An Attempt At A Defensive Puck Possession Metric

I’ve been scratching my head for a while now, trying to come up with a puck possession metric that focuses on defense rather than offense-driven numbers. I mean, if you’re Anton Volchenkov, you might be responsible for preventing a lot of goals, but there’s always the chance that last year’s New Jersey forwards were incapable of putting shots on the opposition’s net, and that your Corsi numbers look bad. What’s a defensive defenseman to do? There’s got to be a way of measuring defensive contributions that isn’t tied to shots or goals.

So here’s my rationale: only the puck carrier can be hit, so if there is a lot of hitting in your end of the ice, the puck must be there a lot, right? Even if your team has the puck, and is getting hit a lot in their zone, doesn’t that mean that you’re struggling to get out? Leafs fans should certainly remember last year’s broken-record recording of "to the line, but not out", so anecdotally, anyway, this makes sense. 

Advertisement - Continue Reading Below

I’ve done some work similar to this before, where I calculated whether teams as a whole were hitting more often in the offensive zone or defensive zone, but only recently, with the help of the wonderful Eric T (follow him @BSH_EricT) from Broad St. Hockey, was I able to get individual numbers for players.

What we did was to tally every single hit that each player was on the ice for – that is, not just when each player was being hit or was hitting (that would only reward players that hit a lot) – and determine which zone it came in. Theoretically, if players’ teams hit more often in the defensive zone when a given player was on the ice, they’re working more frequenty to recover the puck there.

To see whether or not this makes sense as a puck possession metric on an individual player level, I’ve graphed my new statistic called Zone Hit Ratio (ZHR) to see if there is any correlation to Corsi numbers. Now, I never wanted too close a correlation, because we’re trying to measure something different than Corsi is, but the results were nevertheless underwhelming. This graph shows each player’s ZHR compared to their Corsi in 2010-11:

Advertisement - Continue Reading Below

So, looking at our ZHR axis, anything over 0.5 means that more than half of the hits a player was on the ice for were in his own end. There’s actually a very small correlation between having a good Corsi and hitting more frequently in your own end. Of course, it’s such a slight connection that it’s more than likely just statistical noise. Given that p=0.174 in this case, I don’t think there’s much to write home about.

On the other hand, ‘who is doing the hitting’ has a stronger correlation with Corsi than where the hitting is going on. Here, we’re looking for a relationship between a player’s Hitter/Hittee ratio (again, for all players while he was on the ice). In other words, "was your team out-hit, or did your team hot more than the opposition while you were on the ice"? All data from the 2010-11 season, once again. 


Here, we have a significantly stronger correlation, with a "p" value of 0.021. It’s not the greatest thing I’ve ever seen, but it’s worth talking about. It appears that if your team is controlling the play, they’re going to be the recipients of more hits. Makes a certain amount of sense, right?

The only downfall of this is that it’s not zone-specific, so we can’t get at who is doing the most work in the defensive end, but it is a way of looking at who controls the play without examining shots or goals.


I confess to being somewhat surprised at the lack of connection between Corsi numbers and ZHR, which should measure attempts to regain puck possession, but then, there are a number of points to be considered, here. Maybe that there isn’t any connection isn’t so perplexing, after all.

Advertisement - Continue Reading Below

First, is the unreliability of the data. If we can’t even count on the NHL to get faceoff locations right, deciding what is/isn’t a hit would be a nightmare of subjectivity.

Second, hitting an opposing player who has the puck is only one way of regaining puck possession. In fact, a well-timed stick lift or poke check is often a better option, because these other options don’t take you out of the play. So measuring hits isn’t anywhere close to measuring every single puck contestation that occurs in the game, and thus paints an incomplete picture of where defensive work is being done. Of course, puck contestations are far from the only way that puck possession changes, so this also neglects turnovers, dump-ins, and the like. My hope with this study was that hits happened frequently enough (and thus, with a big enough sample size) to measure puck possession in each end of the rink. Not so.

That there is a stronger relationship between the Hitter/Hittee ratio and Corsi is interesting, and merits further investigation.

I’d love to hear your thoughts/ideas on how to go about developing this, or other related ideas, so drop me a line on Twitter (@JPNikota) or leave your comments here.

This site uses Akismet to reduce spam. Learn how your comment data is processed.

  • There’s got to be a way of measuring defensive contributions that isn’t tied to shots or goals.

    Yes, there has got to be, and there are.

    Here’s a quick list

    Briefly, defensive GVT and defensive PC are great ways. So is DeltaSOT.

    Iain Fyffe once had Point Allocations, which looked at how much ice-time a player should have based on his offensive contributions, with any excess therefore being defensive.

    You can also look at a player’s OZQoC – Offensive zone starts and Quality of Competition.

    I’ve learned over the years that looking at shots and goals allowed isn’t ideal, because the great defensive players are matched up against the league’s best, and in the toughest situations, meaning they generally allow a lot more shots than their more defensively-challenged teammates (who get carefully sheltered).

