• Taylor Hall, Fancy Stat-ist

    by  • March 12, 2014 • Hockey • 16 Comments

    TSN’s Ryan Rishaug did an interview with Taylor Hall that TSN aired before the Oilers game in Minnesota. It was pretty interesting for a couple of reasons. First, it’s still kind of weird to see a hockey player dropping Corsi and Fenwick correctly in an interview. Hall made what I consider to be an entirely correct observation about the challenge that teams will have with turning data into things that are actionable:

    Whether they’re useful or not (referring to data or, to use the phrase I hate “advanced stats”), I do think they are for sure, but the thing for a hockey player, if you’re an advanced stat guy and you’re describing to a hockey player, you have to have like some kind of end point. Like, what does he have to do better to get this stat better. That’s the thing that I’m lost on, with Corsi and Fenwick and all this stuff, how do you improve a player by it, what do you tell him?

    That’s sort of what I’ve been hung up on for the past year, thinking about how to do that. It was another comment that he made that I want to poke into a bit though. Rishaug asked him about his Corsi% this year and this exchange ensued.

    Hall: This year my Corsi hasn’t been as good as last year.

    Rishaug: And does that make sense to you based on your play, does that add up?

    Hall: It doesn’t, no. But my chances, for and against, are about the same as last year. So what that means, I’m not really sure. I know we have an advanced stat guy that does it all for our team and I asked him, I said ‘So why’s my Corsi not as good?’ and he really didn’t have an answer for me.

    In all fairness to the Oilers stats guy, it’s a hard question – I’ve fiddled with it a fair bit and I have some theories, some ideas, but not much more than that. It’s a question that’s answerable with data, I think, but the data doesn’t exist to do it. Someone needs to generate it. I imagine some teams do and more will in the near future.

    It’s the bit about the scoring chances that really caught my attention though. Hall doesn’t really explain the statement and, in particular, whether he’s talking about the chances when he’s on the ice or whether the Oilers have a metric like Neilson Numbers. Neilson Numbers essentially involve assigning credit/blame for scoring chances for or against. The problem with them is that they don’t credit a player who creates the context, say by breaking a cycle in his own end, or making a quick outlet pass. Taylor Hall’s going to have way better Neilson Numbers with Brian Campbell and Jeff Petry behind him than he’d have with me and Dennis King. Basically, Neilson Numbers suffer from the same flaw as Corsi% does; they just pretend to deal with it by discarding information.

    David Staples has counted scoring chances for the Oilers over the past few years. Unfortunately, he doesn’t present the time stamped information that would let us see whether the totals when Hall’s on the ice have changed. He does, however, present totals. In 2012-13, according to Staples, the Oilers got 44.9% of the ES chances. They had a 44.5% Corsi%. This year, the Oilers have had 46.6% of the ES chances and they’ve got a Corsi% of 44.6%. So they’re getting a few more of the chances but not that many.

    Keep in mind that Corsi% is generally presented solely on the basis of 5v5 data – that’s what I’ve used here. Staples’ data is ES, which I assume includes 4v4. I’d guess that that accounts for some of the difference – the Oilers have been a much better 4v4 team this year. Overall though, the scoring chances seem to track the Corsi% pretty well, which is what we’d expect given what Eric Tulsky’s found in the past.

    What if we look at Hall’s Neilson Numbers? While they’re not particularly helpful data, they do give us a sense of how many scoring chances Hall was on the ice for, assuming Hall’s rate of contribution to scoring chances remained the same. Staples calculates these on the basis of contributions to scoring chances per 15 minutes. We’ll contrast that with the number of shot attempts for and shot attempts against that he’s been on the ice for per 20 minutes, just to see if the change is similar.

    The decline in contribution to scoring chances is basically bang on with the decline in the shot attempt rate when he’s out there. According to Staples, he’s contributing to scoring chances at about 80% of the rate he did last year; the Oilers are getting shot attempts with him on the ice at about 83% of the rate that they did last year. The increase in the scoring chance against rate is way higher than the increase in the shot attempts against when he’s on the ice, which is sort of odd.

