Off the Charts: Grading the Sharks through 52 games

A B C D E F G...

The All-Star break is over for some teams, but it is still depriving us of San Jose Sharks hockey for little longer. As a result, it is an apt time to examine what each player has brought to the team at this point in the season.

In developing an assessment like player grades, it’s important to establish what, exactly, it is we’re measuring. Are we grading a player’s performance above expected? If so, how are we establishing a baseline measurement against which to compare? Are we measuring each player’s impact on the on-ice product? If so, what is our barometer?

These questions are tough to answer and aren’t always present when we begin discussing things like player evaluations at a certain point in the season. Today, we’ll use measurements of on-ice impact to try to assess how each Sharks’ player has contributed to the standings points — all 65 of them — the team has amassed to date.

To do so, we’ll use a few tools. First, Josh and Luke Younggren over at Evolving-Hockey have developed a regularized adjusted plus-minus (RAPM) metric. Don’t let “plus-minus” confuse you; this has nothing to do with hockey’s useless plus-minus statistic. The RAPM process shows you how certain variables impact a target variable: the thing you’d like to understand better. In the case of hockey, the twins have used the process to help explain the relationship between things like teammates, competition, zone starts and game score to a player’s on-ice statistics. The end result is a given player’s isolated impact on a certain portion of the game.

The Evolving-Hockey RAPM model (written about in more detail here) provides a measurement for every player in the NHL that describes their impact on goals, expected goals and shots (Corsi) at even strength, shorthanded strength or power play strength. This next bit is arbitrary, but the cutoff makes some sense. In order to assess every Shark’s impact on the game, we will separate out the top number of players at each position for each game state. Here are those divisions:

  • 412 forwards at even strength (about each team’s top 13 forwards)
  • 184 forwards at power play strength (about each team’s top six power play forwards)
  • 127 forwards at shorthanded strength (about each team’s top four penalty-killing forwards)
  • 220 defensemen at even strength (about each team’s top seven defenders)
  • 79 defensemen at power play strength (about each team’s top 2.5 power play defensemen)
  • 130 defensemen at penalty kill strength (about each team’s top four penalty-killing defensemen)/

The goal with these divisions was to focus our efforts on NHL regulars. Within each group, the Sharks players all have RAPM impact measurements (provided they played enough time in a given game state to qualify). To assess how each Shark has stood up to the rest of the league, we found the equivalent percentile of each player’s RAPM measurement for goals, expected goals and shots. However, we want one measurement per player for each strength state, so we had to create a weighted average for each of the three components.

To figure out the appropriate weight, we turn to Garret Hohl’s article in the Athletic, “A closer look at Corsi, how much it matters, and what it tells us about the Jets’ season so far.” In the article, Hohl uses empirical research to establish just how much impact each component of the game has on a team’s standings points. His findings are as follows:

  • Shot volume (Corsi): 44 percent
  • Shot quality (expected goals): 10 percent
  • Luck: 33 percent
  • The remaining 13 percent is made up of goaltending, shooting and other. /

However, what else is goal-scoring in the NHL than shooting, goaltending and a bunch of puck luck? We’ll wrap those items, along with the “luck” category into a “goals” percentage: 44 percent.

We have our weights, and we have our RAPM measurements. Now we can take a weighted average of each player’s RAPM impact percentile to see how he impacts the Sharks’ standings points in each game phase.

For both power play and shorthanded game states, RAPM includes measures for a player’s impact on goals for and against, expected goals for and against and shots for and against. We’re only concerned with player impact on events against the team for the penalty kill and events for the team on the power play. For even-strength measurements, we are looking at a player’s total impact.

In the case of defensive metrics, measurements are inverted. So, a -0.01 measure for goals against shows a better defensive impact than a 0.01 measure. The percentile column shows what percentile among the range of players the Sharks skater’s measurement represents. In Brent Burns’ case, his -0.01 goals against impact is in the 71st percentile of our sample of 130 top penalty kill defensemen.

