THE BOOK--Playing The Percentages In Baseball

The three-worst OF are all with Seattle, a combined -65 runs?

Ibanez everyone believes.  Guillen, maybe.  But Ichiro?  At mid-season he was setting the record for most putouts ever, wasn’t he?  Are the pitchers at Seattle FB pitchers?

Can you show us their numbers at Safeco and away?  And maybe all their opponets as a group, at Safeco?

On the other hand, everything else looks good, especially Jeter.

AL LF: No one above zero...!?

Carl Crawford’s ZR is .934 and is the highest between all LFs (according to ESPN stats), is his overall performance that bad? (only 3 assists)

Is that Aramis Ramirez as a +6 at third? That and Ichiro are the only real surprises there.

Is the typical AL leftfielder 10 runs worse than average? That’s remarkable.

JinAZ uses the THT/BIS data to come up with his numbers:
http://jinaz-reds.blogspot.com/2007/10/2007-fielding-data.html

The process he does is, IIRC, identical to the way I was doing it.  Since he published first, I won’t bother to put out something that is likely identical.

Ichiro here comes out waaaaay different than UZR.

The Mariner fans have told me that Ichiro’s been getting to every single discretionary play (and even non-discretionary).  That is, it’s possible that a TON of those out-of-zone plays are plays that a normal LF or RF would have made.  But, Ibanez may be the worst LF around.

Definitely a great case study of the Mariners OF can be made.

Ian Kinsler comes out as +13 on jinaz’ sheets, and -9 here.  Glaus shows a big discrepancy too.  He’s someone that a case study should be made of.  He looks about as mobile as my grandmother, with slightly more injury-prone legs, but he’s clearly either fooling me or some of these fielding rating systems.

The Fans see Kinsler and Glaus as both a bit below average.

Among 3B, Glaus is the second-lowest of all 3B.  However, he’s got a plus arm.  The ONLY other 3B with minus speed and a plus arm are Mike Lowell and Aramis Ramirez.  Of those 3, he’s the worst, by virtue of his minus first step.  (Those guys are average).  Glaus, compared to those two, is about 10 runs worse according to the Fans.

jnaz’s sheets have Grady Sizemore as the 2nd worst centerfielder in baseball, and UZR has him as by far the best.  They also still love Andruw Jones, who is always rated really differently by different fielding systems.

Yikes, looks like Pat the Bat lives up to his name here.

The totals for each league per 150 are as follows:

1B

NL = -.3 AL = -.07

2B

NL = -2.45 AL = 1.95

3B

NL = -.19 AL = -.41

SS

NL = 3.9 AL = -5.2

LF

NL = 1.98 AL = -1.2

CF

NL = -1.76 AL = 2.3

RF

NL = 1.1 AL = -2.2

The above numbers and the ones in original post are NOT zero’d out per NL +AL for the year, as I originally stated.  They are zero’d out for the last 4 years combined.

Crawford is exactly zero.  In the AL, the only plus LF’ers are the ones who did not qualify.  Strange.  Guys like Botts, Brown, Damon, Gathright, Johnson, Lind, Mackowiak, Michaels, Perez, Podsednik, Tyner, and Willits.

Fans have Sizemore as an above average CF.  Here are all the CF who are big-plus (score of at least 70) in instincts, first step, speed, and hands (i.e., catching the ball):

Suzuki, Ichiro
Beltran, Carlos
Pie, Felix
Hunter, Torii
Wells, Vernon
Cameron, Mike
Granderson, Curtis
Duffy, Chris
Sizemore, Grady
Crisp, Coco

The order of the above list is from best arm to worst arm.

***

Just comparing MGL’s list to the Fans’ list.  Just for 1B, Pena and Berkman come out much better with the Fans than UZR.  The rest were ok.  From 2003 to mid-season 2007, Pena and Berman’s UZR were consistent with MGL’s 2007 final numbers.  This might mean that Fans might be missing something with these two guys. 

(Remember, unless a stat is biased, at some point, the data overrides the observation.  However, we don’t know if UZR or the other stats may be biased.)

