I’m putting a little bit of hockey analytics history at the top of this one; if you know it already, or don’t care, feel free to scroll down to the first “* * *”. If you’re already aware of the “Everyone got awful at 5v5 except Hall/Eberle/RNH thing this year, you may skip down to the second “* * *”. As always, you may also close the window or otherwise choose to leave this page.
I’ve been interested in hockey analytics for about a decade now. Over that time, there has been an extraordinary explosion in the volume of data available about NHL games. Simultaneously, there’s been a growing body of knowledge that’s developed about how and why things happen in hockey games.
For me, this has been not unlike examining an iceberg. A decade ago, I was interested in stuff like goal difference and seeing how tightly it was related to wins and success in the standings. It kind of became widely accepted in sensible corners that (subject to some caveats such as the influence of the NHL’s silly OT/SO rule and, I suspect, the odd team like the Ottawa Senators that didn’t call off the dogs in blowouts) there weren’t really teams that consistently outperformed or underperformed their goal difference.
From there, things moved towards examining how and why goals happened. At this point, we’re starting to get below the waterline. My own views on this have changed dramatically over the years – I used to think that players could have a significant impact on a goalie’s save percentage and not realize just how prone shooting and save percentage were to random variance over the course of a season. Now, I think I’ve got a better grasp on how, precisely, that works.
Logically, attention turned to shot based metrics: shots for/against, Corsi and Fenwick. I started noticing a few years ago the extent to which some corners of the internet had adopted these metrics when I’d see something, think “I should write about this” and then find, before I got around to it, that someone else already had. That wasn’t a problem for me five years ago. These metrics are now so widely accepted that you’ll hear the panel on Hockey Night in Canada dropping references to them. Sportsnet Magazine ran a graph recently illustrating a point about Fenwick Close (not just a possession metric, but one designed to exclude score effects!). This isn’t a couple of guys in a dark corner of the internet anymore – this way of thinking about hockey has become mainstream.
Over the past few years, people have started to get further into examining the part of the iceberg that’s underwater. Dennis King has done a lot of work on scoring chances, as have others, which have indicated that scoring chances are pretty tightly connected to shot attempts. More recently, people have started counting Zone Exits and Zone Entries and have found interesting stuff. I think we’re starting to get to the point where better individual metrics relating to players are going to be developed and where the thoughtful use of data is going to start enabling the separation of the difference makers from the guys who are along for the ride. This series of posts from me is an effort to kind of assist in pushing that along while simultaneously shedding some more light on what went wrong with the OIlers at 5v5 in the 2012-13 season.
* * *
One of the things that Corsi doesn’t do is give us a lot of information at the team level about how and why things have changed. People can nod to it, say it was good or bad, and then fill in their own narratives about how and why things are the way that they are with a hockey team. I strongly suspect that, twenty years ago, we’d be hearing that the Oilers needed to learn how to play winning hockey because they had so many young players. In fact, that wasn’t the case.
The Edmonton Oilers took a step back in their 5v5 Corsi% in 2012-13, falling from 47.3% in 2011-12 to 44.5%. However, it wasn’t waiting for Ryan Nugent-Hopkins, Taylor Hall or Jordan Eberle to develop that held the Oilers back.
Hall, Eberle and RNH were generally on the ice with one or both of the other two guys, but the point stands: those guys were at the top of the batting order and, while they weren’t an elite first line in terms of their Corsi%, they weren’t bad. When one of them wasn’t on the ice (keeping in mind that they usually played together), things were awful.
The horrific Corsi% of the Oilers when Hall, Eberle or RNH weren’t on the ice was a new feature this year. Here’s the same chart for the 2011-12 season.
I mean, it’s not particularly great, but it’s not nearly as godawful as this year’s numbers. At the start of the season, when I was foolishly saying that I thought a playoff spot was more likely than not, a big part of my thinking was that the Oilers would put up a similar or better performance when one of Hall, Eberle or RNH wasn’t on the ice. With some improvement from Hall/Eberle/RNH, I figured that they’d be in with a puncher’s chance to make the playoffs and that surely the hockey gods would look favourably on our six years of penance. Make it seven, as they used to say in certain offices at RIM.
