THE BOOK--Playing The Percentages In Baseball

I posted this over at Crashburn, but thought I’d add it here as it is directly related to the authors that run this site:

Here is what MGL wrote in his May UZR primer at fangraphs:

“UZR tries to record a player’s likely true talent and estimate his future performance based on the nuances of the batted ball and the player’s response to those nuances. It is not trying to capture exactly what happens on the field according to some arbitrary categories, like most of the offensive metrics…”

He states clearly that UZR is not descriptive (not in this actual quotation, but in another part of the primer), and here he goes so far as to say that UZR is actually trying to make claims about a players true talent.

So this really becomes a question about the WAR “framework” Crashburn and Tango have been talking about: why does the framework of a descriptive statistic contain a major non-descriptive component? And as it does contain this component, what is WAR actually telling us? Because from here, it looks like WAR is partially descriptive, while being grounded in some kind of true-talent estimations.

Maybe it’s because I follow several Phillies fans on Twitter, but it seems as if Ryan Howard is the #1 lightning rod for WAR critics.  Which indicates to me that much of the criticism of WAR can be summed up in three letters: RBI.

If MGL is regressing the observed UZR, then this would explain why the range of UZR is not as great as we’ve seen in the past.

We’ll have to wait to hear from MGL as to exactly what he meant by that quote.

Part of the RBI argument is Howard’s clutchiness.  And as Tango wrote recently, you can modify the WAR framework to use any measure of offensive production.  Use WPA.  Use RE24.  Both will put Howard’s hits and outs into more context than wOBA.  Pick what context you like and use that metric.  Both are better choices than straight RBIs, because they account for his large number of chances (in some ways).

Right.  Howard has not been clutchy this year, but last year he was enormously so.  You could add almost 3 wins to his total, that’s how good he was.  It was one of the all-time seasons in clutchiness.

But, and this is important, if you do it here, you have to do it all the time.  You can’t pick and choose to add clutch or not, depending on how you feel that day.

Personally, I would definitely add it to some extent.  WPA told us about Ortiz v ARod.  I think it would be ridiculous to call ARod an MVP if every time the Yanks had a chance to win ARod blew it, but he managed to pile up alot of HR during blowouts.

Maybe you want to add clutch/2 or clutch/5, to mitigate some of the uneasiness you have in the stat.  That’s ok.

I think the other thing many Phillies fans would disagree on is Howard’s -10 UZR at 1B.  They see him as at least league average.

As people are introduced to WAR and deal with their initial reaction to it disagreeing with their preconceptions, I think the “framework” aspect of it can be really important.

“Ok, you don’t think Howard’s a -10 fielder; but what should that number be?”

“You think he deserved credit for clutch?  How can you measure that?  How much is it actually?”

“You’re surprised by the large penalty for position?  Why?  Would you want Howard playing 3B for your team?  CF?”

The WAR framework forces people to make the attempt to put everything on the same scale.  Whether you pick numbers objectively or subjectively (UZR vs eyes), WAR forces you to translate from “great hitter, solid glove, good speed, missed time to injury” to “+40 bat, +5 fielding, +5 baserunning, +10 playing time”.

Sky, excellent, well-said.

Right, the WAR framework does exactly that.  It’s broken down by components, so now we can debate those components. We can have a healthy discussion or debate and easily reach a solid ground of understanding, if people were to simply get involved.

WAR to baseball analysts is like donuts to Homer Simpson.

The reaosn why people assume that “higher WAR equals better player” is because that’s how the vast majority of peoiple use/misuse it. People do not have innate understanding of applications ... they follow the crowd.

IMO, people also need to understand that the production from a 1B per WAR is going to be more dramatic than any other position (except maybe DH). So, the difference between a 3 WAR 1B and a 4 WAR is going to be greater than a 3WAR 3B and 4 WAR 3B. I’m not a big fan of how 1B defense is measured either. Teams don’t ask a whole lot out of 1B’s and defense, and a 1B’s defensive metric could be heaviuly influenced by the quality of 2B playing next to him. They ask 1B’s to catch routine grounders, snag some foul pops, but most importantly, dig balls out of the dirt (which is somewthing that is not measured by UZR, correct)?

Ryan Howard is a lightning rod for EVERYTHING. He’s the traditional power hitting 1B who goes deep A LOT, and as such, drives in A LOT of runs, while striking out A LOT, and hitting for a so-so average ... all in the same era as the greatest hitter of the generation, who does everything right ... and plays the same position.

