A look at line and pairing success rates through 28 games

The Rangers are currently on a one-week break, so this is a good time to take a step back and assess which lines and pairings have been effective, and which have not. Given the large number of injuries, specifically to the forwards, only five lines and nine forwards have logged enough ice time together to make the charts (courtesy of Sean Tierney).

First things first, this will be in expected goals, not Corsi. It’s a better metric.

While the lines are most certainly not complete due to injuries, it’s clear the Rangers have one line that is a cut above the rest, and that is the Chris Kreider-Kevin Hayes-Filip Chytil line. That line has been spending a lot of time together lately, and they are just flat out good. On the opposite end of the spectrum is the Jimmy Vesey-Brett Howden-Ryan Strome connection, which has been relatively atrocious. The rest are neither here nor there, but it’s worth noting that Kreider with Mika Zibanejad and Jesper Fast seems to trade chances.

This passes the sniff test, since the Kreider-Hayes-Chytil line has been visibly the best line out there recently. We also enjoyed watching the Kreider-Zibanejad-Fast line when they were together, hence the lower right “fun” section. The injuries to Pavel Buchnevich and Mats Zuccarello have altered what lines we see here. When they come back, I’m intrigued to see what sticks.

As for the blue line, well, that’s ugly. Like….really ugly.

While injuries have decimated the forwards, the blue line has been healthy. The best thing we can say is that the Fredrik Claesson-Kevin Shattenkirk pairing is, at the very least, good defensively. I guess we can say it’s good that no single pairing is in the “bad” section. But there are none in the “good” section either. This is just a mess of mediocrity. On the bright side, they can’t get any worse…probably.

Despite that 9-1-1 run in November –which was fun while it lasted– the Rangers are what they are. We expected them to be mediocre, leaning towards bad. So far the numbers are showing exactly that. The production has been mediocre to bad, with some flashes of good. Sometimes stories aren’t exciting, and the predictions are somewhat accurate.

The Rangers have probably banked enough points to keep them out of the bottom-five in the league, which is probably better than any of us expected. That said, they aren’t a playoff team. As the Rangers sell off more assets, it’s expected that these numbers and results will only get worse. Rebuilds are great on paper, but these are the tough times that were expected.

Show More
  • I really don’t get people’s commitment to the expected goals stat, or see why it’s “better” than Corsi. One thing that has made me dislike XG% is that it’s a hypothetical statistic. It justifies good efforts in losses with a nebulous “oh, they should’ve won”, which to me is a fairly baseless stat. In a league with as much parity as the NHL it’s arguable to say that every team has a relatively equal chance to win a game.

    A sport like hockey that has so much randomness (deflections, bad bounces, bad ice in certain arenas etc) built into its everyday play can’t truly be quantified at some levels, and XG seems to be a stat that is attempting to do just that. Can’t understand why so many people are enamored with this stat, to me it should be called the “should’ve/could’ve but didn’t” stat, as it justifies performance with on-ice events that don’t actually occur. And since it is a statistic I’d think that by definition it has to deal with real past-time events; anything that justifies probability is quantum physics, and not really a sports statistic.

    Feel free to have at it; I’d like to hear from defenders of this stat, and find out why expected goals are something more than the latest flashy new fancy stat.

    • First of all, I question how these really detailed stats can be accurately tracked for every player on the ice, every shift, with players all over the place? Who is tracking it? Seems like an impossibility to me.

      But anyway, the idea I guess is that everything eventually falls to the mean and “unlucky” play is corrected down the road, if the overall trend is on an upward trajectory.

      We don’t have to worry about that with our team, lol. This is a really bad team. For all intents and purposes, they need, in no particular order or priority:

      1) A sniper that can fill the net and take attention away from other teammates.
      2) Another elite forward after the sniper.
      3) A true top pair D man.
      4) 2 other quality D men (Skjei, who I have defended and preached patience for, appears to be a bust, based on expectations. Shatty is a fine 2nd pair complement or top pair D man if there is a strong partner. Claesson has been their best D man which in itself tells you the state of the D corps.).

      Other than that, they’re fine, lol.