When you look at the 5v5 Corsi% for the players who were with the Oilers in 2011-12 and 2012-13, the differences are stark for most of them. It’s all the more striking because the Oilers returned an awfully similar team in 2012-13 to the one that finished out the season in 2011-12 and, while some guys are getting on in age, others who aren’t still experienced massive crashes in their numbers.
Unfortunately, this is where the answers that Corsi gives us dry up. We know that virtually all of the Oilers had a worse Corsi% but that doesn’t tell us why or, really, all that much about the how. Punditry abhors a void and wherever factual information is absent, people will fill in explanations founded on their own particular biases. They will demand more gritensity. They will explain that facepunchers are needed to enhance courage. They will tell you that the lineup isn’t big enough and that more size is needed at the bottom of the lineup because size has transitive properties and that a large mammal at the bottom of the lineup will make RNH bigger. Oddly, they’ll tell you that the Oilers need a better goalie even though it wouldn’t seem that goaltending really has much to do with a problem like this. They’ll talk vaguely about systems.
Basically, the people who are in the business of having opinions are in the business of having opinions. They aren’t necessarily in the business of carefully formulating those opinions or formulating on the basis of data that isn’t all that easy to acquire and that they already understand. GAA is easy to acquire. The stuff on BTN is easy to acquire, if outside the area of knowledge with which a lot of opinion providers are comfortable. Beyond that…forget it. It’s easier to just look at the camera with a steely eye and say: “GRITENSITY.”
* * *
A mix of curiosity about why the non-Eberle/Hall/RNH portion of the team collapsed to such a degree this year and general revulsion towards fact-free opining led me to dig into this more deeply, to try and see more of the iceberg. Earlier this year, I wrote a piece on some data that I thought was pretty fascinating dealing with Sam Gagner and Ales Hemsky on a shift by shift basis. I was struck by the fact that they seemed to have a different profile of SAF (I’m going to use Shot Attempts For, or SAF, in place of Corsi For in this piece) than that of RNH/Eberle, who had played together almost exclusively to that point in the season. Their profile of SAA (Shot Attempts Against, rather than Corsi Against) was basically identical, which suggested to me that the issue that Gagner/Hemsky were experiencing lay further from their own net than people might think.
I thought I learned enough from the Hemsky/Gagner piece that it was worth getting a lot deeper into this data. This involved both looking at the entire team and comparing what happened last year with what happened this year. This required assembly of a spreadsheet with all 43,383 5v5 shifts (give or take a few) taken by players on the Edmonton Oilers in 2011-12 and 2012-13. I screened out shifts in which one team did not have a goalie in an end of the game situation, in order to avoid skewing the data.
I didn’t bother screening out shifts on which the goalie was pulled on a delayed penalty, on the grounds that those will be spread more evenly throughout the team than empty net chances at the end of the game, which will be concentrated amongst specific players. In addition, the only shot attempts in my data base are those with two goalies on the ice – basically, including that time will bump my percentage of shifts on which nothing happened ever so slightly but not enough to be worth the time picking through the data to remove those shifts. I’ll do what the business types would call a sensitivity analysis on this assumption at some point but for now I’m pretty comfortable with it. Once I had the database of shifts, it was simple enough to have Excel count the number of SAF and SAA on each shift.
A preliminary word about shifts and why it makes sense to break the game down this way. Frustratingly, NHL.com does an exceedingly poor job of separating hockey into game states: 5v5, 5v4, 4v5 etc. They just mash a lot of data together, which makes it kind of useless for analytic purposes. One of the interesting things about digging into this data is that you can see how tightly grouped players are in terms of their 5v5 shift length.