Howard also got a ridiculously big contract, so now, saberists must continue to blast Howard just to let everyone know he doesn’t deserve the money and to continue the fight against RBIs being a meaningful stat.

Anyway, back to Howard, and here is his greatest value ... being an elite power hitter. This is great because, despite the slams against him, it is not entirely dependent on the quality of lineup. Howard hits 45 bombs a year. That’s 45 RBIs right there. Suppose half of those include a runner on base (fair assumption). If we leave it as just those homers, he has driven in 78 runs, almost equalling the Mariners team leader ... and that’s just runs off homers.

This is why a guy like Howard has “isolated” (it goes with him wherever he goes)value. He could drive in 100 runs hitting 8th on a bad team, simply because he hits so many home runs. There are guys that hit home runs, and there are home run hitters.

Note: I am NOT a Ryan Howard fan ... at all. But the assault on him or his perofrmance type by certain groups of fans is just ridiculous at times.

Sorry for all the spelling mistakes and typing errors. I was in a big hurry and should have just waited to post.

If I have time later, I will look up the average # of RBIs for players that have hit 45 HR’s in a season. Many fans are quick to just hand wave off Howard’s contributions as “he’s in a great lineup, almost anyone could drive in 140 runs in that lineup” ... and I don’t think that’s the case ... especially when Utley is clearing the bases 30 times a year and driving in 100 runs of his own.

My guess is that Howard’s RBI total sare pretty much in line with other guys that have hit 45 HRs in a season .... meaning that guys that hit a lot of HRs drive in a lot of runs, and not by grounding out to 2nd with a runner on 3rd and 1-out.

CircleChange, perhaps people do pile on Ryan Howard too much.  But it’s because he’s an example of someone who seems to be heavily overrated due to traditional stats that sabermetric-inclined people get worked up.

Ignoring WAR as a whole, just take the batting component of WAR (which is derived from wOBA and amount of playing time).  Howard comes in 7th in the NL this year in that among 1st basemen. 

Take 2009, he comes in 6th. 2008, he comes in 6th again. 

Now okay but you want to go with clutchness.  So lets take WPA over the last 3 years on fangraphs.  Howard comes in ....6th among NL 1st basemen.

I could go on and on.

Philly fans like to believe that Ryan Howard is a top notch 1st basemen, who is among the best in the league.  And they argue this religiously. 

And yet by nearly any measure, this isn’t true.  And that’s why we get a little crazed about this.

"IMO, people also need to understand that the production from a 1B per WAR is going to be more dramatic than any other position (except maybe DH).”

The first thing you say I take is position adjustments - a 1B will hit more than other positions who have the same WAR.

“So, the difference between a 3 WAR 1B and a 4 WAR is going to be greater than a 3WAR 3B and 4 WAR 3B.”

But this I don’t have any idea what you are talking about.

"Howard comes in 7th in the NL this year in that among 1st basemen.”

Among his peers (NL 1B) he’s only 5th in homeruns. 8th in OBP, 5th in slugging.  Pretty hard to get elite out of that.

How does he do in number of runners on base?

@Rally

[1] Yeah, I’m basically saying that 1B’s will hit more ... and their D is less important.

[2] As for Howard and “elite”. I call him an “elite power hitter” and to distinguish that category from “elite hitters”. In terms of power hitting, specifically home runs, he’s been elite over the last 5 years, even counting this year.

Howard’s never going to be a 5.5-6.0 WAR guy consistently because he doesn’t walk enough or have enough range.

I would say that he is an elite player in terms of stardom, award recognition, etc. But in terms of actual production, I limit the term “elite” to just describing his home run hitting ability (which is important).

I have already stated that i am not a big fan fielding metrics for 1B or at least the ones we use now, but how does Howard go from 0 to -9 in fielding runs in one season ... and one season where he came in trimmed down, no less?

Also, there are some years where Pujols is 30 fielding runs above Howard. I love the cardinals and Pujols and Musial are demi-gods in my house, by how the hell can AP5 really be saving 30 more runs down at 1st base? Are these diving plays? Not booting routine plays? Outstanding range on foul pops?

From playing and coaching the game, it is very hard for me to imagine how a 1B can take away 30 more runs than another 1B. Sorry, I’m going off track a little bit.

but how does Howard go from 0 to -9 in fielding runs in one season

The same way Ryan Howard goes from +61 runs to +33 runs to +20 runs to +36 runs to +19 runs in hitting.

