 'from the sky down' rather than 'from the ground up.'

What James is saying is that there has to be a well-defined zero. If I say point A is 4% higher than point B -- that tells me nothing about whether we're discussing mountains or a desert, nor whether we're measuring from mean sea level, from average ground level, or from the center of the earth.
   63. JJ1986 Posted: June 24, 2020 at 11:31 AM (#5959225)
I was reading a book last night and a man was described as "shorter than average" height and somehow I was able to understand what that meant.
   64. PreservedFish Posted: June 24, 2020 at 12:09 PM (#5959240)
BR gives Greg Maddux zero direct fielding credit. He led the league in either assists or putouts about 20 times. In range factor 14 times. Obviously a lot of that is his pitching, not his fielding, and it's likely impossible to distinguish between the two, and that his own fielding ends up improving his pitching numbers. Interesting thing to look at. I haven't given it much thought.
   65. DL from MN Posted: June 24, 2020 at 01:11 PM (#5959249)
people keep saying that replacement level def. is easier to find


This is true for the easier defensive positions simply because you can use your utility infielder at pretty much anywhere except C and P. P and C have a scarcity issue because their defensive demands are entirely different from all the other fielders. SS and to a lesser extent CF also have some scarcity but generally you can find a AAA player who can play defense at SS or CF but not hit because finding athletes who can field is easier than finding players with that athletic skill who can hit.

I do find it interesting that the defensive demands on fielders have almost certainly been shifted to pitchers and catchers over the years as strikeouts have increased but if you look at WAR the measurement capability of defense has improved in the opposite direction which makes it look like Andruw Jones was much more valuable than Willie Mays or Max Carey despite handling fewer chances.

Regarding catchers, I think the primary defensive responsibility that makes them different is handling the running game. The difference in linear weight between a triple and a single is massive. You can't have a big league catcher who has a pop time that is so poor that all walks and singles turn into triples in two pitches. Catchers with bad pop times get moved off catcher.



   66. villageidiom Posted: June 24, 2020 at 02:01 PM (#5959271)
And perhaps more importantly, I personally don't think you should put all of your fielders into the same bucket. I think that positional adjustments are something that you want to be able to measure explicitly, not bury them in the fielding numbers and hope they work out okay.
But we suck at measuring positional adjustments. So why not consider doing it without them?

We don't make batting-order adjustments, even though hitters are selected for batting order spots based on their abilities. Actually, using batting order adjustments could be a way to attempt what James is talking about here. For example:

1. Build defensive value metrics independent of offense, but relative to some number (N).
2. Include a batting order adjustment to account for variations in defensive value by batting order. For example, defense provided by a #4 hitter is more rare, and thus more valuable, than the same level of defense provided by a leadoff hitter.
3. Measure offense, relative to the average for their batting order position.
4. Aggregate all of the above, and calibrate N such that the aggregation of value is equivalent to what is done today.
5. Toss out 2 and 3, and sum the calibrated defensive value metrics with the existing offensive value metrics (without position adjustment).
6. If you want to know differentials in value by position, derive it from summaries of 5.
   67. Mefisto Posted: June 24, 2020 at 02:05 PM (#5959273)
We don't make batting-order adjustments


We do, but it's implicit. Players who bat higher in the order get more PAs and thus more chances to accumulate WAR. Same with players who don't get pinch hit for.
   68. Kiko Sakata Posted: June 24, 2020 at 02:15 PM (#5959274)
But we suck at measuring positional adjustments.


We really don't.

So why not consider doing it without them?


Because they're needed. Alan Trammell and Mark Grace were comparable as hitters over their careers. They were both moderately above-average defenders at their respective positions. All other things being equal, if you can get their level of offense from your shortstop, you're better off because it's easier to find a decent-hitting first baseman than a decent-hitting shortstop. If you want to compare the value of Alan Trammell to Mark Grace, you have to figure that out somehow. You can't just throw your hands in the air, say, "we suck at measuring this so let's just ignore it" and leave it at that. I mean, you can, but if you're not going to think carefully about positional adjustments and try to adjust for them as best you can, then your system is worthless for comparing players at different positions.

And I would think that somebody of the belief that "we suck at measuring positional adjustments" would agree that it would be better to explicitly present the positional adjustments, so that you can judge them for yourself and adjust them as you see fit, rather than throw all fielders into a single bucket, try to build fielding value "from the sky down" and hope like hell the final results make sense.
   69. Karl from NY Posted: June 24, 2020 at 03:12 PM (#5959289)
Infinite runs have infinite value

Do they? I don't think so. Infinite runs in one game are worth only one win. Infinite runs in a season are worth only 162 wins.

