Looking At Corsi And Game-By-Game Results

There’s been a lot of discussion about puck possession lately, particularly how Calgary is succeeding despite some pretty awful metrics in that area.

With another game up tonight, let’s take a gander at how, specifically, they’re doing.

The Flames get out-shot and out-attempted quite a lot. How often?

Here’s a table.

Opponent
Attempts
More
Shots
Attempts
About
Even
(Within 5)
Flames
Attempt
More
Shots
Total
Flames Win 24 6 7 37
Flames Lose 13
(2 OT)
8
(1 OT)
9
(2 OT)
30
(5 OT)
Total 37 14 16 67

So, yeah. They get out-attempted 55% of the time, yet have won 55% of the time anyway. That’s crazy!

Breaking it down more specifically, the team has earned 67% of the available points when they’ve gotten out-attempted. When they’re breaking even, they earn 46% of possible points. When they’re out-attempting their opposition and driving the play, they earn 50% of available points.

Someway, somehow, this is a team that does better on the scoreboard and in the standings when they’re getting out-played than they do when they are out-playing the opposition.

And they get out-played a lot, based on possession metrics.

Screen Shot 2015-03-12 at 9.49.05 PM

(That’s a rolling five-game look at their Corsi For percentage, by the way.)

In terms of games that are outliers there are a bunch that stick out on both ends.

OUT-ATTEMPTED, WON

  • Won by 3 goals: Oct. 9, Oct. 19, Nov. 29, Dec. 2, Dec. 27, Jan. 27, Feb. 2, Feb. 9, Mar. 6, Mar. 11
  • Won by 2 goals: Nov. 8, Nov. 15, Nov. 26, Feb. 4
  • Won by 1 goal: Oct. 15, Oct. 31, Nov. 4, Nov. 18, Dec. 22, Dec. 29, Jan. 10, Jan. 19, Mar. 3, Mar. 5

OUT-ATTEMPTED OPPONENTS, LOST

  • Lost by 3 goals: Dec. 9, Dec. 16, Jan. 21
  • Lost by 2 goals: never
  • Lost by 1 goal: Oct. 28, Nov. 20, Nov. 25, Dec. 6, Dec. 11, Feb. 18

So, in other words, when the Flames get wildly out-attempted, they tend to win big or squeak by when they do win. When they lose despite out-attempting their opponents, they tend to lose squeakers.

I’m not in any way convinced that these distributions are anything but random, but they’re interesting to think about, particularly given how the season’s been going thus far and how little is left on the docket.

    • SoCalFlamesFan

      What’s terrifying is your and others’ knee-jerk responses.

      First off, getting out-attempted does not equate with getting “out-played”. I think a better, more consistent definition is winning equities to outplaying.

      Secondly, if you look at your own numbers there seems to be a clear correlation to lower Corsi=more likely win. Instead of just defaulting to the High Corsi=Better perhaps you should dig deeper…

  • mk

    I remember with Toronto a couple seasons ago, there was an attempt to explain their terrible possession stats by saying that they were winning too much: i.e. the fact that they were leading often meant that they would sit back & give up more chances as other teams pressed to catch up. My first thought here is that it might have something to do with it, except for the fact that that claim was shown to be inaccurate via looking at stats when leading/close/tied/trailing & at the amount of time the team spent leading/trailing.

    This season is a strange one. Loving it!

  • Rockmorton65

    Nice to see the factors determining the outcome of games are still hard work and discipline.

    First article about possession I’ve actually enjoyed reading. Thx

      • Rockmorton65

        Lol. Thanks for the props, but I wasn’t taking a shot. I was being genuine.

        There’s a lot of debate about advanced stats these days. I think they can be a valuable tool, just like “regular” stats, the “eye test”, and a coach/GM gong wth their gut.

        I recently read an article on Sportsnet asking why are the Flames successful even with bad numbers. It seems pretty simple to me.

        They Just. Keep. Coming. The Flames don’t stop working. Down by two with 3 minutes to go. Doesn’t matter. Down by three on the first 10 minutes? Doesn’t matter. It’s like they’re saying “if you’re going to beat us, fine. But you’re going to have to BEAT us. And to do that, you’re going to have to outwork us”. Unfortunately, for the analytics community, there’s no stat for that.

  • Ok I’ve been a long time lurker of this site and I’ve always been a bit surprised about Corsi being treated as though it’s the be all end all of possession stats. While PDO does in fact revert to the mean and teams with a high PDO tend to have an unsustainable style of play, the fascination with Corsi is kind of weird. Do you guys remember that unreal Rangers team in 2011/2012 with Tortorella where they blocked like 20 shots a game and won 50 games? Not to mention they reached the Eastern Conference finals.

    They were 23rd in Corsi. 23rd. And they were extremely successful. People weren’t calling that unsustainable back then they called it good solid defensive hockey.

    I’ve actually been looking for an explanation for this and stumbled on this:
    http://rinkstats.blogspot.ca/2013/12/why-popular-advanced-stats-are-bad-at.html

    It shows that on a game by game basis, Corsi is pretty much useless. It actually has a negative correlation to winning. You’re much better off just going by shots for and against.

    Over a large sample size I can see how it would paint a picture of how a team plays, however as a measure of future team wins, Fenwick is much better but still misses a bunch of factors.
    I actually think Corsi is much more useful on an individual player level than it is on a team, however again, all Corsi tells you is if they can bring the puck from the D zone to the O zone.

