Bc people argue (totally not without base or facts to support it tbh) that Val Nuke>Drai, then people look at points and say that can't be true (even though Drai scores nearly 50% of his points on the PP and is just brutal defensively, even by eye test) also Fox>Makar, Fox isn't impressive by eye test, but I've actually watching games focusing on him, and he makes great plays shutting down teams at the blueline with just a poke check or jump on a pass (plus his RAPM is insane... Why I've went back and watched him over the past few weeks)

Those are for lists for the most part though, I could definitely get behind them

yeah like Draisatl offensively is very good, but Panarin has been an absolute monster all over this season and that is what makes him the most valuable player in the league

Cause no advanced stat model is perfect. Some are good indications but they never tell the whole story. Different metrics can be tweaked and improved but they will always be approximations of the skill/ability they are trying to measure.

IMO you need both. If you rely on just one of the other you’re not 100% in the right. The problem is too many people don’t understand advanced stats so they don’t know how to use them or what they mean. And on the other side too many people blindly follow advanced Stats like they are then end all be all.

As I said, you need a good mix of both to form then best lists of players.

So many "advanced stats" try to distill a players value down to one number. That's a mistake, IMO.

Also, everybody and their dog can come up with a method of evaluating players, just because it's out there doesn't make it valuable. Most of them fail to account for quality of competition adequately, so that 3rd pairing defensemen often have better GAR (for instance) than 1st pairing defensemen on the same team. It doesn't mean they are better.

I prefer to look at the raw data from NST myself, and look for trends and patterns in a player's results. NST also has non-"raw" stats like xGF and GSAA which is great, but if you look you'll notice that these stats differ from site to site. They can't all be correct, so I understand why people mistrust them.

What's really funny is that the analytics community watches the most hockey of anyone because they're natural skeptics. The typical person who is actually deeply trying to understand hockey is super concerned that the data itself matches up with the game on the ice. We're constantly tweaking and adjusting to make sure that the numbers themselves are as close to resembling the entire point of it all: winning games. It's all about that.

It also informs how you watch the game. You look for things like possession through zone exits and entries, how different defensive structures inform the the type and volume of shots a team gives up, etc.

Eyes and Data are not 50 / 50. They're both inherently flawed. Most people who watch the game are prone to puck and hit watching and frankly don't really know what to look for while watching the game, especially away from the puck because it's fast and keeping track of 10 guys on the ice is tough. They remember the big stuff and forget the small stuff, and are prone to massive recency bias.

So you start with the data and use the game tape to fill in some of the gaps in the data.

The other problem is people utilizing the data poorly. Some common misapplications: overvaluing quality of competition (it means almost nothing in aggregate) and undervaluing quality of teammates (it means a ton), extrapolating too much from small samples, looking at things in aggregate that should really only be evaluated on a /60 basis, etc.

But the weirdest thing is people saying "watch the games!" and then using point totals to make arguments. If you want to make arguments with goal or point totals, you're making a statistical argument - clearly you're admitting that you value using data to evaluate players, as do we. So don't get angry that people found a better predictor for future goals than goals themselves lol.

I'm pretty sure natural stat trick has a program that counts the stats, locations and everything necessary. I think sites like Hockey Reference just use the stats that the NHL records and those are almost always off. The people that the NHL employ to record everything are horrible, people that teams employ are also terrible. Both New York teams are notoriously bad at bumping up their own teams stats. A few months ago I kept track of a NYR game and they gave the rangers almost 15 more shot attempts for. They're also very lackadaisical about shot locations.

Well, what does tell the full story then? Certainly not the eye test as fans of the same team have different opinions of the same player.

Are you the EvolvingWild guy? lol

Nothing does. There's always an element of luck that you can't quantify with stats and can't predict using the eye test.

I think adv. stats do kind of have those kind of stats in some sense, such as PDO and shooting % which seem rational enough, but I do appreciate you gave a prrtty clear and reasonable answer overall.

I use NST almost exclusively, so that's good to hear. Thanks.

Good post, I just have an issue with this one point. Common sense tells us that comparing (say) the GF% of a defenseman who gets put out against the opponents top line with the GF% of a defenseman who gets thrown over the boards against the opponents bottom-6 is not going to even out in aggregate. A 1st pairing defenseman (by QoC) has far more value to a team than a 3rd pairing defenseman. QoC matters and should be taken into account.

Another critical component that most people miss when thinking about QoC is OTF (On-The-Fly) shift starts, especially for defensemen. Weaker defensemen will get put out against top opponents only when the puck is heading into the offensive zone. The defensemen always change with out-of-zone possession, so having a weaker defense pairing come out at that point (and when the opponent's top line is in the second half of their shift) is less dangerous. The 1st pairing guys did the heavy lifting.

Watch how a coach uses each player, you'll see that they don't use them equally because these things don't even out in the aggregate. They'll almost always use their best, most trusted defensemen against the opponents top lines, even if they have one of their energy / checking lines out at the same time.

Here's a solid article that addresses almost specifically the point you just made:

https://www.vancourier.com/pass-it-to-bulis/why-quality-of-competition-doesn-t-matter-to-analytics-experts-anymore-1.23414544

The context of that article is strictly TOI-based, which in itself is flawed. And in any case, the writer of that article recently submitted this:

https://www.vancourier.com/pass-it-to-bulis/quinn-hughes-is-playing-big-minutes-against-tough-competition-for-the-canucks-1.23975140

"What’s more impressive is who Hughes was matched up against: Connor McDavid. More than half of Hughes’ even-strength minutes came against the two-time Art Ross winner and no Canucks defenceman spent more time on the ice against McDavid at even-strength than Hughes. That doesn’t happen by accident."

"Against the Calgary Flames, Hughes once again racked up minutes: his 17:45 at even-strength was second only to Tyler Myers. A big chunk of those minutes came against one of the best forwards in the NHL: Johnny Gaudreau. No one played more against Gaudreau at even-strength than Hughes in that game."

The manner in which a coach uses a player is significant.

...I'm not sure why it wouldn't be TOI based... conceptually you're talking about "how much time or what % of the time do you go against the oppositions top 6", that's entirely a function of TOI, you can't control how good your opposition is, your coach can only control how much time you spend against your opponent's top 6.

The concept of QoC is certainly applicable on an individual game basis (which is what the writer - who is not himself an analyst, just a writer) conveyed. But over a full season, the matchups get messier. The OTF line matching you described is an ideal, and the reality is a lot sloppier. In practice, teams only like to change dmen with offensive zone possession as you said. and defensive zone exits with possession only occur about 35% of the time, fail around 23% of the time and end up in a turnover or dump out the rest of the time. So because of that, over the course of a whole season, you end up going against all 4 lines a bunch. Not perhaps 100% equally, but enough to render the concept of QoC kinda negligible when evaluating over a full season.

Individual games though are different.

No, I meant using TOI as a proxy for Quality is flawed. There are lots of players that get fairly high TOI while simultaneously being sheltered. It's not a results-based proxy.


Here's an excellent site which gives the breakdown between different tiers. Looking at how a player does against "Elites" is notable, but you have to take into account their FO/60 also (because OTF isn't available on this site). You'll see a marked distinction between top-4 defensemen and bottom pairing guys in CTOI% and FO/60. Or, in the case of a team that hard matches it's top pairing (Boston), you'll see a wider gulf between the top-2 and the rest.

http://puckiq.com/woodmoney?season=20192020&tier=Elite&positions=d&team=bos&group_by=player_season_team