There's never a point where a marginal run has zero value, but the sum of this infinite series does approach a finite limit of value.
   70. Bring Me the Head of Alfredo Griffin (Vlad) Posted: June 24, 2020 at 03:29 PM (#5959297)
Nobody cares about a pitcher's ability on offense in discussions of value.


I care about that, and it’s actually a significant source of irritation to me that Fangraphs doesn’t incorporate pitchers’ offensive value into their WAR totals.
   71. Karl from NY Posted: June 24, 2020 at 04:10 PM (#5959306)
More thoughts:

You can't "start from the sky" on preventing runs (or linear weights or any other discrete events), since subtracting any counted amount of those from infinity still leaves infinity. What really happens is that each of the three outs in an inning also prevents a third of the infinite potential, so after you record three outs you're back to finite. But then every out has all its value in the infinity-prevention, but of course the finite runs are what decide a game.

I think this is intractable, there is no way to talk about runs saved against zero that makes sense.

Think about evaluating this for cricket in terms of a single batsman. That takes away the contextual dependency between multiple hitters and between the three outs in an inning. How many runs does retiring a cricket batsman save? The answer can only be given in terms of expected value. A play that retires Donald Bradman saves more runs than a play that retires J. Random Bowler. You could decontextualize the retirement to an average across the lineup or the league, but whatever you do has to be relative to some nonzero and noninfinite expectation.

This might be the simplest way to describe it: You can judge offense relative to 0. Defense is the inverse of offense. The equivalent is trying to judge defense relative to 1/0.
   72. Walt Davis Posted: June 24, 2020 at 06:49 PM (#5959336)
Too much to catch up on. Minor point I noticed on the way by, sorry if already addressed: One reason not to worry about pitcher defense is that you will (on average) capture its effect in the credit you give to the pitcher for preventing runs. If Greg Maddux fields better than David Wells then, all else equal, Maddux will give up fewer runs than David Wells. It's a black box approach and potentially you could separate out the pitching and defensive values but you're still capturing the total value.

On Kiko's interesting positional stuff ... I thought the 2B-3B thing was mainly a 19th C and certainly pre-1920 thing, in the age of bunting. There' snot enough pre-1920 info there to find that shift I wouldn't think. As to modern roster construction ... sure there are a couple of reasons to think that the offense/defense mix at SS has changed and/or that SS have declined in absolute defensive value -- lower BIP rates mean there are fewer plays to make and the emphasis on HRs and EV suggest you want larger, more muscular hitters which suggests heavier SS** which suggests less range.

As I've pointed out many times, the veracity of weights aside, Banks was listed at 6'1", 180 lbs. Javy Baez is listed at 6'0", 190 lbs. The physics of hitting HRs hasn't changed and Wrigley has barely changed (the ball may have changed a lot), so there's no particularly good reason to think that Javy shouldn't be hitting 30 HRs a year. That there's a difference across eras, it's that Corey Seager is listed at 6'4", 215; so is Carlos Correa; Lindor is roughly the same size as Banks; Xander listed at 6'1", 210. Kessinger was 6'1", 170; Belanger listed the same (he seemed shorter); Bud Harrelson 5'11", 160; Concepcion 6'2" 155. (Sure, those may be rookie weights and Concepcion probably added 20 lbs through his career ... still leaves him 40 pounds shy of Seager and Correa at the start of their careers.)

You need to adjust for contact and preferably G/F too but, in his prime, Concepcion was getting 450-500 assista per year. Xander has been pretty healthy and he usually doesn't touch 350, career high of 429. Lindor's career high is 447 and he's got a couple in the high 300s. As a SS, Javy has averaged about 425 per 150 starts. So we seem to be talking about 50 to 100 fewer plays made by a starting SS. That isn't necessarily reflective of absolute defensive ability but it certainly would seem to speak to value of a player's defense. (That's something I don't get about Kiko's numbers ... the number of defensive plays has declined so it seems to me that defense relative to offense should be down at all positions.)

** At someplace like 1B, we might have seen replacement of fat with muscle so the weight could stay about the same over time. But the SS of my youth were genearlly either short or really thin.
   73. villageidiom Posted: June 24, 2020 at 06:51 PM (#5959337)
Alan Trammell and Mark Grace were comparable as hitters over their careers. They were both moderately above-average defenders at their respective positions. All other things being equal, if you can get their level of offense from your shortstop, you're better off because it's easier to find a decent-hitting first baseman than a decent-hitting shortstop.
This comparison tells us nothing about how valuable Alan Trammell and Mark Grace actually were. It tells us how players who switched positions generally fared, but nothing about those two players specifically. It doesn't penalize Trammell for being right-handed, nor Grace for being left-handed, neither of which would have been an asset at the other's position. It does, however, penalize Alex Rodriguez when he went to the Yankees, with an implicit assumption that he suddenly was likely not as competent at SS as Derek Jeter. After all, if he were competent at SS he would have been deployed there.

