FanGraphs Baseball


RSS feed for comments on this post.

  1. Another awesome article. Craig Kimbrel is beyond good

    Comment by TimBrownU — February 26, 2013 @ 9:33 am

  2. Really dig this work. How hard would it be to add a column to player pages, whereby wOBA / FIP / whatever gets a 20-80 grade? Now that the legwork (and presumably math behind it) is complete, this could be another tool in the arsenal for discussions / arguments, as well as “who’s the most avg at …” articles. Make it happen!

    Comment by Request Box — February 26, 2013 @ 10:36 am

  3. I know Kimbrel is good when I go on a baseball game, max out all his stats and STILL can’t do what he does in real life.

    Comment by Wil — February 26, 2013 @ 10:39 am

  4. Could you publish your standard deviations and means?

    Comment by philosofool — February 26, 2013 @ 10:44 am

  5. Are you including pitchers in your hitting samples at all? If so, they’ll skew the numbers down a bit, won’t they?

    Comment by Jason461 — February 26, 2013 @ 10:51 am

  6. Regardless, he is only including players with 200 PA so the results are going to be biased anyway. Typically players need to be well established or do well in a small sample to get 200+ PAs, which will skew everything upwards.

    Comment by guesswork — February 26, 2013 @ 12:07 pm

  7. Again, I am going to recommend you redo this analysis using sample quantiles instead of a normal approximation. The normal approximation is nice, but may give unrealistic results for the players at the extremes. I think it would be interesting to see how the results differ as well. Chances are they will be very similar except at the tails.

    Of course, there are issues, namely a larger sample size would be needed (2008-2012 may suffice) and it is restrained to the sample. Thus it assumed that both a 20 and 80 player exist, but at least it makes no distributional assumptions.

    An even better approach? A nonparametric Bayesian approach that uses scouts’ actual grades as priors. No sample size issues or distributional assumptions! Of course, acquiring scouts’ grades is not easy as far as I know.

    Comment by guesswork — February 26, 2013 @ 12:16 pm

  8. Your math is wrong on the WAR/600.

    Mike Trout has had 774 career plate appearances for a total of 10.8 WAR. The proper way to calculate this would be to divide the 10.8 WAR by 774 PA = .013954 WAR per PA. Now multiply that by 600 = 8.37 WAR per 600 PA.

    I did not check the other players but the math may be wrong on those as well.

    Comment by Brandon — February 26, 2013 @ 12:50 pm

  9. My calculation had the same amount. I was just saying that he fit into that category, but I probably should have put his actual WAR/600.

    Comment by Mark Smith — February 26, 2013 @ 1:11 pm

  10. I thought 0.25%, rather than 0.1%, were above/below three standard deviations from the mean. (Or, stated another way, I thought 99.5% was within three stadard deviations of the mean.) But I may be mis-remembering basis statistics. At any rate, it doesn’t change your basic premise about the uniqueness or rarity of these types of players (and ‘two-and-a-half-in-a-thousand’ doesn’t have the same ring as ‘one-in-a-thousand’ does anyway).

    Comment by Jason B — February 26, 2013 @ 1:49 pm

  11. I think it’s 0.25% total, so that’d be .125%.

    Comment by Nick C — February 26, 2013 @ 2:15 pm

  12. This is Great!! Thanks for your work!!

    Comment by SC — February 26, 2013 @ 2:50 pm

  13. I’m surprised that Burriss ranked that high.

    Comment by Baltar — February 26, 2013 @ 4:01 pm

  14. Okay, thank you for clarifying. Great article by the way!

    Comment by Brandon — February 26, 2013 @ 5:05 pm

  15. Just prospects, but has some.

    Comment by Dave — February 26, 2013 @ 10:23 pm

  16. He did essentially. Those can be inferred from each table with reasonable accuracy. Just look at what the stat value associated with the 50 score for each league is and that’s basically the league mean. The difference between the 60 score value for the league and the 50 score for the league is basically the standard deviation (notice that every 10 point change amounts to the essentially the same change in the stat, which is nearly the standard deviation).

    Comment by reillocity — February 27, 2013 @ 12:16 am

  17. Yasmani Grandal, of course! Had him on the tip of the tongue.

    Comment by Jon L. — February 27, 2013 @ 2:33 am

  18. I’m not a fan of the WAR/600 results. I think it needs to have a minimum PA cap for position players; maybe 1500? The first two players on that list: Trout and Ellsbury stand out as having 1 great season only. Trout may never repeat it and Ellsbury probably won’t repeat his 2011.

    Comment by Jason — February 27, 2013 @ 11:04 am

  19. There are explicit tests for whether a given distribution is appropriate for data. For most of this, given the sample sizes, assuming Gaussian is probably sufficient, and I’d imagine that the data would hold up. The one case where it may not is perhaps something like fastball velocity in the last article, where there would be expected to be a skewed distribution since pitchers can always throw slower whereas biophysics constrains upper velocity.

    That said, I think the real contribution here is the use of a distribution that makes “great” a function of average and variance. Nothing says that the best and/or worse of a population is something we should be in awe of, and all too often people try to force players into what should be rarified territory

    Comment by JBImaknee — February 28, 2013 @ 9:22 am

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.

0.087 Powered by WordPress