In 2011-12, the average 5v5 shift length for Oiler forwards (min. 200 5v5 shifts) ranged from 38.2 seconds (Eric Belanger) to 45.5 seconds (RNH and Teemu Hartikainen). Belanger’s number is a bit misleading – he had a disproportionately high share of shifts that were less than 18 seconds in length. Presumably, this is because he would be put on the ice for defensive zone draws and then change as the puck left the defensive zone. In addition, he would frequently be credited with a 5v5 shift as he left the ice after the end of a penalty kill. Darcy Hordichuk was the only other Oiler forward in 2011-12 to average fewer than 40 seconds a shift. It’s a very tight spread.
In 2012-13, the same phenomenon is apparent. Belanger again averaged the shortest shifts amongst forwards, at 37.4 seconds. His share of short shifts was even higher this year; again, that’s the lot in life of a defensive specialist. RNH again had the longest average shift on the team, at 44.6 seconds per 5v5 shift.
The defencemen are even more tightly grouped than the forwards. In 2011-12, their shifts averaged from 39.9 to 43.6 seconds in length. This year, their shift length was up slightly but they were again tightly grouped: Corey Potter and Nick Schultz took the shortest average shifts, at 42.1 seconds; Jeff Petry and Justin Schultz took the longest average shifts at 44.8 seconds.
Given how small the differences are between shifts, in terms of average length, are and given the fact that the shift is kind of the building block of a hockey game, it makes sense to me to look at things through this lens and see what there is to find out about about how the shifts that the Oilers played in 2012-13 differed from those that they played in 2011-12 in terms of SAF and SAA. I’ll start with the forwards and a look at SAA in this post and move on from there is subsequent posts.
* * *
When Corsi% craters, it’s common to figure that a team got worse defensively and spent more time with the puck in the defensive zone. This is generally a reasonable thought – Corsi% is just SAF/(SAF+SAA). As a first step towards testing that with the Oilers, I’ve prepared graphs of the percentage of shifts for forwards who played on the team in 2011-12 and 2012-13 on which there was at least a single shot attempt towards the Oilers net.
This seems important to me because it’s kind of the first level of defence. If the other team isn’t getting an SAA during your shift, it means that you either had the puck or that you played well enough defensively that they weren’t able to accomplish anything with possession. For ease of reference, I’ve split these guys into three groups: guys who had basically the same proportion of shifts on which there was an SAA, guys who saw a significant increase in 2012-13 and guys who played fewer than 100 shifts in one or both of the years.
First, here’s the group of players who saw basically no change in the proportion of their 5v5 shifts on which the opposition got an SAA.
Teemu Hartikainen played only 288 shifts in 2011-12 and 278 shifts in 2012-13; in light of that, the big shift in his numbers isn’t that noteworthy. The presence of Ryan Smyth, Shawn Horcoff and Ales Hemsky is somewhat more interesting. These three players saw their Corsi% fall this year; it wasn’t because they were more likely to have a shot against during a shift than they were last year. Eberle and RNH ran in place; given that their Corsi% was up slightly, this doesn’t seem unusual.
The second group is forwards who played at least 100 shifts in 2011-12 and 2012-13 but saw a significant increase in the number of shifts on which the other team had an SAA. Everyone in this group saw an increase of at least 3.1 percentage points in terms of the percentage of shifts with at least one SAA.
By far, the most surprising name on this list is Taylor Hall. Hall had an excellent season but did see this number move up from last year. Gagner’s presence is, in one sense, unsurprising, given the discussion about his defensive zone coverage that took place this year but it seems somewhat odd that he and Hemsky had such different seasons, given that about 45% of Gagner’s ice time was with Hemsky. One possibility, which I’ll explore in the future, is that Gagner’s numbers are affected by his changed role this year. Gagner basically didn’t kill penalties on the 2011-12 Oilers; he did on the 2012-13 Oilers. A penalty kill that ends becomes a 5v5 situation, generally in the defensive zone and you’re generally effectively at 5v4 for a few seconds while the the player who went free races back to the defensive zone.
