How to build a contender – Part 2: the Impact of Aging

Photo Credit: Bill Streicher/USA TODAY Sports

Some of the most important work in the field of hockey analytics has focused on how player performance changes over time.  

In a salary capped league teams need to pay attention to, and understand how, the aging process impacts future performance, since these factors can and should influence contract decisions. A bad bet on a player whose performance is declining can cripple a team’s cap structure for years to come. 

In part two of this series, designed to take an objective look at how contending teams are fashioned (read Part 1 here), we’ll dig a bit deeper into the aging process and see what successfully constructed Stanley Cup winning machines can tell us about how contending teams navigate the age-old conflict with father time.

Advertisement - Continue Reading Below

Some of the most influential work in this area was performed by Eric Tulsky, who looked at aging from a number of different angles, including the impact on player’s shooting percentage over time, and Gabe Desjardins who looked at points per game

Both analysts found that players tend to peak from ages 24-26 in terms of the specific statistics they researched (shooting % and points per game), but I thought it would be worthwhile to see how overall player performance, as measured by goals-against-replacement, changed over time for forwards, defensemen, and goalies. A summary of the methodogy I’ve borrowed from Eric Tulsky can be found here


Advertisement - Continue Reading Below


As we might expect, goals against replacement shows a trend consistent with other prior research. On average, forwards are in their prime between 22 and 26. Of course, we all remember the exceptions like Ray Whitney and Martin St. Louis, but players like this are just that – exceptions. 

To apply this concept, I thought it would be worthwhile looking at an example of a recent signing from this summer – the extension of Brandon Sutter for a 5 year contract at an AAV of $4.375M/year. 

Sutter just finished his 25 year-old season, and over the past four years his GAR has ranged between -0.55 (2013-14) and 5.5 (2014-15), so I decided to look at all forward who scored in that range in their 24 and 25 year-old season at look at their average GAR over their 26 to 30 year old seasons to get a sense as to what to expect from Sutter throughout the term of his contract. 

b sutter

While we can see that there are definitely examples of players who performed very well in this peer group, specifically Filppula, Bergenheim, and Nielsen, the majority of the players in this group saw their performance decline significantly, and only 14 of the 26 players played all five NHL seasons between the ages of 26-30. 

Advertisement - Continue Reading Below

Of course, Sutter could buck the odds, like Valtteri Filppula, but chances are what we’ve seen from him is what we should expect going forward, which is a player who GAR ranges between -1 to +4, before his games starts to deteriorate with age. Expecting him to be a foundational part of the team going forward is a bit of a stretch, but there’s nothing wrong with this level of contribution in and of itself over the course of the next 5 season. That said, there is a valid question with respect to whether this is an efficient use of salary cap, but that’s a blog for another day. 

On the other end of the spectrum, a definite bright spot for the Canucks is Bo Horvat, who had a GAR rating of 0.23 in his rookie year last year. Because of his age, and the fact that the GAR ratings only go back to 2005-06, we have a relatively short list of comparable players for him, but that list includes the following players (GAR of -1 to 1 in 19 year-old rookie year): 




This graph surprised me somewhat, in that I had expected to see defensemen develop a bit later than forwards. In fact what we’re seeing is very similar, in that there appears to be a definite peak, on average, at age 24.

Advertisement - Continue Reading Below

The most significant recent Canucks defense signing, at least in financial terms, was the 3 year $3.6M AAV extension given to Luca Sbisa, so I thought I’d use that as a test drive for GAR. Sbisa is an interesting case because he’s had a negative GAR for each of his 7 NHL seasons, meaning that he’s never been above the level of a replacement level player over the course of his entire NHL career. As there were no other defensemen that met this criteria – most who had trended in this direction had their NHL dreams halt abruptly – I used defensemen with a negative GAR in their 22, 23, and 24 year seasons to develop Sbisa’s peer group: 

l sbisa

Of the 6 players who fit the profile, four went on to play the next three seasons (i.e. the Sbisa contract period), whereas two did not play in the NHL, having seen their dreams end after their 24 year-old season. The upside for Sbisa is that he turns around his negative GAR trend, like Matt Greene who developed into a reasonably useful defenseman in his age 25-27 seasons. 

