How do we define Possessions and why should you care?

Hockey stats have come a long way. From the debates about whether using shot attempts has any use at all to Zone Entries to using Passing/Pre-Shot Movement to better quantify shot quality and playmaking. And for the most part I've kept along with my analysis along the same lines. Starting out with simple shot differentials, working my way to Zone Entries all the way to tracking passes. Now you don't need to have tracked thousands of zone entries to notice that while these microstats certainly have lots of uses, there's still a gap between having these stats to providing tactical feedback.. Bad zone entry numbers don't have to indicate mistakes or bad reads in the neutral zone. They can be the result of suboptimal breakouts, trapping opponents, etc. Not all zone entry situations are alike, after all. So how do we get more similar situations to compare to another?

Possessions as context

First, something on the terminology used here: We want to use "possession" the way it is used in Basketball or American Football, i.e. a string of events where one team has control of the puck. And while there is some value in simply adding "per possession" stats to add to the regular set of zone exit/entry/shot assist data, further categorisation can be advantageous. Just like there are fast break possessions and half court possessions in Basketball, we can now start to define certain subtypes of possessions in hockey. This will allow us to not just get insights into the overall ability of teams to enter the opponent's zone or get good shot attempts but also give some context to these actions. Where did the attack originate? Who was involved in the actions that precede Exits or Entries? And most importantly: Which situations present the most problems for the team? Do they have trouble getting back into positions after chipping the puck out? Do they have a bad breakout scheme when the opposition backs up into a trap?
Possessions get you closer to actual answers to tactical questions while still using the established indicators of exits and entries as building blocks.

So, what kinds of possessions are there? There are lots of different ways of doing this of course, so I in no way claim to have found the "right" or "best" way to classify possessions, but here it goes:
The basic categorisation I've chosen is based on the location of the puck recovery that starts the possession:

  • Defensive Zone Recovery:
    • Faceoff Win
    • Quick Breakout
    • Set Breakout
  • Neutral Zone Recovery
    • Faceoff Win
    • Regroup
    • Deep Regroup
    • Counter Attack
  • Offensive Zone Recovery
    • Faceoff Win
    • Loose Puck/Turnover

Possessions start with the recovery of the puck and last until a break of play, a loss of possession or a shot, unless the following shot is a rebound shot. So a dump-in where the puck is recovered by the offensive team would continue a possession, while a dump-in where the opponent gets possession of the puck and turns it over would not (the ensuing offensive possession would be classified as an OZ Recovery).

To illustrate what these possessions look like in practice, here are some video examples from last season's Gagarin Cup Final (I've excluded Faceoff Wins):

Quick Breakout
I define a quick breakout as a possession that starts with a puck recovery in the DZ where an attempt is made to transition the puck into the Neutral Zone quickly. Usually, attacks like these get out of the DZ in 6 seconds or fewer.

Set Breakouts
Set Breakouts are possessions that start in the Defensive Zone and the team sets up a breakout deliberately (think Mighty Ducks' Flying V), getting the opponent to usually (unless trailing) back off.

NZ Regroup
A NZ Regroup usually follows an uncontrolled Zone Exit by the opposition. The team recovers the puck somewhere in the NZ and tries to get the puck back into the Offensive Zone. Usually a D-to-D-pass follows.

Deep Regroup
Deep Regroups are NZ Regroups with the added condition that the recovering team pulls the puck back behind their own faceoff dots (regrouping deeper in their own territory) before trying to re-enter the opponent's zone.

NZ Counter Attack
Neutral Zone Counter Attacks follow turnovers  by the opponent in the Neutral Zone. A turnover here is defined as a loss of possession where the opponent immediately gets control of the puck in a close-by position. These are usually quite dangerous because the opponent gains possession of the puck while your team is still moving forward. 

OZ Recovery
Any type of puck recovery in the offensive zone. Usually a recovery of a loose puck after a shot or a turnover by the opponent (remember Dump-in recoveries would not count as a new possession).

Now while the video should show that these definitions describe game situations that at the very least look different, it's worth pointing out how they differ in certain per possession statistics. So let's have a look at the average number of shots (shots here referring to all shot attempts) per 100 possessions*:

cfp100.png

The territorial advantage of a successful Forecheck become clearly visible here. Recovering the puck higher up the ice obviously leads to more advantageous attacking positions. Although since OZ Faceoff Wins - where one would assume a larger proportion of shots to come from the blue line - appear to result in the most shots, maybe shot quality should be considered as well. Let's take a look at the expected goals:

xGp100.png

This confirms that Shots after OZ FO wins are at least comparatively worse than shots after Turnovers/Rebounds, which intuitively makes sense, considering the defending team should be in a better position to prevent good chances after a lost draw.

While there is a lot more to be said about this, I thought I'd rather get into some other topics surrounding possesions. The two I'm going to cover here is the question of "why separate quick and set breakouts" and then a quick introduction into pass clustering.