Alan Trammell would likely have been a worse first baseman than Mark Grace. The notion that there are loads of decent-hitting first basemen out there makes it even worse, as it means Alan Trammell would have been a worse first baseman than loads of guys. In fact, had Alan Trammell been put at 1B his entire career, never playing SS, you would argue that he was a less valuable player than Grace. After all, if he was at 1B, it must mean he was incapable of playing SS, just like Grace. He hit about the same as Grace, and would have been a worse defender at 1B, therefore he's a worse player. Can you imagine taking a player who's a worse 1B than Mark Grace and deploying him at SS?? Clearly you'd be better off deploying Mark Grace at SS!

Before you say, well, Trammell wasn't put at 1B, because he was capable of playing SS... OK, then, was Lou Whitaker capable of playing SS? He played there about as much as Mark Grace did. If we're going to ascribe value to the players who could play SS competently because of positional scarcity, how can we justify only considering the players who played the position as the ones who would be competent at it? Doing it that way likely overstates positional scarcity.

When we determine value of offensive achievement we do not consider context. A solo HR and a grand slam are considered to have the same contribution to offensive value. We care about what the player did, and remove the context from which he did it. That context matters, of course, but we've chosen to ignore it because what was in the player's control was how far he hit the ball, and that has measurable value. Well, on defense a player has no control over what position he's assigned. He can control his reactions, sprint speed, sense of direction, glove work, arm strength, arm accuracy, instincts/knowledge, and maybe his positioning. Those have defensive value to a team regardless of the context in which he actually used it. If a team has two players who hit well and can play SS equally well, one of them will be playing somewhere else. But he's not less valuable. He is producing less value in the context he's assigned. So is Mike Trout if he were batting 9th. But if Mike Trout hit the same from the 9th spot in the order as he does higher up, he will have contributed less to run-scoring, and to wins... but his WAR would be essentially unchanged. I'm arguing Bill James is pushing toward the same concept for defense.

This is going down the path of "Ichiro could have played a capable SS if he wanted to," but... maybe he could have? And if so, doesn't he warrant greater defensive value to a team than another RF who couldn't - even if Ichiro was never put in the context of SS? As you noted, a player is more valuable if he can produce offensively as a competent SS. But you're only acknowledging it if he does. You might as well apply a batting order adjustment to defense, because if you can get SS-level defense out of a cleanup hitter that's more valuable than getting it out of the 9th spot. Cleanup hitters are hard to find; 9th-spot batters are so easy to find you can throw a pitcher in there.

The positional adjustment assumes too much about what a player is capable of, by virtue of the position he was assigned. That's not measuring defense. It's assuming defense. Again, why not measure defense by measuring defense? Why start with a positional assumption? I get why we started with it long ago, and it made sense then. We're better than that now.


Do they? I don't think so.
The way we determine WAR, they do. We don't contemplate in WAR the clustering of offensive events in one inning or one game as being any different from spreading them out over a longer period. A walk contributes a specific fraction of runs, regardless of the context in which it happened. If it happened in an infinite inning, we still say it contributed to runs - and wins - the same as any walk in any other inning.

Obviously at the extreme that's wrong, as you note. But I think that's part of James' point. A team could have given up infinite runs, if not for the efforts of pitching and defense to minimize it. But that's impractical as a starting point. If we want to count runs saved, we shouldn't count down from infinity; but choosing a ceiling as a reference point is harder than it is to choose a floor like we do for offense, so let's pick a ceiling for defense that, when combined with offense, produces sensible numbers in the aggregate. We can use the offense-first method to calibrate a defense-first method's ceiling.
   74. Karl from NY Posted: June 24, 2020 at 08:59 PM (#5959348)
We can use the offense-first method to calibrate a defense-first method's ceiling.

That's all correct -- but that ceiling isn't the zero that James is looking for, it's some kind of nonzero baseline.
   75. Walt Davis Posted: June 25, 2020 at 12:53 AM (#5959381)
It does, however, penalize Alex Rodriguez when he went to the Yankees, with an implicit assumption that he suddenly was likely not as competent at SS as Derek Jeter. After all, if he were competent at SS he would have been deployed there.

