Comments / New

A look at rate statistics and ice time

In a perfect world, NHL players would be given ice time directly proportional to their skill. Coaches wouldn’t be biased in any way, shape, or form, and the best players on any given team would also get the most ice time for said team.

Unfortunately, this does not happen, and there are plenty of times when players are stuck in limited roles simply because their coach has some kind of bias against them, even though they would more than capable of succeeding in a second or first line role.

If you’re just looking at raw point totals (or even points per game), these players will look worse than they actually are, because they won’t be getting the playing time they deserve (it’s also possible that lesser players get more ice time, and look better as a result).

Analysts have accounted for these differences in opportunity by looking at rate statistics, which calculate the rate at which events occur based on ice time, giving players that don’t get a lot of ice time a level playing field against players that see the ice more often.

This isn’t an end-all, be-all statistic, of course, but it does give us a way to identify players that are potentially under (or over) valued by the league due to their point production. Take Adrian Kempe and Alexander Radulov, as an example. Radulov played 1288 minutes at 5v5 last season, while Kempe played just 943. Radulov tallied 40 points, while Kempe tallied 30. This gives Kempe a points per hour of 1.91, and Radulov a points per hour of 1.87. Though Radulov managed 10 more points than Kempe over the course of the season, the two actually produced at a similar rate.

Radulov, however, played on Dallas’s first line alongside Jamie Benn and Tyler Seguin. His most frequent teammates are great players, which led to their line facing the opposition’s best players on a nightly basis. Kempe, on the other hand, played a middle six role for Los Angeles, facing weaker opponents as the number two/three center behind Anze Kopitar/Jeff Carter.

So, how much does a player’s role impact their point production, especially from a rate statistic perspective? If Kempe and Radulov saw their roles switched, could we still expect them to produce around 1.9 points per hour?

In order to answer this question, we first need a baseline to work with. Given the effect that bounces, random variance, and plain old puck luck can have on a player’s production in any given season, there’s no real guarantee that point production per hour is even a repeatable statistic.

With that in mind, I pulled all forward seasons from 2007-2018, and compared each player to their previous season’s production. To keep small sample sizes from playing a role, I only kept players that played at least 500 minutes in both the “current” season and the previous season. To keep secondary assists from creating noise, I only looked at primary points (goals and first assists) Here’s a scatterplot of the numbers.

As we can see, there’s a generally positive trend. This means that the statistic is relatively repeatable – though random variance certainly plays a role, the better players in the league typically have a higher rate of points per hour across multiple seasons, while lesser players stay on the lower end of the spectrum.

So, we can assume that primary points per hour is a fairly repeatable metric. Next, let’s test to see if ice time has an impact on how repeatable of a metric primary points per hour actually is.

Taking the chart from before, I only looked at players that saw an average ice time per game increase or decrease by one minute. Here’s the updated chart.

There’s virtually no change between the two charts. The math backs this up, as the two models are virtually identical, with virtually identical R-squared values (0.19 vs. 0.21).

A one minute change in average ice time per game may not be significant enough, however. What if we only look at forwards who had a two minute increase or decrease in ice time?

There’s not much change in R-squared between the model for players with a two minute difference and the players with a one minute difference, but there is a slight gap between the model for players with a two minute difference and the model for all players (0.17 vs. 0.21). It seems that there might be a slight impact from ice time, so let’s go one step further and look at players that have a three minute increase or decrease from one season to the next.

We’re down to just 50 observations, but the data here is pretty clear. With a three minute increase or decrease in ice time per game, there’s virtually no relationship between past season primary points per 60 and current season primary points per 60.

Overall, we saw that the correlation between previous season primary points per hour and current season primary points per hour declined the more change a player saw in his average ice time. Changing roles did have an impact on primary point production, though the effect wasn’t necessarily positive or negative.

I’ll let Brian MacDonald summarize for us (note: the linked article is behind The Athletic’s paywall).

For a guy like [Riley Nash], if he’s put in a different situation, ideally, his numbers would stay the same — (that is) his advanced metrics would stay the same if they’re accounting for teammates and opponents. But in reality they’re probably going to change. One thing most of these models ignore chemistry or synergy between players.

It’s also most of what people talk about out in the open “Well, this guy is getting this amount of production in only 11 minutes of ice time per game at 5-on-5” and then they basically make the assumption that if you increase playing time that the production rate is going to stay the same. Which is almost certainly not true. There’s no way. That’s a huge assumption.