How did Ryan Braun not end up on the “worst” list?  It appears that he had enough innings at 3B to qualify.

MGL, where does John McDonald come out as a SS?  His name’s being thrown around as the best in the AL (and Justin’s THT +/- agrees).

Are you going to publish the full UZR list like in the past?  A full comparison between UZR, Fans, and THT’s +/- would be interesting.

John Dewan published BIS +/- leaders & trailers at shortstop last week at Stat of the Week.  Thought they’d be interesting for comparison:

Tulowitzki +34
McDonald +26
Vizquel +19
Bartlett +18
Pena +18

Eckstein -15
Young -15
Harris -19
Ramirez -35
Jeter -35

Tulowitzki, Visquel Pena, & Bartlett all show up on MGL’s top-3 lists, while Jeter, Young, Ramirez, and Eckstein all show up on the bottom-3 lists, so there’s good agreement overall.

The simple THT+- conversions I did at shortstop had a correlation of 0.95 with Dewan’s data (biggest discrepancy was Harris).  RZR, of course, is based on the same raw data as Dewan’s more advanced +-, so you’d expect a fairly high correlation.  But I was encouraged by that.

THT vs. UZR tend to have weaker correlations with one another based on past data, which is discouraging.  But as Michael Humphrey’s study at THT showed, there seem to be differences in the data provided by BIS and STATS.
-j

If you’re working with zone rating data, my advice would be to do the THT ratings like JinAZ did (I prefer innings instead of BIZ for my OOZ denominator, but I’m in a minority and its not that big of a deal) and also do Stats zone rating like Chris Dial did, and average the results.

I compared both systems at the team level to DER, and also to the Hardball Times defense rating on the team page, which is basically DER adjusted for the batted ball mix fielders are facing.

In both cases, averaging the two zone ratings gives me better correlations than using either one by itself.  The design of RZR is better (the in zone/OOZ stuff) but STATS is data is very good and certainly adds to our understanding of defense.

Having watched as many Ms games as I did, I can buy Ibanez being that bad, I can even buy Guillen being bad (though worst in the league seems a bit harsh), but Ichiro? There’s just flat out no way. Something’s wrong here.

Braun would certainly be the worst, if he qualified.  Even using zero UZR to make up the games he “missed” he is still probably the worst.  That is a shame since he is such a good hitter, and his poor defense takes most (75% or so) of his hitting value away.

McDonald would be a top SS if he qualified, +21 per 150 for 89 games.  However, he was never a great SS (UZR-wise) before this year, so I would hesitate to call him “one of the greatest” as opposed to “he WAS one of the greatest this year.” Everyone tends to call players “this and that” based on their latest performance, whether that be this season, the last 3 months, or last 10 games.  What always cracks me up is when a player makes a great play in the field or hits the ball extremely well and the TV commentator goes on about great he is (which he isn’t) as a result of that ONE play.

As fas as Ichiro, put it this way.  UZR never liked him in RF, so he was likely going to be quiet a bit worse in CF.  Who knows why.  I can live with it either way (Ichiro great, good, average, or bad) though.  It’s only one player.  Keep in mind that as I always say, if there are NOT some surprises (and even some shocking ones) when comparing the data to what we think, what is the point of the data (and either the data is not very good or the thing we are trying to measure is easy to observe, which we know defense is not). I’m not saying that UZR is right on Ichiro, but you never know.  One thing that can sway us one way or another, of course, are the other PBP defensive metrics (the ones that have similar accuracy).

One more thing. Tango’s last line in #9 is important and profound.

Any system rating Ichiro at -10 plus runs and Ordonez at +10 plus runs has significant problems.

From eyeball observation across maybe 35 games, Ichiro ran down and properly handled 30% of the balls that most every other center fielder leaves to LF/RF.  And Pierre of the Dodgers (based on 80 plus observations) does the opposite, both as to range and proper handling once reached.

Defense is way, way, way under-understood.  Both how to measure--I have grave doubts about all these ‘systems’--and its significance to pitcher performance.

