THE BOOK--Playing The Percentages In Baseball

Interesting stuff

If Jeter has 219 WS and 86 LS, does that give him a personal winning percentage of .718? (219/(219+86))

That seems pretty high. The Lee Sinins Complete baseball encyclopedia has him with a career offensive winning percentage of .630. To get his overall pct up to .718, he would have to be a great fielder. But he is not thought to be one.

Offensive winning percentage tells us what a team’s winning percentage would be if it had 9 identical hitters and it gave up an average number of runs. Can a similar statement be made about a WS & LS pct? Or should we not even turn it into a pct? Maybe I am missing something

Why not use percentage of outs instead of PA to determine the game slices?  I actually like this method quite a bit, it’s simple and certainly seems accurate enough for most purposes.

His .718 rate is only for offense, so fielding doesn’t enter this particular picture.

I think the best way to interpret the .718 is not as if there are 9 Jeters.

Let’s step back as to what we did.  Jeter is +23.6 wins in roughly 2000 games played.  (Take his PA and divide by 4.3 PA per game.) So, he adds +.012 wins per game.

A team of one Jeter and 8 average hitters and average defense would win .500 +.012 = .512 per game.

If you do .500 +.012*18 = .718

Recognize the number?  That’s all that is.  So, .718 represents how a team of Jeters would do (on hitting and pitching/fielding), *IF* their accomplishments were linearly additive.

But, more important, is that this representation should be taking too literally, since the real number is the +23.6 wins and his nearly 2000 games played.  Those are the numbers we care about.  Everything else is really just window dressing or lipstick.  It pretties everything up, but underneath that dress, it’s +23.6 wins and 2000 games, and that’s what we really care about.

Thanks for explaining that. I thought that Win Shares included all aspects of the game, not just offense.

It does, but I only quoted the offensive part of Win Shares.

I am not very familiar with WS, but since WS assumes that the a team’s total wins are commensurate with underlying performance, any disconnect between your system to convert LW and WS could be due to a difference in “third order wins” (pythag wins based on underlying performance) and actual wins, no?  If James used pythag wins, his win and loss shares would be much closer to what linear weights would come up with.  If he used “third order wins” they would be exactly the same, I think.

Doesn’t James use some form of linear weights to apportion the offensive share of team wins in the first place?  So of course the two methods will be similar.  As I said, the only difference is that James accounts for actual team wins and lwts assumes that team wins will be the sum of all players’ lwts contributions.

In fact, that is really all that win shares does.  It takes all the regular metrics and “shapes” them to actual team wins.  James didn’t like the idea that “context-neutral” underlying performance may not reflect a team’s actual number of wins. He doesn’t care that the difference between a team’s actual wins and their underlying performance is mostly luck.  WS is NOT a “talent” type of metric. It is an MVP type of metric, which includes talent and luck (all metrics to some extent include luck, but WS includes A LOT of luck).

If a team of average players with average underlying performance wins 100 games (say they have a great record in 1 run games), each of those players gets credit for a total of 300 WS, no?  And they all look like great players, which they are not, right?

Do I have that right?  If I do, I don’t see why anyone would use win shares in research or studies that rely on player talent.  Way too much noise.  In the long run, it is not bad of course for nailing true talent, but we can say the same thing about any bad metric.  In the long run, ERA (adjusted for park) is perfect for pitcher true talent.  Even a pitcher’s w/l record, one of the worst metrics imaginable for reflecting pitcher talent, is “perfect” in the long run if spread out over many teams (where the collective offensive talent of those teams is league average).  Even RBI and runs reflect an offensive player’s true talent “in the long run” and assuming that he plays for a multitude of different teams and bats in different spots in the lineup.  Etc.

James says that a player will have virtually the same WS whether he plays for a team with a good W/L or a poor W/L.

It’s not how many wins the team has, but whether it outperformed or underperformed is Pythagorean won-lost record.

Right to Dave and Charlie.  If a team’s W/L record matches the pythag, a player’s Win Shares will be identical regardless of the team’s W/L record.

If a team’s pythag doesn’t match the W/L, then the gap between the two is apportioned to the players.  That apportionment of luck is a point of contention of many people.  In my view, we should be creating a “luck bucket” until you can figure out who it is that deserves it (i.e., that was actually involved in the walk-off HR, or the closer who was lights out in high leverage situations).

***

The main point of my post is that Linear Weights ~= Win Shares minus Loss Shares divided by 6, and I don’t know how many people realize this.

The main point of my post is that Linear Weights ~= Win Shares minus Loss Shares divided by 6, and I don’t know how many people realize this.