      • Each arena has a team of people that track player location data for shots/goals/hits/etc. There are 5 individuals watching the game and recording these events, with a manager to ensure quality (to the best of their ability):

        Furthermore, stats websites such as Corsica and NaturalStatTrick all offer “adjusted” metrics of the raw numbers. This means arena bias is adjusted towards league average based on historical weights of how the NHL employees tend to track. For example, MSG is notoriously biased to record shots closer to the goal than they actually happened. So anytime I write my stats-based articles, I’m using adjusted data for the league in order to remove some of that bias.

        • Isn’t the concept of “adjusted” data something that contaminates the data itself? To pull out another quantum mechanics idea here, the act of observing an experiment or event influences the outcome of any experiment or event. If the raw data is being manipulated after it’s been compiled, then it’s not really an accurate picture—it’s more of a probable picture. That’s a pretty significant difference, and it makes the fancy stats crowd look like they’re exaggerating the correlative accuracy of their metrics in terms of the reality of any individual hockey game.

          • No, it’s just weighted averaging that improves the correlation to goals for % based on the large sample size of now 10+ seasons of shot location data. And it’s not just improving the data to look good – there are arenas where the bias tends to record shots from too far away, so the net effect is positive in that case. All in all, the sample size is large enough about 1/3’rd of the way to the season to start washing out the noise to become statistically significant. This is done across the board in all major sports with these statistics.

          • Sorry, I’m not convinced here. Data is data, to play with it is to tamper with its original information. And that goes for stats, mathematics, or science. Just because there’s an agreed-upon bias applied to the data by a large number of analysts, it doesn’t make the data more legitimate. In my eyes it’s less legitimate, because it’s being shaped by outside conditions being imposed upon it by analysts.

          • So did you think your grades were unfair in school because some assignments were worth more (weighted) than others?

          • I don’t think that’s an appropriate comparison. Weighting assignments doesn’t imply messing with data, it implies that some events are more important than others, which would imply that a statistical measurement like this is a probability-based one. That’s what I’m arguing here—XG% is pretty subjective, especially if the original data set is nudged in one direction or another.

        • Thank you Rob, well that puts more context on how the stats are arrived at. Must be a very difficult process.

      • One can track all of this, although there is occasional sloppiness. Player A makes a bad line change and his replacement is charged with being on the ice when the bad thing happens. But there are problems with everything. There is no proof AFAIK that it is legitimate to assign stats to individuals.

        Regression to the mean poses a major problem. Some characteristics can be measured more accurately than others. The implication (true) is that the less reliable stats will regress more than the less reliable ones. However, what this necessary undervaluing of the less reliable stats leads to is the undervaluing of the strengths measured by these stats.

        For a shooter, for example, both quantity of shots and % success matter. But because we can measure quantity more accurately, we overvalue the prolific weak shooter over the more accurate one who takes fewer shots.

      • Skjei possession stats are below his career norm, and he is predominantly in on d-zone starts which may help explain it. His skating and puck handling are good, he has size, his confidence IS NOT GOOD,IMO, he still has upside, I think. He’s far from a bust, I think he’ll snap out of it.

        • I want to believe that Brady can snap out of it and I would be more hopeful if he didn’t have a bad year last year, that I attributed to the coach.

          and when I said “bust” it was based on his contract.

      • You just described a lot of teams. We aren’t that bad, we aren’t as of yet that good. We have some TOP top top quality coming down the pike in the next year or two (Miller and Kravtsov) and our D’ isn’t quite as bad as you suggest (I mean have you even watched ADA play?). Stay the course and move a couple of forwards (Zucc and Names) and one or two of Smith, Staal and McQuaid — although I think Staal is staying.

        • Oh, I want to move them all, Hayes, Zuc, Names, Smith, and McQuaid. And I agree that Staal is not only staying but now may not be bought out at all on top of it.

          I will hopefully feel better after the trade deadline passes and hopefully the GM does the right thing.

    • A lot to unpack, so I’ll try to keep it concise:
      1) Expected goals are only a “hypothetical statistic” in the fact that an expected goal event rarely equals 1, otherwise known as an absolute chance at a goal. This makes intuitive sense, because even if you’ve deke’d the goalie and have a wide open net, things can still go wrong and have gone wrong in the past from that location on the ice. The Evolving-Wild twins do a cool thing every week on Twitter where they take the highest xG events of the week and put video to them. If you watch the videos, you’ll get a sense that basically the higher danger the chance is, the higher the xG is, which is the whole idea of expected goals:

      2) Randomness is taken into account in expected goal models, as shot types are classified by the NHL. To your point, rebounds and deflections often have higher xG counts because they’re very hard to stop. Also, every location in the zone has a historical shot %, so that is applied as well.