Clem/16: don’t just say that defense is “way way way under-understood”, thinking that your three “way"s is any more important than 2 or 1. 

Proclamations without anything to substantiate it is not a way to get a conversation going. We’re not politicians!  Those guys are quick to tell us what is wrong without: (a) telling us the reason, (b) or offering any kind of improvement or alternative.

Here is MGL explaining UZR:
http://www.baseballthinkfactory.org/files/primate_studies/discussion/lichtman_2003-03-14_0/

http://www.baseballthinkfactory.org/files/primate_studies/discussion/lichtman_2003-03-21_0/

Take an hour (or two), and read those if you haven’t.

I posted this at USSM, and will repost here:

Correlation of UZR is r=.50, when BIP=400 (100 games). That compares to component pitching ERA of r=.50 when PA=300, and RC or LWTS of r=.50 when PA=200 (50 games).

In short, you need 2 years of UZR or other fielding metric to be as reliable as 1 year of hitting stats. How reliable is Andruw Jones’ hitting stats this year? Right, so you’d at least like to have 2 years of hitting stats, if not 3. So, you’d like to have 4 years of fielding stats, if not more.

That’s just the nature of the beast though.

I’ve put together a spreadsheet with both RZR ratings, similar to what JinAZ did, and Stats zone ratings, using the Chris Dial process, and an average of the two.

http://home.comcast.net/~briankaat/zonecombo2007.xls

Ichiro is by far the biggest disagreement.  Stats ZR has him below average, -7.  Its close enough to the UZR total, not surprising since they use the same data source.  His RZR rating is +36 mostly due to all the OOZ plays.

Could be that we need to rethink the way we give outfielders credit for OOZ plays, if Ichiro is getting credit for taking an easy pop fly away from the corner outfielders.

Also, I almost wonder if the scorers from the two data companies are watching the same player.  According to STATS, Ichiro allowed 63 fly in his zone to fall in. From BIS, the number is only 39.

In general STATS appears to have larger zones, but some players actually have more in zone hits counted against them by BIS than STATS (like Grady Sizemore)

It looks like the Stats data is just much more conservative than the BIS data.  If you take Rally’s sheet and look at all disagreements of 15 runs in either direction, you basically see all the guys that BIS is in love with.  Ichiro, Granderson, Jack Wilson, Andruw, Wright, Pujols, Glaus, and Tony Pena rank as a combined +240 on RZR and +25 on Stats ZR. 

It’s not quite as extreme on the opposite side, but there’s still a big difference.  Braun, Manny, Bautista, Atkins, Hanley, Jeter, Burrell, and Encarnacion are the bottom eight by RZR for a combined -211 compared to a -113 for Stats ZR.

The total combined runs saved in Rally’s sheet for RZR is +16 and +1 for Stats ZR.

Anyway, getting back to Rally’s point, it seems clear that the OOZ plays are the difference.  All eight guys with 70+ OOZ plays by BIS data are ranked massively better by RZR than Stats ZR.

Am I the only one who think Rockies’ fielding is funny?

Helton +18 (#1 in NL)
Matsui (not qualified)
Tulowitzki +20 (#2)
Atkins -23 (second to last)
Holliday +13 (#2)
Taveras (not qualified)
Hawpe -15 (second to last)

BTW, Torrealba’s .197 CS% also ranked second to last in NL.

OOZ plays are a relatively small sample for outfielders, and don’t tell you much unless you know about hang time.  If Ichiro is making those OOZ plays with an average hang time of 3.5 seconds, he had a great defensive year.  If the hang time was an average of 5 seconds, he did not.

Seasonal zone rating of all types has greater accuracy for infielders than for outfielders, for the simple reasons that location is a more important determinant for ground balls than it is for fly balls of difficulty, and that the sample sizes are larger.

Tango/MGL,

Where do UZR and the fans have Nick Markakis? 

He looks like an excellent fielder to my eye.  He makes diving catches frequently and almost never misses, and he has a strong, accurate arm (13 assists).

It looks like other fielding systems have him all over the place - from well below average to well above average.

