THE BOOK--Playing The Percentages In Baseball

Great data MGL, thanks for sharing. 

Mientkiewicz, D 3 1345 74 8
Dunn, Adam 3 569 32 8
Pujols, Albert 3 1112 52 7
Lee, Travis 3 738 34 7
Olerud, John 3 875 36 6

I assume you just mislabeled?  On FanGraphs, Dunn has had a -17 UZR/150 since 2002 at 1B.

Was gonna say the same thing re Adam Dunn.
Also, Bill Hall seems to be a mistake. He has appeared in 809 games in total, spread over all positions except catcher and first base. He has not played a full 500 games at shortstop, nor does he have over 900 defensive games.

Sorry, I originally screwed up the tallies.  The correct lists are now posted.  Of course Dunn is not on the list, even as a first baseman!  Thanks for the catch guys!

Wow, those names are decidedly non-controversial.  Amazing what happens when you have a decent sample size. 

Although I guess you’d still get yelled at for not having Torii Hunter.

MGL - Am I correct that your methodology would have each position zero out for total runs saved and lost?  If so, the top 5 in total runs saved seem to have a particularly low total runs saved of 220 runs with the number 5 fielder in runs saved at 13.  Are the vast majority of left fielders between -13 and 13 total runs saved?  I have Ramirez, Dunn, Matsui, Lawton, and Dunn as close to -280 runs from 2001 to 2009 in left field from Fangraphs UZR.

When you put a list out like Rob did with my top WAR of the decade, you are always going to get yelled at for something.  “How come my player X is ranked 67?  He’s way better than Y at 54!” or people complaining about a current star who’s played the last 4 years ranking below a guy who was good for the first 6 years of the decade and now is out of baseball or a scrub.

I imagine there will be flak for Barry Bonds and Juan Pierre ranking with the good fielders.

I have two questions on UZR and the PBP data, one small and one big.

Small:  Do you check to see if there is a “ball hog” problem with any of the highest-rated OFs?  For example, is there any tendency for ATL corner OFs or middle infielders to field fewer than expected flyballs during Andruw’s tenure?

Large:  One problem I would anticipate with PBP data is that coding will be influenced by the proximity of the fielder and outcome of the play.  Take two identical GBs to 3B-SS hole.  In one case, the SS is shaded toward 2B, gets a bad jump, and never touches the ball; in the other case, the SS is positioned toward 3B, moves on contact, and gets the runner by at least a half step.  It seems likely to me that the first GB will tend to be scored as harder hit and in a tougher zone than the 2nd GB, despite the best efforts of the scorer.  (I’m sure I would do that). We’ve talked about this potential problem with respect to LD/FB scoring, but it’s really broader than that.  And to the extent this does happen, it will tend to “regress” the PBP data, making great fielders look less effective and lousy fielders look better.

So what steps are taken to help scorers not be influenced by fielders’ positioning or success/failure?  And what analysis of the data has been done to establish whether or not this is a problem? 

(I thought about looking at the correlation between UZR/150 and expected outs per inning, to see if there was a positive correlation.  But there will be a countervailing negative correlation, since players with very high expected outs will tend to have low UZRs (since UZR is overestimating how easy the chances really were), and vice-versa.  So I don’t think that will answer the question.)

My question: BIS is only since 2002… Is MGL’s data based on STATS for 2000-2001, and BIS for 2002+ ?

As I’ve noted, the player with the hugest difference between bUZR and sUZR is… Andruw Jones.  Something like over 100 run difference.  That’s huge of course.

Rather than the “best” list, I’d rather see more of the difference between bUZR and sZUR.

Is the Jeter gap of a similar magnitude?  He was terrible under sUZR, but is tolerable in bUZR, if I recall correctly.

I second Tango/8.

I wouldn’t have thought that Barry Bonds would be the fourth best LF during 2000-09 based on runs per 150 games.  Not that one surprising data point can disprove the usefulness of the data.

Pierre is at +84, but Fangraphs has him around +60 for range but -40 for arm.  Is there an error (subtracting insteaad of adding arm?), is his 2000-2001 sUZR great enough to make up the difference, or are you using sUZR for some of the years where bUZR is published?

WOWY hates Juan Pierre.  I have him at -50 to -100 runs (depending whether I control for batter, pitcher or park).  Rally has him at -14 runs.  MGL has him at +84 runs.

Fangraphs UZR totals 2002-2009 for left fielders:

Reed Johnson - 42.6
Coco Crisp - 35.7
Ryan Langerhans - 21.8
Alfonso Soriano - 18.2
Eric Byrnes - 16.0

Perhaps MGL is using STATS UZR for the entire decade?  If so, I can’t imagine that the datasets for STATS and BIS vary that much to create some of the differences that we are seeing.  MGL must have a mistake in one of the methodologies that is causing these differences.

