2019/2020 Advanced Stats / Analytics Tracker

[SnuggaRUDE]

Here's the biggest problem, as I see it.

QoC is an endless loop because theres no reference point. If Eichel plays against barkov (59 xGF%) we cant say we quantified barkov, because that 59 could be coming against other good or bad players. So we have to look at who Barkov played, then we have to look at who those players matched up against, and so on.

That's also the reason it comes so close to 50% so much of the time.

What if we assigned players an ELO rating and adjusted it based upon their xGF? It should converge quickly. Determine the K-value by TOI against each-other?

 

[jc17]

What if we assigned players an ELO rating and adjusted it based upon their xGF? It should converge quickly. Determine the K-value by TOI against each-other?

I dont really know much about ELO. I think linemates would still be a hurdle, but again I dont know much so it might not be an issue

 

[Der Jaeger]

I’m not ready to take much of the advanced stats into account just yet. A ten game sample size is too small. Especially with the Kings game, which skews things a bit.

I’ll start paying more attention around Thanksgiving.

 

[SnuggaRUDE]

I dont really know much about ELO. I think linemates would still be a hurdle, but again I dont know much so it might not be an issue

It's not clean; ELO is really optimized for 1 v 1 competition, or team vs team. But a lot of applications use it; League of Legends has ELO ratings for a 5v5 game.

 

[jc17]

It's not clean; ELO is really optimized for 1 v 1 competition, or team vs team. But a lot of applications use it; League of Legends has ELO ratings for a 5v5 game.

Each player gets a rating or 5v5 with each team getting a rating?

 

[Dubi Doo]

The last few games they've looked pretty meh. I'm assuming the stats reflect this?

 

[Chainshot]

The last few games they've looked pretty meh. I'm assuming the stats reflect this?

Yep.

 

[SnuggaRUDE]

Each player gets a rating or 5v5 with each team getting a rating?

You can do all of those things. A team can have rating fivethirtyeight follows them for the NFL regularly. A player can have a rating (538 tracks them for QBs). Ideally unique to hockey I'm wondering if we could create line ratings. Each time a goal is scored they 'beat' the opposite line. Naturally there would be a lot of ties.

 

[sufferer]

Seems pretty accurate from what I've seen out on the ice.

 

[jc17]

You can do all of those things. A team can have rating fivethirtyeight follows them for the NFL regularly. A player can have a rating (538 tracks them for QBs). Ideally unique to hockey I'm wondering if we could create line ratings. Each time a goal is scored they 'beat' the opposite line. Naturally there would be a lot of ties.

Oh I see. I think the difficulty would still be the fluidity of things. Like if Sabres Line 1 gets on the ice while Leafs L3 is on the ice, but then Leafs switch to L2 15 seconds later, does that count or get thrown out. Or similar scenario, Leafs change lines but the center from L3 gets stuck on the ice, and its a hybrid of players who rarely play together when one of the teams scores, how does that work? D-men also are either in the mix. I think its a good idea, just so many complications because its not as matchup based as pitcher vs batter, qb vs defense, etc.

 

[jc17]

Off topic, but I dont think I love the term analytics being used so broadly in the public, regarding hockey. I think it over-complicates things and turns people off immediately.

Things like scoring chances are not advanced. They aren't something that is numbers vs eye-test. They literally rely on the eye-test to track. So assuming that's not an issue (which it was earlier this season) I wish certain things like that could be discussed more often without being dismissed because the analytics title.

 

[SnuggaRUDE]

Oh I see. I think the difficulty would still be the fluidity of things. Like if Sabres Line 1 gets on the ice while Leafs L3 is on the ice, but then Leafs switch to L2 15 seconds later, does that count or get thrown out. Or similar scenario, Leafs change lines but the center from L3 gets stuck on the ice, and its a hybrid of players who rarely play together when one of the teams scores, how does that work? D-men also are either in the mix. I think its a good idea, just so many complications because its not as matchup based as pitcher vs batter, qb vs defense, etc.

It's not perfect no, but it would be much better than we currently have. To answer the time on ice question, it would also be possible to normalize matchups by x/60. In your example you'd track the time against L3 and L2, performance against each would accumulate in their own matchup.