    If Hall was talking about the overall numbers and intending to convey that the Oilers get about the same chances for and against when he’s on the ice this year and last, and Staples is accurately tracking who makes the play or mistake that leads to the scoring chance (and I don’t really take issue with his ability to do so, subject to some limitations in terms of knowing the Oilers systems that any outsider has), then it means Hall is contributing less frequently to scoring chances and making more mistakes on scoring chances. That seems really unlikely to me. I’m inclined to think that there has been some sort of negative change in the scoring chances when he’s on the ice.

    There’s a point that I possibly haven’t made clear enough: I’ve written a lot about it being a problem when Hall’s on the ice. That doesn’t necessarily mean it’s a problem with Hall – it could be something else entirely that’s resulting in the puck being on his stick with an opportunity to make a play less often. I don’t think that this is the issue but if, for example, you stuck Hall with four guys like me, his Corsi% would crater but it would be nothing to do with him.

    One other point: it’s not just Hall’s Corsi% that’s dropped off, the goal for/against ratio when he’s on the ice has fallen off too. Last year, the Oilers got 53.8% of the goals when he was on the ice at 5v5. This year, it’s 47.4% of the goals when he’s on the ice. His on-ice shooting percentage and on-ice save percentage are virtually unchanged – this isn’t a case of bad luck. If he’s getting the same volume of chances for and against when he’s on the ice, then the opposition must be scoring on way more of their chances this year and the Oilers must be scoring on fewer of theirs, with the opposition adding a bunch of low risk shots and the Oilers not getting many of them. This doesn’t really make sense to me as an explanation and yet, it would have to be true if Hall’s right about the scoring chances.

    In any event, his broader point is accurate – there has to be an end point, some sort of identification of X, Y and Z that has changed for the information to be actionable. At present, it’s good for providing direction as to where things are going wrong/right. I’ve no doubt that the Oilers will figure it out, or that they have and it’s a short term pain for long term gain kind of thing. When data gathering takes the next step – which I firmly believe it will – this sort of thing should become much easier to tackle.

    Update:: The indispensable Young Willis notes on Twitter that he has been tracking scoring chances based on who’s on the ice. Last year, he had the Oilers getting 52.7% of the 5v5 chances when Hall’s on the ice. This year? 47.6%.

    Email Tyler Dellow at tyler@mc79hockey.com

    About

    16 Responses to Taylor Hall, Fancy Stat-ist

    1. Philip
      March 12, 2014 at

      Observations: Hall is trying to do things entering the offensive zone by himself and more often than not, he loses the puck. At times, he makes poor choice in passes and usually intercepted by opposition.
      Hence, his lower scoring chances and goals for when he is on the ice is a result of a combination of poor puck support and bad decision-making that plays end with him.

    2. Halfwise
      March 12, 2014 at

      I like Hall’s comments because they show how a competitive person thinks.

      Indicators are not performance. He wants to know how to improve his Corsi because he believes it is important. And it is, to a point, because Corsi correlates to wins.

      But good coaching and practice habits should address the on-ice behaviors that tilt the ice. Corsi will improve but the real objective has to be wins not better Corsi.

      The value of analytics is in analyzing… what has changed and when, who was involved, where does it happen and not happen, what do the exceptions have in common.

      Until Hall gets specific help with what he is supposed to do differently I don’t see how Corsi and Fenwick make him a better player.

    3. Ian
      March 12, 2014 at

      I’m actually impressed by the maturity in Hall’s response. He can seem somewhat petulant, and I would have expect him to downplay the downward trends in possession numbers and point to his counting stats.

      I’d agree that it reflects a competitive mindset and a desire to get better. I don’t think Hall is trying more solo efforts, and while the Oilers are poor at collecting dump ins I’m not sure they’re worse than the past. To me, it looks like the system (that is, the rules players are asked to follow regarding rushing the puck into the offensive zone vs dumping it in) is too unpredictable (i.e. players aren’t sure what others are going to do, leaving them stationary when they need to be moving or waiting for a pass when a dump in is coming). There is a balance between allowing for creativity (necessary for goals on the rush) and predictability (allowing teammates to get in position to retrieve a dump in) and the Oilers clearly have not found it.

