WARNING to regular LHH readers: This is an in-depth analytics article without a clear Islander focus. If you're interested in learning more about hockey, read on. If you come here for Isles analysis and prefer not to deal with #s, this might not be for you.
Last year, the guys over at BroadStreetHockey, particularly Eric T (@BSH_EricT) and Geoff Detwiler (@GeoffDetweiler), began the Zone Entry Project, where they tracked "Zone Entries" during 2011-2012 Flyers games. What are zone entries? To quote my own primer on the subject:
Zone Entries are the name given to each entry made by each team into the offensive zone from the neutral zone. In effect, I'm going through each game and tracking each time the puck travels from the neutral zone to the offensive/defensive zone.
What do I mean by tracking? Well, ,what I mean is that I'm tracking who gets the puck over the blue line, how they do so (via dump, tip, carry-in, or pass), and whether it's even strength or not.
In essence, the goal here was to try and track performance by players in the Neutral Zone, where a good deal of hockey is played. Conventional hockey statistics and even so-called advanced statistics ("fancy stats", if you will) don't provide any way of tracking performance by players int he neutral zone. This is of course a big oversight given how much of hockey ISN'T played in one end or the other.
When the entire season's worth of data was collected, Eric T took a look at the data and found several interesting findings. First:
It appears that how a team gets the puck into the zone is as important as how often they do it. Maintaining possession of the puck at the blue line (carrying or passing the puck across the line) means a team will generate more than twice as much offense as playing dump and chase.
The net result here: our surprising results in the offensive and defensive zones appear to be based on a not-particularly-reproducible metric. It may still be true that some players have skills that help the team get more shots per zone entry, but at the end of a year we can't reliably tell which players those are -- we know who did well in the offensive zone this year, but don't have strong reason to believe they'll do well in the offensive zone again next year.
The neutral zone is a different story. The split-half reliability there is 0.44, high enough that we can be 97% sure that this is a real correlation and not just random results. Given a half-season of neutral zone data, we can make a decent guess at what will happen in the other half-season.
Moreover, the neutral zone results look like they are not just statistically significant, but meaningful in practice as well. It's pretty reasonable to see Jagr and Claude Giroux leading the forwards and Timonen and Matt Carle leading the defensemen.
These findings are incredibly significant. The first finding, if true, suggests that teams are often being way too conservative in dumping in as attempting entries via possession (by carry-in or pass) is much more effective such that it's worth teams attempting riskier entries rather than simply attempting a dump.
The second finding is even more significant. Hockey Analytics research has suggested that one of the more important parts of hockey play is the ability to maintain possession, and metrics such as fenwick and corsi have evolved as a way to measure individuals' abilities to drive possession. Not only does possession lead to winning in the long run, but it is far more sustainable than winning via hot shooting or goaltending, and thus is a more effective strategy for long-run success. Eric's conclusion, if true, suggests that possession is near completely driven by neutral zone play, and that offensive and defensive zone play is mainly dictated by the play in the neutral zone. Moreover, fluky possession #s are possibly driven by "lucky" offensive or defensive zone performances, which are unlikely to continue.
Now Eric obtained data from the Minnesota Wild from 2011-2012, and found again that neutral zone performance was repeatable to some extent by players. However, the Wild data suggested that defensive zone performance was NOT random over the course of a season and in fact was a repeatable skill. This suggested perhaps that Eric's thesis was too broad, and that offensive zone and defensive zone performance may be a repeatable talent for players/teams. More data was needed in order to test Eric's thesis.
Fortunately, over the past lockout-shortened season, several trackers have duplicated Eric's work in tracking the neutral zone entries for a few teams: namely the Islanders, Oilers, Canes, Kings, Ducks, and Sharks (in addition to Eric's tracking of the Flyers). Thanks to Corey Sznajder (@ShutdownLine) and Jonathan Willis, I was able to obtain complete neutral zone entry data from this past season from the Isles, Canes and Oilers.* And thanks to Muneeb Alum (@muneebalamcu) of Red Line Station and Eric T, I was able to translate this data into complete neutral zone on-ice data for each of these three teams.
*Trackers for the Ducks, Sharks, Kings, and Flyers have not yet completed their tracking for the season, and I eagerly await their data.
Armed with this data, I set about trying to find if Eric's findings fit these data sets. Now we have smaller sample sizes for each team this year due to the lockout (48 games instead of 82), so our conclusions are going to be a little more limited. Still, the data results are very interesting.
First of all, Eric's first finding, that entries with possession are double or more effective than dump-ins/tip-ins appears to be valid for all 3 teams here, with the overall #s being very close (around .58 shots per entry with possession, .27 shots per entry without possession).* It certainly does seem like teams are being too conservative with their entries.
*Canes data has higher values per each type of entry; however, this is almost certainly due to the Canes tracker (Corey of @Shutdownline) being a much stingier tracker than any of the other trackers, reducing the # of entries and increasing the shots per entry.
However, Eric's second finding - that neutral zone performance is a talent that is repeatable in 41 game samples while offensive and defensive zone performance is not - isn't fully born out by the data. Let's look at each of the three teams for whom we have full data:
These three graphs represent the Split-Half correlations of Islander players in Neutral Zone Play (Left most graph), OZone Play (Middle Graph), and DZone Play (Right Graph). As you can see we DO have Neutral Zone play being somewhat reliable - the R^2 is less than Eric's but at .33, that's probably due to sample size. But we ALSO have Offensive Zone play AND Defensive Zone Play being more reliable - with R^2's of .569 and .387 respectively.
In other words, for the Isles at least, it certainly seems like offensive zone play and defensive play are reliable to an extent - possibly more than neutral zone play! In other words, The Isles seemed to be able to control to an extent the amount of shots per type of entry in each of the zones (Carries were still always far more effective than dumps, mind you, but the shots per each type of entry does seem controllable to an extent!).
Okay, so this contrasts with Eric's conclusions. How about the Oilers and Canes? First, the Oilers:
Again with the Oilers we see some repeatability of Neutral Zone play in split-halves - more reliable than the Isles and more in line with Eric's results. But again we see that offensive zone play is at least partially reliable, with an R^2 of .339. This is less than the Isles' #s and now less than the NZ #s, but that's still very relevant. Oilers' D Zone performance however appears to be completely random, as we'd expect from Eric's thesis, but contradicting the Isles data and the Wild data from last year.
How about Carolina?
Figure 3: Split-Half Correlations between Neutral Zone, Offensive Zone, and Defensive Zone Performance of Hurricanes Players
Here things are just weird. Again we see a Neutral Zone correlation, though it's so small as to actually be insignificant. And now we see what Eric was talking about, with basically no OZone or DZone repeatability. Perhaps the Canes are built in the same mold as the Flyers from last year, but the Neutral Zone sample size isn't large enough to show repeatability.
Obviously further study is needed,, but we now have multiple teams, Edmonton and the Isles this year, Minnesota last year, where Offensive Zone or Defensive Zone performance appears repeatable. So the overall theory that Neutral Zone play may be more reliable than Offensive Zone or Defensive Zone play MAY be true, but the extent of it may have been previously overstated.
I wonder how much these #s are influenced by team systems, rather than simply the game of hockey. As I'll discuss in a future post, the Oilers for example showed incredible consistency in their entry with possession #s, but their opponents' #s at doing so - the defensive aspect of the neutral zone - was nearly ENTIRELY RANDOM. Meanwhile, the defensive zone performance of the Oilers was pretty much random. Could be a coincidence, or could be that certain systems lend themselves to consistent play in one or more zones more than others. Further investigation is required.