How so? In "dWAR theory" he suffers not at all -- he gets hit for moving to 3B (Rpos) but if he does have the skill for SS then he should be an above-average 3B, even moreso than at SS (Rfield goes up). His value might go down a bit due to fewer opportunities at 3B but that was a real world decision the Yanks made in moving him there. (Or as Chris Dial suggested yonks ago, I think back in the Brosius era, the Yanks compensated for Jeter's lack of range by having good 3B who played further off the line so could cover more of the hole.) It's sort of always the case that if you get deep enough into the weeds, you will find ways in which the WAR assumptions about how value works don't fit reality ... but those gaps are in the margin of a couple runs here or there and, for any given player, you probably find as many areas where they're under-valued a bit as where they are over-valued a bit.

Rfield alas bounces around a lot but from 25-27 in Tex, ARod comes out at 0 Rfield (in the not helpful pattern of -9, 1, 8); in his first year at 3B for the Yanks, he was +14. He had as much dWAR at 28 as at 27. But Rfield being Rfield, the next year he (now) is 0 at 3B followed by a -13. So who knows what was going on. He was average-ish from that point on but then he's in his 30s by then and it would be foolish for us to assume that ARod would have still been a good defensive SS in his 30s (though probably better than Jeter). There's also always been commentary that ARod took advantage of the move to 3B to bulk up more -- so he may have traded some defense for power and better aging.

Alan Trammell would likely have been a worse first baseman than Mark Grace.

I don't believe this for a second. I mean if you moved him over there for an inning out of nowhere, sure. But give him a spring to shift, he'd have been brilliant over there. Any SS would. Of course very few guys play that position substantially in the same season and they are mostly utility players. The only player I know of with substantial time at both positions is Banks. He'd been a good SS, apparently even in the season that his knees got so bad they decided he needed to move. His first two seasons at 1B were very average, then he put up a 14. He was in his 30s by that time and had bad knees. His defensive value definitely was substantially lower at 1B but whether that was knees, age or not adapting defensively, who knows?

Miguel Cairo -- in one of the least obvious moves of all-time, Cairo started picking up reasonable amounts of backup/platoon time at 1B in his 30s. He had a lot less career time at SS than I realized (just 264 innings) and is rated average-ish ... TZ thinks he was fine, DRS puts him at a pace of -10. In 972 innings at 1B, nearly all in his 30s, he was 0 TZ, 4 Rfield. From 22-29 (almost exactly half his career PAs) he had 2 dWAR; from 30-38 he had 1. Chris Gomez picked up some time at 1B. Michael Young was an all-time terrible SS in his 20s and all-time terrible 3B and 1B later in his career. Man was a born DH.

Anyway, I'm fine with the idea that the dWAR gap between Trammell and Grace would have been smaller if both were 1B -- fewer opportunities for Trammell; his range less useful (so he probably bulks up some and maybe hits better); loses an out or two due to being RHT; misses an extra throw or two due to being 2 inches shorter than Grace. But he'd have been picking GBs better and showing much better range than Grace ever did.

And this becomes even more obvious as we look through history. "Career 1B" was nearly unheard of until recently. 1B was where you put the already big young guy (who often didn't last much past 30) or aging hitters that you didn't want in LF/RF anymore, partly the DH of its day. There were always exceptions (Gehrig, Hernandez, McCovey) but there's a reason that's what they did -- 1B is easy to play (sorry Wash), SS isn't. And of course that, througout baseball history, SS have always hit much worse than 1B so it's obvious teams are willing to trade offense for superior (or even minimal required) SS defense.
   76. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 03:52 AM (#5959384)
You can't just throw your hands in the air, say, "we suck at measuring this so let's just ignore it" and leave it at that. I mean, you can, but if you're not going to think carefully about positional adjustments and try to adjust for them as best you can, then your system is worthless for comparing players at different positions.


WHy do you insist that we need to make positional adjustments to compare players from different positions? I have been making a case that you dont need to if you simply compare the offensive levels of player X to players of the same position. On the assumption that if the def. skill needed to play SS does indeed detract from off. skill that will already be accounted for since the players in that bucket will have lower off. on average.

Since we havent given it a name, lets give that concept a name: Self Contained Value (SCV) or perhaps Buckets by POsition? Defined as: compare the players off and def. value soley to players at the same position. No need to adjust for position.

I dont see why this would not be sufficient to meet your concerns. So if Trammel is 50 runs better on off then the avg SS and say 10 runs better on def. He's at 60 or say 6.0 WAA. And McGuire is say 60 runs better at off than the avg 1b. and 0 on def. So he's 60. So that's a tie. There's no need to worry about a positional adjustment cause that's already BUILT INTO the SCV system.