Well, of course it will (although the formula to convert one to the other is not obvious, which was probably your principal point).

James says that a player will have virtually the same WS whether he plays for a team with a good W/L or a poor W/L.

I think he says, “regardless of how good or bad the team is,” which is of course true.

Where win shares go awry, at least in terms of player talent, is when a team’s w/l record is significantly different than their third order (not necessarily pythag) win total.  Which is why it is crazy to use win shares for studies and research that need to utilize player “true talent.” And let’s also not pretend that a player’s win shares tells us who is the better player on a level playing field.  Win shares are like RBI and runs and w/l for a pitcher.  They are a player’s talent in the (exact) context of his team’s w/l record.  Sometimes that works (when the team’s w/l record is commensurate with its underlying performance), and sometimes it doesn’t.  Same with a pitcher’s w/l record.  Sometimes it “works” when he has an average and balanced number of no-decisons and when his team is of average offense, and sometimes it doesn’t.

A perfect and simple example of how it is NOT true that a player’s win shares will be the same regardless of his team’s w/l record, is this:

A team of exactly equal and league-average players with exactly league-average stats (and the team has league-average defense and pitching, although that does not matter) play for a team that gets lucky in 1-run games and has a 100 and 62 record.  All players will share equally in those 300 wins shares according to their PA (or outs or whatever he uses to apportion win shares).

The same team gets unlucky in 1-run games and loses 100 games.  Those same league-average players with league-average stats will only split up 186 win shares. Obviously a gigantic difference for players with exactly the same stats and talent.  Win shares is NOT a metric for talent.  It is a hybrid metric.  It combines talent and the “luck” associated with the difference between a team’s w/l record and its underlying performance.

I hate hybrid stats, other than you can look at them and say, “that’s nice.” You can’t really do anything with them.  As I said, if you use them to conduct studies on talent, your studies are flawed (if there is a large enough sample of players and teams, it is no big deal of course).

"Win shares are like RBI and runs and w/l for a pitcher.  They are a player’s talent in the (exact) context of his team’s w/l record.”

No, that’s not a good comparison, and badly undervalues the usefulness of Win Shares.  RBIs, runs and W/L are biased by at team’s entire performance record—which has a huge range in every season.  Win Shares is biased in a similar way but only by the much, much narrower gap among teams in varinace from pythag projection.  Most teams vary only a few games from thier projection and even the largest variances are huge only in rarest of instances.  Over multiple seasons of a player’s career, the variance from pythag in his team’s performance will disappear almost entirely (whereas many players will spend entire careers playing for good teams, or for bad teams.  The actual bias from the pythag variance, once spread around among all players on a team is very limited in its effect over a full season, espeically if you avoid decimal point Win Shares and stick with numbers rounded to the nearest integer.

I have other issues with Win Shares—most notably that offensive performance by middle infielders and catchers is effectively compared to replacement level hitters rather than replacement level middle infielders and catchers, which undervalues their offensive value—but the allocation of the pythag varinace has never troubled me much.  You could refine WS by re-allocating the pythag variation based on some WPA formula, and that would be a nice theoretical improvement.  But it wouldn’t move the numbers much.

I have other issues with Win Shares—most notably that offensive performance by middle infielders and catchers is effectively compared to replacement level hitters rather than replacement level middle infielders and catchers, which undervalues their offensive value

Win Shares also gives them more defensive responsibility.  You can’t have your cake and eat it too, or you will overvalue players at defense-first positions.

Whether WS handles the defensive values correctly is another argument, but if it shortchanges 2B/SS/C, it’s because of how defense is evaluated, not how offense is evaluated.

"Win Shares also gives them more defensive responsibility.  You can’t have your cake and eat it too, or you will overvalue players at defense-first positions.”

More defensive responsibility than what?  Than a first baseman?  Of course—that’s because a shortstop makes more skill-dependent plays than a first baseman and will thus contribute more in terms of preventiving runs than a first baseman.  So that’s appropriate on the defensive side.  But that’s independent of the contribution of a shortstop who in addition to being at least a decent defensive shortstop can also hit well.  The Win Shares formulas from James’s orignal book, if I understand it correctly, are based on an assumption that a shortstop who hits as well as an average first baseman adds value on the offensive side of his team’s performance at a level equal to an average firt baseman.  But that is patently not an accurately reflection of the offensive value to a team of an offense-capable shortstop.  Men who can play at least replacement level defensive shortstop (i.e., they can play defense at short well enough to stay in the lineup are a very small sample of people, compared to say guys who can play a half decent first base.  To have a guy who can play short and also hit like a decent first basemen adds a disproportionate number of runs to a team’s ability to generate offense.  Or to put this more technically, a shortstop’s offensive value should be compared to a replacement level shortstop’s offensive value, not a replacement level first baseman’s or outfielder’s offensive value. I know that James discussed this issue specifically in his bookl and rejected my argument—he said something like “a run is a run regardless of whether it is scored by a shortstop or a first baseman”.  But I think that fails to recognize that a team can’t have nine first basemen in the lineup, it always has to have somebody who can play at least replacement level defense at short.  If it happens that that guy can also hit, the runs he adds are in fact more valuable than the alternative on the bench or in the minors than a the runs added by a similar hitter who can only play first base.

