• OZ Wins and the Corsi% Thereafter

    by Tyler Dellow • October 5, 2013 • Hockey • 9 Comments

    In the comments to the previous post David Johnson is busy suggesting that I haven’t disproven his oft-repeated theory that faceoffs don’t really create a Corsi% effect but that what we’re seeing is the effect of better players starting in the offensive zone and weaker players starting in the defensive zone, with corresponding teammates. David’s never done the work to show this – he’s just seized on this as something that would justify his finding that the adjusted ZS Corsi% that he calculates seems to work ok.

    David’s adjusted ZS Corsi% does work well in many cases but the problem that I’ve always had with it is that his various theories as to why it works are obviously nuts. As always, how you get to a solution matters because if you don’t understand why something works, you can have false assumptions that lead you into error in other matters or cause you not to understand your model’s weaknesses. For whatever reason, David’s never really understood this criticism of what he’s doing.

    What the previous post showed is that there are substantial effects associated with winning or losing a faceoff and where it took place. Duh. What I’m going to show now is that that the effect of, for example, winning an offensive zone faceoff, is structural and that even players who are the fringiest of NHLers don’t do much worse than all-stars when it comes to turning those OZ faceoff wins into Corsi%.

    There were about 1106 forwards in the NHL between 2007-13. I say “about” because the NHL record keeping isn’t what you’d hope and there is an Alex Ovechkin and an Alexander Ovechkin. There aren’t too many of these and it doesn’t matter too much for my purposes. These people were collectively on the ice for 253105 OZ faceoff wins with the goalie in.

    I have sorted this list by the percentage of a team’s offensive zone faceoff wins that the players were out for. I have then created two groups. One group has 29 players. This is the top 10% of players, in terms of having been on for about 10% of the total OZ faceoff wins during this period. These are the players who were out for the greatest percentage of their team’s offensive zone faceoff wins. By extension, these are excellent offensive players who the coaches want on the ice. Sedin, Sedin, Tavares, Lecavalier, Nash, Naslund, Kane, Joe Thornton…people like that.

    The second group is the bottom 10% of players. This group has 535 players, the vast majority of whom you’ve never heard of. These players were the least likely to be on the ice for an offensive zone faceoff win. By and large, they are fringe NHLers. They have names that sound like an RV brand (Carter Camper), off brand stars (Brock Trotter), a character from Hogan’s Heroes (Carl Klingberg) or they are JF Jacques.

    How will our group of NHL All-Stars, Olympians and Hall of Famers match up in a comparison with our group of, inter alia, failed Swiss draft picks (Luca Caputi) in terms of Corsi% in the 37 seconds following OZ faceoff wins?

    Screen Shot 2013-10-05 at 3.44.24 AM

    Wow, what do you know? Whether it’s people who can’t go outside in Canada because they’re too famous or a guy who you might see in your beer league in a year or two, there’s a massive offensive advantage that is created by an offensive zone faceoff win regardless. Whether it is a superstar or someone you’ve never heard, about 75% of the shots in the 37 seconds are going to belong to the team that won the faceoff.

    This has all sorts of implications, which I’m sure people will come to on their own. The critical point for me at this time is this: offensive zone faceoff wins don’t come with a ~75% Corsi% over the next 37 seconds because they’re being taken by stars. Full stop.

    Email Tyler Dellow at mc79hockey@gmail.com

    About Tyler Dellow

    9 Responses to OZ Wins and the Corsi% Thereafter

    1. October 5, 2013 at

      “David’s adjusted ZS Corsi% does work well in many cases but the problem that I’ve always had with it is that his various theories as to why it works are obviously nuts.”

      I have always been clear in how I developed my method and I have always stated that beyond 10 seconds there is no significant impact on a players 5v5 statistics.

      You say my work does well in “many cases” and you have often stated my adjustment is “wrong” but you have yet to give me any cases or a set of cases where it systematically fails. What are the practical benefits of your method over mine? If you can show me that there is a practical benefit, I’ll gladly make the change in my code and on my website, but what I am not going to do is listen to all your crap about how wrong my method is when you haven’t provided one single player where there is a significant difference in the resulting overall statistics. I am all for advancing our knowledge and improving the statistics I provide but all I am asking is that you prove to me that the additional adjustments you make are meaningful and beneficial and need to be taken into consideration.