    There is a much better stat out there called xDiff that is basically Fenwick but instead of counting every single shot as 1 they count every shot as a shooting percentage. The shooting percentage they used is based on a graph of 7 years worth of shot data that they plotted shooting percentage (amount of times they score) vs. distance from goal. Based on how far a player gets a shot off they can predict what his expected shooting percentage is. So the shooting percentage in the “Fenwick” tally becomes like 0.14+0.1+0.05+0.23/(other teams expected shooting percentages). So now instead of just counting Fenwick/Corsi as 1+1+1+1+1/1+1+1+1 etc you get an extra factor of accuracy.

    For the record, the 2011/12 Rangers team that made it to the ECF was 8th in xDiff.

    At the end of the day, my point is that there are much better “fancy stats” out there than Corsi (TSN all of a sudden spamming it should tell you that it’s old news) and the usual blah blah the flames blow cause there Corsi sucks is a bit mind numbing for someone that checks this site every day!

    Anyways sorry for the wall of text! GFG

      • DoubleDIon

        Xdiff sounds like what I’ve been wanting. Corsi is a deeply flawed stat. Why does a shot that’s blocked at the point and leads to a breakaway against count as a positive?

        Also, I love Backlund but his wristers from the blueline side boards shouldn’t count for the same amount as the Gaudreau one timer last game.

        Xdiff, so much more relevant than corsi.

  • Burnward

    And I guess for those that are interested here’s the xDiff stat site.

    It shows that the flames are 23rd in xDiff, which in all honesty is probably a much more accurate description of where the team is. (Like can someone honestly say with a straight face that the Oilers, Blue Jackets, Coyotes, Leafs are better than the Flames??)

    Think of xDiff as Fenwick but actually accounting for *gasp* shot quality

    http://fourthlineheroes.com/#teams

    • DoubleDIon

      My new favorite poster. I agree wholeheartedly. I think the Flames are absolutely in that range and with the good fortune we’ve had it explains our playoff position.

      Anywhere in the 20-24 range passes my eye test. 28th does not.

      Blocks are important. Corsi missed the stat entirely, doesn’t account for penalty differential and doesn’t measure shot quality.

      There is a reason certain guys consistently shoot a high percentage.

  • PrairieStew

    @bookofloob

    100% agree which is why sometimes the Matchsticksandgasoline game by game Corsi analysis is a bit frustrating but all I’m saying is that Corsi is probably actually the worst “fancy stat” out there in terms of predicting future team wins because it lacks so too many factors. The biggest one being the whole shot quality thing.

    I just remember watching an Oilers game earlier this year that we just dominated in the sense that we were getting chances in the slot and way better chances but somehow got “Out-Corsied” and that’s kind of where my Corsi hate got started haha

    • Sure. The thing with corsi is over time it shows where the team has been, which, for better or worse, is probably one of our best methods of determining where the team will go. Obviously it’s not perfect, but patterns that emerge long term do tend to stick around.

      you’re right, a game by game corsi analysis is not particularly helpful because everyone has games out of character from time to time, but it’s possible to compare a season long trend vs. a one game metric and be like “well isn’t that interesting” (ie. watching just how low of a CF% TJ Brodie had in his first game being paired with Engelland. Was that Brodie having a bad game, or having his play brought down by a historically bad defenseman that is also not a Norris Trophy candidate. It’s possible it was one bad game, but it’s more likely to be cause and effect based on the pairings. Again, not perfect, but you can always infer)

      So I agree and disagree, I guess. How you use the information provided is everything

      • Burnward

        I’d argue xDiff is better than Corsi because it factors in shooting percentage as a function of distance from goal.

        And yeah I agree with the TJ Brodie analysis which is why I said above that I think Corsi is interesting on a personal player level, I just don’t really think something like the rolling game by game Corsi chart has much value in terms of team performance

  • Burnward

    2 teams hired the BEST advanced stats experts in the business last summer.

    Those teams are Edmonton and Toronto………..how is that working out for them?

    WW

  • redricardo

    It’s possible to buck the odds and win, despite having bad Corsi.

    That being said, the playoffs are generally a who’s who of the top possession teams in the NHL.

    If we want to rebuild, and become a dominant powerhouse team year after year, we need to continue to add and develop. As we do that, our Corsi and Fenwick will represent that, and we’ll turn the corner from “lightning in a bottle” to “genuine good team”.

  • RedMan

    Elliot Friedman notes that the Flames try the hail mary long bomb breakout pass more then any other team – a high risk high reward play that sometimes ends in a glorious opportunity but more often then not, ends in a change in puck possession.

    this long stretch pass drives me nuts as it misses so often, but then, given their penchant for doing it, other teams need to play back a bit, creating space. I don’t know – is this a thing?

    • Burnward

      They use it too as a quick dump in. I kind of like it.

      It seems to be a solid way to stretch out the D and helps that first forward to get in quicker on the forecheck.

      Can bite them in the ass though as they aren’t all that great on the dot.

  • beloch

    The sooner the stats community recognizes that shot attempts does not = wins the sooner they will be taken seriously. It’s a good metric that provides useful information. But it isn’t a magic formula.

  • DoubleDIon

    Here’s a question, how is that graph a 5 game rolling of the flames Corsi numbers it has easily over 50 points on it? A 5 game rolling average would only have 13 points on it. No?

  • ‘Time of possession’ may show a team is short on ‘quick thinking innovative scorers’ as their guys are plodders who need to hang on to the puck longer to give their brains time to appraise a quick scoring chance that great scoring lines make almost instinctively, ie. Johnny Hockey, Monahan & Hudler.

    Skating around forever – looking for a situation that matches ones brain speed in order to score is not my idea of ‘fun hockey’ no matter what the corsi guy says. Maybe ‘Time of Possession’ metric is too ‘one dimensional’ for a game as fast as hockey?