Fielding Update: Arms and Double Plays

All the UZR stats on the site have been updated thanks to Mitchel Lichtman’s outfield arm and double play ratings!

– UZR now inlcudes outfield ARM runs and Double Play runs.
– ARM and DPR (Double Play runs) are broken out separately in the fielding sections.
– All player Win Values have been updated to include these additions.
– There have also been some slight changes to UZR that adjust for how fielder’s choice plays were being calculated.

The changes are available in all sections of the site including the player pages, leaderboards, team pages and my team pages.




Print This Post



David Appelman is the creator of FanGraphs.


19 Responses to “Fielding Update: Arms and Double Plays”

You can follow any responses to this entry through the RSS 2.0 feed.
  1. J.J. Hardy says:

    Why did I just lose two million in value? Is it because Rickie Weeks turns double plays at a fourth-grade level?

    Vote -1 Vote +1

  2. Alex Rios says:

    I’m pretty awesome.

    Thanks again Fangraphs, you continue to raise the bar.

    Vote -1 Vote +1

  3. Xeifrank says:

    Is it possible to project a 2009 UZR? Is it even possible? If so, how would one go about doing it given the previous year(s) UZR? Thanks for this added information.
    vr, Xei

    Vote -1 Vote +1

    • You could weight the past 3 years worth of data to get some sort of decent projection. Basically do Marcel for UZR and it should be a fairly good projection.

      Vote -1 Vote +1

      • Xeifrank says:

        That’s what I was thinking but figured you would need to do some sort of regression based on total chances or something along those lines. Perhaps the author of this metric could offer an opinion? Thanks!
        vr, Xei

        Vote -1 Vote +1

  4. Xeifrank:

    Why not just use UZR/150 to equalize for playing time?

    Fangraphs continues to rule. Not sure how you guys manage to do this for free and without ads, but maybe I shouldn’t mention that the goose’s eggs are golden…

    Vote -1 Vote +1

    • Xeifrank says:

      What would a -6 runs in 25 games equalize to (UZR/150)? If the answer is -30 runs then I am not crazy about that method. I would think this metric would need some regression, more for smaller sample sizes. Extrapolating out to 150 games keeps the same runs/games ratio.
      vr, Xei

      Vote -1 Vote +1

      • Xeifrank says:

        Duh, should read -36 not -30.
        vr, Xei

        Vote -1 Vote +1

      • Terminator X says:

        You’re absolutely correct, simply extrapolating UZR to UZR/150 does nothing to cope with small sample size. . While I’ve never done any hard mathematical projections of UZR, I usually look at the most recent ~3000 innings with an informal weight to the more recent data, and apply modifications/regression for any outside circumstances (hamstring injury, hasn’t played the position for a couple years, etc).

        Vote -1 Vote +1

  5. Aaron B. says:

    Thank a bunch David, MGL, and everyone else who contributes to this site.

    Vote -1 Vote +1

  6. Freen says:

    And it somehow gets even worse for Brad Hawpe. He can’t really be that bad, can he?

    Vote -1 Vote +1

  7. Bill says:

    Wow – Hawpe makes Manny look like Rios. It can’t be some crazy park effect, either – Hawpe looked just fine in the same right field 2 years ago.

    Vote -1 Vote +1

  8. TangoTiger says:

    If you want to regress VERY QUICKLY, add in a league average (0 runs, natch) of 100 games.

    So, if you have someone as +40 runs in 80 games, you count him as +40 in 180, which is +33 in 150.

    Do that, until someone comes along and tells you to do it the right way.

    Vote -1 Vote +1

  9. MGL says:

    There is enough data on the site for someone to run year-to-year (or whatever time period you want – it does not matter) regressions to get a resultant “r” (correlation coefficient) which tells you how much to regress a certain sample. As Tango indicates, a good rule of thumb is to regress 50% for 100 games. That means a formula of around:

    regression = 100 divided by (number of games plays + 100).

    Of course, you really want to use “chances” to do the regression since that represents a fielder’s sample size. As a rule of thumb, a first and third baseman get around 1.5 chances per game, CF, 2B and SS, 2.5, and RF and LF, 2.0. Remember that “chances” in UZR are the number of outs made by an average fielder at that position given the same distribution of balls in play (the reason is that the way UZR is calculated, almost every ball in the vicinity of a fielder is a potential “chance” for that fielder, as long as ANY fielder at that position fielded the same type of ball at least once during all the years that UZR uses as its “baseline”).