You’re looking at the far right column. This table shows how we are taking an average of each player’s impact to goals against, expected goals against, and shots against on the penalty kill. However, rather than averaging the percentiles into which each score falls, we are weighting the percentile by using the 44 percent, 10 percent and 46 percent weights we assigned to each component of on-ice results, above. The final score is how each player’s RAPM measurements compare to the other top NHL players when we weight for how important a given component is to standings points.

We do that for each game state: even strength, power play, penalty kill, and bring those weighted percentiles together. Finally, we can take a weighted average of each player’s percentiles based on the percentage of ice time he spends in a given game state. If a player spends less than five percent of his ice time in a certain strength, we zero that out in our weights, because chances are the impact during that time is way out of wack, due to a small sample size.

To determine each player’s letter grade at this point in the season, we will assign a grade depending on his final weighted impact percentile. The American school system’s grading (where a “C” is 75 percent or so, but also considered “average”) is no good, so we won’t be using those same percentile and letter grade matchups. Instead, this is how we’ll grade each player:

Here are the final scores for each position: (Rourke Chartier and Tim heed have not played enough to make the cut for any of the sample sizes. We are using the percentiles, measured against the league’s best, of each player’s impact so far.)

That Timo Meier is atop the forward chart should come as no surprise. Despite playing through an 18-game goalless streak, he has consistently provided offense for the team around him. He is one of the premier offensive threats in the league and, at just 22 years of age, should continue to be for years to come.

Hertl’s grade might seem low, especially with two-hat-tricks-in-one-week performance just before the break. However, there are a few things to keep in mind, here. First, while a player’s impact on on-ice goals is weighted fairly heavily in these ratings, scoring points ones self doesn’t necessarily mean that player is contributing to the offense for those around him. Second, he’s been more impactful on the power play, which only represents 15 percent of his total ice time, than he has at even strength. Finally, if we broke players down into offensive versus defensive impact, Hertl’s offense would be higher. His impact on the team’s ability to limit goals, chances and shots against hasn’t been great this season, so his overall even-strength impact is lower as a result.

Brenden Dillon is having a career season. This is somewhat buoyed by luck — he’s had an impressive impact on on-ice goals — but he’s also having a great year in terms of his impact on shots and expected goals.

Erik Karlsson might raise some eyebrows. Truth be told, he isn’t great on the power play. His impact on the Ottawa Senators’ power play the three years before this season wasn’t all that special, and it remains above-average but nothing fantastic this season. Karlsson’s even-strength score is borderline A/A+ material, but the 12.6 percent of his game spent as a 66th-percentile power player is bringing his overall average down a bit.

The process for evaluating goalies is a bit different, since each measurement varies depending on which site one uses. However, both goalies, when compared to the top 70 or so goalies in the NHL this season in terms of ice time, have struggled. Whether looking at percentiles for MoneyPuck, Corsica, or Evolving-Wild’s goals saved above average or expected, neither goalie has played better than about 42 percent of NHL goalies. The following percentiles show each player’s all-strengths performance.

Cole Anderson, of Crowd Scout Sports’ model also shows similar performance tiers.

No matter which expected goals metric we use to evaluate the Sharks goaltenders’ play this season, it isn’t pretty. Luckily, in front of those goalies, the majority of this Sharks’ team is comprised of guys who have been better than half the league at their position. With even league-average goalie play between the nets this year, people are probably talking about the San Jose Sharks the way they talk about the Tampa Bay Lightning.

For fun, here is what the Sharks lineup would look like if it was arranged according to player impact this season:


Evander Kane — Joe Thornton — Timo Meier
Lukas Radil — Tomas Hertl — Joonas Donskoi
Barclay Goodrow — Logan Couture — Marcus Sorensen
Rourke Chartier — Antti Suomela — Kevin Labanc

EDIT: A commenter brought to our attention that Joe Pavelski was omitted from this hypothetical lineup. We regret the error. Here is a re-written forward lineup based on those grades with Pavelski

Kane - Thornton - Meier
Radil - Hertl - Donskoi
Goodrow - Couture - Pavelski
Sorensen - Suomela - Labanc


Brenden Dillon — Brent Burns
Radim Simek — Erik Karlsson
Joakim Ryan — Tim Heed


Aaron Dell
Martin Jones