Apologies if I’ve managed to take this thread off-subject.

#13 - I agree.

One of my economic theories is that compensation for labor should be determined by two things - how much revenue does your labor create, and how hard are you to replace?

Janitors may do valuable work, but how hard is it to train a new one?

Likewise, A-Rod is more valuable at short because how hard is it to find a replacement who can hit as well as he does, and also play SS? On the flip side, you know you have a problem when a light hitting middle infiielder is playign 1B, because how hard should it be to find someone who can hit better and also play 1B? Almost anyone can play first, and even poor defensive play at !B will not hurt as bad as poor defense at SS

There are two ways to try to handle total player evaluation.  One is to measure offense relative to the position, and give fielding credit based on a comparison to an average fielder at the position.  This is the approach that VORP uses, that TPR uses, and that I use (although I readily admit that I don’t think it is optimal).

The other is to attempt to measure everyone’s offense and fielding contributions relative to an average player (or to another baseline), regardless of position.  This is what Tango does, implicitly, by comparing a player’s offensive contribution to the overall average, comparing his defense to the average player at his position, but then adding in a positional factor based on the defensive responsibility of each position.  This is also what Bill James does in Win Shares, although he goes about it by giving more defensive “claim points” to the tough positions.

If you try to mix these two, as you want to do, you will be overvaluing the players at the defensive positions.  You want to give the average shortstop more fielding credit than the average first baseman--which of course is perfectly logical.  But you also want to give the average shortstop equal offensive credit to the average first baseman--despite the fact that the average first baseman makes a much larger offensive contribution, in terms of actual runs and outs.

Or to put this more technically, a shortstop’s offensive value should be compared to a replacement level shortstop’s offensive value, not a replacement level first baseman’s or outfielder’s offensive value.

This is the same problem that Clay Davenport’s WARP has--valuing against a hypothetical player who can’t hit or field.  A true replacement player
is one whose total contribution is such that it can be freely replaced, not one who is that bad in both phases of the game.

"A true replacement player
is one whose total contribution is such that it can be freely replaced, not one who is that bad in both phases of the game.”

I think that gets to the heart of the discusion.  In real life, “replacement level” offense and defense are each different for shortstops and first basemen.  A shortstop and first baseman who are each just barely good enough in the field to stay in the lineup (if they can also maintain a replacement level offense) will have very different replacement level offenses.  Any formula which assumes the replacement level offense of these two players is the same is missing an important piece of value.

I know that the first baseman is, as you say, making an actual larger contribution to run scoring than the shortstop.  And I know that we have to avoid making the mistake, which Bill James has eloquently exposed in his work, of overvaluing contributions by players whose contribution is naturally limimted by rule or strategy.  You can be the greatest holder on place-kicks in football history, or the greatest LOOGY in baseball history, but your overall “value” is still limited because there is only so much value such a player can add.  But that is not what we are talking about in comparing shortstops and first basemen, because unlike the rules that allow place-kick holders a very narow place in football, the rules of baseball require all fielders (AL pitchers aside) to bat once and only once each time through the lineup.  So shortstops and first basemen, despite the very different demands of their respective fielding positions, are forced to participate on offense a very similar number of times.  The result is that a realistic approach to valuing first basemen and shortstops has to reflect different comparison levels for shortstop offense and first base offense.  Otherwise you are failing to reflect the actual value contribution of a good hitting shortstop, who basically has to take about the same number of PAs as the first baseman despite the greater defensive demands of the shortstop’s position.

I don’t claim that the adjustment I am proposing is easy to do.  As you point out, patriot, in real life “replacement level” analysis is not a separate analysis of defense and offense, but rather a combination of the two. A team can keep Shortstop X in the lineup with a little less performance on defense than it would tolerate from any other shortstop in the world so long as Player X provides a little more performance on offense than they could get from any other shortstop in the world.  Replacemeent level as you say is always a combined matter (except for pitchers and DHs).  But it has to be combined taking into acount position on both the offensive and defensive sides, otherwise it is not an accurate reflection of real life value.

