Parallel Paths


Objects on a parallel course give us an interesting example of unification and divergence. On the one hand, they are travelling along the same path, moving towards a common destination, and they never diverge from that path. On the other hand, they never converge; that is to say that despite all the similarities of their path they remain as far apart from each other at the end of their course as they did at the beginning.

What on Earth am I talking about here?

The funny thing about advanced statistics is that the ones I find most useful – the ones I consult frequently when evaluating players – are representative of items that any qualified, diligent observer would pick up just from watching a lot of games.

Take Quality of Competition and Quality of Teammates. There probably isn’t a fan of the game that fails to understand the relative value of playing with Ales Hemsky versus the relative value of playing with Steve MacIntyre. It’s an axiomatic truth: line-mates impact performance. Similarly, down through hockey history there has always been a certain cachet that comes with shadowing top players. Don Cherry preached it as a coach and it goes back well before his day.

What QualComp and QualTeam do is put a number to these values. This is important, for a couple of different reasons. The first is that the average fan isn’t especially good at picking out who plays top opposition, all by themselves. For years now – certainly the entirety of the post-lockout era, probably longer – power vs. power (where the top players face each other) has been the preferred matchup for most NHL coaches. Despite this, fans have persisted in identifying third-line as the guys facing the stars. Sometimes it’s true, especially in the past – Todd Marchant held the role for a time in Edmonton, while Anaheim and New Jersey have both had extended periods where they used a checking line – but as a rule it doesn’t happen. Carefully watching matchups is one way of doing it, and in my experience the matchups rather closely mirror QualComp.

The second reason is that even if a fan is careful about watching how a coach runs his lines, it is all but impossible to watch enough games to properly evaluate all 30 teams, particularly as injuries start accumulate and lineups get jigged around.

QualComp is a help in both areas – it serves as a sanity check for personal observations on a team one follows closely, and it provides data for teams that one person simply doesn’t have time to follow in-depth. It shouldn’t be regarded as something alien: it simply condenses a task that any competent observer should be doing in the first place.

Zone Starts are a similar statistic. A competent observer who watches a lot of games can generally tell you which players are relied on for shifts in their own end, and which players get used in a lot of offensive situations. He knows that these things matter – players on the ice in their own end a lot not only have a higher chance of getting scored on (especially off a lost faceoff) but to get a scoring chance of their own they must travel 200 feet, penetrating increasingly sophisticated defensive schemes, all the while knowing that every deke and every pass could be picked off and lead to a chance against. A player having a perfect shift might spend 30 seconds chasing the puck in his own end and 10 seconds getting the puck up ice, but be forced to dump it in to get the line change. He just played a perfect shift but never got an opportunity to score.

The reverse is true as well. A player starting in the offensive zone has a higher chance of scoring a goal (especially following a successful faceoff), and even if his team loses possession he now has 200 feet to reclaim the puck, and any mistake by the opposition can lead to a scoring chance. Off a won faceoff, this player could spend 30 seconds moving the puck around the offensive zone but never threatening, chase it up ice for 10 seconds, and then get the change while the other team is changing – his line played a miserable shift, but they got away with it because the centre won an offensive zone faceoff.

These contextual statistics matter a lot when it comes to evaluating players. Coaches know them without needing to glance at numbers – they’re the ones deciding when and where to use these players. Competent observers that watch a lot of games involving one team know them too, and appreciate the difficulty or ease of the role each player plays – and they probably have a very good idea for teams in the same division and a good idea of teams in the same conference. A casual fan, the guy who watched 15 games this year and was sober for the first period (special exemption for fans of the Oilers/Leafs, where the games encourage drunkenness as a lifestyle choice), probably doesn’t have a clue.

Corsi and Fenwick are a little more esoteric, but they do fit with something coaches have been doing for decades: collecting scoring chances. Even coaching staffs that don’t physically count scoring chances (and these days, I suspect those are rare indeed) consciously evaluate a player’s two-way game. In a battle between Corsi and scoring chances, I’ll always defer to scoring chances, but the fact is we simply don’t have scoring chance data for every team – and both Corsi (shots, missed shots and blocked shots for minus the same against) and Fenwick (shots and missed shots for minus the same against) closely mirror scoring chances – something that’s been shown time and again.

All those fancy numbers really boil down to a very simple concept, something coaches have been doing since the game turned professional – evaluating a player’s two-way game and noting context while doing so.

Getting back to the introduction, I find a lot of the ‘advanced stats vs. watching the game’ style discussions to be unhelpful and rather pointless. If one watches the game closely, think there’s more to playing hockey than scoring goals and adding assists, and believe that how a coach uses a player has a big impact on how successful he is, they won’t disagree with the statistics on much – at most, it will probably be a stylistic argument, on the relative value of hitting and fighting in today’s game, or an argument of degree (‘sure X played tough minutes, but he still ought to have produced more’), or an argument about specific situations (‘Y is a good player generally, but can’t elevate his game when it counts’).