Is he just a smooth-looking player who gets bad jumps a la Jeter, or is he a legitamately good fielder who gets unfairly penalized by the Camden Yards scoreboard?

I provided the complete list in the other thread (look for a thread titled “preliminary” from the main page).

Markakis is #2 in RF, IIRC.

Rally/19, that’s awesome!

Dave/20,
Anyway, getting back to Rally’s point, it seems clear that the OOZ plays are the difference.  All eight guys with 70+ OOZ plays by BIS data are ranked massively better by RZR than Stats ZR.

This is probably due to how OOZ plays are handled in STATS Inc’s zone rating.  They calculate it as (plays+ooz)/(biz+ooz).  This massively devalues OOZ plays.  The approach I used is basically (plays/biz) + (ooz/biz), which is the same as (plays+ooz)/biz. 

I compared these approaches here last spring using the “raw” THT data:
http://jinaz-reds.blogspot.com/2007/03/how-should-we-calculate-zr.html
Including OOZ in the denominator is results in almost exactly the same estimates as leaving OOZ plays out altogether.  Furthermore, I did some later work:
http://jinaz-reds.blogspot.com/2007/03/how-should-we-calculate-zone-rating.html
...that indicated that a 1:1 weighting of BIZ and OOZ provided the best match to PMR at all positions except 3B.  That, of course, doesn’t mean it’s the right way to do it.  Just that it’s consistent.

I guess we could exclude OOZ plays altogether from outfielders because of the problems associated with ball-hogging.  But I tend to think that they provide a lot of useful information, so I’m hesitant to do this.

For example, Griffey is rated as an average fielder in RF if you look only at balls in zone.  It’s the out of zone plays where he’s well below average.  Given his performance in CF the past several years, I tend to think that a below-average rating in right field is the more correct conclusion, and you only get that by including ooz information (this also conforms to UZR ratings).
-j

Thanks Tango, I found the link.  Markakis did indeed get high marks from the fans, with a high agreement %.

Another interesting Oriole is Miguel Tejada. 

Most of the fielding systems have Tejada as roughly average to slightly above average.

However, virtually every article written about him recently talks about how he is an atrocious fielder who’s switch to third base is long overdue. 

Tango’s survey indicates that fans agree that Tejada is terrible.  Whether that’s from their own observations or from reading those articles is another story. 

I think he looks about the same as he’s always looked - slightly below average range, well above average arm.

It will be interesting to see how this will affect the trade market for Tejada this winter.

IIRC, Tejeda’s done very well with the Fans over the years.

To me, the most fascinating is when the Fans drop so fast on a fielder, like with Betancourt this year, or Junior last year.  This either indicates that the Fans know something real that suddenly showed up (Betancourt’s accuracy went down the toilet), or that they finally accepted the inevitable (their hero was no longer that good). 

Any big drops of guys over 33-34 years old is Fans realizing that their rose-colored glasses should have been cleaned two years earlier.  Any big drops of guys under 30 is, in my opinion, something very real.

I know you ask fans not to consider stats, but I wonder how much of it is fans knowing the stats are bad on Yuni and looking at him more critically?

My glasses are never rose colored towards the Mariners.  From watching 20 or so games against the Angels the last two years he doesn’t look any better or any worse to me.

I’d bet that 90% of the change in Betancourt’s rating is due to his error total.  He had a well documented throwing problem during the first half of the year, missing first base badly on a lot of routine plays.  This is probably the easiest thing for fans to pick up on, and I’m pretty sure it influenced how the Mariner fans about his defense overall.

And in fact, that was where most of the change was on Betancourt.

In Fi Sp Ha Re St Ac
87 89 78 90 91 81 85 2005
86 85 77 84 85 77 74 2006
77 78 72 74 57 73 23 2007

Remember that a 1 point overall difference = 0.7 runs.  For simplicity’s sake, that means for each of the seven traits, a 1 point difference = 0.1 runs (or 10 points = 1 run… gotta love easy rules of thumb).

