"Advanced" Stats and the Problems Around Them

As some of you may remember, Zach Parise came out last year against the dump-and-chase and in favor of carrying the puck into the offensive zone. This was followed by the NHL adding some "advanced" stats to its website.

On the other hand, Paul Bissonnette wrote an article for the Players' Tribune in which he states: "Ultimately, teams don't draft based on Corsi or possession numbers. You draft a player because you've watched how they perform on the ice and then consider their potential to improve."

In short, Biz-Nasty isn't in favor of the advanced stats. For this response, Bissonnette was lambasted by much of the hockey community (including myself) for his lack of understanding and conformity. Similarly, Patrick Roy has been an outspoken opponent to advanced stats, and has been similarly derided for it by writers (again, including myself).

So, we have some players and administrations in favor of the advanced metrics, and some against. The issue is more complex than that (considering sometimes they say they are against it but are really for it), but the question becomes: should NHL players and front-offices care about advanced stats?

This question has a few facets, the first being what the point of statistics are. Statistics serve two purposes: describing what has happened, and predicting what will happen in the future. While a set of data can't completely predict future performances, both kinds of stats can be useful for an NHL team, but they need to be used in the correct ways.

Descriptive

Statistics that describe what happened in detail don't have a lot of use for teams in the NHL. On a basic level, they know what happened, and that's what's important. If a team is outshooting their opponent and losing, they can probably say that they got unlucky, or the other team was lucky. Perhaps the goaltender had a bad night... whatever the case, the team already knows it.

On a slightly deeper level, there are some descriptive stats that can point out trends to teams that are helpful. If, for instance, you notice that your team is taking 5 penalties per 60 minutes of even strength play on average, that's bad and you probably want to mention it. At the same time, a GM or coach probably already noticed that they were getting called a lot and will bring it up.

There's an old Yiddish expression: "To a worm in horseradish the world is horseradish." To an NHL team, hockey is their horseradish; basic things like that probably aren't going to surprise them. In other words: trends that I or others may find through looking at statistics are inherently visible to the team, simply because they're paying attention all the time.

Predictive

Here is where coaches and GMs can really use some of the advanced stats that are out there; knowing who to sign and who to let go. There are a few problems, but ultimately it's rather easy to tell who is worthwhile and who isn't. Bissonnette hits on one such snag: someone playing with Jonathan Toews might be a good player, or he might just be benefiting from Toews' presence. There are ways to isolate players, but it is something GMs need to keep in mind.

Indeed, predictive stats have become ubiquitous enough that there is a team every year who seems to be magic before a collapse. Last season's Flames consistently performed poorly in statistics that are meant to predict future success before collapsing out of the playoffs. The year before, the Avalanche were the darlings of the NHL before losing in the first round. In 2011-12, the Minnesota Wild collapsed from leading the league in December to missing the playoffs.

Predictive stats are not as useful as they might be, however. While superstars are easy to spot, in the NHL teams maintain a good amount of control over a player until later in their careers - often past their prime. While it's nice for GMs to know whether the players they have as a restricted free agents, this doesn't do a lot of good in terms of signing free agents in their prime. Furthermore, true superstars are probably sticking out in more than just the advanced stats, rendering predictive stats unnecessary.

The real benefit of predictive advanced stats would be in identifying good players before they are drafted. While CHL Stats are being gathered, it's not to anywhere near the same extent as the NHL. In other words: while we are evaluating NHL players by metrics which have proven to be more reliable than just goals, wins, and sv%, we aren't picking players for the NHL team using those same metrics.

It is entirely possible that NHL teams are tracking shot attempts in minor leagues; the NHL is (for some reason) more secretive than other leagues about their "advanced" stats. There has been a recent influx of advanced statisticians into the NHL which indicate teams are paying more attention to numbers beyond wins, losses, and goals.

Rosterbation

One effect of these numbers being readily available is easy to point out: as soon as a trade is made, fans everywhere point to it either as a good or bad move based on per-60 stats, or possession metrics. When Chris Stewart was signed, he was bemoaned as a waste of a traded draft pick based on his predictive stats. When Sean Bergenheim was signed, there was a lot of excitement over his possibly being a "diamond in the rough."

I didn't choose those examples by accident. There was significantly more excitement over Bergenheim than Stewart, yet who were people more reluctant to let go? This illustrates an important point about the NHL's "advanced" statistics: they are far from perfect.

Obviously there's no way to predict success or failure perfectly, but there are some major holes in the NHL's current stats. Firstly, luck still plays a huge role in generating a players' numbers. A players' shooting percentage can vary wildly from season to season (see Pominville, Jason). Which is the true player? Without a long career, we can't know, and by the time there is enough data to say for certain, it's too late to do us any good.

Context

In the end, the problem always comes back to context. Sometimes, a stat can hide a player's flaws because of something it doesn't take into consideration, like a linemate. We've developed WOWYs (With Or Without Yous) to help isolate linemates and eliminate that problem. There is one context that there's almost no way to adjust for, however, and that is coaching.

Coaching makes a huge difference in the NHL; more than many people realize. Mike Babcock famously spurned the Red Wings to move to Toronto recently. Though roster changes have been made, the team is largely the same as the previous year. Therefore, we can say with some confidence that any changes in the team's performance are largely due to coaching. Last season, Toronto's possession stat was 4th-worst in the league (46.4% shot-attempts for). This season, they are at 11th-best in the league with 50.9% (better even than Minnesota).

Toronto has not had a great start to the year, but at least part of that is due to their low PDO- a rough measure of luck- which is even worse than last season's.

This coaching effect can matter on an individual level, as well. Perhaps a player's skill set is perfect for one coach's system, but terrible for another's? When Thomas Vanek came to Minnesota, there was much hand-wringing over his lackadaisical defense in such a sturdy defensive system- that Vanek seems to be doing just fine speaks to our misunderstanding both of coach's systems and how to evaluate a player's "fit."

Yet another confounding factor in the whole coaching mess is that coach's are not robotic creatures immune to change, but can adjust to make their system work for the players they have.

In short: while the stats we have are very useful for some things, there is a long way to go before we can rely solely on spreadsheets. Even if we could, that would miss the fun of the game we analyze so much.