NHL EDGE: Skating Speed, Part 2
Role/usage and fatigue
As mentioned in part 1 when discussing the distributions of top speeds, burst rates, and average skating speed, there appeared to be clear differences between forwards and defensemen in both raw ability (forwards tended to be faster) and in playing style (forwards exceed 18 MPH much more often). Furthermore, glancing at the names near the top of the leaderboard for 18+ MPH bursts suggested that a player’s role affected how often they reached that threshold, with depth forwards being expected to reach that speed with regularity (as an energy or checking line, for example).
Relatedly, when looking at the distribution of when top speeds were set, there was a clear trend to when skaters set their top speeds, tending to do so earlier in the game. This trend was mirrored within periods, indicating that fatigue does have an impact over the course of a game.
In this article, I’ll look first at how a player’s role, in terms of line/pairing, is related to their skating output before exploring fatigue.
Role/Usage
In this section, I limited my analysis to players with at least 20 games played in a season, the same cutoff I used in part 1. The analysis here is relatively straightforward, just one-to-one correlations between average ice time, which I use as a proxy for role. For clarity, I included average ice time cutoffs for what constituted a first line/pairing player, second line/pairing, and so on. I reached these cutoffs by taking the top 288 forwards1 by ice time and the top 192 defensemen2 and calling them “first line/pairing,” then repeating this for each additional line/pairing. This may result in an uneven distribution between teams/season, but the purpose was to give a frame of reference for what the ice time means.
First up, I’ll look at 18+ MPH burst rate.
This confirms what I hypothesized previously, which was that depth forwards are expected to fill an energy role, and skate faster more often. Defensemen, on the other hand, display basically no trend at all. This makes sense as depth defensemen are not typically thought of as an “energy” pairing.
Moving on the 20+ MPH burst rate, things get interesting for defensemen.
The correlation for forwards weakens though there is still a clear trend towards depth forwards “bursting” more often. This also seems to confirm what I suspected in the first article of this series, namely that 20+ MPH is a sufficiently high enough speed that most players do not reach it regularly. Defensemen, on the hand show a slightly positive trend. That is, defensemen who are given more minutes tend to burst more often. My explanation for that would be that as the game has gotten faster, being a good skater is a requirement for a top defenseman. A slow defensive specialist can be protected for 15 minutes a night, but if a player is going to play 20 minutes or more, they have to be able to skate well to do so.
Next, I looked at average skating speed.
The relationship for forwards is remarkably similar to the 18+ MPH bursts rate. Defensemen also display a similar, although weaker relationship. This, to me, is the clearest indication that top pairing defensemen are conserving energy. Despite bursting more often, top pairing defensemen skate slower than depth defensemen, meaning their time between bursts is spent skating slower.
Lastly, I looked to see if these trends held up when looking at average even strength skating speed.
At even strength, the trend for forwards almost disappears, while it is slightly weaker, but remains essentially unchanged for defensemen. I interpret this to mean that special teams are the main driver in variation between overall average speed since teams generally skate more slowly during special teams play.
One limitation of this approach, particularly for defensemen, is that ice time alone may not fully capture a player’s role and expectations. For instance, a depth power play specialist defenseman likely has different expectations than a depth penalty kill specialist. My hope is that this evens out to some extent, but I may revisit this sometime in the future with more descriptive roles to see if the pattern change.
Fatigue
To investigate the effects of fatigue, I took a more complicated approach. The EDGE data included game logs for distance, which I was able convert into average MPH for each game, which mean I could compare the effect of time on ice on average speed.
However, simply looking at the correlation between speed and time on ice would ignore several potential confounding factors, such as individual ability, role, and score situation.3 To account for these variables, I created two simple models, one for forwards and one for defensemen, using linear regression to predict average speed at even strength in a given game with terms for individual ability (top speed), role (average even strength, power play, and shorthanded time on ice), and score effects (average score differential from the skater’s perspective). Score effects data was calculated using the play-by-play data available from Evolving-Hockey.
Next I took the residuals (difference between predicted and actual speed in a game) and created a linear model for each skater looking at all their games in a given season to predict their average speed for that game, using the residual and their time on ice as predictors. Then I used the coefficient of the time on ice term as a “fatigue” rate. A negative value indicated the more minutes a player played, the slower the skated and vice versa.
First, let’s look at the distribution of fatigue rates by position.
Interestingly, the range of forward coefficients is much, much smaller than the range for defensemen. I’m not sure if this is because conditioning is more consistent for forwards or because there is a flaw in my methodology. The pattern makes sense to me, with a slight rightward-skew and a slightly stronger skew for forwards. This is what I would expect if skating were a selected for skill, especially for forwards, although I might have guessed the skew would be more apparent.
Next, I’ll look at how fatigue rates correlate to burst rates, which seem likely to wear players out.
Surprisingly, the correlation is positive for both forwards and defensemen. I’m not really sure what to make of this. The relationship is fairly weak, but is statistically significant for both positions. My best guess is that to “burst” that often a player must have good conditioning, so they see less of a drop-off as the game goes on, but this was an unexpected result. This is just a guess though. It would be extremely useful to more granular shift data, or even game-by-game burst rates here.
Next I looked at 20+ MPH burst rate compared to fatigue.
The relationship here is almost negligible: not statistically significant for forwards and only just barely so for defensemen. I attribute this to the rarity of 20+ MPH bursts. Most forwards will only reach this speed a four or fewer times a game, not often enough to really burn out their legs. Defensemen reach this threshold even less often. It’s not really surprising there is no relationship here.
Last, I looked at average time on ice versus fatigue rate.
Here we see two patterns that are illuminating. First, there is a clear negative relationship between average time on ice for forwards and their fatigue rate. The more a forward is expected to play, the faster they fatigue. This seems intuitive to me. In my own experience, fatigue is not linear, but rather builds as you continue to exert yourself. The longer a forward plays, the more effect fatigue has on their ability to maintain their speed.
However, on the defensive side, the relationship is reversed. We’ve already seen indications that top defensemen are stronger skaters than bottom pairing or depth players, and this provides more support for that idea.
Finally, for completeness, I also looked at fatigue rate compared to average even strength time on ice.
The relationship is almost the same (slightly stronger for forwards), suggesting that even strength play is the driver for fatigue. Not surprising considering that’s when most of the game is played.
Conclusion
Role definitely affects how forwards skate. Bottom line players burst at 18+ MPH significantly more that top line players. However, the pattern is not as strong at a 20+ MPH threshold. For defense, role is almost a non-factor in skating style
Similarly, fatigue seems to be a bigger factor for forwards than defensemen. In fact, top pairing defensemen seem to get stronger as the game goes on. This suggests that skating ability and conditioning is a sought after asset for top pairing defensemen, similar to depth forwards.
Next time, I’ll attempt to determine if any of this matters for actually winning games.
32 teams multiplied by 3 forwards on the top line multiplied by 3 seasons.
32 teams multiplied by 2 defensemen on the top pairing multiplied by 3 seasons.
This will be explored more in part 3, but score effects do appear to exist for average speed.