Right? SO McGuire has to compare his off value vs other powerful hitters with minimum def. responsibilities. . And so if Trammel had to move to 1b, and assuming he still hits for the same value, then his Off contribution in SCV terms would decrease. Instead of hitting 50 runs better than avg, he's now down to 30. So he's been adjusted for moving to the left of the defensive spectrum. He's been giving a -20 downgrade for the shift in position.

One of B James's concern mentioned above is there are seeming aberrations in the spectrum, like LF hits worse than RFs. Would it help to consolodate some of the buckets? SO instead of 8 positional buckets we get say 5:

1b > corner OF > 2b/3b > CF > SS/C

So Trammel compares his hitting to the combined average of SS and Catchers? ARod now compares his hitting to both 2b and 3bmen. It shouldnt change too drastically since presumably I Have consolodated positions that have similar offensive levels.
   77. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 04:44 AM (#5959387)

I wasnt able to find a tabulation for hitting by position, for the meantime this 2007 article in Hardball Times provides some interesting tabulation:

https://tht.fangraphs.com/historical-hitting-by-position/

He breaks down the difference in hitting at each position, so in terms of runs created vs MLB avg:


Catchers: -8
First basemen: +13
Second basemen: -5
Third basemen: +6
Shortstops: -5
Left fielders: +8
Center fielders: +2
Right fielders: +7
Designated hitters: +13

Some things of note:

So I should probably lump 2b in with SS in my table above as they seem to hit comparably.

Bill James made a big point in the article about the difference between Rf and LF but here it hardly seems much difference if any, 1 run. Big deal. So that doesnt seem like much of an argument, if that's his big point. Maybe its gotten different in the last 7 years.

Perhaps even more interesting is the author shows several graphs depicting how off levels at certain positions have changed historically. He compares the hitting of SS vs catchers in the first graph, and I think its pretty clear that the changes in their relative positions have a lot to do with the running game. Pre 1920 Cs were hitting a lot worse than SS, but with the advent of the lively ball their positions quickly trade places. With less emphasis on SB, Cs now get a chance to sharpen their bats. Somehow this gap is alleviated by the mid 30s. I dont know what is going on there.


Ok that's pretty interesting, but then comes perhaps biggest mystery in positional hitting, and that is WHAT THE HELL HAPPENED TO MIDDLE INFIELDERS IN THE MID 60s?

The gap between Cs and SS again widens in the mid 60s and this time its SS that suffer. This is biggest gap on the charts. TIs already been alluded to above: The Eddie Brinkman Era or whatever you want to call it. Was it because DPs were on the increase? No idea.
   78. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 04:51 AM (#5959388)
So based on the Self Contained idea, you could use simply the difference in off. production at each position as a proxy for positional adjustment. So using the 2007 numbers above, the spectrum might look like:

1b > corner OF > 3b > CF > 2b/SS > C

The adjustments would be based on this scale:

13 > 7.5 > 6 > 2 > -5 > -8

So comparing a 1b to a corner OF would result in adjustment of: 5.5 runs.
   79. Rally Posted: June 25, 2020 at 08:07 AM (#5959394)
Using offensive stats as a proxy for position adjustment is reasonable, and will get you most of the way to where we should be.

From the table in #77, 2 things stick out:

SS and 2B equal, with a big gap between 2B and 3B. This has fluctuated over time, but in general the defensive skill of 2B and 3B is not nearly that big, and SS should be the infield position with the highest adjustment. After all 2B are generally quick middle infielders who don't have enough arm to be shortstops.

The other is the 1B/DH split. For some time periods you'll find 1B out-hitting DH, mostly an age issue, as for example prime Pujols can play 1B but old Pujols can't handle playing in the field so much and has to DH. It makes no sense to me at all to give a guy who is good enough to play MLB average defense at first exactly the same defensive value as a player who only wears a batting glove.
   80. Mefisto Posted: June 25, 2020 at 09:32 AM (#5959403)
I've never liked using offensive stats as a proxy for position adjustment. I think the adjustment ought to be based off of defense only. Otherwise, we're just picking up potential mistakes by managers in the relative importance of the positions.
   81. jmurph Posted: June 25, 2020 at 09:55 AM (#5959408)
I've never liked using offensive stats as a proxy for position adjustment. I think the adjustment ought to be based off of defense only. Otherwise, we're just picking up potential mistakes by managers in the relative importance of the positions.

I have also never understood that approach. I understand it's easier, but it doesn't actually make any sense to me.
   82. DL from MN Posted: June 25, 2020 at 11:06 AM (#5959419)
Pre 1920 Cs were hitting a lot worse than SS, but with the advent of the lively ball their positions quickly trade places. With less emphasis on SB, Cs now get a chance to sharpen their bats. Somehow this gap is alleviated by the mid 30s. I dont know what is going on there.


