THE BOOK--Playing The Percentages In Baseball

For MGL:

In total that equals -21 plays to average 1B. At 0.8 runs a play thats -16.8. However, his range UZR is -7.3. Why such a large difference?

If those balls he is not getting to in R,S,T & U are successfully made by the 2Bman, do they count as missed plays for Hosmer.

The data presented at that blog is based ONLY on one parameter: recorded batted ball location.

UZR has many more parameters (park, base-out, GB/FB tendency of pitcher, recorded speed, among others).

The blogger basically presented “adjusted zone rating”, not UZR.

But, given that the location is the #1 determinant in UZR, then trying to explain Hossmer’s low UZR would in part be based on looking at his adjusted zone rating.

#1 only GB where looked at, not LD or FB. He also regresses the data bit.

The question on Hosmer’s has been discussed among us 5 or 6 Royals fans for a while.  Seeing Ryan Howard and Prince Fielder as better 1B was not adding up.

Under UZR, wouldnt a 1Bman who hung out in Zone T inflate his overall UZR, given the average outrate in that area is so low on average. I understand this is the need for knowing positioning first, but I seem to remember this being an issue with Utley’s rating. He would shade Howard so much that he would miss the easier middle zone balls but rack up a high out rate on the far right side of his zone. Perhaps a ‘one zone’ rating has some value (ie: Adj Zone Rating).

Is there a simple explanation for why this adjusted zone rating has Hosmer above average in zones V-Y but the Fielding Bible has him well below average to his left:
“Hosmer, however, is bad overall. In 2011, he’s -6 on balls hit to his right, -6 on balls hit right to him, and -5 on balls hit to his left.”
I’ve read up on the differences between the systems, but I can’t find anything that would explain this.

Per #5, my guess is that the further UZR adjustments that Tango talks about (park, baseout etc), and perhaps a ball hogging adjustment on the 2Bman (if there is one), will lessen the negative impact on the R-U zones, and perhaps decrease the positive impact on zones closer to the bag.

"He would shade Howard so much that he would miss the easier middle zone balls but rack up a high out rate on the far right side of his zone.”

That strategy shouldn’t inflate UZR.  Yeah, you will rack up a lot of positive UZR by fielding balls that other 2B don’t get to.  But you’ll get a lot of negative UZR if you fail to get to plays that most 2B field easily.  If you position yourself in a way that doesn’t maximize the number of balls you get to, you are going to lower your UZR.

But is it possible he’s getting more balls that would be doubles down the line, while giving up balls that will either be singles OR be successfully fielded by the 2nd baseman?

As an opposite example, Pujols fields quite a few balls to his right that would be fielded by the 2nd baseman because he plays so far off the 1st base bag, and UZR seems to think highly of him.

He only saw 23 balls down the line vice about 60 for each of the other zones.

Kevin: sure, and that would explain partly why his runs allowed is not as bad as his hits allowed shows him to be.

Darren, no player gets “docked” any runs if another player fields a ball.

If another (adjacent) player “hogs” a lot of balls, it is not real clear to me off the top of my head whether that would hurt the other players (not fielding the ball). On one hand, it robs the other player of getting to field some balls that they might otherwise field (IOW, they gets zero debit, but they also gets no credit - had they fielded the ball, they would have gotten some credit), but on the other hand if that fielder is a very good fielder he helps adjacent fielders since when a ball that would have been a hit gets caught, those adjacent fielders are not getting docked any runs (which they would have been docked had the ball not been caught).  IOW, in UZR a good fielder helps his other (adjacent) fielders in UZR and a bad fielder hurts him, unless the good fielder is hogging a lot of balls that adjacent fielders might catch or bad fielders are allowing other fielders to catch balls they would not have caught.  So it is not real clear whether bad or good fielders help other fielders in UZR, and it might depend on how good or bad, where they are good or bad, and where they position themselves. 

Probably the biggest way a fielder can effect other (adjacent) fielders, whether they are good or bad, is to position themselves in an unorthodox manner.

Jeff, pop flies and LD are not tracked for 1B.

