The evolution of a player is usually on a curve, each section of that curve is described in a multiplicity of ways. "Old School" hockey people will refer this to the rookie, sophomore, journeyman, veteran, & aging vet terms. In recent years, new terms have emerged with a focus towards "Prime Years" those years where a player is at their most productive and most valuable to a team.
Analytically, the curve shows three separate phases to any players' (excluding goaltenders) career. Please note that this is my interpretation of data, but needs to be defined for the sake of the post. Please note that while I use different sources (and whenever possible I will provide sources of information) the bulk of the data comes from two sources:
I would also highlight:
As a good companion piece.
I would note that I haven't had time to go through an exhaustive review of the foundational documents nor do I have the ability to "check the math". Some of the conclusions align with their own, some are mine. Any error in analysis or interpretation is my own, and as a result you can assume that it is guided by my own personal bias rather than a substantive review of their work...which would likely take 100's of hours I just don't have to spend.
Here's a brief outline of certain terms I will use and some of my own conclusions.
Phases of a players' career
I have identified three phases within a player's career - Development, Prime, and Decline.
1) Development - 15-21 - This is a phase in the player's career where they're still not fully physically mature, and still are generally refining their skills. While this does not prevent a player from being productive...it shows a time where the players generally don't have it all together. This of course includes pre-NHL years, and this is traditionally where the potential is realized. To quote Jerry Maguire - It's like corn in the pan; some pop...some don't. Statistically the "late bloomers" are not captured very well...but I couldn't find enough data to draw a significant conclusion.
2) Prime - 21-29 - This phase in a player's career where the mix of experience, physical maturity and confidence come into play. This is not to be confused with "peak" which is a period within a player's physical prime where they perform at their best. A player's prime does include both a certain amount of both development and "decline" but do generally highlight the most productive years in a player's career.
3) Decline - 29 onward - This phase of a player's career is defined by the diminishment of a player's skillset over time as their body begins to degrade. How quickly this occurs varies greatly from player to player. But in the end is an inevitable period within a player's career.
Peak - ~26 to ~28 - This is an amorphous period of time where a player where physical talents, experience, and confidence allows them to play at their best, and not see a significant variation year over year. Some players will not hit their peak until their mid-20's and see that peak last as little as a single year; while other players will reach their peak sooner and will be consistent for longer periods of time. While some studies have identified a specific "peak year" the graphs show a relatively flat line for anywhere from 3 to 5 years in a players' performance. When I refer to peak, I mean the amount of time where a player's performance will be roughly flat with limited improvements or decline, rather than a year which represents the average career peak performance.
Plateau - As players decline, there appears to be a period of time in the player's mid-30's where their decline slows, after which there is a more precipitous drop. This is especially notable for players who are elite at their respective position. In the case of this post, I will use plateau to highlight this timeframe in a player's career. Although may not specifically apply to an elite player.
Let's start with an overall view of NHL players' traditional career progression. What I would note that the above graphis is that this measures not performance, but rather "Average Change in Overall WAR". This indicates that a players' initial performance is significant to their career path. What this chart doesn't take into consideration is the age of entry and development in other leagues. Another issue is that the Journal of Quantitative Analysis focuses on Plus-Minus which we know to be a highly flawed figure...that being said, they do try to adjust for selection and data bias as much as possible. See below:
While the data is somewhat different, the curves are fairly similar. What is clear in both texts is that the overall skill level actually increases as players get older as marginal players retire earlier in their careers. This impact of this turnover can be seen in the second graph starting at the age of 35. This could explain the mid-30's plateau effect as marginal players retire and high skill players become a higher % of the overall formula.
What is interesting is that both studies indicates that if a player doesn't "fall off a cliff" in their early 30's that they are unlikely to have a precipitous decline...at least not a "natural decline" as things like injuries certainly play a factor which cannot be controlled for. Another item which is "clear" from this data is that defensemen generally decline more slowly than forwards, and tend to be better into their later years than forwards tend to be.
Another fairly easy conclusion to make it to assume that a player will be at their best statistical performances occur between the ages of 21 and 28. While their most productive years are most likely to be between 24 and 32. This also brings in an interesting question as to how a player's "peak year" being on average at 28, when WAR shows that the average player has already begun to see a statistical decline at the same time as their peak performance year. While the data is unclear, I would have to assume that the statistical decline is linked to player role. A player's role is likely to be more expansive starting at 25, which impacts their WAR impact, but the player continues to evolve with this more emphasized role. Essentially fewer "easy" minutes mask the continued progression of the player. This could also explain the decline as a player transitions from early to mid-30's and sees their productivity decrease being as much a result of a decreased role as a result of a skill erosion. To elaborate, a declining player is given more minutes, they'll be more able to perform in spite of their decline to statistical performance. A good example of this is Kovalchuk's performance in Montreal. While given top line minutes, and high-skilled linemates he was able to perform better than with limited minutes and less skilled linemates. This indicates that decline is as much a factor of deployment as it is natural erosion of skills.
