At the 2nd Annual Hockey Analytics Conference at the Rochester Institute of Technology, Matt Cane and I presented on new ways of measuring and evaluating defensive play. We used the passing data from my project to take a look. I believe Matt will write up his half of the presentation, so this post will focus on my half of the presentation. You can access our slides and view our presentation here. I will go into greater detail in this post since I am without time constraints. All of the data we used can be accessed here. Everything in this post deals with 5v5 play.
The trouble with defensive metrics is that there is so much happening on the ice that it becomes difficult to measure a single player’s defensive contributions. We use Shots Against and Shots Against Rel to evaluate a player’s defensive abilities, but are those the best ways? A quick look at the league leaders in CA RelTm (Shots Against relative to a player’s teammates) gives us some familiar names: Patrice Bergeron, Hampus Lindholm, and Drew Doughty, but we also get some surprises like Chris Neil, Jiri Sekac, and Matt Martin. What data can we use that offers analysis and pinpoints specific areas of a team/player’s game to improve upon?
I’ve already started doing that this past summer in performing tactical analysis of offensive zone decisions as well as neutral zone decisions, so this will be a similar piece, but focused on defensive play.
Passing Project Origins
I began this project a little over three years ago out of a desire to have better data that could tell us more about player positioning, shot-generation, and how teams pass the puck. While much of the project’s focus has been on the offensive side of the game, I knew I would eventually get around to looking at how teams defend against passes.
Why I think this is a better approach than simply looking at the end product – shots against – is because there are a lot of implied hockey skills contained in a team’s shot assists against. Getting sticks in lanes, pressuring the opposition, effective checking and coverage, those are all traits of good defensive play to prevent the opposition from completing passes. Since I’ve established that teams score goals a a higher rate from shot assists than not, defensive passing metrics should account for all of the hockey skills mentioned above. However, judging a player solely based on metrics for body checks, stick checks, deflected pucks, etc. can be a fools errand.
The book Soccernomics goes into why getting too granular with defensive data can be a distraction. Below is an excerpt from the book in which Mike Forde and Gavin Fleig, performance directors for Chelsea and Manchester City respectively, are discussing this and also how data that seemed important became less so:
Yet by the mid-2000s, the numbers men in soccer were becoming uneasily aware that many of the stats they had been trusting for years were useless. In any industry, people use the data they have. The data companies had initially calculated passes, tackles, and kilometers per player…However, it was becoming clear that these raw stats – which still sometimes get beamed up on TV during games – mean little.
Forde remembered the early hunt for meaning in the data on kilometers. “Can we find a correlation between total distance covered and winning? And the answer was invariably no.” You might know how many passes a player had given, but that didn’t tell you whether they were splitting through-balls or sideways shoves into a teammate’s feet.
Tackles seemed a poor indicator too. There was the awkward case of the great Italian defender Paolo Maldini. “He made one tackle every two games,” Forde noted ruefully. Maldini positioned himself so well that he didn’t need to tackle. That rather argued against judging defenders on their number of tackles, the way [Alex] Ferguson had when he sold [Jaap] Stam. Fleig said, “Tackles to me are a measure of being under pressure.”
Forde reflected, “I sat in many meetings at Bolton, and I look back now and think ‘Wow, we hammered the team over something that now we think is not relevant.'” Looking back at the early years of data, Fleig concluded, “We should be looking at something far more important.” – page 154, Soccernomics.
I think the same thing is happening in hockey right now. Data companies are selling lots of relevant-sounding metrics to teams without answering “how much impact does this have on winning?” Goal-scoring? Goal-suppression? “Analysts” are being hired that contribute no original work of their own. I even have detractors on Twitter because we don’t track everything that happens on the ice. I simply don’t want to repeat mistakes people made in soccer ten years ago. Lots of things don’t matter. It’s okay to admit that once you find value in repeatable and predictive metrics that can also inform what actions a team should take.
Below you will see a table with split-half R^2 values for repeatability and predictability for a variety of defensive metrics. If you’re unfamiliar with my work, know that a shot assist is the final pass preceding the shot, and an entry assist is a pass made in either the defensive or neutral zones that is followed by a controlled entry and subsequent shot.
For the top four (Shot Assists Against, Shots Against, Expected Goals Against, and Entry Assists Against) this is from twenty teams that we have at least thirty games tracked on: thirteen from the 2015 – 2016 season, six from the 2014 – 2015 season, and one from the 2013 – 2014 season. The bottom three metrics are strictly from the thirteen teams from this past season as we did not track to this level of detail in prior seasons. The p-values for each of these metrics is below the 0.05 threshold, except for Shots Against (0.08), Corsica’s Expected Goals Against (0.20), and Royal Road Shot Assists Against (0.42).