So, we are perfectly willing to accept that Ryan Howard’s performance change in offense averages nearly 20 runs of difference in his 5 full years as a hitter, but we are unwilling to accept that his fielding went from -4 to -2 to -1 to 0 to -9 over those same 5 seasons (average difference of 3 runs).

Really?

Sh!t happens.  Sometimes the breaks fall your way and sometimes they don’t.

The snippet from my Primer is either a little (or maybe a lot) misleading or is taken out of context.

I do NOT regress (nor does FG) the UZR numbers, so they are NOT a reflection of (and estimate of) true talent or future performance.  That would be nutty anyway - to regress one-year stats and ignore other years.

What I was trying to convey - and this is important - is that, unlike offensive stats, UZR does NOT represent exactly what a player did on the field!  It is what he did plus noise.  How much noise?  We don’t know, but we can estimate (with a large degree of uncertainty) that noise.  How do we do that?  Here is an example:  Say Howard’s UZR is -10 per 150 for 2010.  It is likely that he actually fielded better than that in 2010 (and hence his WAR is better than it actually is).  How much better (IOW, how much noise is in that -10)? I don’t know exactly, but maybe -5 and -5.  IOW, he actually fielded to the tune of costing his team 5 theoretical runs (per 150 games) as compared to an average fielder and the other -5 was simply noise/errors in the data and perhaps in the methodology (neither being perfect of course).

Now, the amount of noise, in this case, -5 of the -10, we estimate is NOT the same for all players!  Let’s say that Howard was -10 last year and -10 the year before.  Well, our estimate of his true talent might be something like -8 (after 3 years of -10).  So now it is a different story. It is much more likely that most of his 2010 -10 was actual performance and only a little was noise.  How much was which?  Again, I don’t know.  Maybe it is now -9 performance and -1 noise.  I think it is always more likely that the performance is near the estimated true talent, but that there is always noise in what is left over.  But that is all I know, quantitatively.

I hope EVERYONE understands that!  I really do.  UZR is NOT the same as offensive stats in terms of retrospective value.  A single is a single no matter how lucky the player may have been in getting it.  Not so with one run of UZR.  One run of UZR does not mean that the player did anything in particular. It just means that the UZR engine THINKS that the player saved a theoretical run by making one or more good plays.

But, if a player has more good runs or less bad runs than our estimate of the true talent, then it is likely that UZR made a mistake.

Which is why I always say, and I’ll say it again and again, “Please stop thinking of sample defensive stats (other than things like error rate) as meaning anything in particular!  They don’t.  They are not designed to! Just like offensive stats such as linear weights or wOBA are NOT designed to tell you how many runs or wins a player ACTUALLY created.  They don’t!  They tell you that he got a certain number or combination of singles, doubles, etc., and they tell you how many theoretical runs those events are worth, but they certainly don’t tell you how many actual runs or wins a player created!

In a similar way, UZR does not tell you how well a player played in the field!  It only estimates it.  And thus if you combine sample offensive statistics with defensive ones, like UZR or PM, you are in some sense comparing apples and oranges.

If you want to do that (combine, say, lwts and UZR), and you want to try and make it apples and apples, you need to regress the UZR portion.

Now, in order to estimate actual performance rather than true talent, you don’t want to regress as much as you would for true talent, so in the end, it is not THAT big a deal to not regress.  For example, if Ryan Howard is -10 and you know nothing about any past defense and you have no (reliable) scouting reports on him, you might break that -10 down into this:  -4 estimate of true talent (and therefore the most likely number you expect next year or tomorrow), -7 estimate of what he actually did this year, and -3 noise. It might be -6 and -4 noise or -8 and -2 noise. I really don’t know. I know the regression to create the true talent estimate pretty well, but I don’t know the regression to create the actual performance very well at all.  I am just using WAG’s…

"how does Howard go from 0 to -9 in fielding runs in one season ...”

How did Howard go from a .313 AVG to a .268 AVG from 2006 to 2007?

