FanGraphs Baseball


RSS feed for comments on this post.

  1. As soon as I saw the article title, I knew I wasn’t gonna get any work done for atleast 15mins.

    Comment by Chris — May 19, 2010 @ 6:15 pm

  2. Wow.

    Just a quick note – in the aging section there is a typo I believe. Players performance generally decreases in their early to mid 30’s. The article says 20’s. Either that or I am totally confused about when defensive regression hits.

    Comment by BJsWorld — May 19, 2010 @ 7:19 pm

  3. Very good. I’ve been wanting to see the complete details of UZR all in one place for a long time. People are using UZR all over the place now and many don’t really know what it is. This article will be linked frequently I’m sure.


    Comment by Lee Panas — May 19, 2010 @ 7:34 pm

  4. And one way to make it more accurate in a pure fielding sense is to average out the buckets with the batted ball percentages of the AL or NL, therefore creating a Pitching and Hitting Independent UZR, the subject of my community blog post currently under review.

    Comment by Nik Oza — May 19, 2010 @ 7:34 pm

  5. How is UZR/150 calculated? Ostensibly, the 150 refers to 150 games, but do you use actual games played, or do you use innings as a proxy? I’ve tried a couple of times for a couple of minutes to calculate UZR/150 from UZR and games or innings played at a position, and it never matches.

    Comment by misc — May 19, 2010 @ 7:36 pm

  6. It’s calculated using innings and averaged to 150 games. This makes it unreliable because not every fielder gets the same frequency of batted balls per inning, unless the inning batted balls are averaged to the percentages of the AL or NL.

    Comment by Nik Oza — May 19, 2010 @ 7:45 pm

  7. No it uses Defensive Games for both Range+Errors, ARM, and Double Plays. Innings are not used at all.

    Defensive Games – The number of outs made by an average fielder at his position given the exact distribution of balls in play for that player divided by the number of outs an average player at that position makes per game.

    Comment by David Appelman — May 19, 2010 @ 7:48 pm

  8. And that’s why my numbers never matched. But that makes sense. Thanks.

    Comment by misc — May 19, 2010 @ 8:04 pm

  9. I believe that is not a typo. Overall player performance doesn’t usually decline in the mid 20s, but that is because offensive performance is usually improving while defensive performance erodes. Note his following remarks:

    “his overall defensive value is likely already declining by the time he gets to the majors. If you find that hard to believe, look at the aging curve for a player’s triples rate. It too declines from an early age.”

    Comment by Newcomer — May 19, 2010 @ 8:05 pm

  10. Nik, not sure what you mean. You can either explain here or I’ll wait for your community blog post.

    BJ, no typo. As far as we know, fielding starts to decline in a player’s early to mid 20’s, depending a little on position. That is because, as I said, fielding depends mostly on speed, agility, and recklessness. Those skills peak early, as you can partially see from the aging curves for triples rates and stolen base attempts. Fielding probably doesn’t decline as much as the decline in those skills would suggest by itself, because there is probably some learning curve associated with fielding, but not enough to prevent an overall decline at an early age. Obviously your results for individual players may vary, and a good gauge for how much you think a player’s defense might be eroding as he ages, especially in the OF, is a player’s speed and overall physical shape. A player like Ichiro, who has appeared to retain his health, speed and agility (and his weight), is likely going to decline a lot less and later (or not at all) than a player like A. Jones (before he lost all the weight this year) or Griffey.

    Comment by MGL — May 19, 2010 @ 8:14 pm

  11. Yup, FG used to have defensive games listed with the UZR numbers. Maybe David will put them back. Basically, if you see that a player played in 50 defensive games, that means that an average player (for that position and that year and that league) would have had to play exactly 50 games in order to get the number of fielding chances that this field got. So, this fielder may have only played in 40 games, but got lots of chances for whatever reason, or he may have played in 60 games and got fewer than average number of chances. Again, we do that because we always want to try and reflect a fielder’s performance and/or talent in a completely context neutral environment.

    Comment by MGL — May 19, 2010 @ 8:18 pm

  12. Question:

    If UZR attempts to compare players to league average, isn’t it possible that data could be skewed by the quality of other players in the league that season? And could this not lead to some questionable comps?

    For example, can you really compare players UZR from different time periods?

    If player X had a +10 UZR from 2004-2006 and player y had a +11 UZR from 2007-2009, couldn’t that just as easily be a product of the play of OTHER players around the league rather than a statement of player Y’s superiority?