    4. March 12, 2014 at

      My Corsi stat thoughts:

      1. Corsi really needs some subcategories. Let’s pick:
      a. Shots from a distance from a puck carrier
      b. Shots from a distance immediately after a pass
      c. Scramble shots with a scrum around the goalie
      d. Deke / shots where the shooter and goalie are 1-on-1 at close range
      e. Shots where the goalie has “no chance” i.e. team defense breaks down, back door tap in – an error

      Each of these scenarios result in different save percentages for every goalie, and different levels of danger for the defenders.

      2. Its like counting runs in baseball. You’ve got to breakdown homeruns (c,d,e) from small-ball (a,b), from stolen bases and errors.

      3. What is therefore more important than tracking a,b,c,d,e events is what is AVAILABLE to the attacking team. For example, if the attacking team enters the zone, and spends 20 seconds moving the puck to players with several opportunities for a,b and c shots, but no actual shots are attempted, that is far more dangerous than having a single shot on goal from the puck carrier from a poor angle immediately upon entering the zone.

      4. In other words, great attacking teams create sequences where a,b,c,d and e may become available, but only shoot the puck when the expectation of a goal is far greater than the risk that a turnover ends the sequence. Conversely, poor teams can create the same number of shots, but are not forcing the defending team and goalie into the most dangerous and low save percentage subcategories. The poor teams have fewer sequences, and less dangerous sequences.

      5. Track ‘Corsi Availability’ as above and you’ll answer Hall’s question.

      • March 12, 2014 at

        For example, if the attacking team enters the zone, and spends 20 seconds moving the puck to players with several opportunities for a,b and c shots, but no actual shots are attempted, that is far more dangerous than having a single shot on goal from the puck carrier from a poor angle immediately upon entering the zone.

        Not it’s not. The first part of that scenario has a 0 percent chance of resulting in a goal. The latter, while probably a miniscule percentage, at least has some chance of going in.

        • March 13, 2014 at

          No no. The low percentage shot, once saved, results in a turnover and ends the sequence. The sequence is over and there is zero chance of scoring. If possession is retained, a new sequence can begin, and the opposition is without the puck. Darryl Sutter’s comments from a couple of days ago are on this point.

          • dawgbone
            March 14, 2014 at

            Who constantly plays like that though?

            Corsi is a proxy for puck possession. That’s it’s role.

      • Rob
        March 14, 2014 at

        As I understand it, Corsi was borne out of the following chain of logic. We observe that almost all goals scored occur when the attacking team is in the offensive zone (obvious). A team that spends more time in the offensive zone than its opponents do will score a proportionately larger number of goals. How do we measure offensive zone time without actually sitting down and watching every game with a stopwatch? Well, some bright person figured out that “Corsi events” correlate very strongly with offensive zone posession time. So let’s count Corsi events.

        The subtlety here is that Corsi events are not being used as a measure of goal scoring chances, but of zone posession. If they were being used as the former, then measuring how “dangerous” each Corsi event would make sense, but they’re not (which is not to say that that’s not a useful exercise, it’s just not the exercise that’s being performed with Corsi). Corsi is kind of like measuring how dangerous a driver is based on the number of speeding tickets he gets a year. Yeah, a ticket for going 10kph over the limit is not as bad as getting a ticket for going 100kph over the limit, but the assumption is that the more dangerous driver gets more of both kinds of ticket (and everything in between) than the safe driver. However, if you wanted to measure which of those particular incidents was more likely to result in an accident, then you certainly would be better off treating the 100kph ticket as more dangerous than the 10kph one. This is also why it doesn’t make sense to pay attention to small slices of Corsi like per game data. Anybody can get a speeding ticket on any particular day, that doesn’t they’re necessarily a dangerous driver. It’s the pattern over the long haul that’s meant to tell the story.

        This is one place where the old saw of “corellation is not causation” can be put to good use. Corsi is not causal to zone time, it’s correlated. Thus, doing things to artificially increase your Corsi by, e.g., taking bad angle shots or shots from 100 feet away, do not help win games. Increasing the underlying metric that correlates with Corsi, i.e., offensive zone time, should help win games and, as a byproduct, should lift your Corsi. Contrast this with scoring chances, which (I think most people would argue) are causal with respect to winning games. Increase the number of scoring chances you get and you will score more goals and win games.

