Analytics in hockey isn’t really about numbers. It’s about trying to break a hockey game or team or player down into little bite sized pieces that permit us to compare apples to apples and understand what separates teams, games and players. The numbers aren’t important because they’re numbers; they’re important because if they’re gathered properly and understood correctly, they can help us isolate the differences that make players better than other players, make teams better than other teams and win hockey games. It’s about finding a way to organize the terabytes of information in a hockey game.
One of the criticisms that people who are into data get is that hockey isn’t a game that’s susceptible to this sort of analysis. I don’t buy it. The more I dig into this, the more it just seems like constant repeating to me. The puck starts in the neutral zone. There’s a zone entry. Defensive/offensive zone time. Zone exit. Repeat thousands of times. If you used this as your basic framework for understanding 5v5 hockey, I don’t think you’d go far wrong.
* * *
I’ve been a bit obsessed lately with Taylor Hall’s season. There seem to be two lines of thought on this. The first is that Hall has scored at an 87 point pace so everyone who thinks that there’s something wrong with him is insane. The second is a little more nuanced. Hall’s scored an extraordinary amount of points on the 5v5 goals that he’s been on the ice for – north of 100%. Points are valuable information because of what they usually mean. They’re generally pretty tightly related to the number of goals that a team scores when a player is on the ice. If a player has a lot of points, his team probably scores a lot of goals when he’s on the ice.
Keep in mind though – points are an imperfect reflection of goals. Just because a player has X points, it doesn’t necessarily that his team will have Y goals scored when he’s on the ice. Hall’s having a bit of a season like this. As of a week or so ago, he was at 2.8 5v5 PTS/60. That’s a pretty incredible number. If he held it for a full season, it would be the 36th best season (min. 500 5v5 minutes) since 2007-08. It’s elite 5v5 point production.
There’s something goofy going on though. Points are important because they mean that goals have been scored. I took the 114 players since 2007-08 with a) at least 500 5v5 minutes and b) at least 2.5 PTS/60 as of about a week ago (confession: I grabbed the data, saved it and I forgot the precise date – it’s very recent) and then I graphed their points/60 against their GFON/60.
Unsurprisingly, as a rule, more points means that a player is on the ice for more goals for than a player with fewer points. I’m misrepresenting the causal connection here – we hand out points after goals and only if a goal is scored, so the connection runs GFON —-> points. The implicit assumption that we make that a player with a lot of points has been on the ice for a lot of goals for is usually a good one.
Not always though. See that dot between 2.7 and 2.9 on the x-axis? The closest one to the bottom of the y-axis? That’s Taylor Hall’s 2013-14. He’s accumulating 5v5 points at a very impressive rate (2.8 PTS/60) but the Oilers aren’t scoring very many goals when he’s on the ice. We’d expect someone with 2.8 PTS/60 to have been on the ice for 3.5 GF/60, a fantastic number. Hall’s at 2.6. If the Oilers were scoring 3.5 GF/60 with Hall on the ice, they’d have an extra 12 goals, which probably buys you another four or five points in the standings. It’s a big loss.
AND there’s more! Guys who score points like Taylor Hall is scoring points are usually on the ice for many more goals for than goals against. Collectively, the 114 players in my group that I’m talking about were on the ice for 6,222 goals for and 4,031 goals against. That’s about 61% of the goals scored when a 2.5+ 5v5 PT/60 player is on the ice. Usually, when we see a player scoring like Hall is scoring, his team is crushing the opposition when he’s on the ice. As of the date of this data, the Oilers were getting 44.6% of the 5v5 goals when Hall was on the ice. That’s up to 47.9% as we enter the Olympic break which is nice but still not what we’d expect from a guy piling up points like Hall.
What’s going on? Well, the Oilers are getting murdered in shot attempts when Hall is on the ice. Put another way, his Corsi% sucks. It sits at 43.5%. The percentages have actually been a bit kinder to Hall than they’ve been throughout his career – the Oilers are shooting 9.8% and getting a .914 SV% with him on the ice at 5v5 this year compared to a 9.0% and .912 from 2010-13. Despite that bump in the percentages, which would usually pay off in a bump in the GF%, the Oilers are scoring just 47.9% of the goals with Hall on the ice at 5v5, down from 50.2% in his first three seasons.
