Evaluating offensive strategies in the CAN v. USA Rivalry Series
An analytics breakdown by Mikael Nahabedian
 |
Note: Today’s guest post was written by Mikael Nahabedian, a video & statistical analyst for the McGill Martlets. Follow him on Twitter at @hunterofstats.
In preparation for the 2020 Women’s Hockey World Championship, which would eventually be cancelled due to COVID-19, Team Canada faced its American rivals in a five-game Rivalry Series that took in place between December 2019 and February 2020.
The United States won four out of the five games, leaving Canada with only one win thanks to Victoria Bach’s OT goal in Game Three of the series.
After getting outscored 6 to 2 by their opponents in the first two games of the series, Team Canada replaced head coach Perry Pearn with Troy Ryan, who was previously Pearn’s assistant. One of Ryan’s main challenges was to find ways to help his team generate more offense at even strength (i.e. generate more and higher-quality shots).
With this in mind, we will analyze and compare the shot generating ability of Team Canada and Team USA using manually tracked data from the 5 games of this series. The goal here will be to show how both teams differ in their respective approaches to generating offense, while evaluating the effectiveness of these strategies.
Prior to starting the analysis, to better understand the play-driving abilities of both teams, we can compare the Corsi and Expected Goals (xG) differentials of the two rivals.
Team USA significantly outperformed Canada both in terms of quantity (Corsi) and quality (xG) of offensive output generated throughout the series, topping it off with a dominating performance during Game Five in this regard.
A piece of good news for Team Canada is the positive difference between their xG and Corsi differentials, which means that the chances generated by the team slightly prioritize shot quality. However they are still being outplayed overall at 5v5.
In order to understand the nature of these discrepancies, we need to break down the offensive sequences of both teams. Consequently we will be looking at the combination of events that led to shot attempts for each team.
The offensive sequences, focusing on events directly leading up to shots, are divided into five categories:
- Shots generated “Off the Rush” – directly following a Controlled Entry
- Shots generated “Off the Rush with a Shot Assist” – directly following a Controlled Entry, which is then followed by a Shot Assist
- Shots generated “Off Shot Assists” – during an offensive zone possession, when a Pass is directly followed by a Shot Attempt
- Shots generated “Off Rebounds” – when a Shot Attempt is directly followed by another Shot Attempt
- Shots generated “Off Other Events” – Turnovers, Failed Exits, Dump Entries & Recoveries, etc.…
Both teams seem to have similar approaches to generating offense. While Team USA’s strategy puts more emphasis on generating shot attempts off shot assists (which is consistent with their strong possession ratios throughout the series), Canadian players seem to focus slightly more on generating offense off the rush compared to their rivals.
Approximately 65% of the shots generated by both teams can be attributed to a combination of offense generated “Off the Rush”, “Off the Rush with a Shot Assist” and “Off Shot Assists”. Further analysis should therefore be focused on these 3 categories.
I. Shots Generated Off the Rush
In order to determine shot generating trends off the rush, a clustering algorithm, that groups similar rushes together, was used to identify 4 types of offensive sequences used often by each team.
Off the rush, Team Canada prefers entering the offensive zone from the left side, outside the dots, while Team USA prioritizes middle entries, as exhibited in three of the four trends identified.
This leads to Team USA players carrying the puck all the way to the top of the circles, or sometimes even to the slot yielding high quality shots off the rush. In contrast Canadian players, entering the zone outside the dots, generate weaker shots in terms of quality, often times being angled and pressed to shoot by defensive players before the top of the circles, which results in a bad angle to generate quality chances.
From time to time, Canadian forwards are able to generate quality shots after entering the offensive zone from outside the dots. But the higher overall quality of US rushes in comparison to Canada’s is confirmed by the quantification of this analysis when comparing the average xG/rush for all shots generated off this type of sequence: USA xG/rush = 0.03 and Canada xG/rush = 0.01.
The following clips show the offensive trends identified with the clustering analysis.