    5. Tom Benjamin
      March 12, 2014 at

      Whether they’re useful or not (referring to data or, to use the phrase I hate “advanced stats”), I do think they are for sure, but the thing for a hockey player, if you’re an advanced stat guy and you’re describing to a hockey player, you have to have like some kind of end point. Like, what does he have to do better to get this stat better. That’s the thing that I’m lost on, with Corsi and Fenwick and all this stuff, how do you improve a player by it, what do you tell him?

      Bingo. Even if I hold my nose and pretend that individual Corsi rates are not hopelessly polluted by context, there is no place to go with the data. At best they are a result of playing hockey well or poorly. They will take care of themselves if a team skates hard, checks tenaciously, plays well positionally, and moves the puck.

      Getting better as a team involves getting better players, or getting your existing players to play better, to correctly execute thousands of tiny details correctly over and over again.

      Your post on Petry and Marincin provides an excellent example. Petry misread the play thinking Marincin was in control and turned the wrong way. When Marincin erred, Petry is out of the play and the puck is in the net. It is not incorrect to blame Marincin for the turnover, but players make mistakes all the time. Petry should have been in position to bail him out. It took both small mistakes to produce the scoring chance – which explains why hockey appears to have so much randomness. If Marincin gets the puck out, nobody notices Petry’s mistake. If Petry turns the right way, Marincin’s turnover is forgotten.

      (And this is the biggest problem with the Staples statistic. Most of the time Petry gets away with his mistake and Staples only charges him when he doesn’t get away with it. We don’t know whether a bigger or smaller percentage of Petry’s errors end up in the net than say what Andrew Ference’s mistakes cost him. Ference may make far more errors, but get away with it far more often.)

      Petry surely heard about his error on the goal, but the key to improving Petry as a player is for a coach to point out the error even if the error is irrelevant because Marincin gets the puck out. The game is so fast and chaotic misreads and mistakes are inevitable. It is the attention to little details that prevents mistakes from becoming a chain of errors and the same attention forces errors on opponents.

      (While I disagreed with your take on that particular goal, I certainly agree with your assessment of Petry generally. When I look a young defensemen – young players generally – I am interested in their skill. I know nothing about Marincin and his Corsi rates don’t tell me much. I like Petry’s tools enough and even if this sort of mistake is relatively common for him, it would not be a good reason for me to discard him. This is the sort of thing that can be fixed. I can’t fix a guy who is too slow.)

    6. FastOil
      March 12, 2014 at

      Great piece Tyler. Are you any older than ‘Young Willis’ I wonder? Kid Dellow?

    7. CalgaryOilBaron
      March 12, 2014 at

      I’m not really a fan of the “advanced stats” moniker either, but at least everyone knows what I mean when I use it.

      If we want information derived from data to be “actionable”, then we need better data. The fact that a player can’t figure out how to improve his Corsi% is quite telling because the data doesn’t isolate any specific behaviour. The more behavioural measurement we can do, the more players will be able to learn about what they need to do to improve.

      Here are some things I’d like to see data on:

      Giveaways – yes this is already measured by the NHL, but I understand that there is a lack of consistency in how a giveaway is recorded. A giveaway can be defined as a player, who has possession of the puck, loses possession by either forced or unforced error. The focus here is to improve the reliability of the data.

      Turnovers created – this one is the flip-side of giveaways – when a player (on Team A), who does not have possession of the puck, causes a member of the opposing team (Team B) to lose possession of the puck to the opposition (Team A). This is different from a “takeaway”, which is both unreliable and misleading.

      Passing success – passing the puck is a possession vulerability. It’s a crucial, but unmeasured skill in hockey. There are three possible outcomes of a pass: 1) completed pass, 2) incomplete, loose puck, and 3) interception, change of possession. I’m less interested in cases where missed passes are recovered because the passer missed his target or the receiver missed the pass. Either way that stat is captured below.

      Passes attempted – total number of passes attempted

      Passes completed (%) – percentage of attempted passes that reached their target – possession maintained.

      Passes received (%) – This is the measure of a player’s ability to receive a pass. There are two possible outcomes: 1) pass received – possession maintained, 2) pass missed. There is no reason to measure whether possession was maintained after the pass was missed because we’re measuring the player who missed the pass, not the team’s possession.