In real life, “replacement level” offense and defense are each different for shortstops and first basemen.

The only reason why this is the case is because shortstop is a more demanding defensive position.  Once you have accounted for the fact that the average SS is much more valuable defensively than the average 1B, there is no need to give them an extra credit for an offensive handicap.  The offensive handicap is a result of the difference in defensive responsibility, not a seperate phenomenom. 

The unfortunate side-effect of what could be called the easy way out approach to player valuation (the VORP/TPR use of offensive position adjustment plus fielding value above an average player at the position) is that it really has gotten a lot of people thinking that a shortstop who creates 100 runs is a better offensive player than a first baseman who creates 120 runs.  Then when a different (and more realistic) valuation approach is used, people think that it is shortchanging the offensive value of a shortstop.

Patriot has perfectly captured both sides of the argument.

On the “tango” side: A win on offense is a win on offense.  It is not worth more or less because Guillen happens to be playing SS, 3B, 1B, DH, or LF. 

His overall value however (offense + defense + position) is virtually identical, regardless of the position he plays.

I prefer my approach, since it models reality better.  It doesn’t presume that, overall, the average 2B = average 3B, even if you have a sudden influx of ARod’s, Rolens, and Beltres.

We’ve discussed this at length and many times in this blog.  I highly suggest those interested spent an hour or two going through the archives, so that we don’t have to repeat too much of what is being said here.

I also recommend going to the wiki, and going to Patriot’s site.  That’ll give you a good 3 or 4 hours of reading time.

The question for me is what is the reliability/ tradeoff between Tango’s approach and the other, even if Tango’s approach is theoretically better? How reliable are those DPAs, applied to the typical who doesn’t switch positions much? In the other way, which of course does not account for having a sudden influx of better players at a certain position, at least the OPA is easy to calculate. So, where is the ‘break-even’ point which makes one approach superior, in the real world?

There is no doubt that every player has a defensive value that is equal to his value at one position relative to the “average position.” That is the same for hitting.  IOW, if one player is an average defensive player at SS and another one is an average defensive player at 1B, their relative defensive value is simply what each would do at the other’s position, or more generally, what would each player do at an average defensive position.  It is a “defensive position” adjustment, just like we do any other adjustment, such as a park adjustment.  We are always faced with the question, what is the value of an average player at Coors versus an average player at Shea, or what is the value of an average hitting NL player versus an average hitting AL player?  No different.  And that is the way it should be approached - normalizing every defensive position to “position-neutral” defense.

The problem of course is two-fold.  Even if we “should” be doing it that way, it is not so easy in practice.  The first part of the problem is that we don’t really know how to normalize defense from one position to the other.  Tango uses a certain set of adjustment factors, which are admittedly approximate.  The second part of the problem is that it is not necessarily apples and oranges when we compare or try and adjust for defensive position.  For example, catcher is a unique enough position that we can’t just throw it in the spectrum and adjust to and from the position.  To some extent that is the same for all positions.  So the only thing we can do there is to look at the scarcity of the players at that position.  And the only way to to do that is to look at offensive production.  We have to assume that over the long run, the position in which offensive production is weakest are the positions which are most difficult to play defensively, or perhaps the positions that are most desired, most likely a combination.

Like I said, we went through this already.  I think I showed that over some 20yr period, the offense of 2b = offense of SS.  Since we know that virtually all 2B played some SS in their lives, it’s ridiculous to think that the avg 2B = avg SS on offense, implies that the avg 2B = avg SS on defense.  There’s no way around this one: in that 20yr period, the avg SS was superior, overall, to the avg 2B.

If 20 years is not long-term enough, then that theory falls on its face.  Ain’t buying it.

However, IF/OF, as a group, I can *possibly* buy.  I can definitely buy the C/nonC.

The theory further is unsupported in college and high school: the avg SS is far superior, overall to the avg 2B.  Just look at all the first rounders selected at 2B (if you can find any) and SS.

Is the average RP = avg SP?  No. Just as all 2B are converted SS, all RP are converted SP.

Is the average DH = avg RF?  Obvious answer.

We know the theory is b.s. in the NFL (avg QB is far more valuable than the avg OT), and the NHL (avg Center is a bit more valuable than the avg Winger).