Key Takeaways from these clips:
Speed through the neutral zone is key to generating high quality shots off the rush. As seen in Clips 2 and 3, players are able to generate high quality chances when they use the neutral zone to gain speed as they are entering the offensive zone. Clip 4 also illustrates how to effectively enter the offensive zone through the dots, gaining speed through the NZ, combined with a good attack position (puck positioned on the side of the body) once in the offensive zone to generate high-quality shot attempts off the rush.
Using crossovers to accelerate through the neutral zone with the puck can give puck carriers an advantage when they are entering the zone, resulting in a better entry position or helping them escape from angling attempts of defensive players.
The speed that American players develop through the neutral zone prior to entry is the main reason for which they are able to generate higher quality shots than Team Canada off the rush.
II. Shots Generated Off the Rush and Shot Assist
Shots generated with shot assists off the rush (Shot Assist + Controlled Entry) represent approximately 10-15% of a team’s total offensive output. Despite this small proportion, this type of offensive sequence has important implications in getting a global understanding of a team’s offensive trends.
Using a similar approach to the previous part, with clustering algorithms, trends in this type of offensive sequences can be identified for both teams.
For the most part, Team Canada puck carriers angle themselves towards the boards after an entry. Puck carriers, now close to the boards, then proceed to pass from low to high to open up a shot lane.
In contrast Team USA focused on having a middle presence after the entry and playing the puck laterally from the outside to the middle support player.
Both teams have one very high value yielding controlled entry & assist sequence (the 0.16 xG one for USA and 0.09 xG one for Canada). However, the cluster size of these particular sequences is smaller than the size of all the others, which translates into a lack of repeatability of these sequences.
The following clips illustrate the offensive trends identified with the clustering analysis.
Key takeaway from clips:
Shots taken off the rush following a shot assist occur in various situations, as seen in the clips above. The puck carrier often has the option to either shoot or pass the puck. It would be interesting to analyze if the puck carrier made the right decision given the circumstances.
Linking this idea back to our offensive sequences, the goal would be to quantify the value of each shot assist to evaluate the payoff of decisions made by players with the puck. In other words, we are trying to answer the following question from a quantitative standpoint: did the passer make the right call when choosing to pass the puck?
III. Shots generated from shot assists
a. Introducing Shot Assist Expected Value (xV)
To keep our calculations simple, when evaluating a shot assist, we will assume that the passer only has two options: she could either pass or shoot the puck.
As presented during the ISOLHAC conference by David Yu and Sam Forstner earlier this year, prior studies on passing value consider calculations based on expected goals and opportunity of moving with the puck. In the future, adding a movement opportunity component could be an interesting addition to our calculation.
For now, in order to analyze the value of shot assists, we will consider two components in our calculation of the Expected Value (xV).
Expected Goal on Reception
xG on reception (xGr): as shot assists lead to a shot attempts, our calculation considers the xG value of the shot attempt resulting from the shot assist.
Opportunity Cost of Passing (instead of Shooting)
We are trying to determine whether the passer made the right decision when choosing to pass (instead of shooting the puck). Therefore, when evaluating a shot assist, we need to consider the incremental value created from making this pass.
To illustrate this point, we can take the example of Blake Comeau’s infamous breakaway miss. Even though this failed pass is not at even strength and is technically not a shot assist, it is a good example for explaining the rationale behind our opportunity cost calculation.
In a normal game setting, if put in Blake Comeau’s position, most players will shoot the puck (not pass it).
Moreover, from that position in the offensive zone, a player has a good chance of scoring a goal if she shoots the puck.
Logically, passing the puck in this particular scenario is a terrible idea.
Expanding this idea to all shot assists, our estimation of the opportunity cost of a shot assist is based on the two particular factors mentioned above:
1. Would other players shoot or pass if put in the same position as the passer?
2. What is the probability of scoring from the position of the passer (if the passer had shot the puck instead of passing it)?
