Determining NHL player peaks has frequently focused on production and, occasionally, wrinkles are added to account for the steeper fall-off for goal-scoring as opposed to playmaking. Generally, the peak appears to be around the ages 23-25, with some skills like shooting exhibiting fairly early peaks and others a bit later.
Poking around some spreadsheets, I came across data that I’ve always meant to get to: time per shift. The NHL has been keeping a measure of average time per shift for players going back to 1997-98, so I licked my chops over the robust data set. The “Why?” for looking at it, I think, takes us to an interesting place. To some degree, time per shift can allude to a player’s stamina and overall physical fitness; it can also allude to the coaching staff’s assessment of their performance — though there are plenty of shifts ended on the fly in a hockey game. What’s more, we simply haven’t had a lot of player peak estimations using time on-ice, and when done carefully, I think we can capture something like a total physical peak for players.
To build my curve, I focused on year-to-year change for players, age 19 to age 38. To help a bit with survivorship bias and what I’m calling “rookie bias,” and to keep us comparing apples to apples, I focused on consecutive player seasons of 20 or more games played, and removed the changes from a player’s rookie to 2nd season (“first change”) and 2nd-to-last to last season (“final change”). This artificially creates a data set of players with at least four consecutive 20+ GP seasons (seasons #2 and 3 would provide a data point). Rather than chucking the removed data entirely, I decided to include it in a chart alongside the line from what I’m calling “the middle years.”
The number of data points in the middle years was about 5,300 year-to-year changes; the other two groups were around 1,600 apiece.
As you can see, the first and final change years introduce some fluctuations that can affect our curve in a way we might not want. For instance, the first change has a marked bump right about the time players get out of rookie contracts. Because it’s near peak in the studies I cited above it might go unnoticed, but its relationship to the other data (and its own) is just too divergent to not be capturing something outside peak evaluation. The final change line is instructive, too; unlike the middle years, the final change players are losing playing time from one year to the next even in their early-to-mid 20s, which could drag our player peak curve down prematurely. The general idea is to avoid these more drastic effects on the curve, effects that might be attached to decisions beyond player skill and fitness. By doing so, the middle years provide us with a nice, smooth line, before the small sample strikes at 38.
Taking the year-to-year change of the middle years, and using the typical shift length for a 19-year old as our starting point, we get this player peak estimation for shift length:
The raw data, of course, is for an atypical player who gets real time at 19, but the curve is the important thing here. The peak starts at 23 and continues till roughly when the player turns 26. At that peak, the player is averaging 49.74 seconds a shift. They’ll have gone from a start of 47.4 seconds per shift, to 49.74, down to 45.55 seconds by age 38. While that doesn’t seem like much of a difference, it adds up pretty quickly when you have 20 (forward) to 27 (defenseman) shifts in a game. Take the average NHLer, who has a shift length of 46.2 seconds — about 3.5 seconds less than the player’s peak above. If they’re a forward, that’s a loss of a minute of ice time per game; if they’re a defensemen, a minute-and-a-half. By the end of the season, that’ll be over a full game, or more accurately to the individual player, about 3 to 4 games’ worth of ice time. Long story short, the curve is real for shift length, the differences over time are substantial, and the peak is defined.
One final thing: shift length is highly repeatable year-to-year, with a correlation of 0.78 (r^2 = 0.614) from one year to the next. So in addition to capturing this curve, it’s fair to say that players rarely fluctuate wildly from their previous performance by this measure.
While attributes like muscle mass, maturity, and “hockey IQ” might peak at a later time, the physical peak for players seems to be pretty firmly entrenched in the ballpark of 23 to 25 years old.
5 thoughts on “NHL Player Physical Peak Estimation”
suggestion, change units in chart to seconds
Curious if the result is the same when you look only at home games. As the home team gets the final personnel change, you eliminate the short shifts that occur due to line matching by the visiting coach. I wonder if certain players are substituted for more often when there’s a line mismatch that might (or might not) skew the results.
That was something I thought about too, but I decided it shouldn’t really impact things much with this amount of data. Minimum 20 GP (most are comfortably higher than 20, b/c we’re looking at players who played at least 4 such seasons) is enough to get a decent amount of home and away games. Also, we’re looking at hundreds of season-to-season changes per data point; with the way NHL scheduling works, this population should have a balanced impact from home vs. away, if there is any.
First amazing article.
This data doesn’t necessarily represent peak performance. It’s one thing to have longer shifts & another to be most productive. Theoretically 27-30yo players could have slightly shorter shifts due to physical limitation, but produce more due to experience & “hockey IQ”.
Would be interesting to see at what age players peak in terms of performance.