Given the Corsi% posted by the Oilers fourth line this year, it’s unsurprising that Ryan Jones, Anton Lander, Eric Belanger, Lennart Petrell and Ben Eager saw the percentage of shifts on which they were on the ice for a SAA spike over last season. A more interesting question is why that happened.
I’ve put together a data table with this information and included an extra piece of information: the ZoneStart, or percentage of shifts that started with an offensive or defensive zone faceoff that were in the offensive zone. I would expect that this would have a significant impact on the likelihood that there would be a SAA during a 5v5 shift – if a player is starting in his own end of the ice significantly more often or less often, it affects the number of defensive zone faceoffs that he’ll lose, which tend to lead to SAA. I’ve highlighted a few of the more interesting players.
RNH and Eberle got much more favourable ZoneStarts in 2011-12 than then they did in 2012-13. With that context, the very slight increase in the number of shifts on which they allowed at least 1 SAA seems impressive – the ice got steeper for them and they didn’t show any significant decline in the number of shifts on which they allowed a shot.
Belanger, Petrell and Eager would appear to have at least some explanation for the increase in the volume of shifts on which they allowed at least 1 SAA. All three of them suffered a drop of 9.5+ points in their ZoneStart – that’s a significant change that will affect Corsi and, presumably, affect the number of shifts on which a player suffers an SAA. It has been correctly pointed out to me that, statistically, this idea doesn’t seem to hold up, in that other players who suffered big drops didn’t suffer big increases in their percentage of shifts with one or more SAA. I’m going to look into this in some more detail at some point but for now, I simply note this as a potential partial explanation for the big fall in their numbers.
For the sake of completeness, here are the Oiler forwards who didn’t have at least 100 5v5 shifts in both of 2011-12 and 2012-13.
Let’s go to the next step. So you’re in your own end and you’ve given up a SAA. Now what? Well, ideally, you’ll stem the bleeding and not give up another one on that shift. Regain possession and move the puck out. How can we look into this? Well, my idea is to look at the number of multi-SAA shifts as as a percentage of shifts with an SAA. If a team always regains possession and moves the puck out after an SAA, they’d have zero multi-SAA shifts. If they never regain possession, every shift would be a multi-SAA shift. It’s effectively a measure of ability to stem the bleeding.
The change in these numbers is of particular interest to me this year because of a post that Derek Zona wrote at Coppernblue. Zona was of the view that the system in the defensive zone was the problem for the Oilers. This is the gist of his theory;
Since the numbers weren’t working, I kicked around some theories and started focusing solely on the struggling centres while watching night in and night out to see if I could recognize anything amiss in the zone.
What I picked up over a number of games was that I believe Ralph Krueger’s defensive system is asking too much of the Oilers’ centres in their own zone. The pivots, all of them, are playing the entire ice from sidewall to sidewall and through the slot in a combination with the two defensemen. The wingers, both of them, are allowed to stay very high in the zone, much higher than in Craig MacTavish or Tom Renney’s schemes.
Typically Renney used one forward high in the zone and a second supporting the puck or patrolling the middle of the zone. In Krueger’s system, the forwards were literally watching the centres lose 1-on-2 puck battles along the walls, sometimes just a stride or two away while they ferociously guarded the point shot most times and the boards and glass others. Even when the forwards were in the correct spot to support puck movement, they weren’t getting to the puck. And they were leaving the centres on an island.
It sounds compelling. There was video and everything. The problem with that, of course, is that we’re all susceptible to having our eyes fool us. There’s so much stuff happening in a hockey game that even cautious observers can see what they want to see. If Derek’s theory is right, I would expect the percentage of multi-SAA shifts to have generally risen this year for the Oilers, as wingers who weren’t in a position to support the puck let the opposition stay in the offensive zone for a longer period of time due to an inability to obtain possession and exit the defensive zone. Is that what happened? Let’s look at the 14 forwards who played at least 100 shifts in 2011-12 and 2012-13.