What we quickly learn in this analysis is that the way in which a team defends against shot assists (remember from above – this will likely account for positioning, getting sticks in lanes, giving your opponent time on the puck, etc.) is both highly repeatable and predictive of future goals against. When we compare it to a team’s overall shots against (Corsi), we see a much stronger relationship in both repeatability and predictability.
We also see that passes originating from behind the net or in the center lane (between the faceoff dots) in the offensive zone have strong correlations as well. There is a relationship at the team level with regards to Royal Road passes, but it’s not a terribly strong one in terms of how well it predicts goals against.
Identifying traits at the team level that suppress goals against can help us better evaluate both how teams should play defense and which players suppress these types of chances. Keying in on where these chances originate (behind the net, center lane) expedites video review and preparation because now you have data that answers an important question: “How can I improve my defense and reduce goals against?”
Looking at the shot assist against rates for each of the thirty teams we’ve tracked from last season (each team has at least twenty games tracked) we see the top five made up of the Florida Panthers, Carolina Hurricanes, Los Angeles Kings, Washington Capitals, and Tampa Bay Lightning. That mostly makes sense, though as the samples grow, we could see some shuffling.
The teams at the bottom make sense well: the Colorado Avalanche come in as the worst defensive team and, counting upwards, we have the Calgary Flames, Arizona Coyotes, Dallas Stars, and Vancouver Canucks. Teams that weren’t great defensively.
We can also look at how teams ranked based on neutral zone play and the rate of entry assists they allowed.
We again see Florida leading the pack, but we see changes immediately after. The Columbus Blue Jackets are now second after being in the bottom half of overall shot assists against/60. What this implies is that the Blue Jackets may have been successful at breaking up entries or simply not allowing them, and then were horrendous in their own end by comparison. Conversely, we see the Kings drop from the second spot to around where Columbus was, suggesting there are areas in their defensive game teams excel at and could improve upon.
The reset of this article deals with video review and translating this valuable data into tactical analysis. Everything comes back to how well an analyst can relate the meaning of the data to the coaching staff. Here, I’ve made it easy and offered evidence of what defensive strategy should look like.
Defensive Zone Coverage
Now, we’re going to take a look at evidence of why Florida is a good defensive team, and why Colorado is a bad defensive team. I went back and tracked and analyzed Florida’s 4/9 game against Carolina, looking at both team’s defensive styles, as well as Colorado’s 4/3 tilt against the St. Louis Blues.
Here was Florida’s standard defensive zone coverage: a half-ice overload that combines man and zone concepts. Naturally, a team’s F1 and F2 will be the first two forwards back into the zone and will take up positions on the boards where the puck is – the strong side. F3 will occupy the position in the high slot and will watch for cross-ice passes. D1 covers the low corner position and D2 covers the net and backside pressure, similar to F3.
The tactical advantage of playing this way is it shrinks the space with which the opposition has to work with. You see this in soccer all the time: teams will expand the field of play on offense to give themselves as much area to work with as possible. Without the ball it will be just the opposite: they will play with a high and aggressive back line to squeeze the field of play. It’s not as easy in hockey due to different offsides rules, but the same principles are at play here. Let’s see how the typical sequence plays out for Florida in this game.
You see that he three strong side players (F1, F2, & D1) each have a zone they are responsible for. However, they also have the freedom to assist with pressuring the opposition and supporting their teammates when the puck is in a 50/50 situation. The Hurricanes’ winger loses possession and the Panthers player sends the puck up the ice to try and clear it.
Once the ‘Canes defenseman at the point receives the puck, his only option is to fire a shot from the point, which we know isn’t preferable, and then the process repeats itself, only this time the Panthers are able to get possession of the puck and move out of the zone.
Let’s contrast this with Colorado’s approach.
Colorado plays much less structured and this leads to a lot of chasing and open ice. They are often “at sixes and sevens” in their own zone. A failed pass up the boards leads to a shot, possession down low, and then another shot. As the puck goes to the corner from the initial shot from the point, Jaden Schwartz and Patrick Berglund are able to play a 2-on-1 against Tyson Barrie. Barrie is essentially monkey-in-the-middle here chasing Schawartz, then back to Berglund, which leaves Schwartz open behind the net.
Barrie finds himself in this position because Mikkel Boedker doesn’t fully commit. Once Barrie moves to go with Schwartz, Boedker fails to step up onto Berglund to deny the return pass. Boedker appears concerned with…well, I’m not sure. He’s not in a lane to deny a pass back to the point and any pass to Backes at that point would be a tough pass to make. So, he’s occupying space, but doesn’t have any work to do. Had he jumped onto Berglund, it would allow Barrie to stick with Schwartz, and Duchene could fill and defend against the likely outlet to the point. Instead, lots of chasing.