This is something by Brian likely has looked into in the past, but I couldn’t find anything online confirming that it was true, so I investigated. It’s rather clear, I think, that point production rates are prone to change as a player’s ice time changes.

It’s likely that there are a huge variety of factors at play here, though I suspect most of them stem from the fact that we’re missing a lot of the context that goes into ice time changes.

Here’s a brief and incomplete list of scenarios that come to mind where a player could see his ice time change (and his production change as a result).

  • A young player starts out playing fourth line minutes with unskilled teammates. The next season, he’s bumped up to a second line role alongside two excellent players. His production increases, as he now has skilled teammates to work with.
  • A veteran player signs with a new team, downgrading from a second line role to a third line role. The quality of his teammates doesn’t change much, but he’s now playing against much weaker competition, and his production increases as a result.
  • A skilled player with key flaws finds success on the third line, but is forced to move up to the first line after a player leaves in free agency and another falls to injury. His flaws are exposed when he’s faced with stronger competition, and his production suffers.
  • A veteran player racks up points playing alongside his team’s top center, but an up and coming rookie takes his place on the top line. Between old age and the new teammates, the veteran struggles, and he sees a decline in scoring.  /

Going forward, it would be interesting to dig into the context a bit more, to see if there are specific groups of players more prone to an increase or decrease in point production based on factors such as projected ice time, projected teammates, or even projected opponents.

At the end of the day, though, you most likely clicked on this blog post for Panthers analysis, not analysis of the NHL at large. Here’s a look at primary points per hour for Panthers forwards last season.

Florida Panthers P1/60 2017-2018

Player GP TOI Goals First Assists TOI/GP P1/60
Evgenii Dadonov 74 1072.32 20 19 14.49 2.18
Nick Bjugstad 82 1080.13 18 16 13.17 1.89
Aleksander Barkov 79 1222.45 13 17 15.47 1.47
Frank Vatrano 41 410.17 7 3 10.00 1.46
Jonathan Huberdeau 82 1193.83 16 12 14.56 1.41
Jared McCann 68 791.22 7 10 11.64 1.29
Denis Malgin 51 624.20 10 3 12.24 1.25
Jamie McGinn 76 884.98 9 9 11.64 1.22
Connor Brickley 44 475.52 4 5 10.81 1.14
Colton Sceviour 76 807.45 10 5 10.62 1.11
Vincent Trocheck 82 1216.87 9 13 14.84 1.08
Radim Vrbata 42 471.97 3 4 11.24 0.89
Derek MacKenzie 75 660.87 2 6 8.81 0.73
Keith Yandle 82 1440.25 6 11 17.56 0.71
Maxim Mamin 26 267.20 3 0 10.28 0.67
Mike Matheson 81 1424.27 7 6 17.58 0.55
Mark Pysyk 82 1316.87 3 9 16.06 0.55
Micheal Haley 75 575.80 3 2 7.68 0.52
Aaron Ekblad 82 1450.65 9 2 17.69 0.45
Alex Petrovic 67 945.52 2 5 14.11 0.44
Ian McCoshen 38 523.58 3 0 13.78 0.34
MacKenzie Weegar 60 831.22 2 2 13.85 0.29

We can see that player role impacts point production for the Panthers right from the start, with Evgenii Dadonov and Nick Bjugstad coming ahead of Aleksander Barkov. Seeing as Barkov’s usage is very defensive-minded, it’s not surprising to see him after over 15 minutes of 5v5 time per game, but produce points at a lesser rate than his counterparts.

Frank Vatrano is also interesting, as he’s averaged about 1.16 primary points per hour over the course of his career. His uptick in production last season is notable, though it’s questionable if he can sustain it. He clearly played a fourth liner’s role last season, averaging just 10 minutes at 5v5 per game, but if he can maintain his rate of production in a third line role next season, it would help the Panthers develop one of the league’s stronger third lines.

Vincent Trocheck’s low rate of production at 5v5 was offset by his astounding power play production last season, so people aren’t talking about it. Still, it’s noticeable, and one can’t help but wonder if he would benefit from having less defensive responsibility.

Micheal Haley is an enforcer who would have made a great fourth liner… 15-25 years ago. Unfortunately for Haley, the league is changing. If the Panthers want to hang with the juggernauts of the East, they should look into icing a fourth line that has a bit more skill.

Again, this isn’t an end-all-be-all statistic, and it only tells part of the story. Still, it’s good to know that context matters, even if that context is something as basic as ice time.

Data courtesy of Natural Stat Trick. R code, R plots, and source data can be found here.

Talking Points