One would really have to know how often and when Hosmer is holding a runner on 1st base. That is the biggest determining factor for a first baseman’s simple ZR. Of course UZR adjusts for that, but just looking at the out rates in the various zones tells you little about the 1st baseman’s range unless you know whether he was holding a runner or not on each play.

Jeff, I don’t know if I gave you that data for each play, but if I didn’t, you really need to know that and include it in your chart.  For example for each zone, give us the out rate for all fielders when they are either holding a runner or not, given what Hosmer was doing on that play.

If Hosmer happened to be holding more runners than average on all of his plays, that would help to explain the data in Jeff’s chart, but it wouldn’t explain his poor UZR since UZR adjusts for positioning (based on the outs and runners).

I don’t think that plus/minus (Dewan/Fielding Bible) adjusts for whether a 1st baseman was holding a runner or not, when it talks about a player’s range to his right and left (and at him).  I think it assumes that all players play the league average position or positions. I could easily be wrong about that.

There is evidence (by me) that the only position in which players get better with age (experience most likely) is first base.  That could explain a little or a lot.  IOW, Hosmer may have the raw skills/tools to be a good first baseman but he doesn’t have enough experience at the MLB level.

Nice work by Jeff so far, and I can’t wait to read about the video review.  I hope it is thorough and extensive. I have always wanted to video review an entire season (or large chunk of a season) to see where UZR might be wrong on a small sample.  Obviously some of the zones will be wrong and some of the balls that are missed by Hosmer will be a lot tougher than that implied by the description and some of the balls caught will be a lot easier.

Warning: important concept - please commit to memory!

All you can hope for with any metric that relies on a limited (and not necessarily accurate and perhaps even biased) description of the data (batted ball and positioning of the fielders) is that everything evens out as quickly as possible.  Obviously, the larger the sample the more likely things will even out (assuming no major biases) and the larger the number of players looked at the greater the chance of finding a player or players for any given sample size for whom things don’t even out.

I am having trouble interpreting Jeff’s chart. [The chart uses STATS Inc. zones, and his article shows a STATS zone diagram, but the data used is from BIS.] There seem to be way too many expected outs for first basemen in the second base hole (zone U) and zones closer to second. And there are too few expected opportunities and outs close to first.

When I see the chart indicate 55 total ground balls in zone T, for example, with 12 expected plays for the average first baseman, that number of total ground balls is maybe a little lower than I would have thought, but it must be total ground balls in that zone with Hosmer on the field; it can’t be ground balls with successful plays made by the second baseman (or the pitcher) subtracted out. So taking that number as the total ground balls hit to zone T, including plays made by other fielders, 12 expected plays made by an average first baseman should leave in the vicinity of 35, maybe 40, plays made by the second baseman, with a few plays not made by anyone. That would be a ratio of 3 expected plays made by the second baseman to 1 by the first baseman. That’s wrong. That is way too low a ratio in a zone which is pretty close to the second baseman’s average starting point. By similar argument, that chart implies that first basemen should be expected to make twice as many plays than second basemen in zone U. That’s the most difficult zone on the right side, and I’d expect about 25-30 plays not made by anybody per 62, and if 26.5 were really the correct expected outs for the first baseman leaving an implication of no more than 10 expected outs for the 2nd baseman.

For anyone who is not comfortable with what these zones are supposed to represent in reality, in Jeff’s linked article there is a picture of a field diagram showing zones T and U and tiny dots for the nominal fielder positions.

I can think of four explanations. Either the raw BIS observations are very biased, spreading the location data on ground balls away from first and toward second, or something has gone wrong in translating the raw data from BIS vectors to STATS zones, or something went wrong in aggregating the numbers shown in the chart, or else something is wrong in the computation of expected outs by first basemen. But if zone is ignored, in total the expected outs for 1st basemen from UZR look plausible, so I incline toward the first or second possibility.