The reality of it is that players belong to pools:
A: potential catchers
B: pool A, plus potential infielders (2B,SS,3B)
C: pool B, plus potential outfielders
D: pool C, plus potential 1B
E: pool D, plus potential DH (i.e., everyone else)

MLB is not in equilibrium after 1 year, or 5 years, or even 20 years.  Let’s not pretend that it is.

Anyone who does their adjustments on an annual basis (meaning avg 2B = avg 3B in 2007, avg DH = avg RF in 2006) is merely fooling everyone.  Just look at Bonds… by himself, just by himself, he adds +3 to +4 runs on offense to the league average LF!  So, if LF are +10 and RF are +8 for the league, it looks like the LF must be worse fielders (according to the equilibrium theory), when in fact, they could have similar fielding, and have an extreme player.

We know that MLB players move easily between SS and 2B, SS and 3B… that has to be considered.  They don’t move easily between C and the other positions… so, we can accept that the average C = average nonC… maybe.

I agree!  I am just pointing out some of the potential problems with trying to assign relative defensive value.

Let’s say that SS is clearly a more difficult position than the other infield positions.  And let’s say that there are fewer players who possess that ability to play SS, as we would expect if something is “more difficult” (in fact, that might be the only definition of “difficult” that works in this case).  Let’s also say that the few people who possess that ability to play SS also hit better than those that play other infield positions because the same skills that are necessary to play SS are the same skills that make one a better hitter.  This is not true in MLB of course.  It is true in Little League and even some higher levels of play.  How does one reconcile that?

I think that defensive value all boils down to scarcity, but I am not sure.  For example, let’s say that all position were unique.  IOW, they were NOT fungible with a certain fixed adjustment factor when going from one position to another.  IOW, let’s say it were like a lawyer and a doctor.  One is not more “difficult” than the other; they are completely different.  How would we assign relative defensive value then?  I presume it would be by scarcity.  Out of 100 eligible players, if there were 10 at a certain hitting level that could play catcher and 15 that could play SS, then catcher would have to be more valuable, no?  Regardless of what their relative offenses would be?

For example, what if there were only 20 people in the whole world who could play catcher with any skill whatsoever, but whatever skill they had that enabled them to be a catcher also enables them to be great hitters (let’s say that you needed a rocket arm to be a good hitter, such that catchers were the best hitters in baseball in the long or short run.  Let’s also say that these catchers could not play any other position.  What relative defensive value would you give the catcher?  Would it be the best defensive position with the highest defensive value (because there are so few of them in the world)?

I don’t know the answers to these questions.  At least without thinking some more about them.

The NFL metaphor (QBs have more value than OTs) would only be useful if the NFL had a rule requiring that QBs and OTs throw an equal number of passes in a game.  This would be the equivalent of baseball’s lineup requirement, in which players who can play the most demanding positions in the field have to bat once every time through the the lineup just like those playing less demanding positions.  If the NFL had such a rule, an OT with a good throwing arm would be enormously valuable—and the increment of value over a QB with a similar throwing arm would not be limited to the mere additional value of the OT as blocker over the QB as blocker but would also include the increment of the OT’s throwing arm over the weak throwing arms of most of the big lugs who are capable of playing OT.

Baseball’s demand for versatlity in an athlete, and the resulting complexity of each player’s profile, is a reason why I enjoy baseball so much more than football (and dislike the DH).  Such versatility mandated by the lineup rule in baseball means we can’t just compare player offense to player offense and player defense to player defense in a total player evaluation system in baseball; player offense has to be evaluated to some degree in the context of the defensive task to which the player is assigned.

mgl: you are asking interesting theoretical questions, but they are merely theoretical in a context in which we need not be so theoretical know a lot about the actual relationship between defensive and offensive skills at the major league level, which is the context most of us are interested in.

We know that year after year, in contempary baseball, collective offensive performance (average, median, replacement level, whatever meausre you care to use) by catchers, shortstops and second basemen is lower than that of first basemen and corner outfielders, with third basemen and center fielders somewhere in the middle.  These relationships do evolve glacially over MLB history but are stable enough in any particular era to rely on with some confidence.  Such consistency seems to show, again with a significant level of condience, that the defensive tasks of playing short second and catcher in the majors is sufficently demanding that the available pool of players will not have the same offensive skills as a class as those capapble of playing less demanding positions.  That means a team lucky or clever enough to find a player who can both play short and perform on offense at a level that matches the best offensive players at other positions has achieved an increment of value in its likelihood of winning games and pennants that exceeds merely the increment in value of this great-hitting shortstop’s value in the field over the comparable-hitting first basemen’s performance in the field.  The lucky team has an asset that if lost would have to be replaced by a shortstop from the pool of normal available middle infielders (which we know has a lower offensive performance level than the average ballplayer), while the comparable first basemen if lost could be replaced by virtually any major league ready ballplayer.

birtel: you ignored my college and hs examples.

