Fuzzy Analytics

[diceman934]

A little food for thought:

A coach can have a large affect on a players advance stats....if the coaches system is anti advance stats a player who can be a good possession stat player may look below average.

Stats like plus/minus giveaway/takeaways etc have a place when used within a team .....but not when comparing players from one team to the next. The same can be said with advance stats as systems play a huge role in individual stats.

I think if you compare Gardiner with Carlyle and Gardiner with Babcock you will see a pattern that you can see with all Leaf players who played under both coaches.

There are no stats that we as fan are not exposed to.....it is their applications and how each stat is used and the context that a team uses that separates us from them.

Data is also only as good as the person collecting them. Giveways are a stat that I have collected for as long as I can remember and I apply my own standard for them. A giveaway is not simply a bad pass.....a pass to an area that under the system we are playing should have been a completed pass, is not a giveaway, this is a system break down. If a player misses a rotation this effects giveaway, possession stats as well as corsi etc. It is these stats that teams collect that you will not see published as they are collect for the team only by a person who is applying context to each stat.

 

[Nithoniniel]

I've been watching hockey long enough to trust my own eye vs a spreadsheet littered with untrustworthy data. I don't need a computer to tell me Peter Forsberg was a good player. I could tell by watching him dummy teams.

If you could just go by stats, you wouldn't need scouts or even GMs for that matter. You would just enter the data into a computer and it would tell you if a trade is fair or which players are the best. Everyone would have the same list and value the players the same way.

Like I said, stats are fun for us hockey nerds, but I guarantee there will be a long list of "analytics guys" on the unemployment line when they realize their faulty data led them astray.

Data is data, it doesn't inherently have traits such as "untrustworthy". Just like having watched a lot of hockey rids you of flaws that every human share. As long as you are human, you have human flaws such as cognitive bias.

Nobody has talked about going "just by stats". I don't think I've ever seen anyone advocate this. It's about whether or not you include more information in your process.

The only way data would lead you astray is if you don't understand what they represent. What do you base your opinion on that the analytic personnel hired by NHL franchises can not understand the data that they make a living on, and in many cases have an education on?

 

[Mr Hockey*]

A little food for thought:

A coach can have a large affect on a players advance stats....if the coaches system is anti advance stats a player who can be a good possession stat player may look below average.

Stats like plus/minus giveaway/takeaways etc have a place when used within a team .....but not when comparing players from one team to the next. The same can be said with advance stats as systems play a huge role in individual stats.

I think if you compare Gardiner with Carlyle and Gardiner with Babcock you will see a pattern that you can see with all Leaf players who played under both coaches.

There are no stats that we as fan are not exposed to.....it is their applications and how each stat is used and the context that a team uses that separates us from them.

Data is also only as good as the person collecting them. Giveways are a stat that I have collected for as long as I can remember and I apply my own standard for them. A giveaway is not simply a bad pass.....a pass to an area that under the system we are playing should have been a completed pass, is not a giveaway, this is a system break down. If a player misses a rotation this effects giveaway, possession stats as well as corsi etc. It is these stats that teams collect that you will not see published as they are collect for the team only by a person who is applying context to each stat.

There is some interesting points in here. If I told you Gardiners advanced stats were better with Carlyle, than with Babcock would you believe me?

 

[blueberrie]

I've been watching hockey long enough to trust my own eye vs a spreadsheet littered with untrustworthy data. I don't need a computer to tell me Peter Forsberg was a good player. I could tell by watching him dummy teams.

If you could just go by stats, you wouldn't need scouts or even GMs for that matter. You would just enter the data into a computer and it would tell you if a trade is fair or which players are the best. Everyone would have the same list and value the players the same way.

Like I said, stats are fun for us hockey nerds, but I guarantee there will be a long list of "analytics guys" on the unemployment line when they realize their faulty data led them astray.

You can use both.. it's not either or.

I watched Peter Forsberg score a lot of goals in the NHL. Does that mean there is no value in keeping track of them to know he scored 249 over 708 GP?

 

[cookie]

For hockey teams, the numbers and models they use need to have a predictive value associated with them. I don't think that teams nowadays are making judgments too heavily based on the eye test and what people are saying regardless of whether they're bloggers, panel experts, or even scouts themselves. Game has changed quite a bit from when many of the front office players played. Nevertheless, if everyone is using similar models and arriving at similar values, one of the ways to gain an advantage over the competition would be to get the best numbers.

Personally, I don't think that it's possible for a wholly predictive model and as such, the issue spills over to the quality of information that's being generated. There's a saying that misinformation is worse than no information at all and I think that that's where the eye test becomes increasingly more useful.

I think that's where the issues are based. Honestly, how predictive are these figures? What value is there in hitting the smaller (and ultimately near useless) signings out of the park and missing out on the bigger signings? How relevant are the end results at the end of the day? How does one account for player-coach and coach-game differences (ie some players playing better for some coaches due to systems implemented, the role given,etc., and some coaches' decision-making being better than others) when a ton of the models are player and team-focused.

 

[Warden of the North]

The fact you think there's some sort of magical formula underscores your green-ness with analytics. There are so many different metrics and so many contextual variables to account for. Analytics help inform decisions, that's it and that's all.