For first base, -9 and 0 aren’t all that dissimilar.  One way that UZR’s advocates and critics both misuse UZR is that they take what it says as gospel.  So you’ll see mainstream bloggers scoff at UZR because it says Mark Teixeira has a “negative UZR.” The implication is that he’s a bad defensive player.  Of course, as MGL just explained, the story’s way, way more complex than that simplistic conclusion suggests.  And as Tango said above, when a player’s offensive production fluctuates, we don’t bat an eye.  Because it’s not “exact” like offense, defense is inappropriately held to a higher standard.  We accept that, over the course of a season, one player may run hot on offense for longer than usual, may get more breaks than usual, and may face easier opposition than usual.  Why cannot a player get better or worse breaks on defense in a given season?  In fact since the sample size is smaller, we should expect greater fluctuation season-to-season than we would with offensive statistics.

MGL, thanks, I really appreciate the detailed response. And I certainly didn’t mean to suggest you were regressing UZR numbers, and I’m sorry if the chunk I took was misleading (I actually thought the snipit I took matched up rather well with your response above).

I think my confusion came from Bill at Crashburnalley calling WAR a descriptive statistic. Based on what you have written, “descriptive” doesn’t really seem to be the right word. First, if I’m reading you correctly, UZR does need to be regressed somewhat if we want to use it as an actual performance measure. Second, even the batting component of WAR is an estimation, as opposed to what a player “actually” created, runs wise.

My end goal is to have the clearest understanding I can of what WAR means when I look at it. Based on what’s been written here, WAR is more a performance estimation than a strict description of what a player actually produced. If that’s not right, I’ll just need to refine my questions and look for the proper time to ask them.

Thanks again.

UZR is a hybrid stat in terms of whether it is descriptive or theoretical, I guess.  I don’t like to get hung up in semantics though.  It is as I explained it.  Whatever someone wants to call it is fine by me.

I’m glad someone brought this up.  The point to remember, I think, is that it is problematic to add defense and offense in order to come up with WAR for the reasons I articulated. You want to combine them?  First regress the defensive number.  How much?  I am not sure.  That means that WAR is going to be even more variable than before, and people don’t seem to like that. 

My response to that is, “Why are you expecting a stat that has a large uncertainty level to be one, intractable number, and why do you care if different people have different, but usually similar, numbers?” IOW, if I say that so-and-so’s WAR is +2.3 for 2010, but it could be 2.0 or 2.7, why would you care that source I says 2.1 and source II says 2.5?  You shouldn’t.  It is like when someone reports an uncertain number to 3 decimal places.  Tango hates that and he is correct.

If CC has a WAR of 5.0 and Felix has a WAR of 4.7, all that means is that it is like 52% likely that CC pitched “better” than Felix.  Does that mean that he deserves the CYA?  If your answer is yes, you are going to be wrong in giving it to him 48% of the time!

Smooth Jimmy Apollo:  [explaining his poor prediction]
Well, folks, when you’re right 52% of the time, you’re wrong 48%
of the time.
Homer: Why didn’t you say that before!!

http://www.snpp.com/episodes/8F12.html

"If CC has a WAR of 5.0 and Felix has a WAR of 4.7, all that means is that it is like 52% likely that CC pitched “better” than Felix.  Does that mean that he deserves the CYA?  If your answer is yes, you are going to be wrong in giving it to him 48% of the time!”

Are you just using those numbers for illustration or have some method of estimating the error involved? 

I don’t know how you’d do it, but if there were a reliable formula, then we could look at it like this:

1. I want to be 90% sure I’m considering the best candidates for an award.
2. 90% confidence is acheived when WAR difference = 1.8
3. Look at the leader and everyone within 1.8 WAR, and use that as your candidate list.

I’m making that number up (although it is probably close to that). If you knew the uncertainty level (SE) of your WAR estimate then, sure, you could set the level such that you are 90% sure, or whatever…

"For first base, -9 and 0 aren’t all that dissimilar.”

That would be relevant information. It could very well be my own fault for not knowing that -9 and 0 were somewhat similar. i was looking at it as “1 WAR difference” or $4.5M as FG puts the value on it.

I know that production fluctuates, but defense should fluctuate far LESS. At the plate, the pitcher and oppossing defense can do numerous things to affect the batter’s production, but in the field, I would assume that fielders generally get the same types of chances on a regular basis, unless something drastic happened like turnover or injuries to pitching staff. For example, it’s not like teams can say “Howard has trouble with one-hop liners to his non-glove side, let’s try to hit more of those”

A better exmaple, IMO, would be Albert Pujols who had 30 fielding runs one season to 0 3 years later, without adding a bunch of weight, or experiencing an injury, etc.