      3) Just a thought to your “should’ve/would’ve” analysis: you’re right to an extent. If the Rangers win the xG battle 3-2 but lose the game 2-3, it is fair to say (to an extent) that they should’ve won the game and luck/randomness had not gotten in the way, they would’ve. That being said, the reason Corsi and Expected Goals are used in this manner is because over the course of 82 games and the playoffs the actual goal outputs regress towards both the shot share and expected goal outputs (regression can be positive or negative, of course). A recent example of this is the Sabres, wherein during their 10 game win streak they had an expected goals % of 44.8 (all situations), aka not good. A big part of why they ended up winning those games, though, is that they shot 9% (above average) and got 0.925 save% goaltending (top 10 level). Since then, they’ve lost 4 straight, with an xGF% of 38.7, a save % of 0.929, but a shot % of only 8. It doesn’t seem like a major drop, but the 1% in shooting puts them closer to 2 goals a game rather than 3, which ends up in more losses than wins.

      • Thanks for the thorough explanation, Rob. Much appreciated. I still ultimately disagree with most folks on the inherent value of the XG stats.

        To me, this stat is like many fancy stats in that it’s offered up as set-in-stone statistical evidence by many writers and fans, when it is something that should be viewed as a probability metric only. And as such, I don’t see its value as much as Corsi/Fenwick and the other first-generation fancy stats. To each their own though I guess.

      • Suppose you have a 2 on 1 break. Is there really anything in xGA or xGF that takes into account how good the three players involved are? A good defenseman tries to give his goalie a predictable shot, but what is the difference between a predictable shot and one the goalie cannot anticipate.

        • To my knowledge no NHL Expected Goals model takes into account which players are involved, and while this of course matters in hindsight as to whether the puck went in or not, it would negatively influence the purpose of the model. Regardless of who is shooting the puck for the scoring chance, the expected goal level is based on all of the history of that location of the chance – both good and bad (meaning good and bad players). Over time, better players accumulate more expected goals for because they continuously get more quantity and quality scoring chances.

          The difference between a predictable shot and one the goalie cannot anticipate is the nature of the game, or put another way, randomness and luck.

          • “The difference between a predictable shot and one the goalie cannot anticipate is the nature of the game, or put another way, randomness and luck.”

            You could not be more wrong. Two on ones are easy to understand, but all hockey is like this to a certain extent. The defenseman is basically helpless to prevent a shot. Playing perfectly though, he thwarts the pass so the goalie knows who will shoot. He also hinders the shot a little so that the goalie has a better chance. And this is about skill.

            So here is the problem with xGA. A two on one will be assessed the same way whether Vlasic or Yandle is defending (a superficially great shot will happen). The xGA charged to Yandle and to Vlasic will be the same. So this play at least tells us that Yandle and Vlasic are equal. And yet, it is not an accident that the puck ends up in the back of the net when Yandle is defending and doesn’t when Vlasic is.

            Another point is that while a breakaway is a breakaway, a good defender is far more concerned about allowing a Crosby breakaway than a Tanner Glass breakaway. More generally, one might allow a really good shot to an Oveckin teammate rather than a good shot to Ovechkin.


            All of these stats are predictive, but they are also defective. What you are saying is that everything you can’t measure is luck. Yes, there absolutely is luck in hockey. The third goal in Philadelphia scored on hank would have been prevented by most goalies and by Hank on most days. The rebound bounced into a defender and into the net. Nine times out of ten the rebound goes somewhere else. However, the expected outcome of an opportunity depends on many things that are tough to measure. By design, some players are “lucky” and some are not.

            xGF, xGA are attempts to remedy the weaknesses of Corsi and I do not know how good they are, though I know they are not as good as their adherents think. However, I will give a thorough explanation of the weakness in Corsi – if you are willing to listen (read).