    • What?

      “For ValD, the defensive component of GVT, team defense is measured by looking not at how many goals are allowed by a team, but rather how many shots were allowed (currently without regard to shot quality). The team defense is then assigned to players based on how many goals were allowed when they were on the ice.”

      And then, Delta SOT is a Corsi-based metric, so once again, that’s offence-based.

      Looking through the GVT explanations once again also tells me that it’s a combination of goals/shots.

      Even PCD relies “on the statistical link between wins and goals” and a comparison to replacement-level goal-scoring talent.

      I’m well aware of QoC/QoT, Zone Starts, and the good work done on faceoffs, but I don’t see how these other metrics are NOT based on offensive numbers i.e. shots and goals.

    • To be clear, I think it’s all very interesting and worthwhile work, and I don’t want to come across as adversarial. I’m in this to learn, not to be dismissive.

      But GVT, defensive PC, and DeltaSOT are all offense-based.

      QoC and QoT aren’t possession metrics at all, until it’s paird with Corsi, which is, of course, shots based.

  • BobB

    Before I even read this, I’ll say:


    Attention to creating a metric for defenders/defence, which actually reflects the dynamics of the game has long been needed.

    Far too long. And if it wouldn’t require me to quit my job to put a dent in it, I would have tried starting.

    It’s frustrating to see hockey boiled down to 1. dozens and dozens of metrics for offence (forwards), 2. sv%, and 3. Time on Ice (or a few other meaningless stats) for defenders.

    It turns Hockey into baseball: pitching and hitting.

  • Sometimes offense is the best defense. The lack of scoring on Wayne Gretzky, was that because he was great defensively? Or was it because the other teams were spending time in their own zone and/or so worried about his scoring that they put on defensive players in defensive roles?

    How about looking at how well the team does over-all with particular players on the ice? If you really want to separate the two, and if you’re happy with how to measure offensive contributions, then just subtract one from the other and you’re left with defensive contributions.

    Ultimately I do think defensive contributions boil down to preventing opposing offense, which is usually best generated in shots (or attempted shots/Corsi, or goals, or scoring chances, or whatever), but can also be represented by time in zone, or puck possession – which both correlate well with shots anyway.

    Of course it’s tricky, because not all situations are created equally – there’s a big difference between a shift in your own zone against the Sedins, and a shift in the opposing zone against 4th-liners.

    Even if you were to calculate expectations based on situations (like zone starts and quality of competition), and measure defensive contributions relative to those, it’s still tricky because in one case you’re comparing yourself to those who typically play against Sedin in their own end (i.e. great defensive players) instead of the league as a whole.

    But I digress. And I apologise if none of this is proving helpful.

    • I think JP’s concern is that there are players that are effective at limiting possession in their defensive zone, but that team effects arising from teammates having difficulty generating shots is masking their effectiveness.

      I don’t find the hitting metric very compelling for most of the reasons discussed above. That being said I do think a metric measuring just prevention of possession could be helpful. I would think something like the differential between corsi events against on ice versus average expected corsi events against would render some useful data.

      Some one smarter than me would have to sort that out though.

  • Just a couple of points:

    The reliability of the recording of hits is as far as I understand not there. I get that you used a more detailed recording of a single team, but it can’t be expanded using NHL data, because it sucks (misattributions, bias etc). It also completely ignores positioning, stickwork, and speed.

    “So here’s my rationale: only the puck carrier can be hit, so if there is a lot of hitting in your end of the ice, the puck must be there a lot, right? Even if your team has the puck, and is getting hit a lot in their zone, doesn’t that mean that you’re struggling to get out?”

    False premises. Hits recorded still only rewards and/or penalizes teams who hit or are hit often. Hitting may or may not have a correlation to good play. You’d have to prove to me that hitting actually measures anything at all that reflects on scoring/scoring chances/prevention of scoring chances/winning games before assuming that it measures something meaningful.

    The problem with hitting as a metric is that its effects, if any, are primarily attritional and psychological, as there is no guarantee that a hit will strip the puck or allow your team to gain possession. I don’t even think it has the virtue of mirroring zone time, as you won’t be getting hit a lot in your own end if you’re facing say Lidstrom and Datsuyk, or the Sedins, you just won’t have the puck.

    The classic ‘finish your checks’ etc. if consistently applied throughout the game may cause increased turnovers late in the game with hard forechecking etc. when the stress, fatigue and the accumulated hits cause rushed plays to avoid being hit again. That’s all anecdotal and ‘seen it’, and I have no clue how to measure it or how valuable it is. Is it worth one scoring chance? Maybe 2-3 shots a game? 1 more minute of offensive zone time? I have no clue, the only virtue this explanation has is that it actually matches what is seen in psychological research.

    I’m sure I’ve made some mistakes in my arguments, so have at them.