When Ken Holland announced his interest in re-signing both Darren Helm and Drew Miller, the conversation inevitably turned to one of their much lauded strength: penalty killing. While we've made light all season of the near-mythical penalty-killing abilities conferred upon Helm and Miller by the Detroit press, we've been able to recognize two general truisms about penalty killing:
- No single person (outside of the goalie, perhaps) is responsible for the success or failure of the penalty kill. At least, not to the extent that people will have you believe.
- It's difficult to pin down exactly what makes a good penalty killer because of hockey's inherent randomness and complexity.
In this piece, we'll explore a facet of the penalty kill that may influence a skater's ability to suppress opponent shot attempts. We hypothesize that a player's deployment can significantly impact shots against, expected goals against, and goals against. Specifically, we posit that players who receive more on-the-fly shifts ultimately experience fewer goals against.
To ensure that every player had at least one season of data, players with at least 500 minutes at 4v5 between the 2007-2016 seasons were included. The resulting dataset included 187 forwards and 191 defensemen. From Corsica.hockey, data on each player's 4v5 time on ice, defensive zone faceoff starts (DZS), on-the-fly (OTF) starts, Corsi against per 60 (CA60), Fenwick against per 60 (FA60), shots on goal against per 60 (SA60), goals against per 60 (GA60), and expected goals against per 60 (xGA60) were obtained. For more details on the expected goals model, please see Emmanuel perry's complete write-up here. CA60 was selected as the primary comparator given that the ice time samples in this study are relatively small despite incorporating all available data.
Each player's 4v5 OTF shift starts per 60 minutes (OTF60) were plotted against his CA60, FA60, SA60, xGA60, and GA60. A linear regression line of best fit was plotted for each comparison. The r^2 were calculated for each comparison.
Figure 1 demonstrates the relationship between OTF60 and CA60 for forwards.
Figure 1. OTF60 vs. CA60 for forwards (>500 mins), 2007-2016
As seen in Table 1, the r^2 for the relationship between OTF60 and CA60 was 0.493. Over the last nine years, the deployment of a forward explains almost 50% of the variance in 4v5 shots against, 40% of the shots on goal against, and nearly 23% of the goals against.
Figure 2 demonstrates the relationship between OTF60 and CA60 for defensemen.
Figure 2. OTF60 vs. CA60 for defensemen (>500 mins), 2007-2016
As seen in Table 1, the r^2 for the relationship between OTF60 and CA60 was 0.278. The deployment of a defensemen explains almost 28% of the shots against, 20% of the shots on goal against, and 6% of the goals against. A full listing of all r^2 values for each variable is listed in Table 1.
Table 1. r^2 values for OTF60 vs. assessed variables for forwards and defensemen
On-the-fly deployment appears to have a stronger correlation for forwards as compared to defensemen across all identified metrics. The ability of OTF shifts to explain nearly 50% of the shots against forwards is significant as it alters the way we evaluate forwards on the penalty kill. Skills such as speed, physicality, and toughness are often cited as necessary qualities of a penalty killer, but these findings challenge whether any of these skills are relevant. In fact, one could argue whether true "skill" exists for penalty-killing, or if our perception of skill is altered by how players are deployed. These findings also call into doubt whether one forward is significantly better than a replacement forward, or if most forwards are ultimately interchangeable.
Players with higher OTF60 start more shifts with the puck outside of the defensive zone. With a majority of penalties being two minutes or less, any time spent outside of the zone is a substantial benefit to both forwards and defensemen. Additionally, with the puck starting outside of the defensive zone, the forwards also have an opportunity to defend against the ensuing zone entry.
It is worth noting that a majority of the forwards who outperformed their expected place along the line of best fit are highly skilled forwards. Some of the top performers include Henrik Sedin, Henrik Zetterberg, Joe Thornton, Patrice Bergeron, Pavel Datsyuk, and Brad Marchand. These are players who had significantly lower shot rates against compared to their expected place along the line of best fit. These players had large variations in their OTF60, with Bergeron and Marchand having the toughest deployment of the aforementioned names. A possible theory is that these highly-skilled forwards are more adept at forcing turnovers off of the zone entry and then are more likely to carry the puck up ice as opposed to dumping the puck in. However, this theory has yet to be validated.
The correlation between OTF shifts and shots against was not as strong for defensemen. Nonetheless, the ability of OTF shifts to explain nearly 28% of the shots against a defensemen is still significant. Defensemen utilized on the penalty kill are often selected for their size, physicality, and shot-blocking abilities but these findings challenge whether or not those qualities are significant. A cursory glance at the defensemen graphic shows Nick Lidstrom, Adam Larsson, Erik Karlsson, and Alex Edler as players who outperformed their expected place on the line of best fit while Brendan Witt, Mike Komisarek, Ed Jovanovski, and Luca Sbisa all underperformed.
At this point in time, an overhaul to the way we evaluate forwards and defensemen on the penalty kill is needed. The relative impact that one forward can have on shots against appears to be largely related to how they are deployed. Similarly, the impact an individual defenseman has on shots against also appears to be heavily influenced by their deployment. There are many ideas as to what skills are important for a penalty killer to have; however they remain unvalidated at this time. In the future, we hope to elucidate how these skills impact shot rates against as well as how OTF shift rates correlate with players on the first and second penalty-killing units.
*All data obtained from Corsica.hockey
**Additional thanks to JJ, Asmean, Matt Cane (@Cane_Matt), Mike Fail, Sean Tierney, and the rest of the crew at Hockey-Graphs.com for their help with this piece.