As long as you don’t adhere that the avg 2B = avg SS = avg 1B, overall, then I don’t have an issue here.

"but are stable enough in any particular era to rely on with some confidence”

As I’ve said here, and in more detail elsewhere in this blog: this is a false statement.  Or, the statement is replete with so many conditions (stable enough, particular era, some confidence) as to make it a true statement with such a wide error range as to make the statement irrelevant.

***

I’m begging you, for my sanity, go through the archives so I don’t have to repeat everything I’ve said!

Unless Birtlecom has recanted his previous position, we are discussing different things.  Tango, MGL, and David (and me a bit in my earlier comments) are debating the merits of offensive position adjustments and defensive position adjustments. 

Birtlecom, on the other hand, wants an offensive adjustment AND a defensive adjustment (while WS does not have a defensive adjustment in the sense that Tango’s WAR does, it gives different responsibility levels to each position which is a de facto adjustment).

tango: I agree that one can’t assume that an average 1B = an average 2B = average SS merely because they are each average at their respective positions.  They each have different roles in the game, and in any historical era one may have more value in terms of winning than another.  But that’s not the same as saying that no offensive position adjustment (in addition to the obvious need for a defensive adjustment) at all is needed.  As patriot accurately expresses my view, I think you need both a defensive positional adjustment (to reflect the fact that a shortstop plays a more important role on defense in preventing runs than a left fielder) and an offensive positional adjustment of some sort (not necessarily a full “ave. hitting SS = ave. hitting LF” adjustment, but something more than zero) to reflect the value over replacement-level-SS-offense of a good hitting SS.

for the sake of tango’s sanity, I’ll resist the temptation to comment further on this issue in this context unless invited.

I’m wondering where does supply and demand figure into Tango’s model? In HS, where the SS is much better than the 2B, there are only a few players the team can use as backups. So, the backup MI might be -30 runs (per 162 G) vs the 2B, but -45 vs the SS (made-up numbers).

But in MLB, there are lots of available players to fill the slots. If this results in a repl level for both ss and 2b of say -18 runs, then even if the avg ss is better than the avg 2b, it won’t have any marginal win impact. (I’m not claiming they’re both at -18, just hypothesizing to make a point.)

birtel: I don’t mind that you comment.  It would just be helpful if you can go through the archives, as you are obviously interested in the subject.  We talked about this *alot*.  I do a pretty good job with the archives, and the title of the threads almost always has “replacement” in it.  So, it’s just a modest request, if possible.

***

birtel: as for the “offensive adjustment” and “defensive adjustment"… I don’t know that we need to break it down as two sub-components.  I simply have a “positional” adjustment (in wins per 162G):
+1.0 C
+0.5 SS/CF
+0.0 2B/2B
-0.5 LF/RF
-1.0 1B
-2.0 DH

If you want to work out your numbers separately, and then combine it, if you end up with numbers +/-0.2 wins of mine, then we really don’t have a disagreement.  Until you actually come out with specific numbers (even estimated), it’s hard to really see if we even disagree at this point.

***

David, when you say “-18 runs”, that’s relative to what exactly?  I like to use Willie Bloomquist as my perfect replacement-level player.  (If not him, then a group of players that end up looking like him.)

Maybe use WFB as the illustration in what you are trying to say.

***

Aside: I’m thinking of changing the name of WAR (wins above replacement) to WAW or WAB (in honor of Willie Bloomquist).  Or even WFB (wins from bloomquist), which has the advantage of also standing for willie f---ing bloomquist.

tango: I agree 100% that it doesn’t matter under what name one makes the appropriate adjustments, or whether you describe them as two separate adjustments or one.  My original point was a quibble about Win Shares, not about your WAR (or WAB!).