It just seems absolutely incredible (impossible, really) for a player’s defensive performance to fluctuate THAT much. I am saying that from experience playing/coaching, and from a statistical research perspective. It is just very difficult for me to comprehend how a defensive player, at such a low premium position, could range from 3 WAR (fielding only) to 0 WAR (fielding only) or even negative WAR. This isn’t an “outfielder after knee surgery” or “rapidly aging shortstop” type situation.

If the defensive metrics do range so greatly, why do we place such weight or porecision on it when figuring WAR ... and then treat WAR as if it is THE quantitative metric, and refer to it for player awards, contract values, etc? From what I understand in this thread, the variation or estimation in defensive metrics could vary as much as 1 WAR or more.

The range of the values could be such that the metric only tells us something like the fielder is somewhere between slightly below average to slightly above average, which in the end, doesn’t really tell us much.

Certainly defensive metrics are more complicated than others, and I am not suggesting that I have something better. I am curoius as to why we use such a precise number (when a range is possible) when calulating something that we treat as most important (WAR).

Circle I am with you in regard to the skepticism toward WAR.  I don’t think we should hold any value as holy and inviolable.  One thing that irks me is when people just drop WAR like they used to drop VORP as if that’s all which needs to be said; the discussion cannot be killed that easily.

However, defensive opportunities work can and do fluctuate in regard to their inclusion in UZR. It is true that most plays are easy, and they’re easy for everyone.  Even though, defensively, the majority of plays are easy to make, it is nevertheless the case that a player may be asked to field more challenging plays in any given year, and that this could affect his UZR.  Say there’s one play which the average fielder makes, hmm, 55% of the time.  But suppose for Ryan Howard the play is not as “easy” as it is for others; maybe the ball is hit at a worse angle than it is for other opportunities in that zone, and maybe it is at the edge of the zone, making it even harder to field, or maybe the baserunner is not Jorge Posada and it is Brett Gardner.  Lots and lots of stuff can go unnoticed.  So even though the actual play is classified as 55%, it’s “really” 35%.  It takes sample size to smooth these edges out…

Remember, UZR is relative to an average fielder at a given position.  A bunch of things can change a fielder’s score that have nothing to do with his performance.  Say Derek Jeter in 2012 fields the same number of balls as he did in 2011.  Same plays, same zones, same everything.  His UZR in 2012 is -15, and his UZR in 2011 is -10.  How’s that?  It’s possible because the average shortstop did not field as well in 2011 as he did in 2012, so Jeter comes out looking a bit worse.

Please correct me if I’m wrong someone. 

I’m fine with small variations. It happens, and it could happen as you say ... without the player fielding any differently than they did the prevoius year.

Awile ago, I looked up differences in Pujols and Howard in terms of WAR because IMO WAR was selling Howard a little short.

So, I noticed the fielding metrics:

Pujols 2007: +21.7 fielding runs
Pujols 2010: +1.9 fielding runs.

Looking at the metrics more closely the difference is all in the range component, since AP5 made half as many errors in ‘10 as he did in ‘07.

1B is not really a range position, outside of foul pops. You basically get to balls one step and a dive in either direction.

The one difference between 07 and 10 for Pujols is Aaron Miles in 2007 and Skip Schumaker in 2010. It may be possible that Pujols was playing more towards 2B to help with grounders there, versus being able to play closer to the bag with Schu at 2B.

I have stated before that with Utley at 2B and Howard at 1B, I as a manger would tell Howard not to worry about anything to his right, unless all it requires is a lean in that direction. Chase will get the balls to his left. So, something like that might affect the range component for 1B’s that have really good 2B’s.

My comments are mostly me trying to envision what the difference between 20 fielding runs (2 WAR) and what it would look like on the field for a 1B. Pujols is solid fielder annually, but he doesn’t make a ton of great diving or leaping catches, but does have good hands and quick feet. I don;t often see him ranging into shallow RF for pop ups (aside form the one highlight where he ran up on the tarp), or hanging over the dugout railing or things of that nature.

Let me deal with purely factual data with Pujols. Ready?

YEAR BIP OUTS
2001 806 55
2002 415 35
2003 1,122 78
2004 3,712 294
2005 3,867 295
2006 3,604 270
2007 3,938 309
2008 3,575 289
2009 3,915 351

That’s how many balls were in play when Pujols was playing firstbase, and how many he was he first to touch that led to an out.