            Goal scoring depends on Corsi and PDO. My analysis says 70% PDO, 30% Corsi; I know of no one else who has looked at it. A hockey game produces ten times the Corsi data as PDO data and so it seems reasonable to believe that we can determine statistically how good a player is Corsi-wise, but even say 20 games gives incredibly unreliable PDO numbers. If you simply want to predict which teams will win, relying on Corsi tells you something because it is real while the PDO numbers seem worthless. HOWEVER, it is a fact that there are good PDO players and bad PDO players. The greats excel in both categories. However, there are good Corsi bad PDO guys and bad Corsi good PDO guys. Choosing between them depends on how good is good and how bad is bad. BUT fancy stats will tell you that the good Corsi guy is better when in fact, all things being equal, the reverse is the case.


    • Interesting point, I recently heard Vally on NHL Radio on XM Radio talk about this topic, Vally stated the Corsi will go away and the “Scoring chances” will be the stat replacing CORSI and SOG.

  • Forgot to offer congratulations to Ryan Reaves for taking Tom Wilson to the woodshed last night too. Karma owes Wilson several more visits, I sincerely hope they are not long in arriving.

    • I saw the hit. A penalty yes, but why an ejection? No elbow or hit to the head. Blindside yes.

      Wilson is a guy that I would gladly take a 5 game suspension for.

  • I think we need to break out offensive side and defensive side of the pairs. We just do not have great defensive players. There is also the “stupidity factor” which shows up on no chart. It it when a defenseman makes a bonehead play (see Skjei on the penalty call) or takes a bad penalty (Smith many times). These tend to all to turning points of the game.

    • We also need to look at PP stats vs. 5 on 5 situations. Pionk drives scoring on the PP, but is a negative contributor 5 on 5. In my opinion, people focus on his PP points and ignore his defensive lapses.

  • All I had to do was look at the chart and see Howden, Strome and Vesey say dull and stop reading after that.

    • They’re in the bad category? Vesey has no history of driving play (he doesn’t score enough for it to be acceptable) and Howden, despite a decent start, hasn’t played great as of late. Strome has been just as bad as Spooner thus far as well.

      • Rob L; I have no idea what you are watching. Last 3 games yes Howden has slipped however Strome has been a great trade. I have watched him closely and he is 95% of the time in proper position. He is far better than Spooner and I will boldly say a great trade. Howden is your future Captain

  • Actually, my reading of the charts are that the Staal-Pionk pair has been pretty good. Yes, they are only average defensively and a tad below average offensively, but they are the guys opposing Ovechkin, McKinnon, McDavid, et al. Going up against the big boys and coming pretty close to holding their own. Unfortunately, I presume these are all 5 v 5 numbers and they have struggled on the PK.

    The other defensive pairs have gotten much easier assignments.

  • Hockey has five skaters on the ice, all of whom have roles in what happens offensively and defensively. Separating out defensive pairings from the forward lines that are on the ice with them as if the defensemen are the only players responsible for goals against for example seems to me to be intrinsically invalid. Furthermore, after 28 games, you are still talking about a small sample size. You also don’t capture the quality of the opponent.

    I am not sure that one can put too much stock in the stats and draw firm conclusions from them. Unfortunately, articles that use them usually present the results without what I consider to be necessary caveats.

  • I read the graph and see that Fast could turn a positive group into some sort of negative unless he plays with Howden and Vessy. Though Howden, Vessey and Strome looks like they play better together though the stats don’t demonstrate that. I don’t think they have enough time on ice together to make a call.
    In the quadrants is positive negative or negative positive good? Depends what you are good at and bad at.

    • This graph reinforces that Fast drives down the positives of a good line. In each occasion you can remove Fast and have a better line. Let the math envy people throw the down the down thumb.

  • Don’t think vesey howden and strome was as good as vesey howden fast because fast has more speed then strome getting in on forecheck. And that’s what made that line effective.

  • I watched the Leafs and the Sabres last night. It was an exciting hockey game and both teams have some tremendously skilled young players. It can happen in New York too and I am hopeful that Gorton and the Rangers pull it off. I believe them to be on the right track.

    The stats we are looking at include a lot of young players who are still learning the game at the NHL level. I would remind everyone to keep that in mind, and most of all, to enjoy the games!

  • Back to top button