The second component of the calculation rewards players that create value, with shot assists, by moving the puck from an unfavorable scoring positions to a favorable scoring position. For example: passing the puck from behind the net (an unfavorable scoring position) to the middle of the slot (a favorable scoring position).
To summarize, our Expected Value Calculation is based on the following formula:
xV = xGr – Opportunity Cost of Passing
b. Evaluating Shot Assist Trends
Now that we have explained how we measure shot assist values, we can compare the shot assist trends for Team Canada and Team USA. Once again using clustering analysis, we were able to determine the shot assist trends for both teams at 5v5.
Both teams exhibit similar trends in terms of passing clusters:
Low to High, xV = 0.0
Passing the puck from low to high – usually from the half-wall or the corner of the boards to the players across the blue line – yields a low shot assist value for one specific reason.
While significantly improving the shooting angle vis-à-vis the net, a low to high pass gets the puck significantly further from the net, which is negatively correlated with the probability of scoring a goal.
Distance from the net being the most important predictor of the probability of scoring, Low to High shot assists do not add much value in generating higher quality shots.
However, Low to High shot assists could be very useful when trying to force opposing players to move in order to open up the play. This is particularly useful when facing a very compact defensive structure, with all 5 defensive players “collapsing” towards their own net.
D to D, xV = 0.01
Passing the puck from one D to another yields a very small incremental value when analyzing xG contribution of such a play. Passing the puck from one side to the other of the zone (with similar distance from the net, shooting from a similar angle on the opposite side) does not improve your position for taking a shot, while also not worsening it.
That is why the value of this assist approximately corresponds to the value of the xG of the shot taken off the assist.
Similarly to Low to High shot assists, D to D passes also have a structural purpose, helping the team on offense open up the play by forcing movement in the defensive structure of the opposing team.
Perimeter to Danger Zone, xV ≥ 0.10 xV
This type of shot assist yields the most value. With this type of event, the puck is basically moving from a location on the ice where the probability of scoring is low (i.e. behind the net or in a sharp angle position vis-à-vis the net) to a location where the probability of scoring is high (i.e. the slot area).
This type of assist has a positive dual effect on the xV formula, as it improves both the angle of the shot opportunity and gets the puck closer to the net. The opportunity cost for the passer in this type of scenario is usually very low, as the passer is usually in a bad angle for a shot, in addition to being far from the net.
We will call such shot assists, higher danger assists. For analysis purposes, we will also classify assists yielding an xV between 0.04 and 0.10 as moderate danger assists.
Having explained the inherent characteristics of different shot assist types, we can compare the total xV generated by each team at 5v5.
The first thing we notice when comparing the in-zone offensive performance of both teams is the very large number of shot assists completed by American players. As we had mentioned earlier, this stems from Team USA’s strong possessions/play-driving indicators, which translates into more time spent in the offensive zone compared to Team Canada, at 5v5.
However, in terms of the quality of shot assists, Team Canada seems to be significantly ahead of Team USA, producing an average 0.023 xV per shot assist, compared to only a 0.015 xV average rate for Team USA.
Comparing the number of assists for both teams in terms of the different types of passes, we arrive at a similar conclusion: having only generated 52 shot assists, when possessing the puck in the offensive zone at 5v5, Team Canada completed 19 moderate to high danger shot assists, which is more than the 17 assists completed by American players having produced 89 shot assists in total.
A significant proportion – approximately 80% – of Team USA’s shot assists correspond to low xV shot assists such as D to D or Low to High passes. On the contrary, Canadian players seems to be prioritizing quality over quantity, as shown with 37% of their shot assists being classified as moderate to high xV assists.
While Team Canada should be working on slightly increasing the amount of shot assists, the focus on quality of shot assists over quantity is an encouraging sign when evaluating their overall shot generation performance at 5v5.
*
Postscript: Today’s guest post is part of an ongoing effort to give exposure AND financial compensation to emerging voices in the hockey tactics & analytics sphere. If you have written a high-quality post on a topic you feel passionate about and would like to see it published here, contact me via Twitter DM. - Jack Han