Fielding equipment is almost certainly responsible for some of these shifts.

https://sabr.org/journal/article/the-evolution-of-catchers-equipment/#:~:text=William Barrett in 1927 patented,notable catcher inventions they created.

   83. DL from MN Posted: June 25, 2020 at 11:10 AM (#5959421)
13 > 7.5 > 6 > 2 > -5 > -8

So comparing a 1b to a corner OF would result in adjustment of: 5.5 runs


That compares offensive averages and we have a good handle on offensive variance in performance. It doesn't capture the variance in performance on defense. WRT catchers in particular I think the stats are understating the spread in best and worst performance on defense. The worst catchers used in a season are probably bad at both.
   84. villageidiom Posted: June 25, 2020 at 11:23 AM (#5959425)
I don't believe this for a second. I mean if you moved him over there for an inning out of nowhere, sure. But give him a spring to shift, he'd have been brilliant over there. Any SS would.
You're going on record as saying a 6' tall right-handed player with no experience at the position would be brilliant as a 1B, compared to a player we already know was a great defensive 1B, after maybe a month of practice? And this, presumably, is solely on the basis that he was a good SS, he can do anything?

Pick a great defensive 1B. For now, I'll pick Keith Hernandez, but you can pick who you wish. Hernandez's career dWAR, per BB-Ref, is +1.3. That's cumulative, not a per-year average. (Grace, BTW, is a -5.) Now, let's take Trammell's career dWAR and translate it to 1B using just the differential in positional adjustment. Trammell's career dWAR as a hypothetical 1B would be +51. I did not misplace the decimal. If Trammell's defense translated cleanly to 1B such that the only that has changed is he's being measured against a different average, he'd be 50 wins better than Keith Hernandez at 1B, on defense alone. I suspect the reason you think Trammell would be "brilliant" at 1B is that the positional adjustment told you so. Well, this is what the positional adjustment tells you.

Want to pick Roger Connor, the 1B Hall-Of-Famer having the highest dWAR? He's at +6 for his (19th century) career. Frank Chance? +3. Someone more recent, like Casey Kotchmann or Mark Teixeira? Negative dWAR. The positional adjustment would have you believe Trammell would be +51.

What does the positional adjustment tell me? It tells me that it's really damn hard to add higher-than-average value at 1B. And that's not because 1B is populated by players who move slower than trees. It's because of chances, and of human inability to impact those chances. How often does a sharply hit ball get past the SS in a drawn-in infield? The 1B plays closer to home plate than that routinely. It doesn't matter how much range you have on those balls, because there isn't enough time to use your range. (And "showing greater range" than Grace at 1B would only be true if a player ranged so far off 1B that he's fielding balls that the 2B would have gotten anyway. I mean, Kevin Millar used to do that all the time, and it often cost his team because the pitcher never anticipates having to cover 1B on a grounder to second.) How's the player's arm? On many grounders he won't use it, and on most of the rest it works best to underhand the throw. So it doesn't mean a SS will dominate if you put him at 1B; it just means a SS, like any other player, cannot add massive value above average at 1B. The opportunities aren't there, and the nature of the opportunities are such that few players can stand out - and the stand-outs are not adding much more value than average. Players with all the skills are routinely put at SS, because those skills will make a bigger difference there. It doesn't necessarily mean a player at 1B doesn't have those skills. But the use of the positional adjustment for comparison among positions implicitly assumes it.

My point still holds, though, that no matter how bad or brilliant Trammell would have been as a career 1B, the positional adjustment in that case would have had us believe he couldn't possibly handle SS. And we know that's not true. If the positional adjustment tells us something that isn't true, why is it foundational to defensive value stats?
   85. Kiko Sakata Posted: June 25, 2020 at 11:39 AM (#5959432)
I have been making a case that you dont need to if you simply compare the offensive levels of player X to players of the same position.


I apologize, Sunday. You and I have apparently been talking past each other through this entire thread. This is literally exactly what I do with my Player won-lost records.
   86. Kiko Sakata Posted: June 25, 2020 at 11:45 AM (#5959434)
I think the adjustment ought to be based off of defense only. Otherwise, we're just picking up potential mistakes by managers in the relative importance of the positions.


The problem with using defense only is you get a self-selection problem. The only players who play multiple positions are players whose managers have decided that they are capable of playing multiple positions approximately equally well. Really bad players never play premium defensive positions. But also, really good defensive players never get slid down the defensive spectrum - Ozzie Smith never played an inning at any position but SS, because, if you're going to play Ozzie Smith, why the hell would you play him anywhere but at shortstop. In my opinion, this results in too narrow a spread across fielding positions.