The reason you can’t really see this in Hall’s points – his 5v5 point rate is up significantly over what he accomplished in his first three seasons – is because he has points on 105.7% of goals scored when he’s on the ice this year, compared to a much more normal 77.9% in his first three years in the NHL. If he was getting points on 77.9% of the 5v5 goals that were scored when he was on the ice this year, he’d have ten fewer points, which would put him on pace for about 71 points per 82 games. Nobody would be crowing.
This is all a bit of a distraction because there are a couple of things that are undeniable. 1) Hall’s Corsi% has cratered this year, 2) the Oilers GF% with Hall on the ice has dropped with Hall on the ice, 3) a team’s GF% is tightly tied to winning and 4) you can can’t hope to win big without lines running up a big GF% at 5v5. The Oilers Corsi% with Hall on the ice this year is a problem that needs to be fixed. Fixing it requires answering a question: what’s gone wrong?
We can infer something about the change by looking at Hall’s shot attempt volume:
The bigger part of the change seems to be in his rate of SAF (shot attempts for), which is down a smooth 18% from last year. There’s a slight tick upwards in terms of his rate of SAA (shot attempts against) but it’s not nearly as significant. If you want to read more about this, I wrote about it a few weeks back. I won’t rehash it here – this post is long enough.
* * *
This is a methodology section. If you don’t care about such things, feel free to skip on to the next section (marked by * * *), in which I discuss results.
I wanted to get a sense of how the Oilers were different with Hall on the ice this year at 5v5 than they were in 2012-13. In 2012-13, Hall had a Corsi% of 50.3%. This year it’s 43.5%. The problem I had is that watching 1400+ minutes of 5v5 time wasn’t really realistic for me. In a better world, we’d have incredibly detailed information about what’s happening when a player is on the ice available to us. We’re not there yet, although it’s probably closer than people think. This means that I need to figure out some sort of a decent sample that I can be reasonably confident is representative of Hall’s season as a whole.
Starting with 2012-13, I went through and numbered each of Hall’s 5v5 shifts. This was a bit of a trial and error process, trying to figure out how small I could go in terms of a sample and still produce samples of Hall’s season that were reasonably consistent with the other samples and the whole. I ended up numbering each shift from 1 to 7 sequentially. That gave me the following groups of shifts:
You can see that they’re pretty consistent with each other. I found that if I went much smaller, I started running into problems with sample size. See, for example, what happens if I create groups from every tenth shift:
My five point spread from worst to best Corsi% when I take every seventh shift grows to 14.6 points when I expand it to every ten shifts. Having decided to create groups of seven shifts, I picked group number 4, with a Corsi% of 50.3%, which is basically Hall’s Corsi% for the 2012-13 season.
I repeated this process with the 2013-14 season through January 5, 2014, which was when I first started kicking around how to go about going doing this. I ended up creating six groups for that season, picking one in which Hall was on the ice for 136 shifts in which the Oilers lost the shot attempts 77-97, a Corsi% of 44.3%. The groups looked like this:
I proceeded to knock out any entries when the goalies were pulled. I should have done this beforehand but it doesn’t change things hugely. Hall’s 2012-13 sample ends up with 94 SAF 88 SAA and his 2011-12 ends up with 63 SAF and 91 SAA. That’s a 51.6% Corsi% in 2012-13 and a 40.9% Corsi% in 2013-14. A bit bigger than the real gap but not by much.
Twitter critics were of the view that this post was too long and could be livened up with a cat pic.
* * *
With sample groups of 5v5 shifts from 2012-13 and 2013-14 selected (you really should have read the methodology), the project is pretty easy: go through and watch the shifts making up each group, noting where the puck is in terms of the offensive or defensive zone and how it got there. This gave me ten different types of zone entry:
CS – Successful carry
CU – Unsuccessful carry attempt
DC – Dump and change, used when there was no or negligible pressure following the puck being dumped in
DS – Dump-in where the attacking team achieved control of the puck before it was cleared from the offensive zone
DU – Dump-in where the attacking team did not achieve control of the puck before it was cleared from the offensive zone
SS – Used when Hall came onto the ice with the puck already in the defensive or offensive zone
PE – Used when a penalty expired with Hall on the ice and the puck in the defensive or offensive zone
OZW – Offensive zone faceoff win
OZL – Offensive zone faceoff loss
X – Other. This largely consists of two types of play. First, when the defending team brings the puck into the defensive zone, frequently after a faceoff win in the neutral zone or while regrouping. Second, when the attacking team sends the puck into the defensive zone without any intent of making an entry. This might included failed passes in the neutral zone or pucks being fired out of the defensive zone or before the red line to relieve pressure.