    Comment by GiantPain — May 19, 2010 @ 10:46 pm

  13. GiantPain. Yes! I explained exactly that in the Primer. Did you read it? I could have “baselined” everyone to the entire 6 years of data I use for each year of UZR, but I didn’t. Most metrics are scaled to a particular year and often to a particular year and league. Like UZR, they don’t have to be. If you truly want to compare players across years and leagues, you have to do some other things than just compare their UZR’s. Anyway, using multi-years as the basline creates other problems, so it is 6 of one and half a dozen of another. I probably would have preferred to base line everyone to the entire 6 years of base line data (both leagues combined), but people really do not like metrics that do not sum to zero for a particular year, and I can’t really blame them. If I saw that the entire AL, for example, had an average UZR per player per 150 games of +3, I would be skeptical that defense was that good in that year and league. The data itself could cause something like that rather than the entire league being above average in defense for that year and league. There is no good reason for us to assume that the data we use for UZR is extremely consistent from year to year. In fact, we know it is not. So I am very wary of not zeroing out UZR for that year. Of course I could have zeroed it out for the AL and NL combined and not for each league individually. The way I do it now, as you correctly point out, we don’t know if one league is better than the other in defense (just like we don’t know which league is better in pitching or offense unless we do some more research in IL games and players who switch leagues), and we don’t know if overall defense happens to be better in one year compared to another. It is a dangerous world!

    Comment by MGL — May 19, 2010 @ 11:49 pm

  14. Now I feel like a dummy. Thanks for the response though. I will admit to having skimmed certain sections, heh. I guess UZR isn’t great, but it’s the best we’ve got.

    Comment by GiantPain — May 20, 2010 @ 12:25 am

  15. So you’re saying it’s possible that Andruw Jones completely skewed the metric for centerfielders playing during the 1990s? Is that possible?

    Comment by Bronnt — May 20, 2010 @ 2:25 am

  16. “So, even after regression, there is no guarantee that our UZR number reflects what the player actually did or his true defensive talent over that time period. But, it is the best we can do (not knowing anything else about that player)!”

    I defintely agree about the true talent part. But your metric does reflect what a player does over a short period, since it is measuring actual events, albeit imperfectly and with uncertainty.

    Now say you observe Player A over a 30 game period. Adrian Beltre for example. If you read SOSH, you may have seen this discussion (if you can call it a discussion). UZR had him at 0. Doesn’t say anything really without observational data. If it was +10, it might suggest he was excellent if you did not watch the games. But I did, and he was not great, and I would throw that number out and not use it. If it was -10, it might suggest he was terrible, and if your eyes agree, you would be on pretty good ground to say that he was terrible. But it wasn’t -10.

    Based on my observations, I though UZR had Beltre pretty much right, since among the errors and some mental mistakes, he made some nice plays. For Beltre and per expectations, he played terrible in these 30 games, yet for a league average 3B’man, not terrible.

    Now if I look at the 2010 UZR leaders and trailers 5+ or -5-, I see names that I would expect to be there based on what is known about their true ability, and few that do not. So there is correlation even in a SSS. Of course, if you want to look at a players +2 and another player at -2, and make any comparisons without having seen any of the games, thats stupid.

    “Now, to some extent we are measuring something which is categorized, even though I just said that we aren’t. It is just that it is not particularly evident”.

    I agree with that, it is categorized, and it’s not evident because unlike offensive stats, we can not see it in the box score or play by play. Thats fine though. And I agree with your latter point that offensive stats over short periods do not always reflect how well a player did, just as UZR does not always do so. However, sometimes they do, by random chance or whatever.

    An 800 OPS hitter over 3 years may have an OPS of 800 today after 150 PA. A fielder may have a UZR/150 which matches his true ability based on a 3 year average. The difference between offensive stats and defensive stats is that offensive stats are based on counts, which have essentially no uncertainty (except umpires or official scorers decisions), while defensive metrics have some subjectivity, perhaps observational errors, and are based on probabilities and assigning runs for a player based on league averages (the latter point being the same for linear wts).

    Larger samples help to smooth things out, but there is still some uncertainty. I believe Tango estimated it at 5 runs over 2 years or something like that. Of course, any attempt to estimate uncertainty is uncertain.