Are those important distinctions? Yes. However, in the main, proponents of both schools are looking for the same thing.

  • Quicksilver ballet

    At first i found it difficult to grasp the message behind this article. After skimming it a second time i decided to whip out my trusty Sublimanator 5000. After entering the data, it appears as though the author of this article is extremely excited that Ryan Smyth has been returned to his rightful place. It brings us all great comfort that he reminds us of this fact in this article.

  • In the future I forsee the value of statistics in the evaluation of hockey teams and players will be increasingly recognized by the mainstream fan. However, it will require a visual and graphic basis so it can be easily grasped by the intuition.

    This is exactly what Hans Rosling does in this 20 minute video. The data is not relevant to hockey, but the method of presentation certainly will be. Book it:

  • ChiliChunk

    Excellent article JW.

    One thing that has always bugged me is when people use QualComp to qualify a player’s offensive output. It seems to me that it would be much more relevant to the their defensive ‘production’.

    If I had a choice to play against either the Kovalchuk triplets or the Bolland triplets my QualComp would be much higher against the former but I bet my point totals would be much higher too. Am I out to lunch on this?

  • positivebrontefan

    Uh…WOW…I feel really inadequate right now…or stupid. I kind of understand what you so marvelously put together there Jon but I have no idea how to traslate that into knowing who to put on the ice at a given time…

    *picks knuckles off of ground and looks around self consciously*

    Good article Jon, The brighter ones on this blog will understand what you just showed us… that does not include me. I’m not near smart enough.

    Ps. your advanced stats Prof must be so proud.

  • shau_co

    Willis: lets assume you are right when you assert : “For years now – certainly the entirety of the post-lockout era, probably longer – power vs. power (where the top players face each other) has been the preferred matchup for most NHL coaches”. This gets me to my biggest problem with the groupthink in the current NHL – the dreaded “top 6, bottom 6” (what rubbish) If you are right that the top lines square off against the top lines, wouldn’t it be a differentiator then to have skill in the bottom 6, rather than 2 lines of garbage, roll players who goon it up, slow vets, checkers, fighters, enforcers, etc that don’t create much in terms of scoring chances (and according to you, don’t play against other teams skilled players). Right now you have “bottom 6 vs bottom 6”, or garbage vs. garbage. Especially since every other team has this “bottom 6” mentality: that in my mind will be a game changer for the first teams that figure that you can beat the groupthink. You get a bunch of relative fast skill guys and shooters who skate around the other teams “bottom-6” grinders, and it will take a few years for the other teams to adjust: makes total sense to me.

  • John Chambers

    One last comment (for those few of you who care):

    Sabermetrics are an interesting tool that helped to shape new viewpoints on the value of players in baseball. Basically, instead of measuring a player by Homeruns and RBI’s, Sabermetrics analyzed players to find value in those who didn’t make outs.

    This type of analysis is easier to measure in baseball because the impacting variables are few, eg. L vs R handed pitchers, runners on-base, etc.

    In measurable performance situations in hockey are dynamic and to have any meaning require complex vectored regression.

    My original point is that for the most part, the conclusions drawn from this complex analysis are more or less the same ones drawn from qualitative observation with a few exceptions.

    It’s all fascinating for those of us who like numbers, and all boring as hell for those of us who don’t. It SHOULD be part of the essential kit bag of NHL exectives, and probably serves to reason why Zenon Konopka got a low-paying 1-year deal contrary to how much Gregor loves him.

  • OB1 Team Yakopov - F.S.T.N.F

    Hey JW, maybe you can give us(me) a quick coles notes on Qualcomp.

    Take Vancouver and the Ducks for example:

    Vans centers and qualcomp are:

    Sedin .55

    Malhotra .46

    Kesler .28

    Laps .007

    Ducks are:

    Getzlaf .034

    Koivu .034

    Marchant -.094

    Chipchura -.148

    Now I’m assuming the higher the number, the higher the caliber of competition you typically face (and the lower the number the lower the caliber) with anything above 0 being above average competition and anything below 0 being below average competition. Is that correct?

    If that is correct, it appears that all Van centers face much harder compition then all the Ducks centers?

    Or is the info irrelavant accross teams? ie simply shows us that Getzlaf sees tougher competition then Marchant…. but tells us nothing about Getzlafs compitition vs Sedins?

    • SumOil

      Qual Comp is a flawed statistic. IMo a better way to determine QOC is to use the rel corse qual comp. That means that the average relative corsi of the opponents a player faced. However this takes in the assumption that best opposition players on the team will have better rel corsi. which too can sometimes be debated.
      I find this to be more accurate measure than simple Qual comp which is basically the average of differential of on ice/off ice plus minus of the competition

  • There is definitely a “which came first, the chicken or the egg” side to advanced stats.

    A player is going to have a low Qualcomp or CORSI REL QOC if the coach decides to shelter him from elite competition.