Now, whether you think that problem is bigger or smaller than the problem of using offensive performance to assign defensive value is a personal opinion that I don't think can really be classified as "right" or "wrong".

There isn't really any "perfect" solution out there.
   87. . . . . . . Posted: June 25, 2020 at 11:57 AM (#5959437)
SS and 2B equal, with a big gap between 2B and 3B. This has fluctuated over time, but in general the defensive skill of 2B and 3B is not nearly that big, and SS should be the infield position with the highest adjustment. After all 2B are generally quick middle infielders who don't have enough arm to be shortstops.


No! You guys keep making the same mistake. 2B offensive issues are much more about SIZE than selection for defensive skill.

Here are the listed (i.e., probably slightly inflated) heights of the 10 players who played the most 2B in MLB in 2019. Ordered by innings at the positions:

5'8
5'10
5'11
5'9
5'11
5'10
5'11
6'0
5'6 (yeah, right)
6'2

See a pattern? 2B, for whatever reason, selects for shortness. And short guys can't generate the same amount of exit velocity as tall guys, all else being equal (which admittedly, all else never is).

   88. DL from MN Posted: June 25, 2020 at 12:05 PM (#5959438)
2B, for whatever reason, selects for shortness.


Because 2B are generally players with SS range and no arm. Smaller players generally have weaker arms. Does shortness help turn the DP?
   89. Kiko Sakata Posted: June 25, 2020 at 12:09 PM (#5959439)
No! You guys keep making the same mistake. 2B offensive issues are much more about SIZE than selection for defensive skill.


At which point, the question becomes: do second basemen HAVE to be short? Is there a rational basis for WANTING a short second baseman?

Because that argues for using offense by position as your basis for positional adjustments. Because while it's kind of true that nobody bats as a second baseman, it kind of isn't: every team has to have a guy play second base and they have to put that guy in their lineup. And if having to play a guy at second base means having to play at least one guy who's under 6'2", 200 lb., then finding the little guy who can hit like Joe Morgan or Jose Altuve is a hell of an advantage.
   90. Kiko Sakata Posted: June 25, 2020 at 12:15 PM (#5959442)
And expanding a bit on #89, I think this is an argument in favor of using offense as the basis for positional adjustments even if the use of short second basemen is not rational. Because if it's not rational and I can, instead, get by with playing a 6'5", 230 lb slugger at second base that's a real advantage for my team, because that gives me one more 6'5", 230 lb slugger than other teams and that extra slugger is going to translate into more wins - assuming that he's not an abomination as a fielder, but that's already baked into the value model through his fielding numbers.
   91. Mefisto Posted: June 25, 2020 at 12:18 PM (#5959444)
The problem with using defense only is you get a self-selection problem.


That's the problem with using offense-only too.

In my view, we ought to be able to assess the purely defensive value for each position. That is, we should be able to say how many plays the average SS makes, how many he misses, and the run value of each. Do that for each position. Then base your positional adjustments on that defensive run value.
   92. Hysterical & Useless Posted: June 25, 2020 at 12:27 PM (#5959446)
on defense a player has no control over what position he's assigned

Unless he has compromising photos of his manager...

Pre 1920 Cs were hitting a lot worse than SS

That's mostly Bill Bergen though.
   93. Karl from NY Posted: June 25, 2020 at 01:04 PM (#5959457)
Is there any work that breaks down the components of defense? Range, speed, hands, throwing. A single positional value seems coarse; different types of players have different skills that would translate differently around the diamond. Like Jeter was known for poor reaction time, but good hands and throwing, he could have made a fine outfielder. Handedness may have to be one of those components, since that's what prevented Keith Hernandez from playing 2B or 3B, for example.
   94. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 03:40 PM (#5959489)
Some quick thoughts. I think its quite possible that theres a relationship between smallness and quickness so you might need smaller guy at 2b just because of that. Also you can't just say ss is inherently harder to play than 2b its possible than 2b is fielding more throws on sb and possibly relays from OF. Even if most are routine its still adding to his work load.

I think the criticism in no. 80 is fair and deserves a response. But what is the alternative method using def numbers how would you do that?
   95. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 03:46 PM (#5959491)
going from memory I quickly compared positirnal adustment shown at baseballreference to the values obtained using the off numbers from hardballtimes article. It seems bref is under rewarding catcher and over rewarding CF
   96. Mefisto Posted: June 25, 2020 at 04:25 PM (#5959506)
But what is the alternative method using def numbers how would you do that?


Using Statcast data, we know both how many balls are hit into each "zone" and the run value of those events. I'd use that to establish run values for the average player at each position.