      So, lets put this to an end to this by doing a comparison between my 10 second adjustment and your ‘open play’ adjustments for the 29 ‘Stars’ you identified above. If your theory is right and that the time between 11 and 37 seconds matters significantly we should find that my SAF rates and corsi% are systematically and measurably higher across the group. Let’s once and for all find out what the practical implications of your research is. It really shouldn’t take more than a few minutes to do so I don’t understand why you haven’t done it.

    2. Tyler Dellow
      October 5, 2013 at

      Daivd -

      Amazing. You’ve managed to completely miss the point. Just amazing.

      • October 5, 2013 at

        What’s the point I am missing? You have discovered that winning a face off in the offensive zone has dramatically different results than losing a face off in the defensive zone. I understand that. I get that. It is completely intuitive.

        What I am saying is that your discovery is for all intents and purposes largely irrelevant on how we conduct player or team analysis and until you prove otherwise the research you have done is nothing more than academic in nature. For the most part, I am not interested in academic research such as this if there are no practical implications. What I, and I am sure many of your readers would be interested in, is the practical implications of your work. How does your work impact how we conduct player analysis? How does it impact what stats I put up on my website?

        Ultimately though my beef is that you have consistently called my method for adjusting zone starts wrong. Recall http://www.mc79hockey.com/?p=5790 where you suggested the impact of zone starts is far far more than my adjustment suggests? I am happy that you have progressed to saying my method “does work well in many cases” but I ask that you provide evidence of cases in which it does not work well or else stop suggesting that such cases exist.

        • Tyler Dellow
          October 5, 2013 at

          Move along David. Go miss the point elsewhere.

          • October 5, 2013 at

            I get your point. Just stop suggesting something is wrong with my method with no evidence.

            • Tyler Dellow
              October 5, 2013 at

              Are you out of your mind? I’ve just laid out over two posts why your reasoning for why your method produces results that work in most cases is insane. You are wrong.

    3. Axel Fant-Eldh
      October 5, 2013 at

      My takeaway is, and I might be wrong as I haven’t done any calculations and am just going by my math intuition here, that there is a clear dropoff after 10 seconds (which can clearly be seen in your graphs in the prior article) and even though there are still some residual effects at that point they get diluted in a large sample to the point that they are negligible. Open play Corsi may be more accurate, but isn’t it also more complicated to calculate and gives a smaller sample which creates variance issues?

      I mean 5v5 Close data is arguably more useful than total 5v5 data, but when looking at a single player it usually creates too small a sample so that variance effects to appear. It only becomes really useful on a team level.

      My overarching point is that while faceoff effects aren’t totally absent after 10s, removing those 10s reduces them enough to make them negligible and at the same time creates the largest possible sample. The marginal return on removing additional seconds after 10 diminishes heavily so to speak.

      But I do agree that the player skill argument is bogus, and you proved it here.

    4. October 5, 2013 at

      “I’ve just laid out over two posts why your reasoning for why your method produces results that work in most cases is insane. ”

      From day one I was clear on how I developed my methodology. I showed that beyond 10 seconds there is a negligible impact on a players overall 5v5 statistics. Everything I have said as to why 10 seconds is all that matters was hypothesizing. If all you are debunking is my hypotheses then so be it. If that is what you enjoy doing fine, I don’t care, but when you say things like “David’s adjusted ZS Corsi% does work well in many cases” and that the effect lasts as long as 37 seconds you are inherently implying that there are cases in which it does not. Either prove it by showing examples where it fails, or stop saying it.

    5. October 18, 2013 at

      標準のマップと同様Googleマップが使えて、しかもちゃんとナビゲーションしてくれて(ただし音声案内は英語)、オフラインでも使えます。しかもちゃんと日本語メニューオフライン用の地図データは、xGPS ManagerというPC版アプリでGoogleマップのデータをダウンロードしてiPhoneに転送できますただし、オフラインでルート検索はできません。自宅のWiFi経由でルート検索してから、車に持ち込むことになりますxGPSは、原則、現在位置を表示し続けるので、任意の場所の地図を見るには適していません標準のマップは、任意の場所の地図の参照には向いてますが、ナビゲーション機能はありません用途によって使い分けが必要ですが、オフラインマップを重複して持つとiPhoneのメモリを圧迫するので、どちらのオフラインマップを使うか、しばらく両方使ってみてから決めようと思います。?iPhoneで写真を撮る人に必須のカメラアプリ定番10選

    Leave a Reply

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