On the other end of the spectrum, while his game is quiet and unassuming, Chris Tanev is establishing himself as one of the better young top-four defensemen in hockey. Like most advanced stats, GAR is a big Chris Tanev fan, crediting him with a GAR of +9.7 last season. The Canucks smartly signed Tanev for 5 years at an AAV of $4.45M, which could prove to be a bargain if he continues on his current trajectory. 

Here are his closest peers based on his GAR in his 23 and 24 year old seasons (GAR +5 to +15): 

Advertisement - Continue Reading Below




There are a couple of minor differences in the goalie graph as compared to the two graphs for skaters.

Advertisement - Continue Reading Below

Goalies rarely win starting jobs in their early twenties, so we see a bit of year-to-year variability in their GAR ratings during their 20’s, which is magnified a bit by the lower sample size of goalies in general. Still, the overall trend is consistent, with their prime appearing to extend until age 29 when the impact of aging really start to set in. We see a bit of an anomaly for age 36, where Tim Thomas’ 45 GAR 2010-11 season single handedly brought up the average for that entire age group. 

The Canucks biggest investment in goal is of course Ryan Miller, whose 3 year, $6M AAV contract signed last summer has him signed as a Canuck through to his 36 year-old season. I decided to look at how his peers performed in their 35 and 36 year-old seasons. 

We all recognize that Ryan Miller is no longer the Vezina calibre goalie he was in his 20s. That said, he’s been reasonably effective over the past five years, averaging a GAR of 8.6 over this period, although last year represented the low water mark of his career with a GAR of -3.12. 

So I decided to look at goalies with a GAR between -4 and 15 in their 33 and 34 year-old seasons as Miller’s peer group: 


Advertisement - Continue Reading Below

It’s not a flattering group of cohorts for Miller… 

Five of the six players in Miller’s peer group completed their 35 and 36 year-old seasons in the NHL, although only Nikolai Khabibulin managed to maintain a positive average GAR over this two-year period. Coming off of a significant injury last season, it may be a tall order to expect Miller to bounce back to his pre-2014-15 GAR levels, and with the predictable impacts of aging at play, the onus is on Jacob Markstrom to not only establish himself at the NHL level, but perhaps to take on a larger share of the workload. 

Markstrom has yet to show himself to be a positive GAR goalie at the NHL level, but that said he closest comparables (-15 to 0 GAR in 23 and 24 year-old seasons) include Justin Peters, Brian Elliot, Darcy Kuemper, and Ben Bishop, so there’s some room for optimism.


On one side of the coin, Horvat, Tanev, and to a lesser extent Markstrom provide some degree of optimism – maybe the Canucks at least have some vague semblance of a core to build around. 

Advertisement - Continue Reading Below

On the other side is a grimmer portrait, that of an aging team with too many players-in-decline signed to inefficient contracts. While we obviously hope that Sutter, Sbisa, and Miller turn out to be the exceptions rather than the rule, these recent signings and extensions underscore a concerning pattern that this management team may not be bullish enough in their consideration of the impact that aging has on player performance. 

In the Canucks most significant recent signings at each position, the Canucks have allocated $12.9M, or 18% of the 2015-16 cap to three players that we can expect will have very little positive contribution to the club’s overall team GAR. 

In my next post, I’ll explore management of the salary cap, in the context of goals-above replacement. 

Others in this Series


  • Locust

    “Some of the most important work in the field hockey analytics”

    I assume you mean, field ‘of’ hockey analytics…

    The way you wrote the first line made me think this blog was about [people] running around with curved sticks.

    Which might have been a refreshing change over discussing the cluster*$#@ of an off-season the Canucks have had.

  • Mantastic

    An interesting post in the focus on Sbisa and Sutter in particular; we already know the effect of aging superstars or higher-end players (Sedins,Miller) but it really is a question about the direction of the team to place so much faith (and money) into marginal complementary players. It’s far too early to write off Benning’s approach as some have done but I do worry about the “Boston Model” of stacking the team with high character or gritty players at overpays while sacrificing talented difficult ones.

    I do like the approach of providing some kind of support system for the young talent to develop within and potentially even the term, just not sure about the amount of money spent doing it. In that sense Sutter is a better bet than Sbisa who we could have replaced far better with Corrado and Clendenning.

    • Mantastic

      One factor that could disrupt the trends suggested by GAR is the impact of coaching, or more specifically, the impact of a teaching coach, and the Canucks would appear to have one in Desjardina, given his work history.