      Passes intercepted – This is a measure of players who do not have possession of the puck, but gain possession by intercepting a pass — a skill of positioning and anticipation. Players who give away the puck by causing an interception are picked up by the ” giveaways” measure.

      Dump-ins – used as a global team measure. Total number of times the puck is chipped below the goal line in the offensive zone. The limitation of this one is filtering out dump-ins for line changes. Perhaps differentiating between active (dump and chase) and passive dump-ins (line changes) would help.

      Dump-in recovery – used to measure dump-and-chase success. Total number of times the offensive team maintains possession of the puck in the offensive zone. This too has some issues because possession can change a few times before it leaves the offensive zone.

      Icing – both an individual and team measure. Who iced the puck and how many times (not including PK)

      % of shots blocked – this is a measure of a shooter, not a blocker. It is the percentage of shots taken that are blocked by the opposition. Avoiding having your shot blocked (reading the defence, quick release, move/drag) is just as much of a skill as blocking the shot.

      Performance consistency – this is exactly what is sounds like and can be easily summarized for any measure you wish – how consistent a player is over time – points, blocks, giveaways, passing, etc. It’s ultimately an average of all data in a category plus a variance to show how much fluctuation exists in that category over time. This could be broken down into five-game segments by default, but could be adjusted to any period of games.

      Rebounds – a new goalie stat – how many rebounds went back into the slot (regardless of what happens afterwards).

      http://www.coppernblue.com/2014/3/11/5496228/change-needed-in-advanced-hockey-stats

    8. Shane
      March 13, 2014 at

      I am glad that Hall recognizes that “advanced stats” are beneficial and wants to improve those numbers knowing that will improve his production and team success. The issue is that he went to the wrong guy to findout what to do to change things, that is not the stats guys job, that is the coaches job.

      The stats guy should be gathering all the stats possible in order to come to a conclusion. Then the video guy needs to understand the numbers and look for trends in that player or the line of that player that are both positively or negatively affecting those numbers and communicate that to the coaches and the coaches need to communicate those trends to the player/line in order to implement the change required to improve those stats.

      The more detailed the stats are, the easier it is to reveal the positive and negative plays that affect those stats. It seems to me that Hall knows he needs to improve his Corsi but nobody is communicating to him what he needs to do to accomplish that.

      I don’t care how many stats you track, if there is no process put in place to learn from them and use them to teach and coach your team, those stats are useless.

    9. Bruce McCurdy
      March 13, 2014 at

      David Staples has counted scoring chances for the Oilers over the past few years. Unfortunately, he doesn’t present the time stamped information that would let us see whether the totals when Hall’s on the ice have changed.

      At the suggestion of me and perhaps others, David has started to publish time stamped info this season, e.g.
      https://t.co/rrPZQ96xmT

      (And this is the biggest problem with the Staples statistic. Most of the time Petry gets away with his mistake and Staples only charges him when he doesn’t get away with it.)

      Staples would charge him on both scoring chances and goals. Every player on the ice gets away with a lot of stuff on plays that don’t result in scoring chances.

      In that specific case Staples definitely nailed Petry with what I call an “Error of Omission”.

    10. Tom Benjamin
      March 14, 2014 at

      I understand how Staples derives his numbers. I also realize that every player on the ice makes mistakes that turn out not to hurt the team. Sometimes they even make mistakes that help the team. We’ve all seen out of position players get handed an opportunity.

      That’s the point. “Mistakes per minute of ice time” might be a very useful indictor if it was data collected. What do we learn if we only count mistakes that cost if most mistakes don’t cost? Are we fairly criticising a player for posting a poor number? Is the player really error prone? Or is it that the player is unlucky enough to have a larger than normal percentage of mistakes end up as scoring chances?

      • Bruce McCurdy
        March 14, 2014 at

        What if the player is unlucky enough to have a larger than normal percentage of mistakes end up as shots against / any other negative stat ? The same type of question applies to virtually every stat on the sheet. Whatever we count as 1′s, there are always many uncounted 0′s.

        • Tom Benjamin
          March 14, 2014 at

          I’m not sure that I get the connection between counting mistakes and shots against, but I probably agree. There are very significant problems with every stat that purports to measure individual player quality. I don’t take any of them very seriously.

    Leave a Reply

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