• @s: Is Positive Corsi = more wins 100% accurate?

    by  • December 4, 2013 • Hockey • 6 Comments

    I was asked this question in the context of talking about defencemen. I’m probably going to get more of this when a piece that I wrote about Brian Campbell and Tom Gilbert goes live so this is a sort of pre-emptive shot.

    The short answer to this question is basically yes. As a defenceman piles up minutes, his GF% and his Corsi% becomes ever more tightly tied. I’ve put together a chart that summarizes this:

    So for defencemen with at least 5000 minutes of 5v5 play between 2007-13, 73.5% of them saw goals scored at a rate that was within +/- two percentage points of their Corsi%. That’s pretty astonishingly tight. You can see, if you look from left to right, how the sample size increasing shrinks the difference between the Corsi% and the GF%. If Corsi% comes to equal GF% and we know that GF% comes to equal wins, then it’s pretty accurate to say that positive Corsi = more wins. These players who people imagine with a 55% Corsi% who give up so many ten bell chances that it overwhelms it don’t seem to exist.

    There’s plenty of evidence that defencemen don’t significantly impact on shooting percentage or save percentage, although I confess to wondering if the best two or three defencemen don’t have an impact on save percentage. If that’s generally true, the only sensible way to evaluate their 5v5 play is on the basis of how they impact on the Corsis – the goals are basically noise in the short run.

    This doesn’t mean that good Corsi = good defenceman. You want to try and figure out if the defenceman is causing the good Corsi or not. Sometimes, it’s pretty easy – Chara, for example, has a 54.8% Corsi% between 2007-13 on a team that’s a 52.1% Corsi% team overall. I have no difficulty in concluding that he makes them substantially better. Christian Ehrhoff has a 46.8% Corsi% this year on a Sabres team that is at 43.3% overall. I suspect that he’s a pretty good player.

    There’s a layer of stuff below Corsi% that we don’t fully understand yet. When we do, we’ll be able to say with much more accuracy than we can now which players are driving the play and which are along for the ride. With that being said, we can draw some pretty sound conclusions now by carefully examining data. As long as we look at the right stuff.

    Email Tyler Dellow at tyler@mc79hockey.com

    About

    6 Responses to @s: Is Positive Corsi = more wins 100% accurate?

    1. sacamano
      December 4, 2013 at

      This is my semi-annual comment about how your mission to increase the number of people who see value in quantitative data would be immeasurably more successful if you mastered even some of the most basic data visualization techniques.

      Boxplots would make the relationships in these data so blindingly obvious.

      Grumble, grumble, and get the hell off my lawn.

    2. Tyler Dellow
      December 4, 2013 at

      You should recommend a book on basic data visualization that my GF can buy me for Christmas.

      • sacamano
        December 4, 2013 at

        http://fellinlovewithdata.com/guides/data-vis-beginners-toolkit-1

        Few’s book (Show Me the Numbers) is a good basic intro — probably much too basic for you in parts given that he feels the need to explain some basic stats not only how to display them.

        Given your stats bent, I’d probably start with Tuft’s book (Visual Display of Quantitative Information) and/or the two Cleveland books (The Elements of Graphing Data and Visualizing Data) despite the fact that they aren’t as new or pretty as lots of the newer books.

      • ScrillaVilla
        December 9, 2013 at

        Hey Tyler, awesome work as always. I have to agree though, a good visual representation of the data would really go a long way. You can even make the case that the difference between a hard and soft science is their respective methods of data visualization. This article lays it out nicely:

        http://psycnet.apa.org/psycinfo/2002-18352-001

        The parallels between the current “debate” over the use of advanced stats vs the “intangibles” guys and that of behaviorism and psychology in general are quite interesting.

    3. December 5, 2013 at

      I’ve been trying to figure out which players are contributing to positive and negative Corsi outside of the context they play in a for a while using regression models.

      In the long run similar to the scope you’re using in your GF% vs CF% analysis (5000+ mins) the dCorsi results I’ve been getting are very reliable and the top end guys tend to be extremely good D men… the Charas and Subbans of the world.

      The reliability is also higher than what we see for 5v5 SV% in goaltenders over a similar time frame (which I find interesting as a comparable as people tend to rely on goaltending SV% far more than they should).

      Long story short – I think if you could construct a line up of guys that consistently out-perform expected Corsi results then you’d dominate.

    4. December 6, 2013 at

      Ryan Ellis presents an interesting test case for this line of thinking in Nashville. His Corsi results have been excellent for three seasons in a row, but he’s getting less ice time than ever (13.5 m/gm) because he’s not viewed as reliable. Yes, he looks like a kid sometimes getting out-muscled in front of the net, but the results say he more than makes up for it in other ways.

    Leave a Reply

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