The only exclusions to the data are:
- out of the park home runs
- bunts (or more accurately, plays marked as bunts)
- pitchers-as-hitters

Focus on 2007 and 2009:
YEAR BIP OUTS OUT_RATE
2007 3,938 309 0.078
2009 3,915 351 0.090

So, facing the same number of balls in play, Pujols got 42 more outs.

Now, one could ask: “how is that possible?”.

The answers could be:
1. different hitters or pitchers that tend to get the ball toward the right side more in 2009 and less in 2007

2. same hitters and pitchers, but those guys happened to get more balls to the right side

3. change in talent level or positioning

4. luck

What do we find?  Well, first off, the avg 1B got to more balls in 2009 than the other years.  That by itself accounts for a third of the gap in plays.

Pujols’ pitchers favored getting balls hit to the right side alot more in 2009 than the other years.

Plays were marked more as being groundballs than airballs in 2009.

Add it all up, and we can explain a good portion of the gap, but not all of it.  The remaining gap would be his talent or luck.

Strangely enough, UZR has him at +25 in 2007 and +3 in 2009, pretty much the opposite of what one might expect with no adjustments.  So, with MGL’s adjustments, it actually swung the other way.

And he’s not alone here.  Dewan has him at +29 to +12 from 2007 to 2009, even though Dewan notes that Pujols made 198 plays in 2007 and 217 in 2009.

As you can see compared to my number of plays, I count an extra 100 outs that Dewan does not.  Which goes to another possible difference that Pujols got a whole bunch of easy popups in 2009 that he didn’t in 2007.

TZ went from +19 in 2007 to +11 in 2009.

All to say: this stuff is tough to do, if we are dealing with data that has not been well-classified.

My preference is to use half the Fans Scouting Report and half the data, because, basically, there’s so much we don’t know about the data, and so much we (think we) know about how good he should be fielding.

Basically, add a bit of cold water to that hot water, so you don’t get scolded.  That hot water is still valuable.

Thanks Tom.

Whenever I see large ranges in a metric, my thoughts move toward “What does this look like on the field?”.

So, in this example, my mind wonders if it is due to Albert ranging more right, or taking more pop ups that the C might nor,mally take, etc ... just like with BABIP, and I wonder if Josh hamilton is hitting more liners to the oppo gap, or is he pulling groundballs harder through the 2B/1B hole, or is he just getting those ducksnorts to fall.

The stats represent actual plays (i.e., a spray/range chart for fielders), so I try to envision what they might look like to get an idea of what the differences could be.

Thanks for the explanation.

#26 is right on the money.  And how many times do I have to say this:  UZR does NOT tell you what happened on the field. It is an approximation of what happened on the field!  And technically, you have to regress any UZR if you really want to accurately guess what happened on the field. I probably should have included a regression in the UZR engine, but unfortunately I didn’t. The raw number does not mean anything.

So, not for the last time, WAR with UZR (or TZ, Dewan, Pinto, etc.) should be a regressed version.  The non-regressed version is meaningless (well, not meaningless, but you know what I mean).

Here is the way it works:

I tell you that someone is -10 in half a season. I also tell you that there is a large uncertainty around that number, mainly because the data is not nearly perfect (we don’t really know where the ball was hit, how hard it was to field, and where the fielder was stationed), nor is the engine that computes that number.  You want to know what the most likely number is that represents what actually happened.

Guess what?  It ain’t -10!  That is because the error bar is not symmetrical. It is much more likely that what happened is closer to the player’s population mean. If we have no other info on the player, that mean is zero. So it is much more likely that what happened is closer to zero than -10!  Now, if the player was -15 the last 3 years, then his “population” (all players with a -12 in the last 3 years) mean is probably close to -10, so we can estimate what “really happened” at -10.

So, if Teixera who historically has a good UZR and is considered by the fans and scouts to be a good 1B, has a 2010 (and he is healthy) UZR of -10, his defensive part of WAR is NOT -10!  Please repeat that to yourself 100 times and tell all your friends!

This is a mistake that we have been making with WAR for a long time, and I am trying to correct that mistake.

What effect does positioning have on UZR TZ etc.?  A firstbaseman can be holding the runner on, guarding the line, charging a bunt, double play depth, shallow, etc., etc.  If the player had more plays with a runner on first, where he was holding him on, thus making it more difficult to field a ball at a certain angle away from the line, would he be penalized?

UZR adjusts for the base runners and outs, and treats bunts separately. I don’t know about the other metrics…