Here is my own table showing the distribution of plays made on ground balls by an average team, based on STATS-based data I collected and linked in 2006. What I assembled may certainly contain errors, but I think the results “match” what we see on the field a lot better about how often plays get made, and which players make them. Note that in this dataset, the ratio of plays made for 2nd basemen to first basemen is 25:1 in zone T and 1.6:1 in zone U.

ZN GB | pm1|pm2| pm3| pm4| pm5| pm6| noplay
A 5 | 0 | 0 | 0 | 0 | 0 | 0 | 5
B 10 | 0 | 0 | 0 | 0 | 0 | 0 | 10
C 89 | 0 | 1 | 0 | 0 | 57 | 0 | 31
D 116 | 0 | 1 | 0 | 0 | 98 | 0 | 17
E 106 | 1 | 0 | 0 | 0 | 87 | 0 | 19
F 89 | 1 | 0 | 0 | 0 | 47 | 2 | 39
G 86 | 2 | 0 | 0 | 0 | 16 | 20 | 49
H 81 | 2 | 0 | 0 | 0 | 6 | 49 | 24
I 93 | 2 | 0 | 0 | 0 | 0 | 79 | 12
J 113 | 2 | 0 | 0 | 0 | 0 | 101 | 10
K 95 | 3 | 0 | 0 | 0 | 0 | 76 | 15
L 100 | 8 | 1 | 0 | 0 | 0 | 60 | 32
M 124 | 38 | 1 | 0 | 2 | 0 | 27 | 55
N 102 | 44 | 1 | 0 | 13 | 0 | 4 | 39
O 73 | 11 | 1 | 0 | 35 | 0 | 1 | 25
P 76 | 6 | 0 | 0 | 55 | 0 | 1 | 14
Q 76 | 3 | 0 | 0 | 67 | 0 | 1 | 6
R 83 | 2 | 0 | 0 | 75 | 0 | 0 | 6
S 83 | 2 | 0 | 1 | 68 | 0 | 0 | 12
T 86 | 2 | 0 | 2 | 50 | 0 | 0 | 31
U 69 | 2 | 0 | 13 | 21 | 0 | 0 | 32
V 70 | 3 | 0 | 42 | 6 | 0 | 0 | 19
W 93 | 2 | 1 | 79 | 1 | 0 | 0 | 10
X 73 | 2 | 1 | 59 | 0 | 0 | 0 | 11
Y 7 | 0 | 0 | 1 | 0 | 0 | 0 | 6
Z 2 | 0 | 0 | 0 | 0 | 0 | 0 | 2

Joe, send me your email and I’ll send you the raw data I provided for Jeff.

I can think of four explanations. Either the raw BIS observations are very biased, spreading the location data on ground balls away from first and toward second

I can’t speak to any of your other possibilities, but HITf/x data has indeed shown us that the BIS data is skewed in the manner you mention.  STATS data didn’t seem to suffer from the same problem, at least not nearly to the same extent.

Here are some previous threads on the topic:
http://www.insidethebook.com/ee/index.php/site/comments/how_much_random_variation_in_fielding_on_bip/
http://www.insidethebook.com/ee/index.php/site/comments/does_one_year_uzr_tell_us_exactly_what_happened/
http://www.insidethebook.com/ee/index.php/site/comments/gameday_slice_bias/
http://www.insidethebook.com/ee/index.php/site/comments/our_lab_bias_in_batted_ball_data/

Some early observations, since MGL kindly shared his data yesterday:

I can’t judge translation of BIS locations to STATS locations, because they are already translated in MGL’s data.

The data is not limited to ground balls [361], but includes some “bunt grounders” [25] as well.

The expectancy values on bunts for hits,outs, errors & fielder’s choice do not make sense to me. The total expectation for each combination of parameters (batted ball speed, batter speed, batter handedness,zone, runner held) still adds to about 1.000, as they do for ground balls, but some of the out and error/FC expectations are less than .000, or greater than 1.000.

No park effects are visible in the expectancy rates for ground balls or bunts. For given parameter combinations, they are the same regardless of park.

The ground ball expectancy rates are generally plausible, but do have some small “blips”, especially for the less common parameter combinations which surely is due to sample sizes in the low hundreds (or less) for those combinations. (I don’t think this will amount to anything serious.)