Looking at Win Shares as actually calculated for historical players seems to consistently show almost all the top players in any group are found on the outfield/first base end of the defensive spectrum.  There are good baseball-structural reasons for some of that result.  For example, because catching is so physically demanding, catchers play fewer games in a season than others and their overall potential value is capped by that limit.  Pitchers also have important structural limits on their total value (which have become more severe through history as their individual IP per season totals have declined).  But the dearth of middle infielders among WS leaders is harder to support, given that middle infielders play, and must come to the plate, about as often as other everyday players and teams need them as much they need corner OFers and first basemen.  Logically it seems to me the greatest hitting shortstops should be coming out with similar (perhaps not exactly the same, maybe, but similar) WS totals as the greatest hitting first basemen (even though the greatest hitting shortstops create fewer runs than the greatest hitting first basemen), other things being equal.  But that is not the case with Win Shares as they have been calculated to date.  And I think at least part of the reason is that because James believed (he is crystal clear about this in the WS book) that no positional adjustment reflecting offensive expectations was appropriate. That decision I think short-changes the value of players on one side of the defensive spectrum, with real distorting effects on the Win Shares totals assigned to players. So really, my argument was never with your method, but Bill’s.

----"when you say -18 runs, that’s relative to what, exactly/”

In the case of the SS, it’s relative to the avg SS--and similarly for the 2b. So, if it’s -18 for both, and the avg SS is better than the avg 2b, then a repl SS is also better than a repl 2b. So, the advantage of the avg ss over the avg 2b is negated.

It seems to me that this is all dependent on the supply of qualified available players. When it reaches a certain level, then it’s not scarcity, it’s more the structure of the game that determines, or rather limits, expressed value.

The average team wins 81 games, 52% of which are credited to defense in Win Shares, and then 32.5% of those go to the fielders.  So there are 13.7 wins for the fielders, and the average position gets 13.7/8 = 1.71 Win Shares.  The catchers get a weight of 38, 1B 12, ...., adding up to 200.  Thus, the average catcher would get 13.7*(38/200) = 2.6 wins.  Subtracting the average of 1.71, the inherent adjustment for a catcher is +.9.  Comparing Win Shares to Tango (WS treats the OF as one and then gives more credit to the CF, but Tango’s average for the OF would be -.5/3 = -.2):

C: +.9(James)/+1.0(Tango)
1B: -.9/-1.0
2B: +.5/0
3B: -.1/0
SS: +.8/+.5
OF: -.4/-.2

These actually match up pretty well, with the big difference being the middle infielders...with James giving them MORE credit.

I’m not sure what you want Bill to do on this issue.  It’s pretty hard to finesse the numbers in such a way to make a Ernie Banks the equal of a Jimmie Foxx.

I’m sure WS has (at least, had anyway) a problem with fielding.  If I look at the best fielding SS in terms of fielding+position, that’s about +2.5 wins, +3.0 wins or so per year.  And if I look at the worst fielding 1B that’s about -2.0 wins or so per year.  So, there should be a gap of around 4.5 wins, maybe 5.0 wins.

In WS, that gap would be probably on the order of 3.3 wins.  If I look at studes’ database of players since 1969, and I take the top 50 fielders, their WS is around 11.  The bottom regulars would be at around 1 WS.  So, the gap is 10 ws, or 3.3 wins.  So, I think we can see how depressed the fielding+position is in WS. 

The gap between an average SS and average 1B should be 1.5 wins.  A bad 1B would increase the gap by at least 1 win.  A great fielding SS should add at least 2 wins, if not more, especially if I were to use the same selection process as I did for WS.

Expanding on Patiot/34: while James has the basics of the position-to-position comp ok, it’s the intraposition that he fails at.  He simply doesn’t let the fielders excel enough (basically, he has an almost artificial cap).

However, with the new Loss Shares, with guys allowed to get negative Loss Shares, maybe this has been fixed.

So, while the published win shares let the top SS get up to like +1.0 to 1.5 wins above the average SS, with the loss shares, it may double.  Who knows.  It just needs stretching, that’s for sure.

"It’s pretty hard to finesse the numbers in such a way to make a Ernie Banks the equal of a Jimmie Foxx.”

Foxx and Banks can make an interesting illustration.  Let’s take two seasons exactly a quarter century apart, a Foxx MVP year in 1933 and a Banks MVP year in 1958.  James gives Foxx 37.5 Batting Win Shares in ‘33 and Banks 26 Batting Win Shares in ‘58.  We can see how that happens pretty easily: Foxx created about 181 runs in 1933 and Banks about 139 runs in 1958.  42 runs difference, or about 4 wins worth, or about 12 Win Shares worth of difference.  (All my numbers here are very rough, but I think they are sufficiently accurate to make the point—please feel free to check all my calcs as I did them fairly quickly.) But if Banks had been lost to the Cubs in 1958 they would have needed not a first baseman but a shortstop to replace him.  The pool of MLB shortstops in 1958 created on offense about 66 runs for every 682 PAs (that’s the number of Banks’s PAs in ‘58).  Banks at bat in 1958 created about 73 more runs than that.  Foxx created about 88 more runs than the pool of first basemen in 1933 created per 670 PAs (Foxx’s number of PAs in ‘33).  I would submit that the real difference in Banks’s offensive value to the Cubs and Foxx’s offensive value to the A’s in their respective seasons is not 42 runs or 11 to 12 Batting Win Shares but closer to 15 runs or about 4 or 5 Batting Win Shares.  No, Ernie Banks was not as valuable in 1958 as Jimmie Foxx in 1933, but I would argue they were a heck of a lot closer than Win Shares (as described in the original book) suggests.