Why separate quick and set breakouts?

While tracking Zone Exits, it occured to me that crediting players with "Controlled Zone Exits" on Set Breakouts isn't actually very helpful when trying to evaluate the ability to exit the zone with possession. Because unless the opponents are very motivated to get the puck back quickly (when trailing late, for example), when they see that you have the puck safely and they can't apply immediate pressure, they'll usually back off and set up in the Neutral Zone. So carrying/passing the puck beyond your own blue line is hardly the same achievement here compared to doing it against a proper forecheck right after gaining possession of the puck.

The main differences one would expect as a result of this definition are that

  1. Set Breakout Possessions have a higher Exit Control% than Quick Breakout Possessions
  2. Set Breakouts should usually result in an entry attempt while quick breakouts (which count dump-outs and quick turnovers as well) probably won't
  3. Set Breakouts should probably have a lower percentage of controlled entries since we expect the Neutral Zone to be packed with opponents whereas Quick Breakouts, while generating fewer entry attempts overall, should probably have a higher percentage of controlled entries since you'll need to have played past forecheckers and into a more open Neutral Zone

All of these actually show up in the data. The Exit Control% on Quick Breakouts is quite a bit lower on Set Breakouts so it is probably wise to either look at these exits separately or at least adjust for the different situations.
As expected, Quick Breakouts generate fewer zone entries but a larger percentage of those are controlled. It also intuitively makes sense that Set Breakout Possessions see more Dump-ins but a higher recovery rate, since there are more opponents stuffed into the Neutral Zone and Set Breakouts are often designed for recovered dump ins with this in mind.

table_breakotus.png

Pass Clustering by Possessions

To complete this intro into possessions, I wanted to illustrate one more thing: Pass Clustering. After reading Dustin Ward's article "Finding the Best Pass in the Bundesliga" on Statsbomb, I started thinking about applying pass clustering to hockey. Using possession types to further improve the classification is an obvious advantage in my opinion. While I'm open to the idea of clustering all passes and then filtering them by possession to notice differences in distribution, I chose to split the passes into possession types and then cluster them by start and end locations (so we don't have different frequencies in the clusters by possession types, but rather different clusters for different possession types).
For the possessions shown here, I decided to only use Defensive Zone and Neutral Zone passes, since the ensuing Offensive Zone Passes are probably less indicative of distinct plays that are markedly different between possession types.

First off, let's have a look at DZ Breakouts (the numbers represent the percentage of all completed passes in this possession category belonging to this cluster):

Cluster_Quick2.png

I think most of these passes make sense, at least in the Defensive Zone (the goal here being to get all important and markedly different pass types in there while avoiding unnecessary splitting). Since I'm not an expert in clustering, I had some difficulty getting them right strictly via the k-means clustering, so I took the results and adjusted them in some cases (in case you were thinking "no way they come out symmetrically like that without tinkering"). Feel free to let me know that I made some errors/forgot some crucial pass type, I'm well aware that analyzing one team's play for two years may have biased my ideas of what breakout passes look like.

Let's move on to Set Breakouts, where we again, in my opinion, find passes that match the general look of set breakouts. You start behind the net (where there are lots of "passes" just pushing the puck to another dman before starting the rush) and move out, usually going via one of the outer lanes, into a packed Neutral Zone, where passing is a lot harder and completed passes are much more rare.

Cluster_Set.png

And finally, Neutral Zone Regroups (Deep Regroups and Pass types in the Offensive Zone will probably be covered at some point in the future). The typical, "slow it down" play here is the D-D pass deep on your own blue line after the puck is chipped out of the opponent's zone and then move the play back up the wings.

Cluster_NZR.png
hist_khl_d.png

Separating passes by possession and cluster can also improve the evaluation of a single player. We can improve on raw passing percentage by giving each pass more contextual information and then stop rewarding players for making lots of easier passes.
The following is a distribution of the number of Quick Breakout passes per Set Breakout pass among KHL defencemen with at least 150 Passes in the sample. It shows that the amount of breakout passes in Set vs in Quick Breakout situations can be quite dramatically different among defencemen depending on their own ability and the team's style of play. This is something to be considered in any passing analysis.

Obviously not just the type of game situation can change the probabilities on pass outcomes, the location from where a pass is played and where a pass is played to can also have a large impact and the pass distribution can be quite different from player to player (see below for examples).

These are just a few examples of why I consider possession analysis to be a valuable addition to hockey analysis. While it's obviously intended to be descriptive in nature, focussing on describing a team's style including strengths and weaknesses in certain areas of the game, it also has applications in areas such as passing analysis and analysis of transitional play.

 

Data: The data used for this analysis is detailed play-by-play data (shots, passes, exits, entries, etc.) from InStat.

*Basketball uses per 100 possessions and I thought I'd stick with this convention since the numbers get quite small if you simply calculate them per 1 possession.