Image via Wikimedia Commons
Corsi, Fenwick, PDO and rate stats have come to the NHL’s official website. Sort of.
On Friday morning, the NHL rolled out a new page of advanced team and player stats with their “Enhanced Stats” resource page. The page itself is clean and well designed, and the stats we’ve been discussing for the past seven years are now officially recognized. Although the metrics have been renamed, the NHL is now officially onboard. That matters.
It’s also the big picture takeaway here, and it’s a significant shift, an important moment in the history of hockey analytics. Blogging is all about optimism and overreaction though, so in that spirit, let’s look into what NHL.com’s enhanced statistic site is offering – let’s discuss what it does well, what it could do better – and give our initial reaction.
The Great Renaming
The NHL has decided to rename several core fancy stats, including Corsi (now known as SAT for Shot ATtempts), Fenwick (now known as USAT Unblocked Shot ATempts), and PDO (now known as SPSV%).
I’m not someone who has any particular attachment to the terms Corsi and Fenwick. I came up reading the irreverent oilers fans archives, not participating in the discussions.
In my own writing I’ve been describing these metrics as shot attempt differential and unblocked shot attempt differential since at least 2013. That said, when putting together a table or a graph, I’ve always continued to use the shorthands: “CF%” and “FF%”.
My first reaction to SAT and USAT is admittedly negative, but that’s probably because the new acronyms seem so unfamiliar. It’s hard to wrap my head around the idea of using “SAT%” in a table instead of the standard “CF%” going forward (though as I write this, it doesn’t seem like such a big deal).
Rebranding the shot based metrics with the goal of making them more accessible to the uninitiated makes good sense in my view. The relevant question then isn’t my first reaction to these numbers, so much as it’s “do SAT and USAT make these numbers easier to understand for the laymen?”
If they do, then five years down the road these terms will be ubiquitous and the original nomenclature will be a footnote. My initial reaction won’t matter. If Corsi and Fenwick persist in the lexicon of fans and media having these conversations on-line though, then it’ll be very strange – and counterproductive – to have the league referring to the stats by a different name officially. That’s the league’s gamble here.
On the other hand, I like SPSV percentage quite a bit. Simple, descriptive, the acronym isn’t a word on it’s own. I’d suggest that it’s a major improvement on PDO, with the added benefit that PDO is the sum of two separate numbers, which should always be regarded separately anyway. The SPSV percentage acronym has the added benefit of making that point internally.
Currently NHL.com’s enhanced statistics page for both teams and players list “SAT” (Corsi) and “USAT” (Fenwick) in a variety of forms – as raw totals, as percentages, and as rates. There’s also a variety of other useful stats here, like primary and secondary assists and penalties taken and drawn. All very cool.
One issue worth noting here though is that the pages have sort of divorced SAT and USAT from the numbers that really matter in hockey: goals and goal differential. Right Mike Babcock?
In my view, hockey’s advanced statistics have a relatively high barrier of entry, in terms of people hearing the phrase “Corsi” and thinking – oh cool, I should know everything about that, it probably matters – in part because the importance of out-attempting the opposition isn’t necessarily intuitive. It requires several leaps of logic to begin to understand that shot attempts are like the building blocks of outshooting opponents, and that over time outshooting opponents is at the heart of outscoring opponents. Outscore your opponents, and you’ll win games, which is at the rudiment of winning championships.
By divorcing possession numbers, rates and differentials from goals for and on-ice shots on goal differentials, I wonder if the NHL.com’s enhanced stats page could unintentionally makes it more difficult for new, fancy-stats curious fans to make these intuitive leaps.
Rate Stats Inconsistency
I switched from /20 to /60 to be consistent with everyone else but now http://t.co/pyNcLm12fW reintroduces /20.
— David Johnson (@hockeyanalysis) February 20, 2015
I’ve always used the per 60 minute rate, partly because of the influence of behindthenet.ca and partly because I like the idea that that a per 60 minute rate captures what x-player would accomplish if they were some sort of robot who could skate on the ice for an entire game at 5-on-5.
For an example, over the last three seasons there are 18 forwards who’ve logged more than 2000 5-on-5 minutes and have scored at a rate better than one goal per 60 minutes of 5-on-5 ice-time. For me it’s a cool concept that there are 18 elite 5-on-5 goal scorers that, if they could play the entire game at 5-on-5, would be goal a game players.
