• Evaluating Defencemen With CorsiRel

    by  • May 26, 2014 • Hockey • 13 Comments

    I have, over the past couple years, increasingly come to be of the view that the value of a defenceman at 5v5 lies in how he affects his team’s Corsi%. There’s no real magic to this conclusion – once you conclude that defencemen have a negligible impact on the save percentage and shooting percentage when they’re on the ice at 5v5, this is all you’re left with.1

    The question is then this: how do we do any evaluation of a defenceman by way of Corsi%? We know that it’s fraught with problems. Most obviously, some guys play on good teams and some guys don’t play on good teams. I’ve suspected for a long time that the league, as a whole, doesn’t do a very good job of dealing with defencemen who play on bad teams unless it’s painfully obvious that they’re really good but being held back. Even then, these guys tend to get tagged as being problems.

    The intuitive way to go about doing this is to look at CorsiRel, which looks at how a player’s team does with and without him on the ice. When I write about guys like Tom Gilbert and Anton Stralman, the basic premise of my argument is pretty much always that they seem to make things better when they’re on the ice. Players like that should be players that you want because they make your team better.

    I’ve never seen anyone explicitly test this – it sounds logical, but does it work? In order to take a look, I set out to identify guys who were top four defencemen who played an entire season with one team and then an entire season with another team. I set cutoffs for each year of 41 games played and 19 minutes a night of ice time. I discarded Dustin Bufuglien and Brent Burns, just because of the positional issues. This gave me a list of 50 defencemen who played complete seasons with one team and the next year played a complete season with another.

    A word about CorsiRel. There is probably a difference between a guy who posts +5.0% CorsiRel on a team that puts up a 40% Corsi when he’s not on the ice and a guy who posts a +5.0% CorsiRel on a team that puts up a 60% Corsi when he’s not on the ice. It’s harder to make a really good team better than it is to make a really bad team better. My suspicion is that, all other things being equal, CorsiRel works something like this:

    Screen Shot 2014-05-26 at 4.24.25 PM

    Don’t get hung up on the numbers in that graph – they’re just to illustrate a point. Basically, what I’m saying is that a rising tide lifts all boats but, unlike with tides and boats, with defenceman and Corsi, the rising tide will lift a defenceman but the higher the tide, the lower he sits in the water.

    OK. Let’s take a look at the guys who had a year one CorsiRel of at least +1.5% – guys who made their team better when they were on the ice. Did that continue in the second year?

    Screen Shot 2014-05-26 at 5.27.47 PM

    (Note: the “Corsi%” column refers to the team’s Corsi% when the player is not on the ice. To get his Corsi%, you just add or subtract as necessary.)

    In ten out of twelve cases, it did continue in the second year. Only Kyle Quincey and Ryan Suter weren’t positive CorsiRel guys on their new teams although, in Suter’s case, he was playing on a team that posted better Corsi and without Shea Weber as a defensive partner. I suspect, although I can’t be certain, that Colorado tried Quincey as a first pairing defenceman, which was probably a step up in terms of competition from his year in LA. Overall though, it holds pretty well.

    What if we look at the other side of the coin – guys who had a CorsiRel of at least -1.5% in Year One and then moved to another team?

    Screen Shot 2014-05-26 at 4.38.39 PM

    12 out of 16 were negative CorsiRel guys on their new teams. Of the four who weren’t, three of them (McCabe, Hamrlik and Ballard) went from decent possession teams to bad teams – it’s easier to look good on a bad team.

    So the theory – such as it is – seems to roughly hold. This doesn’t mean that I’d endorse picking guys off CorsiRel alone – there are too many confounding factors. I’d want to know, in particular, who a guy is playing with, who he’s playing against and where he’s starting on the ice. That’s why I plowed through some of that information with respect to Anton Stralman the other day. As the starting point on discussing a guy though, it seems pretty good to me.

    Email Tyler Dellow at tyler@mc79hockey.com


    13 Responses to Evaluating Defencemen With CorsiRel

    1. Aaron Luchko
      May 26, 2014 at

      “once you conclude that defencemen have a negligible impact on the save percentage and shooting percentage when they’re on the ice at 5v5, this is all you’re left with.”

      Not quite as there’s a difference between shots directed at the net and shots reaching the net. This might be particularly relevant for defencemen who often have a reputation for shot blocking.

      Why would Corsi be superior to Fenwick for evaluating defencemen?