Here is a perfect sequence of everything that’s wrong with Colorado in their own end. As the Predators’ forward emerges from behind the net, there is ample opportunity to check him to attempt to remove the puck. Erik Johnson or Nick Holden could attack the puck carrier and the other could cover the front of the net. Instead, both make half-hearted attempts and then Holden goes to the front of the net, while Johnson skates aimlessly higher in the zone. And that’s when Colorado is already beaten
The entire Avs forward line and Johnson follow the puck, which is bizarre because it’s getting farther away from their net. When Roman Josi activates and attacks downhill with the puck after a quick exchange, he has a step on everyone the Avs have on the ice. Holden, still at the net, makes another half-hearted attempt at playing defense and Josi is able to slide the puck over for an easy goal for Mike Fisher.
On the backside, while this play develops, the weak-side winger, center, and Johnson fail to see Fisher coming around. Johnson especially is at fault on this play as he failed at checking the puck carrier when the play began and then failed to cut off the passing lane or get proper positioning on Fisher as the play developed. It’s clear the structure with which Colorado plays in their own end is lacking compared to Florida.
Neutral Zone Play & Forechecking
Florida’s aggression is also a big reason why their forecheck and neutral zone play are so successful.
As Carolina attempts to get this puck out of the zone, Florida has four players inside the zone and the fifth is just outside the blue line. You can see that the depth at which Jiri Hudler (24) is at deters a pass through the center. Jakub Kindl (46) also deters action through the center due him holding a position at the top of the zone. The far side- forward in your picture is Rocco Grimaldi (23) who shuts down the board-side option to get the puck up. Finally, Alex Petrovic (6) pressures just enough as the Carolina exits the zone to cause the turnover and chance going the other way.
Florida essentially applies the same pressure-concepts from their defensive zone coverage at the other end of the ice, shrink the field of play with their positioning, and then squeeze the puck from Carolina as the puck carrier goes the only place Florida wants him to.
Here is an example of how Florida pressures off of a regroup. Carolina sends the puck over and this pass immediately triggers pressure from Grimaldi. The defenseman has no choice but to send the puck back, at which point Hudler moves in to pressure the puck. Rather than go back across to his partner, Michal Jordan (47) opts to send the puck out of the zone. Unfortunately, Kindl steps up and picks off the clearing attempt. This leads to a quick chance.
The reason why this pressure is successful is because Teddy Purcell (26) takes up Kindl’s position on the blue line. The ability to run a successful aggressive defensive system is reliant on players to switch positions and cover when one initiates pressure. If only two or three players can do this, the system eventually breaks down. Kindl reacts to Hudler stepping up to pressure Jordan, and then Purcell steps up to support Kindl moving in. The team works in concert off of the pressure being applied in front of them.
Let’s take a quick look at how Colorado does just the opposite.
After Colorado loses possession, St. Louis is forced to regroup quickly and attack again. Colorado does a decent job pressuring the puck carrier in the Blues’ end, but fail to take away the board-side outlet that we saw Grimaldi do in the previous clip. Here, Carl Soderberg (34) comes inside, when he really should stay on the wall as Blake Comeau (14) is forcing the Blues defender away from the center of the ice. This is an example of a player not reading the situation and support of his teammates and then making a poor decision.
As the play develops, we see Francois Beauchemin (32) with an opportunity to stop this play in the neutral zone. St. Louis makes a pass across and, had Beauchemin read this correctly, he could have stepped up to steal the pass, deflect it away, or make a decent hit on Robert Bortuzzo (41). Instead, the puck continues down the ice where Bortuzzo and Paul Stastny (26) are able to isolate Beauchemin on the entry and Bortuzzo feeds a puck to Stastny for a backhand chance.
By playing as passively as they do, Colorado has no excuse for at least not rotating over properly and cutting off his chance in their end. If they have anyone in that organization analyzing things in this manner, they certainly have their work cut out for them.
Forechecking Immediately Following Faceoffs
Finally, we’re going to look at contrasting styles after offensive zone faceoffs. First up is Florida.
Here, Florida loses the draw and immediately sends Jonathan Huberdeau (11) and Jaromir Jagr (68) after the puck carrier. Aleksander Barkov (16) stays high in the zone to cut off any pass through the center (or, be in a good position to cut off any backside outlet). Carolina has to hurry the pass and it misses its target, bouncing off the boards and out to center for the Panthers to regroup. Had this pass connected, Florida has another defenseman waiting on the boards to make a play on it as well.
And now for the opposite in Colorado.