The problem with that line of analysis is that if Banks went down, the Cubs would be able to find someone who could play an average defensive shortstop (some of the previous threads that Tango alluded to here show that replacement level players are close to average in fielding).  So there is no need for a defensive adjustment and an offensive adjustment.

If you think that theoretically, the adjustment should be at the offensive level, I respectfully disagree, but I also am not going to object.  What I do object to is the double adjustment. 

Re: Tango’s #s 36, just to reiterate my earlier post, I have no trouble whatsoever believing that WS might be off on the fielding evaluations.  In #12, I said:

Whether WS handles the defensive values correctly is another argument, but if it shortchanges 2B/SS/C, it’s because of how defense is evaluated, not how offense is evaluated.

What I meant to say is not that they necessarily would get an average fielder, but that they might get an Adam Everett-type (brilliant fielder, atrocious hitter).  Or they might get a guy who fields like Derek Jeter but hits like Jhonny Peralta.  Or they might get a guy who who is bad but not atrocious in both aspects, like Chris Gomez (I’m actually not up on Gomez’ fielding rankings, so that may be unfair).

But they’re not going to get a guy who fields like Derek Jeter AND hits like Adam Everett.  That is what the Davenport position (which essentially is an offensive and a defensive adjustment) would try to tell us.

Replacement level = level at which your replacement will play (the average guy at the end of the bench, who is capable of playing your position)

For example, Jack Wilson missed a third of the season with a leg injury. He’s above avg on defense for SS, at best avg bat, but which is a little above avg for a SS.

He’s replaced, in about equal numbers, by minor league prospect Brian Bixler (good glove not hit) waiver bait Luis Rivas (above avg glove below avg bat) and crusty vet free agent Chris Gomez (poor glove avg bat)

A non-PBP fielding metric that is accurate is necessarily going to shrink the differences among good and bad fielders.  It is not really the fault of the metric, but the “fault” of the database it uses.  For example, I don’t know if anyone has ever done it, but we can certainly translate BA to runs (say above or below average).  And I assume that the spread of runs using BA is going to be a lot less than if we used OPS (translated to runs) or lwts.

Usually, but not always, when we use a less granular metric, we are going to have less of a spread in performance between good and bad players.  There is nothing you can do about that.

Because of that, since there is a greater spread in the fielding talent among SS and CF (because of the opportunities), a system such as WS, which does not have much of a spread in fielding, will naturally short-change those positions even if it correctly does the positional adjustments.

patriot: Because the potential value on the non-pitching defensive side is always much, much lower than on the hitting side, an adjustment that is limited to adjusting for defensive value variation is never going to be enough to reflect the impact on the offensive side.  To use Banks as an example again, Ernie’s Fielding Win Shares in ‘58 was about 1 Win Share below average.  If he’d broken a leg on the first day of the season, the Cubs might have replaced him with, we might say optimistically, a shortstop who was an average fielding shortstop and an average hitting shortstop.  That would have gained the Cubs an extra Fielding Win Share and cost them about 20 Batting Win Shares.  But because James’ Win Shares system doesn’t compare SS to SS in calculating Batting Win Shares, Banks is not given enough Batting Win Shares to reflect the actual cost in Batting Wins that the team would have suffered.  The inadequacy of a postion adjustment on the Fielding Win Shares side may partly be the artifically small spread that MGL describes above, but it is primarily just the fact of life that batting has nore potential value than non-pitching defense, and Banks is a good example of the problem.

Excellent point MGL.

However, I don’t think this necessarily applies to Win Shares.  I believe it is a system that by design prevents it from topping off too high.

birtel/42: this is purely a Win Shares design issue, and has nothing to do with all the other systems out there.  If we can agree, I can finally move on from this…

tango: I can’t consider myself expert enough on the inner workings of total evaluation systems other than Win Shares, so yes, I can agree that my end of this discussion was intended from the beginning to be directed to a specifically Win Shares design issue.  The degree to which my arguments might happen to also apply to other systems I leave to others more qualified to judge.