      • Luka Ryder
        May 29, 2014 at

        Corsi and fenwick are very closely tied in terms of predictive value — although in the long run fenwick is a better predictor. The reason corsiRel is a good evaluative tool in this instance I believe is that in sample sizes as small as an individual defensemens season it is preferable to include all shot attempts as it simply gives you better sample sizes.

      • Benjamin
        May 30, 2014 at

        I’d love to know this too (whether FenwickRel isn’t in fact a better measure of defensive performance). I think the argument could be made for both sides, the traditional ‘blocking shots is important’ and the newer ‘goalies are so good that the risk of a tip instead of a block is too high’. It’d be really interesting to try to figure out.

        @Luka Ryder: sample size isn’t exactly king, quality data is king. If Fenwick is a better measure of performance and the sample size is still appropriate for the statistical test, it doesn’t really matter that the sample size of Corsi is bigger.

    2. Danny
      May 26, 2014 at

      I would imagine…That for a lot of these guys either moving teams through trade or free agency, that they had a decent season. Considering a team moves players or cannot retain them based on high value. Making the odds of the players personal stats to decline somewhat more likely. It would be interested to see these same players classified by how they left, how they performed in regards to previous play in there careers, and change in personal percentages. Taking into account the position they were used in and moved into some light might be shed on players to target.

    3. May 27, 2014 at

      Thanks for putting in the work, Tyler. The argument seems sound to me. Just wanted to stop in to say that the year you’ve got Quincey in Colorado was the year I counted chances for that team (for the first half of the season), and I can confirm that they had Quincey playing the toughest minutes, mostly with Scott Hannan, which even a few years ago, wasn’t an ideal situation.

    4. May 27, 2014 at

      Pierce Cunneen wrote something a few months ago along these lines: http://www.poweranks.com/blog/how-to-analyze-defensemen-and-the-top-5-defenders-in-the-nhl. But I like how your piece makes the context of these numbers easier to understand.

      In general, I agree completely with this, as it’s stated: if you’re looking for a good “rule of thumb” stat to evaluate defensemen, Corsi Rel is probably the best one. Just as long as you understand it’s a rule of thumb, context matters, etc etc.

    5. Ken
      May 27, 2014 at

      I’m curious how you might jibe this with your Duncan Keith post from a few months back. As you said, Keith’s Corsi #s implicated he was hurting the Hawks, despite observations to the contrary, “Keith looks awesome when I watch the Hawks play, and he’s highly regarded by Professional Hockey Men.”

      Keith finally cracked a positive CorsiRel this past season, though I’d guess it’s mostly related to a gaudy 57%ZS. I know this is a different question than the one you’re trying to answer here, but if certain guys are putting up questionable CorsiRel #s in the first place, and there’s such a wide variance year to year, even for guys who stay with one team, how much can we read into those numbers, even at a cursory glance? I guess Keith could be an outlier, but I feel like Rel #s bounce around a lot, with zone starts, and changes in the surrounding personnel playing a bigger role than anything said player is actually doing differently.

      • Truth Observer
        May 27, 2014 at

        When I compare Keith’s 2013/14 season to his last full season (2011/12), I notice a couple things imediately.

        Keith was paired with Seabrook more often this season (only 184 even strengths minutes were played apart from him, as opposed to 650 in 11/12) and his possession dips without him. Keith also played against tougher competition in 11/12: The six forwards that Keith played against most often this season were Berglund, Landeskog, Eakin, Brodziak, Hemsky and Whitney. In 2011/12? Franzen, Bertuzzi, Nash, Datsyuk, Backes and Pavelski.

    6. Bank Shot
      May 28, 2014 at

      RelCorsi will always favour defensemen that are offensive or not entirely trusted by the coaching staff.

      For this reason it falls down as an evaluation tool.

      A player like Ekman-Larsson for instance has slid down the rankings in Coyotes defencemen in relcorsi over the years as his responsibilities on the ice have grown.

      • MJamesD
        May 28, 2014 at

        Dellow ended his article with “This doesn’t mean that I’d endorse picking guys off CorsiRel alone – there are too many confounding factors.”
        CorsiRel doesn’t fail because of the limitations you have noted, it is not intended to be a “be all stat” or anything of the like.

    7. Pingback: Spectors Hockey | NHL Blog Beat – May 28, 2014

    8. Pingback: Evaluating Defensemen using their effect on both team and opponent Corsi% | hockey-graphs

    9. Pingback: Wie gut sind die Schweizer Verteidiger wirklich? | Emptynet Hockey Blog

    Leave a Reply

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