Colorado fails to get pressure after losing the draw and then immediately retreat down the ice. The Blues have all the time and space in the world in transition here. Colorado at least forces a bad shot, but fail to recover the puck. Alex Pietrangelo (27) activates once inside the offensive zone and takes this puck low.
Watch Alex Steen (20) come into the picture and fill the space Pietrangelo previously held. Colorado fails to account for this space and Shawn Matthias (18) does his best Erik Johnson impression of “looking busy while doing nothing of use” by letting Steen skate right by him for a chance. Matt Duchene (9) goes to pressure Pietrangelo in the corner, but can’t cover Steen at the same time.
Risks Associated with Aggressive Play
While there are many positives in how Florida plays defense, you do wonder if playing this aggressively will lead to odd man situations going against them. Matt Pfeffer asked this very question at #RITHAC and I made sure to look into my data afterwards. See below.
So, we do see that Florida is one of the worst teams at allowing odd man situations to develop, though these are relatively rare situations. Let’s have a look at what happens when their press fails. For clarity, we are only recording an odd man situation in which a pass is made and a shot is taken.
We’re going to look at images here in order to highlight the breakdowns. It starts out very much like the opening goal of the game for Florida (first clip under the Neutral Zone Play Header above). Florida has three players below the faceoff dots and Carolina’s Justin Faulk (27) really doesn’t have many options here. Ideally, you’d want him to slide the puck back towards Elias Lindholm (16) up the boards, but given the pressure he’s facing he may now know that’s a possibility.
Let’s advance a few seconds and we now see the puck is loose in the far corner. Barkov appears to have the inside track to get it and Aaron Ekblad (5) has pinched to eliminate an easy breakout up the boards. Michael Matheson (56) is also well inside the zone, eliminating any opportunity for an outlet pass in his direction. Florida appears to be in good shape here.
The puck comes up the boards to Joakim Nordstrom (42) and he simply taps it back to Jaccob Slavin (74). Barkov really should have taken a more direct approach and cut under Nordstrom rather than above him as Ekblad had that covered. Now, Carolina has an area of space to exploit.
Slavin lifts the puck out of the zone and now there are two Hurricanes at center ice and only one Panther, Matheson, back. Matheson cannot be too aggressive here because if he misses or guesses wrong, it’s a breakaway. Once Lindholm possesses this puck, he can only defend the two-on-one.
Lindholm advances, forcing Matheson to make a decision on how to play the threat that Jordan Staal (11) poses.
Lindholm actually does really well to force Matheson to commit one way and then pass across to Staal. Roberto Luongo ends up making a brilliant save here to deny Carolina the opener.
So, what happened? Where did things break down for Florida? It’s easy to look at this and say, “Well, Carolina flipped a puck out, Lindholm won a race, and it was a bad pinch by Ekblad,” which is what I’m sure people say at times when watching situations like this unfold. But, to understand where things broke down, we go back to the second image.
We know from how the play develops that Barkov ends up almost shoulder-to-shoulder with Ekblad and behind Nordstrom. This isn’t an ideal position and has to do with Barkov’s pursuit angle to the puck. If he comes under the puck and stays between Nordstrom and Slavin, then he cuts off the space that eventually is created later on in this play.
It’s fine margins when a team plays this aggressive style, so these details are important. As I mentioned above, if someone doesn’t support correctly, lanes open up and opponents are given time on the puck.
Measuring how well a team suppresses shot assists is a better predictor of goals against than simply looking at total shots. Teams that do this well do several things that you hear coaches discuss: keeping an active stick, getting in lanes, checking and pressuring your opponent effectively.
Florida’s defensive keys: manipulating the space with which they allow the opposition to play in each zone, and free-flowing positional responsibilities. More teams should employ this type of thinking not only while pressuring the opposition, but also offensively. There are risks associated with giving up these chances, but Florida also is also one of the better teams at generating odd man situations and this is due to their aggression.
At the end of the day, Florida’s style is likely worth the risk. There are some improvements to be made for sure, but the overall impact is clear to see.
Colorado’s passive system that lacks structure leads to, unsurprisingly, lots of chasing in their own end, coverage mishaps, and a style that allows the opposition to transition at ease. Fewer teams should play this way as it confounds logic to do so.
Digging into the cost-benefit of pressing in certain situations is an intriguing topic. Identifying the optimal forechecks to deploy in certain game states and situations is another.
With more and more people tracking data and looking at the game differently in order to try and gain an edge over the opposition, enhancing a team’s tactical and opposition analysis is now possible. Looking at offensive, neutral, and now defensive zone tactical data analysis, there are components of an idealized hockey system that we can now begin building. So, look for that to come with future posts as we build our samples to continue testing our data, in addition to planning out plays and strategies.