      Mind you, so far he has been doing more cutting than teaching (Clendening, Bonino, Kassian) but maybe his approach can change the GAR trajectories for Sbisa and Sutter from how they are illustrated in this article.

  • Mantastic

    To be honest, after a summer of reading this guys (and most of the other writer’s) arrogant and one-sided writing, the last thing I wanna do is read how this guy would build a team.

    I can tell by the lack of comments from these articles, most CA readers agree.

  • Mantastic

    As much as disagree with a lot of the viewpoints being put forward by the writers of this blog these days, I am very interested in how one would go about building a team using advanced analytics vs. the old school eye tests. I’m not sure I agree with this method as I think advanced stats need to be used in conjunction with good ole hockey IQ, but I am definitely interested in this approach. I think we should let this mini-series unfold and allow moneypuck to express all of his talking points and ideas before we pass judgement.

  • Steampuck

    I want to complicate this. It seems we can create fairly sophisticated cohort analysis for incoming prospects, but we revert to a base average for aging veterans. I accept the basic tenet that older players are unlikely to rediscover their youth and productivity can follow a fairly steady decline, post-30. But I want more. To simply acknowledge that Ray Whitney or Martin St. Louis is an exception rather than the rule does a disservice to advanced analytics. Surely, there are ways of measuring players and their metrics against players of similar age, productivity, height, fitness, etc. to better gauge the effects of aging. PCS for geriatrics? Couldn’t we get better at predicting which players will age more gracefully?

    Of course, this is setting aside all the character components and “value in the dressing room” leadership stuff. But there has to be something about the fact that the Sedins are likely going to record the best time up the Grouse Grind (again) in a couple of weeks that factors as heavily as them being elite players. Just as a wayward example.

    I appreciate that this GAR data doesn’t go too far back, but I also want to insert some complexity around the gargantuan steps forward we’ve witnessed in sports science, which should (in theory) enable greater longevity and ease the decline in the latter portion of an elite career.

    As I think about it, running the Sedins through this kind of analysis would be kind of interesting. Points down. But they remain in the top 20 in league scoring, so everyone is down? PDO: Steady decline?

  • Mantastic

    This is an interesting article on GAR (and one that I enjoyed) but this series is designed to “take an objective look at how contending teams are fashioned”. I can’t help but wonder how many articles into the series we will get before Moneypuck actually gets to the meat of his premise and, one would presume, discuss what contending teams have done to build their teams. Both Moneypuck and the series would be well served if the series were more focused and each post tied directly into the primary premise.

    All in all I have enjoyed the series so far.

  • Locust

    This is a very thoughtful article.

    First off – to @Jo mama – way to pick out a typo! I’m sure your 8th grade English teacher is proud of you.

    The r^2 number is near 1 for each of the measures, which does indeed indicate a well-fitted model… and it doesn’t appear to suffer from over-fitting (example being Tim Thomas’ weird season didn’t skew the model significantly)

    The point @steampuck makes about outliers – indeed! There are always going to be outliers to every model. This model is broadly predictive, but it will never be perfect. The Sedins, indeed, are freaks of nature and have incredible fitness levels. However, the VAST majority of players, be they stars or otherwise, don’t… and the vast majority of players do indeed follow this curve (case in point: Ryan Kesler, who started noticably declining 2 years ago…) Just because there’s an exception to a theory doesn’t invalidate the theory. It just means that there’s an outlier.

    Last point: how depressing is it that one of Sutter’s peers is Taylor Pyatt?

    • Steampuck

      Right, but you can measure for fitness. I seem to recall that McCann had off-the-charts VO2 max numbers. Surely a good indication of long-term fitness? Many sports–cycling, for example–can pre-determine an athlete’s potential based on a series of such tests. I would imagine you could do similar analyses to predict longevity.

      Surely, the success of any metric is based not just on what the overall numbers say, but also when to go all in on the exceptions. The data above says to sell on a 30-year-old Niklas Lidstrom. But that’s the kind of decision that would get a GM fired in retrospect. I’m not disagreeing with the data, but looking for ways to enrich it. How to better determine which aging players to keep versus which ones to jettison. Otherwise, the rule is can everyone over 30 and don’t sign Radim Vrbata.

  • Steampuck

    All of this is useful and interesting, but a) doesn’t account for outliers (of which there seem to be many, particularly is you are Swedish) b) doesn’t speak to the right mix of veterans and youth (a team full of 24 year olds? I think not) and c) treats human beings like robots rather than humans.