I wondered that too, Peter, but I don’t think Andruw rated that highly by STATS data.

Peter, imagine away:

http://www.insidethebook.com/ee/index.php/site/comments/suzr_v_buzr/

The result?  In sUZR, he’s a collective -5, and in bUZR he’s the best fielder in baseball from 2003-08, at +112 runs.

So, MGL is probably using sUZR for 2000-01, and bUZR for 2002-09.  That’s the only way to get Andruw Jones on his best-fielder list.

Tango: 
A tangentially related question on WOWY:  I believe the denominator you use is BIP.  Rally made the point, in a recent BTF thread, that WOWY is in a sense including a GB fielded by the 3Bman as an “opportunity” for the CF.  Instead of using BIP, could you use BIP that are not fielded by any other fielder?  So, for a SS, the denominator becomes hits + SS outs + SS errors.  I suppose this reduces sample sizes in a sense.  But you incorporate some useful information, which is a whole lot of BIP which—just like Ks and HRs—your player could not possibly have made a play on (or did not need to).  Would that be an improvement?

I’m not sure it’s an improvement.  A BIP is what it is.

But, a subset of BIP now leads to bias.  If I’m looking at Terry Pendleton, and remove all the outs that Ozzie Smith got, and I’m looking at Mike Schmidt and removing the much fewer outs Larry Bowa got, then I’m reducing the denominator on Pendleton to a greater degree than I am for Schmidt.

At the same time, Pendleton’s numerator has already been pre-shrunk, because outs he might have gotten to were recorded by Ozzie.

So, it’s a tough call to make here.

Yes, I am using STATS for 00-01 and BIS for 02-09.  Let me check Pierre.  His arm is indeed the worst in baseball I think, -4 or -6 per 150, which should negate most of his fielding runs.  Maybe I screwed up again with my tallies. I’ll check again.

I’m using WOWY held for park and bathand. For an outfielder, how many hits and total bases on liner, flies and pops, and how many total bases on ground ball hits. Gameday is the data source. Soon I will have the hitchart linked in so I can use vectors for a more accurate ball placement. Arm ratings on basrunners also coming soon, but not yet.

My ratings for Pierre

   |    2006 |    2007 |    2008 |    2009
   | Opp RAA | Opp RAA | Opp RAA | Opp RAA
LF |         |         | 277  +2 | 230  +4
CF | 676 +14 | 736  +2 |  55  +1 |  75  -4

Very limited opps in CF last two seasons, but overall +5 for the last 3 seasons after a good 2006.

"MGL must have a mistake in one of the methodologies that is causing these differences.”

I took quite a bit of heat over that suggestion earlier, and was pointed in the direction of your study on accuracy of the data, which I then pointed out had a different aspect.  Be careful what you wish for.

MGL’s lists look right to me, witht he exception of Andruw.  Andruw’s ball-hogging is not in the taking away of FBs from LF and RF (mainly, there is some of that), but mostly in the taking popups (shallow FBs) that SS and 2B take on other teams.  One of the reasons for a lower STATS mark and MGL’s stuff is that MGL doesn’t remove those, but STATS does (because they are defined as popups, they don’t count for anyone).

Tango,
I would assume you are already controlling for other fielders, as I assume you do for pitchers’ rates of allowing BIPs (K-rates).

MGL could also overlay the “stated” ZR zones and provide a percentage of “In zone/Out of zone” plays on average for a position.  That is, in a CF’s 500 ZR chances, 450 are in zone, and 50 are out of zone (or 10% are OOZ plays).  Then looking at which CFs are benefitting from OOZ % would be telling.

This may also be the case with WHATEVER is happening in RF in Safeco.

Yup, forgot to add in the arm ratings when I fixed the first set of bad lists.  I updated the lists again.  Please note.  Sorry.  That’s what I get when I do these things in the middle of the night.

MGL - Left field is still wrong. See my list in post #14.

Peter, I don’t think most of those players have at least 500 defensive games in the decade, which is my min requirement.

Plus, in my files (they may be different from Fangraphs), I have Soriano in left field as:

06 +3
07 +4
08 -2
09 -8

OK, for some reason, I did not include Soriano’s arm rating, which is high over the years.

I guess I didn’t include arms for any OF who didn’t have at least 500 arm defensive games either and that is why Soriano’s arm lwts were not included.

MGL - The way you wrote the original post above made it seem that the 500 defensive game requirement was for the UZR/150 rankings and NOT for the total runs saved rankings

Sure, no problem.  I probably should have had no min games requirement for the total runs saved for the decade, but I only ended up looking at players with > 499 games.  I figured that no one with less than 500 games would be a leader in total runs saved, but that is obviously not the case in left field, where the spread of defensive talent is apparently not great and players don’t play as many games there as a general rule.

Are you using actual games, or your “Defensive games” for that cutoff, because one is real, and one is a construct of chances.

Defensive games.