    Last year Sutter sheltered Babchuk in 5v5 situations, thereby giving him one of the lowest CORSI REL QOC ratings in the league. Does that mean that Babchuk sucks? Or does it mean that his coach just chose not to put him out versus elite competition?

    The same can be said of zone starts. Perhaps Sutter gave Babchuk one of the highest offensive zone start ratings in the league (21st overall, among players with > 40 games played) because he gave the Flames the best chance to score? You could also choose to interpret this as Sutter having no confidence in Babchuk in his own end. Perhaps it was a combination. It is difficult to say without actually observing Babchuck play and seeing what patterns emerge.

    My point: It is difficult to use stats in a vacuum. They can often tell you the “what”, but can rarely tell you the “why”. This is not to say that advanced stats are useless, but it does mean that we need to be careful when interpreting them.

  • OB1 Team Yakopov - F.S.T.N.F

    Also, what metrics is used to grade the actual quality of each member of the “competition” or quality of teamate?

    We all know you’ve got a better linemate if you are playing with Hemsky then Stortini, but how is it quantified?

    (ie what makes Hemsky a .5 and Stortini a -.5???)

  • shau_co

    Thanks JW. I also, appreciate your articles, although I am at times skeptical of results that do not include enough of the context because they are limited by insufficient data.

    Overall, I agree with the use of advanced stats to supplement decisions made based on intuition. I think they have to be used tightly together though. You can run into trouble if you are too extreme on either side.

    For example, there are intangibles that are not measured by stats today and affect a players value to a team and there are personal biases introduced on players that affect a players perceived value to a team.

  • shau_co

    It took me a lot of time to understand the advanced stats. Well my first introduction to them was on LT’s site when he used to dole out all the numbers in his player evaluation.
    This is around the 2007 summer sometime in August. Well I had only started watching hockey due to the playoff run(which i missed lol) and watched almost everygame of the 06-07 season. Based on my limited knowledge, I was confident of some predictions i had made regarding oilers/anahiem and others. Many of them came out to be wrong and I realized that there is something I am missing that I am unable to fathom by watching the game multiple times. And there has to be something more than the stats available at which would explain the game better.
    Anywho coming back..So all those stats at LT’s did not make any sense to me, but I am was sure that in order to understand the game and its flow, I need to grasp it. And it wasnt until you JW started that I really grasped the importance of the underlying stats.

    I think for me the defining moment was the ‘Amicus Brief’ in defense of Dustin Penner. My buddy and I were absolutely mad at the fan base/soach making penner the scapegoat when we thought he was helping the team create a lot of chances/goals and that brief did it for us. We were comverted to the stats side.

    IMO people who watch the game and think that they now know it all will never accept the advance stats as they are too stubborn to acknwledge that sometimes a guy not watching the game can know more about a player than themselves. Funny thing is that if a stat guy gets his prediction wrong, they will jump all over him. However if they get it wrong then bad luck. I want to know if anyone who called you an idiot for prediction regression in Brule’s has apologised to you yet.

    However these stats are still incomplete. We dont have time of posession and many other statistics. But these micro stats are a huge step in the right direction and hopefully in the future things will change.

  • A18

    Interesting piece, statistical analysis is great but needs to be accompanied the fundamentals to get a complete picture. I think you tried to do this in your article.

    I can understand the frustration vented by some here. The simple adage that less is more always prevails.

    I work in the finance industry and I can throw around convoluted stats and arguments all day at work but unless I can convey the information in a clear and concise manner the effort is in vain.

    One last note for all those who are defending stats, it’s all historically derived and does not correlate perfectly with what’s going to happen in the future. They are important but stats can be manipulated based on the dataused and averages are misleading. Plus it means nothing unless you can convey it to your general audience without confusing the hell out of them.

  • Another example of stats that are true, but can be interpreted in many different ways:

    The top 3 in offensive zone starts for 2010-2011 were:

    Henrik Sedin
    Cam Janssen
    Daniel Sedin

    I don’t imagine Cam Janssen got 74.4% of his starts in the offensive zone because of his offensive prowess. He was likely put on because he is considered a liability in the defensive zone.

    On the other hand, the Sedins likely got almost every offensive zone start that the Canucks received last year. This makes sense as they are obviously great scorers.

    Without observation (or a ton of other backing stats) we could make these stats prove just about any point that we wanted to make.

  • @ OB1:

    Gabe’s system works in two different ways.

    QC/QT are based on goals for/goals against. Over short samples it’s awful but over a long span (30 games is where I start trusting it) it’s a handy tool.

    Corsi QC/QT is based on Corsi ratings; I like it but use both because I think playing against high percentage shooters counts for something too.

    • SumOil

      Jones was a sinkhole when it came to scoring chances. And really why was he resigned?
      Apparently Smyth made his request after the season was over and Lombardi and ST have been talking since. So if the possibility of Smyth coming over esisted what was the need to overpay Jones?
      Now we will be paying 1.5 million a year to a 4th line penalty killer.