    For instance, as was raised by another poster in a different thread, Miller was adapting to new system and a new goalie coach, was expecting his first child and suffered a significant injury last season. As such, isn’t it quite possible that he could perform better this season despite being a whole year older? That seems plausible to me based on those “unmeasured” factors. Just like it was plausible to think the Sedins would bounce back from the Coach Tortorella experiment.

    I am not arguing against evidence-based analysis. I am arguing that drawing aggregate data down to the level of an individual player has its challenges.

  • Locust

    @jamie E

    Models don’t account for outliers. That’s why they are outliers: they exist outside of statistical models. Put down the Gladwell and spend 10,000 hours reading a textbook.

    • Locust

      Despite your attempt at withering sarcasm (somewhat successful) and sounding like a pompous, insufferable boob (largely successful), I will attempt a polite reply.

      My point on “outliers” (I haven’t read the Gladwell book) is that they aren’t really outliers if there are lot’s and lot’s of examples at hand. It’s not that uncommon for elite level NHL forward talent to remain elite level well into their 30’s. Same with D-men. Same with goalies.

      So using aggregate data to claim that forwards peak in their early to mid-20’s may make sense for “average” players, but doesn’t for higher end talent.

      That was my point, poorly stated I admit.

  • Locust

    Sooooo…. athletes are in top form and are better when they’re younger as apposed to being an old cougar? Gee, who could have known? I was waiting for players to turn 40 before they hit their peak. I wonder if Eric Tulsky can tell me if my fridge can get cold?

  • Locust

    @steampuck point taken. But: these models are limited to publicly available data. I would imagine if there were data about individual fitness etc it would be enveloped in the model.

  • Steampuck

    Kind of off-topic, but did anyone else notice and chuckle about Ehrhoff’s GAR? He had two incredible seasons by this metric, can anyone guess which team he had them with?

    I honestly think that not signing Ehrhoff at the end of the 2011 season was the biggest mistake Gillis made. He made the defense so dynamic, and it’s the kind of space that he can make through his absolutely excellent skating that this team seemed and seems to miss dearly.

  • Double Dees

    This site is sooooooo lame.

    The reason is that no one cares about the Canucks.

    The truth right there.

    Nobody cares about an aging, boring team anymore.

    • Double Dees

      But the PR guys here want everyone to like their aging boring team. And if we say otherwise they’ll say we’re trolls. They want a free cake and they want someone else to pay for it too! lol

      • Double Dees

        Danja, ROM Spacenight, and Double Dees are the same lame person. Actions of a Schizophrenic or at best, someone who has a schizoid personality disorder.

        He will deny it, but that further proves my theory.

        Help can be had, Son…just reach out…

        • Double Dees

          I won’t deny it. I’m every one and everything you say I’am. I believe you just because you said so. I also believe the Canucks have a future and are true contenders…..BAHAHAHAHAHAHAHAHAHAHAHAAAA!

          I really tried to say that with a straight face but I can’t.

          You can’t hear or see me but I’m laughing real hard at you right now.

          The Canucks and their PR fans….WINNING? HATE IT!

  • Steampuck

    I really liked this article. I think this a great way to view the data. Using a value like GAR is a streit forward way to assess player aging.

    I don’t read a lot of analytic blogs to great detail, but this one was particularly interesting… however, I had a couple comments:

    I had to go outside of this article to look up GAR. A quick definition would have been handy so I could have read on. It is easy to understand, so whatever if you don’t define it next time.

    The larger question I had was on the subjects in the analysis. I am assuming in the ‘GAR aging curve’ figures, the points were mean GAR per age of player taken across many players for many years. That’s a hefty sample size. What years did you use to find the mean? I feel like with this analysis you wouldn’t want to go back too far. Also, what was the error for each point? Your r^2 value is crazy tight, but I would like to see the error to determine if in fact those outliers (good or bad GAR) had an effect on the possible range of values as players aged. May indicate what kind of gamble teams are really making.

    Lastly, did you throw out some outliers? I would understand if you did, however without telling us, the results to me are a little suspect.

    Oh, and sorry to be a douche but could you label your axis please.

  • Double Dees

    The fact u guys speak more about us “trolls” tells the true side of the story.

    There’s nothing to talk about re: the Canucks.

    I’m just trying to wake u fools up!!!

    This franchise has NO direction!!! None.

    If I’m mental like some of u say, then what are u?