The "plateau" factor could be explainable by players in their mid-30's being put into a role that's designed to get the best out of their performance. This can be especially telling for defensemen who are more likely to receive overall support . Plateauing can be seen as a factor of a reduced (but stable role) a decent amount of minutes, and being put in a pair/line designed to use the best of the player's ability rather than a slowing of the decline.
So...What does this all mean?
Prime age, and peak ages are fairly consistent across players. Even those players who come into the league later or decline faster can generally be expected to have their best, most productive years be between the ages of 26 and 28.
Players who don't "fall off a performance cliff" by the age of 32 aren't likely to do so, and will likely slowly decline with time.
Defensemen generally enter their primes later, and decline later than forwards...although their peak age remains the same.
For forwards, 32 is a critical age, and there is a significant drop-off in how many of these players are still in the league. Players who are in the NHL at 32 are likely to still be in the league until they are 36/37 at which point only the most elite players of that age group remains. For defensemen, the critical age appears to be 34/35 with only the most elite players remaining by 36/37. This would indicate that that the age of 36/37 is the "bottom" for a significant majority of players. Signing a player beyond that age should be considered extremely risky.
An interesting observation is that it appears that investing into a player in their early-to-mid 30's is actually less of a gamble than investing into a player at 30 due to the "x factor" of precipitous decline. If a player's decline curve is more-or-less in line with expectations by 32, it should be easier to apply the modeling to their performance. Putting this into the perspective of our oldest skaters (Weber and Petry) it's likely that we'll get value from them well into the latter parts of their 30's barring significant injuries.
By these conclusions, a veteran contract should never be signed beyond the age of 37. If a player declined significantly by after the age of 30, it can be assumed that their production will not recover, and should be dealt as quickly as possible. If a team can swing it, avoid having NMCs be in effect before the age of 32 to enable moving the player if they start showing a significant decline during the first couple of years of their deal.
In the cases of the two major sources of information, I feel like the studies merely scratch the surface and emphasize what could be done with better analytics being available. The Journal of Quantitative Analysis' assessment has by far the better model, but reliance on +/- also introduced the largest bias. Revisiting the data using better and more comprehensive data points should produce a significantly better analysis.
I for one look forward to see what will come out of the data being generated in the last 5 years and what it will mean to assessing players and predicting their performance through every stage of their careers.
Finally, without comparing the exact same data-set it's difficult to see a true correlation between the two sets of data. The fact that the graphs are superficially similar could likely be a red herring. It does seem to indicate that WAR is not a predictor of on-ice production. Assuming that a player's most productive years are truly between the ages of 26 and 29, this should be clearly visible in the WAR chart...but what it *seems* to indicate is that although players are at their most productive, they're past their statistical peaks...I can't say that I would say that this fits in with the "eye test" nor does it fit in when analyzing players in general. Some players do see a statical as well as a production decline in their late 20's when others are seeing a peak...but those are in many way outliers rather than the norm.
What I would like to see is a comprehensive attempt to track players throughout multiple draft classes (a dozen or so) and identify each stage of their career. Looking at "Points on the board" results, and comparing them with statistical performance.
The biggest questions I've been asking myself after reviewing these studies are as follows:
1) How many - and which - players are added and removed every year, and how does that impact these calculation. The sample is small enough that even a dozen players being introduced can significantly skew the numbers.
2) How does the normalization for an "age bucket" impact the "0" in the WAR calculation?
3) Why is there such a big difference between the WAR peak and production peak?
4) How is it that WAR calculate a fairly stiff decline after 30, when production seems to be relatively stable until 32?
5) How influential are high-value and negative impact players in the calculations?
6) Using Ovechkin as an example (as per Hockey-Graph's relatively deep-dive) it appears that the WAR methodology isn't a predictor of things like G/60, zone starts or contributions to their lines. Looking at the WAR chart exclusively would lead one to believe that he had a significantly down year in 13-14 and a rebound in 14-15, but the case is that the two years are very comparable in terms of production, and that most of his advanced stats are better in 13-14 than 14-15. This may be an outlier in of itself, but it doesn't indicate that the graph seems to be have a statistical bias I can't identify.
Overall it pains me to admit it, but it seems to indicate that the most robust study using flawed stats (such as +/-) seems to provide better, more applicable data than the limited study using what should be better advanced statistics. I do believe that analytics properly deployed can provide a new level of understanding for hockey in general and the NHL in specific. But it's clear that it's not just about what information you get...but how you use it that really matters.