If you look at his Instincts and First Step, Fans see a 1 run drop.  His Speed barely changed.  His Hands was a 1 or 2 run drop.  His release was a 3 run drop.  His arm strength barely changed.  His accuracy was a 5 or 6 run drop.  Total it up, and his “getting to the ball” categories was a 2 run drop.  His “holding/throwing the ball” was an 10 run drop.

His poor throwing, I think, seems to have been fairly reflected by the Fans.  There may have been *some* fallout into the other categories.

In need to change my COL UZR park factors.  I have not adjusted for the “long-humidor effect” of the last couple of years yet.

Markakis is -5 this year and Tejada is -8.  Yuni is -3.

Last year, they were:

Markakis +2
Yuni -6
Tejada +3

05 was:

Yuni -2 (limited time)
Tejada -4

UZR does not include arm ratings.  It is only range and errors, including throwing (or any other ones, like receiving errors on a throw)errors.

I should note that this is critical: since we know, via observation, that Betancourt regressed by around 12 runs, then if you see a UZR drop of 12 runs, that’s a real drop.  UZR, if it sees a 12 run drop for ANYONE would regress some of that difference.  It must chalk some of that up to sampling error, and some of it to true talent changes.  Since it doesn’t know which is which, it has to treat every player with the 12 run drop the same, and therefore might conclude that the true talent was only a 4-run drop.

On the other hand, if UZR knows that Betancourt truly was subperforming physically, it is more certain that its 12-run drop was real, and therefore has a new regression point to move toward.

HOWEVER, we don’t know if this true talent change is correctable or not.  We don’t know if his laziness (or whatever it is that caused his sudden change) to persist long-term, or if it can be considered a “short-term injury”.

DOUBLE-HOWEVER, given that he has had this “short-term injury”, he may be more susceptible to fall into this than, say, Tulo.  If that’s the case, then you can’t completely discount the change.

Tough call…

MGL and I cross-posted.  The UZR he posted for YuBet shows a slight improvement every year, something consistent with someone who is learning in his 20s.  Those numbers are fairly static.  As far as UZR can see, there’s been no sudden change in performance. 

However, based on the Fans’ observations, the 05/06 UZR shows numbers that are much too low, and the 07 numbers are either right-in-line, or a bit too low.  That is, UZR, going into 2007, would have regressed his 05/06 performance toward the league mean somewhat, and guessed say a -4 performance.  What it should have done, using Fans’, is regress that performance toward a +15 or so, to come up with say a +7 as the best guess.

Coming into 2008, UZR will guess say a -3 performance now.  With the new Fans, it might now guess a +0 performance.

That is, while the UZR values themselves hardly changed, its forecast would go down by 7 runs between 2007 and 2008, because it now knows that there’s been a drop in talent levels.

Speaking of defensive park factors…

Listening to the Col/Phi game right now.  Steve Phillips just commented that an outfielder (I think it was Victorino) had an advantage in his home park because he knows the dimensions and therefore knows how much room he has and where balls will land.  The prediction from this idea would be that the average park should have a positive outfield fielding park factor, at least based on home team performance only (i.e. ignoring visitor defenders).

Anyone ever looked at this?  Just curious.
-j

J, I just looked.  I used TotalZone, my retrosheet based system.  From 2003-06, outfielders were actually better on the road.  The effect is so miniscule that it can’t possibly be significant.

Home Of’s are -0.1 runs per season, road +0.1

Thanks for taking a look!  Always interesting to test claims like that. -j

Here another one that I don’t have time to test right now:

It has been shown that extreme groundball pitchers have more of their groundballs turned into outs.  I know MGL makes an adjustment in UZR for pitcher’s groundball tendency.

Does this advantage show up mainly when the groundballer pitches at home?  I’m watching the D-Back - Cub game and the announcers are talking about things the groundcrew can do to help a groundball pitcher - leave the grass long, dampen the area behind home plate.

You’d think if a Brandon Webb type was pitching on the road the team facing him could tailor the field so that as many groundballs as possible go through.  I have no idea if any of this makes a statistical difference.