Again, I'm nearly positive things like this exist in the matchmaking AI for online gaming. There's a huge incentive to build 50/50 match making online.

 

[Chainshot]

 

[joshjull]

It's not perfect no, but it would be much better than we currently have. To answer the time on ice question, it would also be possible to normalize matchups by x/60. In your example you'd track the time against L3 and L2, performance against each would accumulate in their own matchup.

Again, I'm nearly positive things like this exist in the matchmaking AI for online gaming. There's a huge incentive to build 50/50 match making online.

But L3 and L2 aren’t against each other in a vacuum. There are also d-pairs involved. Since there are 4 lines and 3 d-pairs ice time together will not be even. As in line 2 and d-pair 2 will not play together all the time. That’s of course assuming the forward lines and d-pairs stay the same all game. Which isn’t always the case.

Normalizing by x/60 would be hard to do. No player faces any other player , let alone 3 or 5 player combos, for 60mins due to the schedule. (4 games vs division, 3 vs conference non-division and 2 vs other conference ). Add in the impact of injuries, coaching decisions to change up lines and differences in ice time based on line assignments and you end up with incredibly small sample sizes of any 3v3 or 5v5 matchups.


Using Jack vs the Panthers again.

Jack at 5v5 in that game (4 games projection)

16:08 in total 5v5 minutes (64:32mins)
10:25mins out with Barkov (41:40mins)
6:08 —> VO/Jack/Sam vs Huberdeau/Barkov/Hoffman (24:32)
2:00mins ->VO/Jack/Sam vs Huber/Barkov/Dadonov (8mins)

All of the above obviously assumes things I cant really know like no injuries, lines together, etc.

The 3v3 line samples are very small and thats for two lines of players getting the most 5v5 ice time For their teams in 4 divisional games. The samples against non divisional opponents and for lesser 5v5 players would have even smaller samples to work from. Like Mitts vs any lines on LA.

The League of legends comparable seems off the mark a bit assuming I understand it correctly. If I don’t please correct me. But as I understand it the League of Legends stats would be more “pure” than hockey numbers. As in a 1v1 matchup is a straightforward 1v1 matchup. No other players are involved. Same with the 5v5 matchup. Its those 5 players vs the other 5 from start to finish.

 

[Chainshot]

 

[jc17]

Bump: Wanted to look at evolving wild's GAR before it goes private now that we're a few games in and its odd to say the least. Ultimately I think it shows why the metric isn't good for predictive value.

Here's the RAPM. Couldnt fit asplund but hes 3rd in xG+- in this metric:

 

[Kyndig]

Sabres are 1st in the NHL in crossbars hit with 7. Says we only hit 12 posts which I call BS on because I was keeping track of it at one point but alas..the combined stat has them at 7th in the NHL and 4th in the NHL through the month of November. Unlucky.

 

[jc17]

Sabres are 1st in the NHL in crossbars hit with 7. Says we only hit 12 posts which I call BS on because I was keeping track of it at one point but alas..the combined stat has them at 7th in the NHL and 4th in the NHL through the month of November. Unlucky.

Add another for Sam

 

[Kyndig]

Add another for Sam

and another for Olofsson...and Reinhart again.

 

[Kyndig]

Proof that the stats are a lie. We very clearly hit the post/crossbar 3x today and only got credit for one...(or its a late addition from yesterdays game?)

Edit ~ it was a late addition from the first Leafs game. It was recently updated.

Sabres tied for 2nd with most posts/crossbars hit in the month of November with 15...onto December and hopefully some better luck.

 

[dotcommunism]

I don't know where post/crossbar numbers are coming from or the methodology, but I suppose it could be a matter of what part of the post or crossbar gets hit. For example, a puck that hits the outside of the post isn't going in, but with a puck that hits the inside it may or may not go in based on how it bounces, which is difficult to predict.

 

[Chainshot]

 

[Jim Bob]

Hockey advanced stats primer, Part 2: How can (and should)...

This is a good article for those that subscribe to The Athletic.

 

[Chainshot]

 

[SnuggaRUDE]

FFS Tierny could you make some of your graphs X=Y these axis are so non-intuitive.