The advantage of the per 20 minute rate is that it more closely approximates what a player accomplishes per game. Except that scoring .33 goals per 20 minutes of 5-on-5 ice-time just means less to me, and is less intuitively impressive, than crossing that 1 goal per 60 barrier.
Game State Stats
First stats site thought: purge close numbers and get score-adjusted numbers in ASAP. http://t.co/T093cNwhM4
— Travis Yost (@TravisHeHateMe) February 20, 2015
In November Micah Blake McCurdy in a seminal piece over at hockeygraphs.com laid waste to the utility of using game state statistics when discussing team quality. Prior to McCurdy’s work, many – myself included – had believed that “Score Tied” or “Score Close” Fenwick and Corsi was the gold standard of predictive metrics, since it helped correct for any potential score effects.
As McCurdy showed decisively, score effect do matter and should be adjusted for, but looking at the underlying numbers in a game state specific manner added nothing to the predictive weight of the shot based metric and was actually harmful to our understanding:
Score-close possession metrics are utterly indefensible for any purpose at any time. Raw measures are preferable for conceptual clarity and for predictivity at almost all sample sizes, and adjusted measures are superior for predictivity at all sample sizes. It is difficult to overstate how important it is that they be purged from the lexicon of all right-thinking people. They purport to distill the essence of possession when in fact they do great violence to data by censoring large tracts of meaningful information and magnifying a smallish portion. Adjusted measures, by contrast, apply small nudges to the raw data—their seeming complexity masks how much closer to raw data they are than ‘close’ measures.
Currently the NHL’s enhanced stats page includes SAT and USAT stats sortable by game-state, which they’ve named “Even”, “Ahead” “Behind” and “Close”.
I’m a bit agnostic here in that I still want to be able to access game state qualified shot based metrics, if only so I can figure out which coaches and teams see their play shift the most when playing with the lead or attempting to comeback. In terms of combining data with video analysis and systems breakdowns, game state numbers can be helpful in terms of helping one decide what you want to look at.
In terms of evaluating true talent and team quality though, the adjusted numbers are significantly superior. At the very least, it would be cool if the league included them alongside their inferior cousins.
This is just a basic usability critique.
Currently there’s no way to filter the data for minutes played, or total games played. So if you want to go look at who leads the entire NHL in shot attempts for percentage (SAT%) you’ll find ten players who have played fewer than 20 games – and none of whom the average fan has heard of – before you get to Pavel Datsyuk.
In comparison, if you filter Corsi-on for 30 games played over at behindthenet.ca, you’ll find Patrice Bergeron and Pavel Datsyuk are number one and two at this, oh and there’s Jonathan Toews just outside the top-five. Maybe controlling the middle of the ice and dictating play matters after all!
Without better filters, there’s just no way that I’ll be visiting the NHL’s official enhanced stats page with the regularity that I visit war-on-ice.com, and hockeyanalysis.com.
As it stands now you cannot find the NHL’s enhanced stats site from their homepage (it’s currently unlisted under stats). Even if you’re on the old standard nhl.com/ice/playerstats.htm page, the advanced/enhanced section isn’t mentioned or accessible.
I suspect that’s because the new landing page is going to be simpler and more accessible. Nhl.com/stats is already live and is a massive improvement visual speaking – with all of the same features, plus enhanced statistics and a smattering of appealing graphs.. When that new page rolls out, I suspect the homepage will be updated.
I mention this as context. It’s proof that it’s probably a bit too early to seriously judge the NHL’s first foray into enhanced player evaluation techniques. Presumably things will be tweaked and improved in the days to come.
As such this preliminary evaluation should be taken in the spirit in which it’s intended: it’s a preliminary review from a content producer who spends endless hours looking at hockey stats on the internet for professional purposes. I’m an outlier in terms of the consumption of this data, which makes me not at all the NHL’s target audience here. I’d like to think it provides me with some insight, however, into what’s useful and interesting in this particular realm.
What remains to be seen, for example, is whether the implementation of an enhanced statistics page is part of a larger redesign. Will NHL.com’s player pages include their enhanced statistics (they don’t yet)? Will official boxscores?
Until we know the precise shape and scope of this initiative we can’t accurately judge what the NHL has accomplished, not fairly or at length, at least.
What we can do is provide our informed feedback and commend the NHL for the attempt. That latter point is crucial here. As with shots, it’s the volume of attempts – in this case a willingness to re-think what works and what doesn’t constantly – that probably matters way more than initial quality.