    People do not like the word uncertainty. They prefer things to be black and white. They are not. So just because a number has uncertainty, does not mean it has no value. But using such numbers with certainty without taking into account the uncertainty, especially if you have not watched the player much, may lead to wrong conclusions.

    As for park factors, I saw Jason bay last year and knew he was not as bad as UZR had him to be. Maybe thats because he played at Fenway where being a great defensive LF’er is not as important, but Jason Bay did play in Fenway. In 2007 Bay was troubled with knee problems which may be why he had such a poor UZR. In 2008, he played 1/2 of his games in Fenway and he did not play it well since it takes a bit to know how to play there.

    Comment by pft — May 20, 2010 @ 2:29 am

  17. I should say, the earlier aughts. I wondered about that once, when noticing how he was in the 20s in UZR every year but there were very few CFs in the teens. I’m probably just imagining things, though.

    Comment by Bronnt — May 20, 2010 @ 2:32 am

  18. Double wow. Even as a software developer/DBA I find it hard to imagine the size and structure of the databases you have access to.

    Comment by John — May 20, 2010 @ 10:32 am

  19. Excellent, MGL.

    Personally, I am persuaded that Jeter’s improved UZR numbers in 2009 resulted from a focused effort to improve his mobility to the right during the off-season.

    Incidentally, one thing that could be published that might help make the link between subjective observations and statistical measures of defense would be simply the importance of each position of balls in each location. So for centerfielders, if you divided up the zones, you could get numbers which on average reflected the importance of balls in the alleys, balls in front of the centerfielder and balls over his head. Similarly, with shortstops, balls in the hole vs. balls up the middle would be helpful (I believe that it’s 60-65% up the middle).

    Comment by Mike Green — May 20, 2010 @ 10:46 am

  20. Great read! I was wondering if there is any consideration given to the effect, if any, a great fielder has on other player’s UZR. For example, a great center fielder who can cover acres of extra ground could impact the data of a below average left fielder by taking away out chances and limiting the area the LF has to cover.

    Comment by neuter_your_dogma — May 20, 2010 @ 11:48 am

  21. This is a very interesting read, though whether or not you would call the variation in year to year data “reliable” or not, given that you don’t rely on it to display a player’s true fielding ability, what can you rely on it to tell you? How do you feel comfortable using this data as generated?

    If a player with yearly ratings of -11, 14, and -3 should be considered the same as a player with yearly ratings of 1, 0, and -1, what happens if you measure the three most recent years of the first player one year later and he measures 14, -3, and 0? Then you’d think that he was rated 3.75 instead of zero.

    The data may not be so unreliable as to be useless, but I sure as heck can’t figure out what I’d want to use it for if I were worried about any sort of accuracy.

    Comment by Nutlaw — May 20, 2010 @ 2:20 pm

  22. I hugely miss the DG being listed. It’s like not having BFP for pitchers or PA for hitters; it absolutely needs to be squeezed back in.

    I mean, the current FanGraphs design doesn’t fill my laptop browser with the bookmark pane open. Wider columns, more data!

    Comment by Eric M. Van — May 20, 2010 @ 2:46 pm

  23. so Appleman, does UZR 150 use an averaged batted ball chart and filter it through the fielder’s performance or use the fielder’s performance and filter it through the batted ball chart? I believe the first one would be more accurate for judging a fielder because it is taking the league average ratio of batted balls instead of the actual ratio that the fielder fielded. Or am I just not getting the Defensive Games concept?

    Comment by Nik — May 21, 2010 @ 5:07 pm

  24. Nik, we want to know how a fielder would have done if we received the league average distribution of balls in 150 league average games, so that is the way the defensive games are calculated. Here is an example which should make it clear:

    A fielder makes 300 outs. It does not matter what the distribution of balls looks like that he fields or does not field, but he makes 300 outs, period. An average fielder with those exact same balls would have made 310 outs. Obviously we have a below average fielder to the tune of 10 outs or around 8 runs (-8 UZR).

    Now, here is how we assign defensive games to this fielder: The average fielder in his league and at his position gets 3 outs per game. 310/3 is 103.3. That is his number of defensive games. Literally had he played for a random team in that same year and league and gotten the same number of chances to field a ball and make an out, he would have played 103.3 games.

    Comment by MGL — May 21, 2010 @ 5:28 pm

  25. Ok, thanks, I get the defensive games concept now.

    The aspect I think would be better (or at least supplementary) for UZR is as well as applying to the fielder the average number of outs for that position, apply the ratio of distribution of batted balls. Also, instead of taking the fielder’s effect from the batted balls he fielded and multiply it by the outs (to get the defensive games) multiply the fielder’s effect to the average ratio of batted ball distribution.