    Vote -1 Vote +1

  10. MGL says:

    I wrote this in the comments section of David Cameron’s post about the arm ratings:

    Didn’t see this thread until today. I pretty much do the same thing as THT. Outfielder’s get credit (plus or minus) depending on what the runners do on a hit or a fly ball out. A runner can stay put, advance, or get thrown out. So a fielder will get credit not only if he throws out more than his share of runners, but also if he keeps more than his share of runners from advancing extra bases.

    I adjust for park effects, like LF at Fenway (where the assumption is that the LF’er plays close) or all fields in Coors (where the OF’ers play deep).

    I also account for the type of hit (line drive, fly ball, etc.), the speed, and the location of the ball (where it is caught or lands), as well as the number of outs. And I put different base runner configurations in different buckets – e.g. runners on 1 and 2 are treated differently than a runner on 2 only. Obviously these are important variables.

    I don’t do anything for a fielder who may prevent a player from stretching a single into a double, or double into a triple, because there is really no good way of knowing this from the data with any degree of reliability. (I suppose I could look at how many singles, doubles, and triples an outfielder allows for each location, speed, and type of batted ball. In fact, I think I looked at that a while ago and found that indeed the players with best arms also allowed fewer doubles as compared to singles and triples as compared to doubles, to the tune of a couple more runs a year.)

    I do, of course, include in the “credit” column when a fielder throws a runner out attempting to stretch a hit. And again, all of the credits and demerits are based on the the league-average rate. For example, if there were 100 singles against an OF’er (on a certain batted ball type in a certain location with a certain number of outs) and he threw out 3 (3%) attempting to stretch, and the average OF’er threw out 2% given the same batted ball characteristics and outs, then the OF’er would get some credit equal to the value of erasing a runner on first (like a CS) with that many outs.

    One more thing. When you look at even several years of combined arm data, and you see a spread of 45 runs, like David shows above, the actual spread in true talent is less than the spread of what you see. Obviously the more the number of years, the closer the observed spread is to the actual spread, but the former is always larger than the latter. In this case, a spread of 45 runs in 3 years is probably equivalent to a “true” spread of maybe 35 runs or so, or 12 runs a year (+6 to minus 6). If we assume that the largest spread we see is equivalent to around 3 SD, then I think it is safe to assume that the SD of arm talent is around 2 runs per season. Compared to around 5-7 runs per season in UZR range and errors for OF (5 at the corners and 7 in CF). (These are educated guesses). Which makes arm maybe 25% of total defense in the OF.

    Vote -1 Vote +1

  11. Samg says:

    Does the fielder receive credit is less players run on him?

    Vote -1 Vote +1

  12. MGL says:

    Sam, yes, of course. Here is an example of what the “arm engine” does:

    Let’s say that on a certain type of batted ball single, say, a medium speed line drive to a certain part of LF, with one out, on the average, when there is a runner on first, the runner advances to third 50% of the time, stays at second 45% of the time, and gets thrown out 5% of the time. Let’s say overall, the average run value of those configurations (1st and 3rd one out, 1st and 2nd one out, and 1st 2 outs) is .8 runs (using a “run expectancy by bases/outs state” chart).

    Now, let’s say that on the same ball, a certain outfielder fields it and the runner only goes to second. So the bases/outs state is runners on 1st and 2nd 1 out. Let’s say that the run expectancy of that configuration (using the same chart) is .7 runs. So, this OF’er has “saved” his team .1 runs (.8-.7) on that play.

    The “engine” does that for all the plays and adds them all up. Voila, you have arm linear weights (with some park and other adjustments, like handedness of the batter, thrown in)!

    Vote -1 Vote +1

  13. mike pelfrey says:

    Any chance your team fielding stats can be offered in smaller(and more recent) time frames like your hitting and pitching ones?

    Vote -1 Vote +1

Leave a Reply

Your email address will not be published. Required fields are marked *

You may use these HTML tags and attributes: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <strike> <strong>