I’ve prepared two charts. The charts are sorted from largest increase in multi-SAA shifts to smallest. Here they are.
Right away, something should jump out you: given that a single SAA occurred, the Oilers look to have been generally better at preventing multi-SAA shifts this season than they were in 2011-12. Only five of their 14 forwards who played 100 shifts in each season did worse at this this year and, outside of Paajarvi, the increases are pretty negligible. Shawn Horcoff had the second biggest increase, moving from 36.3% to 37.8%. Just to give that a sense of scale, it’s an extra 3.5 multi-SAA shift over his 31 games – about one extra multi-SAA shift every ten games. It’s not a particularly massive difference.
In fact, most of the Oilers forwards were less likely to experience a multi-SAA shift given that they’d allowed a shot already this year than they were last year. This, I think, casts doubt on Zona’s theory – if he was right, I’d expect it to show up here and it doesn’t. Instead, what we seem to have seen is an Oilers team that, while it was more likely to have shifts on which it gave up shots, wasn’t particularly worse at preventing multi-SAA shifts once they had allowed a shot attempt on a shift.
Just for a sanity check, let’s look at the multi-SAA shift rate using total shifts as the denominator. This changes the question a little bit, from “Given that the Oilers had already given up an SAA, did they get markedly worse at preventing that from turning into an SAA?” to “Overall, did the Oilers get worse at preventing shifts on which they allowed multiple shots attempts?” The numbers for the 14 forwards are pretty stark. Again, I’ve prepared two graphs for the 14 forwards who took at least 100 shifts in 2011-12 and 2012-13 and sorted the graphs from biggest increase to biggest decrease.
Once you get past Gagner, the change is pretty minor. Again, some context using Horcoff: he had an extra 4.5 multi-SAA shifts this year, an extra one every 6.9 games. It’s not a massive change. You can see that there’s kind of a breakpoint between he and Gagner; Gagner experienced an extra multi-SAA shift every 2.6 games this year.
So, as I see it, Paajarvi, Eager, Belanger and Gagner got worse at preventing multi-SAA shifts by an amount that concerns me and warrants further examination. Horcoff, Hemsky and Hall got marginally worse at preventing multi-SAA shifts but not by a particularly huge amount. The rest of the forwards got, uh, better at this than they were in 2011-12, which kind of makes it difficult for me to figure that there was something in the defensive zone causing the problems.
* * *
I’m going to end this first post here – there’s at least four or five of these to come, although they’ll be shorter than this one, which had a lot of preamble that I won’t need to repeat. On the SAA side of things, there are five guys whose seasons I’m really interested in looking at further: Gagner, Petrell, Belanger, Eager and Paajarvi. These guys have caught my eye because their SAA performance got markedly worse in one of the areas examined above. Then there’s a kind of second tier of these guys, basically Jones, Lander and Hall, in whom I have some mild interested given some slight decline but nothing major. Horcoff, Smyth, Hemsky, Eberle, RNH and Hartikainen? I don’t know that there’s a lot to see there as far as SAA goes. They don’t seem to have performed particularly differently than they did last year.
The only conclusion I’m really in a position to draw at this point (and I emphasize that this is my first time poking around in a data set like this and that I don’t really believe anything too strongly at this point) is that I’m skeptical of the extent to which changes in defensive zone systems caused the Corsi collapse for the Oilers. If they did have some impact, it doesn’t seem to have been teamwide. From a forward perspective, the SAA changes from last years seem to have been limited to a discrete group of players. There’s no exclusive characteristic shared by the group who weren’t as badly affected – Horcoff, Smyth and Hemsky are older guys who are thought to be in decline, RNH and Eberle aren’t. There’s nothing that jumps out at me as an explanation.
More generally, outside of Gagner, Belanger, Eager, Paajarvi and, to a lesser extent, Petrell, I’m kind of having a hard time linking the Oilers Corsi% collapse to issues in the defensive zone. Next up, I’ll see what I can find in the offensive zone.Email Tyler Dellow at firstname.lastname@example.org