    I hope the blog post gets approved. There’s much more detail in there.

    Comment by Nik — May 21, 2010 @ 6:49 pm

  26. Great work. Is this proprietary? Why not open it up completely so that others can replicate it.

    Also, can anyone quantify how much change there has been in UZR over the years? E.g., how much difference is there from, say, what it indicated in 2003 about that year’s performance, compared to what the model today indicates about 2003 performance. Because, as a decision-maker, I would want to know how stable the product’s indications have been in the past.


    Comment by Acton — May 23, 2010 @ 10:30 am

  27. I have a couple questions about UZR that I’ve been wondering about. Hopefully these weren’t already answered and I just missed it. Sorry if I did.

    1. Do you get credited for getting to a ball even if the batter isn’t out? Take for instance your a SS and you show a bunch of range to even get to the ball but you can’t throw the runner out because he is too fast. Does that hurt you since you didn’t get the runner out?

    2. For outfielders does it take into account you getting to a ball that isn’t in the air? Like say there is a line drive that looks like its heading to the wall for extra bases but you show enough range to cut it off and hold them to just a single. Do you get any credit for that, or is it just ignored because you didn’t make an out?

    Comment by Dwight S. — September 17, 2010 @ 4:40 pm

  28. Does UZR take into consideration double plays turned, or only double plays started?

    Comment by Mark — November 17, 2010 @ 2:44 pm

  29. consideration double plays turned, or only double plays

    Comment by sdfadfasd — December 3, 2010 @ 10:42 pm

  30. still get confused by this. also, can’t it be a bit misleading?

    Comment by will — January 4, 2011 @ 9:34 pm

  31. How are there UZR stats for players that played before this was tracked? Doesn’t it involve watching each game?

    Comment by Bonzi — January 5, 2011 @ 3:15 pm

  32. Is there a website that lists shows uzr runs awarded and taken away for individual pa and games?

    Comment by Jack — June 15, 2011 @ 7:27 pm

  33. I want to show some appreciation to this writer for bailing me out of such a trouble. As a result of researching through the the net and coming across opinions which are not powerful, I was thinking my life was done. Living without the presence of answers to the issues you’ve fixed by means of your good posting is a critical case, and those that might have adversely affected my entire career if I had not discovered your web blog. Your talents and kindness in controlling all the details was vital. I don’t know what I would have done if I hadn’t come across such a subject like this. I can also at this moment relish my future. Thank you very much for your reliable and results-oriented help. I will not hesitate to propose the website to any person who will need guide on this issue.

    Comment by kominki — August 18, 2011 @ 1:45 pm

  34. obviously like your web-site however you have to test the spelling on several of your posts. Several of them are rife with spelling issues and I to find it very troublesome to inform the reality then again I’ll definitely come again again.

    Comment by gilmorova devcata online — November 12, 2011 @ 10:03 am

  35. I do accept as true with all of the ideas you’ve introduced in your post. They are really convincing and can definitely work. Nonetheless, the posts are too quick for beginners. May just you please extend them a little from next time? Thanks for the post.

    Comment by Latina Natalia Spice — November 13, 2011 @ 11:02 am

  36. This is a great article and it really helped me understand the metric a lot better. It really goes in depth in analyzing the game beyond what we see on TV. But I just have one question. When you talk about regressing towards the mean, you mentioned that you like to “cut the number in half.” Let’s just say that a player has a UZR/150 of +10.0 through 2 months. Would you assume that his true talent UZR would be +5.0, because 10/2=5? It’s probably a stupid question, I got confused in that part, but nonetheless great article.

    Comment by Robert — November 7, 2012 @ 6:38 pm

  37. i really need help trying to figure out this problem… its really hard … the problem is 2+2

    Comment by Payton — April 5, 2013 @ 1:35 pm

  38. lata blablabla said this I do accept as true with all of the ideas you’ve introduced in your post. They are really convincing and can definitely work. Nonetheless, the posts are too quick for beginners. May just you please extend them a little from next time? Thanks for the post

    Comment by Payton — April 5, 2013 @ 1:36 pm

Leave a comment

Line and paragraph breaks automatic, e-mail address never displayed, HTML allowed: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>

Close this window.

0.264 Powered by WordPress