I will assemble some video observations and report back on those ...

The first part of the video evidence:

http://www.royalsreview.com/2011/11/10/2538524/eric-hosmers-defense-part-2-positioning#storyjump

Great roll-up-sleeve work from everyone concerned.

That leaves what side of the plate the batter hit from. Of the 29 balls that went as hits against Hosmer that the majority of other first baseman record an out, 83% of those plays came against left-handed hitters (or switch hitters hitting from the left-side). For the 15 plays 50% of other first baseman field that the Royals second baseman recorded the out, 14 came against LHH. That means 86% of the balls that Hosmer did not get to that most first baseman reach came against LHH.
...
After watching the plays a hit was charged against Eric, I reached two conclusions and one hypothesis.

Hosmer plays too close to the line, especially against lefties.
Hosmer occasionally makes poor decisions in the field.
Hosmer may not see the ball off the bat well against lefties, or has less range than originally believed.

Just fantastic stuff, so thanks to Connor Moylan for the great work!  Sabermetrics needs more people like Connor.

***

By the way:

I prefer to show everyone the video, but that would break copyright laws and the MLB is strict on its video policies. If you have mlb.tv, I encourage you to look at the archives to see the plays yourself. For everyone else, the screen shots after the jump will hopefully suffice.

It would not break copyright laws.  There’s something called “fair use”.  And, if you use the video as part of your own opinion or commentary (and the video is relevant), then copyright won’t apply.  MLB simply scares people with their “no part can be rebroadcast without express permission”.

I’m not a lawyer, so don’t proceed based on my opinion.  Seek qualified legal counsel.  But at the very least, don’t write that you WOULD be breaking a law.  Just say “I don’t know”, or “I’m too scared to try”.

Yup, great work!  Definitely work that a team should be doing.  I doubt many teams do, though.

Someone asked about how UZR “screens” (ignores) shifts.  BIS makes that determination. I assume only somewhat extreme shifts are included, but I am not sure.

The thing about playing too close to the line, like in a no doubles defense, is that you don’t get much benefit since much of the area in which you are improving your range is in foul territory!

IOW, with a lefty at the plate, even though they will (obviously) pull ground balls (almost all players pull ground balls) to the right side and you want all your fielders to shift to the right, the one fielder who should shift the least is the first baseman because once you get a certain distance from the line, there is no reason to shift any more since you are going to field most balls hit down the line anyway.

There is one thing that makes video analysis of UZR data problematic.  Any ball that is not fielded, on the average, is going to be further away from a fielder than the recorded location suggests.

Let’s say that you have 100 balls recorded in slice “X” and 50 of them were caught and 50 of them were not.  The ones that were caught were, on average, more toward the fielder than the ones that were not.  In fact, even if we use the exact (according to the stringer’s estimate of course) location (rather than the “slice"), we would have the same phenomenon.  If those 100 balls were recorded at exactly 20 feet from the first base line, the ones that were caught would average 19 feet and the ones that would not would average 21 (numbers made up).

Same with speed.  Even if all 100 of those were classified as medium speed, the ones that were caught would be softer hit than the ones that were not.  And the ones that were not caught would have had more tricky bounces and spin, etc.

This is important, and why I always say that UZR does NOT usually represent what actually happened on the field, especially for extreme UZR numbers: If, according to the BIS (or other company) data a particular ball was fielded on average by a MLB SS 50% of the time (over a large sample), and that same exact ball (and same handed batter, and all the other bucket criteria that UZR uses), at least according to the stringers, was missed by fielder A, if we truly knew the parameters of the ball (and other factors related to its difficulty), we would find that major league SS only field that ball 25% of the time (again, numbers made up).  Again, maybe UZR should do the “regression” internally (IOW, dock the fielder .25 of a play and not .5), but you can imagine why I chose not to do it that way.

Thank you Tango and MGL, the compliments really mean a lot. And thank you for the advice, I think I fell under the “too scared to try” category.