Funny, that came up earlier with Wang struggling in Cleveland. He’s consistently put up worse numbers on the road. Looking at his career splits, the one huge difference is singles (click on my name for quick & dirty breakdown). Wang’s major league results:

Home: 967 BIP, .264 BABIP
Road: 826 BIP, .318 BABIP

Not sure how much weight we can put on that number of balls in play.

re: #38

Wang pitched 11 of 39 (28%) road games on turfs (TBD, TOR, MIN) in his career, I don’t know what kind of impact it really has, so I just post the numbers:

TBD: 157 BIP, .318 BABIP
TOR:  36 BIP, .500 BABIP (sample size)
MIN:  41 BIP, .268 BABIP

Turf:  234 BIP, .338 BABIP
Others: 588 BIP, .310 BABIP

And I think all teams who face him know their ground crews can do something on their infields to neutralize his sinker. So it seems reasonable that his AWAY BABIP is much higher (vice versa). Is that right?

Derek Lowe’s pitched part of his career in a hitter’s park and part in a pitchers park.

BABIP:  home .284 road .299

Brandon Webb home: .297 Road: .284

He plays in a hitters park.  Maybe the groundscrew needs to do a better job.

Thanks for posting all this MGL. Regarding McDonald this is the first year he’s been an everyday player. Prior to this season he’s mainly been a late inning sub and spot starter at a variety of positions. Do players do better, fielding wise, playing on a consistent basis as opposed to being jerked in/out of the lineup? This could explain the incongruity of McDonalds past UZR’s .

#41, I don’t know.

#40, I have a pretty large infield park factor for Arizona.  The IF is lightening fast.  Part of that I think is a hard, dry infield, and part of that is the altitude.

#34, That was Steve PHILLIPS!  Just take the opposite of whatever he says and you pretty much can’t go wrong.

I don’t find much difference in H/R UZR stats or just normal how often balls are fielded, I don’t think.  HFA is almost entirely in BB, K, and triples rates.  Chasing down flly balls (or fielding ground balls) doesn’t really have anything to do with being familiar with a park or not (with all due repsect to Mr. Phillips, the ex GM).  Playing caroms and things like that does, which is why the 14% or so greater triples rates at home than on the road.

#34, That was Steve PHILLIPS!  Just take the opposite of whatever he says and you pretty much can’t go wrong.

LOL!  Yeah, that’s about right.  I still try to keep an open (yet critical) mind about stuff he says--thanks to you and Rally for taking a look.

I have to say, though, I’m getting incredibly tired of his constant concern about how “into it” the fans are at games.  I just can’t imagine that has a significant impact on anything except maybe communication among fielders on pop-ups.
-j

Are there any plans to post the full year 2007 UZR data?

Not really.

The gap between the best and the worst defensive possible is 327 runs acording to this metric.  That’s a 2 run per game gap in ERA between the bad (5.5) vs. the good (3.49) given an average team ERA of 4.5 over the course of a season.  These numbers are unsupportable by any rational statistical or logical means.  If UZR doesn’t pass this simple sanity test, then how can we take it seriously.  The methodology needs to be rethough.

Eric,

With all due respect, your thinking needs to be rethought.

Please read the WOWY article in The Hardball Times Annual, and that will open your eyes as to the impact of fielders.

That article manipulates the data to its minimum, but it is basically foolproof in its basic conclusions. 

While the merits of UZR are hard to discuss because there’s alot going on under the hood, everything in WOWY is above board. 

Let’s talk about the merits, the actual methodology.  That’s our basis for discussion.

The difference between the best and worst possible lineup is at least 1000 runs, but you don’t see any teams scoring 1300 runs or 300 runs over a full season.

The difference between the best actual defense and the worst, say Rockies vs one of the Florida teams, is about 150-200 runs.  You can check this without any defensive metric, just look at simple defensive efficiency ratings.

Even a good fielding team like the Rockies has poor fielders like Atkins and Hawpe.  And a terrible team like the Rays has Carl Crawford in left.  Replace Atkins with Feliz or Inge, and Crawford with Manny Ramirez, do a few more subsitutions and a 300+ run gap between the best and worst defense seems very reasonable.