For previous eras, I'd use IF assists and OF putouts -- league averages in both cases -- to estimate the number of balls hit to each "zone".
   97. Graham & the 15-win "ARod Vortex of suck" Posted: June 25, 2020 at 05:28 PM (#5959517)
Is it possible that there are few tall second basemen for reasons similar to Bill James' observation* about the lack of fast right fielders? If you have a player who is quick enough to handle second base, but also big and strong, then where are you going to put him? Shortstop of course because his strength presumably translates into the ability to make tough throws from the hole. Cal Ripken, Jr. would likely have been a terrific second baseman, but why would anyone have ever put him at second? The same goes for Troy Tulowitzki, A-Rod, JJ Hardy, Andrelton Simmons, and Brandon Crawford. I realize there are short guys with strong arms, but 1) those short guys still end up at shortstop and 2) I would still think height and arm strength are at least slightly correlated.

*I believe it was in the New Historical Abstract. James noted that any player with the arm for right who was noticeably fast was going to end up in center field.
   98. Rally Posted: June 25, 2020 at 05:32 PM (#5959519)
Ichiro is the exception to that rule.
   99. Rally Posted: June 25, 2020 at 08:42 PM (#5959546)
Derek Jeter LeMahieu at 6-4 is an exception to the short 2B standard. A 3 time gold glover with good metrics. Mike Moustakas played some 2B last year and was signed to play there for the Reds. Not super tall, but a big guy that in the past would not have been tried at 2B. He seemed to do a decent job at the position.
   100. Sunday silence: Play Guess How long season lasts Posted: June 25, 2020 at 08:44 PM (#5959547)
I've never liked using offensive stats as a proxy for position adjustment. I think the adjustment ought to be based off of defense only. Otherwise, we're just picking up potential mistakes by managers in the relative importance of the positions.


Let me ask you this:Is this only rational criticism that you can come up with to the suggested "bucket method.?"

If so, then if we could somehow show that nearly every position player is being used at his optimal position then this criticism would disappear? And the method would be totally logical, yes?
Page 1 of 3 pages  1 2 3 > 

You must be Registered and Logged In to post comments.

 

 

<< Back to main

BBTF Partner

Dynasty League Baseball

Support BBTF

donate

Thanks to
dirk
for his generous support.

You must be logged in to view your Bookmarks.

Hot Topics

NewsblogFacts refute claims Cardinals got COVID-19 at casino
(5 - 8:56am, Aug 06)
Last: Lassus

NewsblogOT Soccer Thread - Spring 2020
(618 - 8:47am, Aug 06)
Last: AuntBea odeurs de parfum de distance sociale

NewsblogThe Lost Career of Luke Easter
(6 - 7:56am, Aug 06)
Last: TJ

NewsblogOT – NBA Revival Thread 2020
(812 - 7:16am, Aug 06)
Last: Thok

NewsblogEmpty Stadium Sports Will Be Really Weird
(8716 - 6:26am, Aug 06)
Last: Never Give an Inge (Dave)

NewsblogOMNICHATTER has a false sense of security, for August 5, 2020
(28 - 1:40am, Aug 06)
Last: Mayor Blomberg

NewsblogAn Olympic speedskater now plays for Miami
(6 - 12:28am, Aug 06)
Last: Russlan thinks deGrom is da bomb

Gonfalon Cubs60 Second Season Preview
(37 - 8:20pm, Aug 05)
Last: Eric J can SABER all he wants to

Hall of MeritMost Meritorious Player: 1929 Ballot
(9 - 4:18pm, Aug 05)
Last: bjhanke

NewsblogWide NBCUniversal cuts include Hardball Talk
(4 - 4:10pm, Aug 05)
Last: Jay Seaver

NewsblogDerek Jeter blames Miami Marlins coronavirus outbreak on 'false sense of security'
(9 - 3:33pm, Aug 05)
Last: Starring Bradley Scotchman as RMc

NewsblogMLB's 'Field of Dreams' game in Iowa postponed to 2021 because of coronavirus
(7 - 3:29pm, Aug 05)
Last: Starring Bradley Scotchman as RMc

NewsblogOMNICHATTER! is its own bubble, for August 4th, 2020
(39 - 3:20pm, Aug 05)
Last: What did Billy Ripken have against ElRoy Face?

NewsblogMLB commissioner Rob Manfred says 'we are playing' but 'players need to be better'
(64 - 2:51pm, Aug 05)
Last: Jose Needs an Absurd Ukulele Concert

NewsblogTwins vs. Pirates game delayed by unauthorized drone flying over Target Field in Minneapolis
(9 - 2:16pm, Aug 05)
Last: What did Billy Ripken have against ElRoy Face?

Page rendered in 1.0900 seconds
48 querie(s) executed