Here’s a google doc’d version of the data.

MGL or Peter:  Any thoughts on the two questions I posed in #7 above?  (ballhogs, and independence of coding from player positioning and play outcome).

Guy - When I first started working on a fielding metric I was concerned about the ball hog effect for outfielders, but when I analyzed the data any effect ended up being very minor.  The reason for this was that there was a countereffect where balls hit between the fielders that were uncatchable would also be fielded more often by the centerfielder because he was usually the faster of the two fielders and therefore the centerfielder was also credited with more chances that ended up being hits.  This is a particular artifact of the Gameday system of recording the position of hits where it is picked up and not where it lands and therefore probably does not apply to UZR. 

Which leads to your other question about independence of coding.  There is much speculation and I think some evidence that the recording of ball position is affected by whether it is fielded for an out or is a hit, both for infielders and outfielders.  The areas between 3d and SS, LF and CF, 2nd and 1st, CF and RF are the farthest away from any visual cues that help locate the ball accurately within the field of play, so I think the natural tendency is to assume that balls that a fielder gets to are nearer to where the scorer thinks he usually plays.  We won’t know for sure if this is happening until we have instruments tracking the ball through its entire flight. 

Which brings me to another problem that I discussed in my presentation at this years Sportvision Summit (and is available at their web site) concerning the curvature of balls that go through the infield for hits.  Gameday gives the location where the outfielder picks up these balls and I believe BIS and STATS do to, but I am not certain of that.  Hit f/x gives the horizontal angle as the ball comes off the bat.  But because the ball will be usually be curving in its flight until it hits the ground and then will also bounce in the direction of its spin, neither the initial horizontal angle nor the angle where it is picked up will necessarily be an accurate measure of the angle at which it the ball passes the nearest infielder.  Again, Field f/x or the Trackman system, which both track the hit ball throughout its flight, are the only ultimate solution to these problems.  Having said this, I don’t either ballhogging or coding error are very large large problems in term of runs over the course of an entire season.  At least not large enough to be causing the differences between sUZR and bUZR.

Guy, the UZR methodology minimizes the problem of ball hogging.  For one thing, if a ball is caught by anyone, no one else gets docked any runs.  For another, ball hogging occurs on easy plays.  If a bucket has a high out percentage, whoever catches it gets little credit.

Concerning your second question, I’m sure there are all kinds of biases associated with the scorers and things happening on and off the field.  Not that that is a good answer....

If a bucket has a high out percentage, whoever catches it gets little credit.

Can you expound on this specifically as it applies to a CF taking popups behind second base?

How little credit, and how many do you have to take to add up to, say, 10 extra runs?

I think we’ve talked about this in the past.  If say on popups behind 2B are caught for an out 96% of the time, with the normal breakdown as:
.45 SS
.45 2B
.06 CF

Then, whoever catches the ball gets +.04 outs (rather than say the CF getting +.94 and the SS and 2B each getting -.45).

But, if the ball falls in, then it’s -.45 for SS, -.45 for 2B and -.06 for the CF.

If, on the extreme, you have 50 such popups and 48 are caught by the CF and 2 drop in for a hit, it’ll come out like this:
CF: +.04 * 48 - .06 * 2 = +1.8
2B: -.45 * 2 = -0.9
SS: -.45 * 2 = -0.9

Not necessarily the “fairest” thing, since the CF gets all the credit for catching stuff, or almost no debit for not catching stuff, but we’re talking about 1 or 2 runs in the most extreme of all possible cases.

I think this is the way MGL does it.  I pretty sure this was the consensus we reached on this blog a few years ago.

Tango,
that seems like it would be really hard to generate any runs saved.  If FBs turn into outs about 80% of the time, a fielder tends to get only +0.2 when he catches one (on average).  That seems odd.

I haven’t done it that way, so maybe that works, but I am skeptical.

Chris - I looked at Andruw Jones for 2005, one of his best defensive years of the decade by UZR.  Gameday hit locations have him catching 13 FBs at a distance of less than 225 feet which is about 75 feet into the outfield grass.  The average major league team CF caught 10 that year.  Jones didn’t play all his teams innings (1366 out of 1462) so give him credit for about 4 more than the average CF.  That’s about 3.2 extra runs.

But was he stealing from the shortstop or was he catching more balls because of his talent for coming in on short balls.  To answer that question I looked at how many balls the shortstop was catching in that area.  Greater than 175 feet the average team shortstop caught about 15 balls that year, 13 between 175 and 225 feet and 1.5 beyond 225 feet.  Atlanta shortstops caught 18, 16 between 175 and 225 and 2 beyond 225 feet.  So it doesn’t appear that Andruw was stealing many, if any of the balls that the shortstop should have fielded.  He was just a very good outfielder in his prime and he earned those extra 3.2 runs.

Thanks, Peter.  That’s very interesting.