I’m going to refer to time in the offensive or defensive zone as “possessions”; you should keep in mind that this doesn’t necessarily mean that the puck was possessed by the team enjoying a possession. If, for example, Edmonton dumps the puck in, the opposition recover and clear the zone, I count that as a DU with Hall on the ice. That’s a possession. If you can think of a less awkward term (particularly given that we talk about shot attempts as “possession numbers”), feel free to share it in the comments.
Let’s get into the results, starting with what the opposition did against the Oilers in terms of how they entered Edmonton’s zone.
Boy – isn’t that pretty much exactly the same? There’s a notable decrease in entries to the Oilers’ zone that start as a result of an Oilers’ defensive zone faceoff loss (which is, obviously, an offensive zone win for the other team) but that’s probably a sample thing – it’s not like the Oilers are a team that wins 73% of their defensive zone faceoffs now. The numbers are just bang in line though, year over year.
Of the ten zone entry types that I’m tracking, four of them are kind of oddballs – penalties expiring, shift starts and faceoffs. If you back those out, these are the splits between the remaining six types of entry.
Boy, there really doesn’t seem like there’s a lot there, does there? The entries into the Oilers end with Hall on the ice in 2012-13 are virtually bang on with 2013-14.
What about if we factor in the shot attempts? I’ve expressed this in terms of the number of shot attempts per entry type. Given that the bulk of the entries are carries and that we’ve got a fair amount of oddball categories with relatively few instances of that type of entry occurring, there’s a small sample caveat to a lot of this. As you’ll see from the table, that doesn’t matter too much – as it happens pretty much everything is identical year over year.
The data off dump-ins is just eerily similar. There is a spike in shot attempts allowed when pucks are successfully carried in to the zone that’s worth noting – it’s up by 15% from 0.62 to 0.72. Let’s look at that a little more closely.
You can see that there’s not really a huge change in terms of how often the other team carried the puck in and got zero shot attempts – it’s down just over a percentage point, from 50.6% to 49.4%. It’s really not that much of a difference – we’re talking about 79 total carries into the Oilers zone in 2012-13 and 77 last year, so it’s a case where one more zero as opposed to giving up a shot attempt and it’d be identical. I don’t see much there (although a lot of people tell me on the internet how good Hall is defensively now and he sure seems to be a lot like last year).
The trouble comes once the first shot is taken. This is easier to show with a table showing the percentage of entries that had X SAA but at least 1 SAA.
You can see that there’s a slight decrease in the number of carries into the Oilers zone with at least one SAA that didn’t see subsequent SAA. There are many more three and four SAA zone entries. In 2012-13, 2.6% of the times that the puck was carried into the Oiler zone with Hall on the ice resulted in 3+ SAA on that possession. In 2012-13, that number has increased to 12.8%.
Again, I hasten to emphasize that we’re dealing with a small sample size. That being said, this is pretty unlikely. There’s a small red flag that comes up here to me, a reason to wonder if something in the defensive zone has changed for the worse. That said, I’m hesitant to conclude anything from it without a bigger dataset than the one that I’ve constructed. All in all, the defensive stuff looks pretty similar year over year, with a slight note of caution about the tendency towards multi-SAA zone entries for the opposition.
A final piece of information about the defensive zone: the average length of possessions by type. I’ve posted some videos of late of Oilers defensive zone gongshowery; the truth is, that sort of thing happens to everyone on occasion. The more time you spend in your zone, the more likely something ridiculous looking will occur. I’d love to have data like this for every team in the NHL (and we will, sooner than we think) but it’s kind of cool to compare Hall’s 2012-13 and 2013-14.
Again – it’s very similar, in terms of the average length of each possession in the Oilers end with Hall on the ice. The average successful dump-in has nudged up a bit but I took a look and there are a few long possessions skewing things this year. I don’t see anything here that says to me that there’s something particularly different going on.
* * *
I’m going to cut this post off here – it’s already too long. I’ve already done the work for the post on the SAF data and it’s far more interesting. That’s where the change in Hall’s game has occurred. What do I take from this? Well, the uptick in the rate of carries that resulted in multiple SAA is a bit concerning but really, there’s not much that catches my eye here. This year looks a lot like last year.
The offensive zone entries and SAF data will be up in the next few days. It’s a much different story.Email Tyler Dellow at firstname.lastname@example.org