Red Wings never had a classical tough guy but they did have some jam. In recent years, gritty guys like Abdelkader, Ericsson, Kronwall, Helm, Miller, Glendening, Cleary all provided some physicality. They weren't a bruising team, nor did they fight hardly ever, but they didn't get pushed around.

Darren McCarty was part of all 4 Championship teams and fought dozens of times over his career. He could play too, but he was prolific fighter.

 

[Bullseye]

Salt thanks for posting the article and renewing this topic. I need a refresher on advanced stats regularly!

Refreshing to hear Dubas put things in perspective. He's a guy going forward the Leafs organization needs to lock up tight - I don't want his brain and personality and management skills going to another NHL club.

 

[Pookie]

That will lead to a point where teams will attempt to find exceptions to the general rules, players that perhaps impact quality more profoundly than quantity and thus would be undervalued in a market that relies on quantity as a measurement.

That doesn't mean that possession metrics will lose their meaning, just that regardless of how complete measurement metrics and systems we come up with, teams will seek an advantage by finding those that are not evaluated accurately.

I don't believe in a formula for success. I do believe there's a formula for team-level improvement though, and I think that goes through the market inefficiencies that Dubas mention.

And those advantages won't be published on the Internet and certainly not on NHL.com.

The data will be proprietary to the team and I think will be heavily focused on player health and peak performance metrics.

 

[saltming]

Salt thanks for posting the article and renewing this topic. I need a refresher on advanced stats regularly!

Refreshing to hear Dubas put things in perspective. He's a guy going forward the Leafs organization needs to lock up tight - I don't want his brain and personality and management skills going to another NHL club.

My pleasure Bullseye, I am the same on the topic. I'm really enjoying the input here. So truly fascinating stuff.

I totally agree with you on Dubas. The kid has a great hockey mind. Imo he is what the future NHL GM will be for all 30 teams. The video guy we just hired, Bean, also looks like an analytics gem. I very happy the leafs are trending towards the future and not to the past as have been the path for too many years.

 

[Mr Hockey*]

Data is data, it doesn't inherently have traits such as "untrustworthy". Just like having watched a lot of hockey rids you of flaws that every human share. As long as you are human, you have human flaws such as cognitive bias.

Nobody has talked about going "just by stats". I don't think I've ever seen anyone advocate this. It's about whether or not you include more information in your process.

The only way data would lead you astray is if you don't understand what they represent. What do you base your opinion on that the analytic personnel hired by NHL franchises can not understand the data that they make a living on, and in many cases have an education on?

You do not use Advanced stats correct?

 

[Bullseye]

My pleasure Bullseye, I am the same on the topic. I'm really enjoying the input here. So truly fascinating stuff.

I totally agree with you on Dubas. The kid has a great hockey mind. Imo he is what the future NHL GM will be for all 30 teams. The video guy we just hired, Bean, also looks like an analytics gem. I very happy the leafs are trending towards the future and not to the past as have been the path for too many years.

Ya, we are going to see more surprise success players in the near future. Marincin and Hyman are early examples but there should be more to come.

 

[saltming]

Ya, we are going to see more surprise success players in the near future. Marincin and Hyman are early examples but there should be more to come.

And advanced draft techniques. Lots more of those, 'who is that guy?... Oh wow that he looks really good!!' draft moments to come in the future.

 

[Kurtz]

A little food for thought:

There are no stats that we as fan are not exposed to.....it is their applications and how each stat is used and the context that a team uses that separates us from them.

That's likely incorrect. Refer to my post on page 2 re: sportslogiq.

 

[diceman934]

That's likely incorrect. Refer to my post on page 2 re: sportslogiq.

I have heard it from the horses mouth.....it is the context of the data that makes the difference.

 

[saltming]

I have heard it from the horses mouth.....it is the context of the data that makes the difference.

Which horse told you?

 

[BoredBrandonPridham]

I have heard it from the horses mouth.....it is the context of the data that makes the difference.

Sure, but exposing context to a stat makes it a new stat. E.g., Sv%5v5 is a different stat than Sv%4v5, as it sheds a different context. When you break those down to high-danger, low-danger, etc... you get other stats with more context.

How you use that stat to inform your decision making is another aspect.

So what are you trying to say exactly? That analytics employees for NHL clubs that used to innovate and crunch numbers and context from NHL games stopped doing that when they were hired, and just started browsing the internet for the existing stats available to all fans?

Or are you saying that analytics employees for NHL clubs are still publishing all this information for fans?

 

[Teeder9]

In my own database (yes, I'm that much of a hockey nerd), I adjusted for arena bias by simply normalize using numbers away from the arena.

You were 20% more likely to get credit for a hit in Toronto compared to away from it. So I used that number to deflate Toronto's home numbers, adding them to the away numbers for an adjusted total.

And just watching the games doesn't just mean that you intentionally use less information, it also puts you completely in the hands of cognitive bias. I also think being aware and able to adjust for your bias is more or less mutually exclusive with the arrogance of thinking that your impressions is all you need to figure out the truth of things.

And that still isn't accurate. Why were you 20% more likely in TO vs another arena? Was it false in TO or true? My biggest problem right now with these stats is they aren't accurate and if they aren't accurate neither is the argument using them.

 

[Nithoniniel]

And that still isn't accurate. Why were you 20% more likely in TO vs another arena? Was it false in TO or true? My biggest problem right now with these stats is they aren't accurate and if they aren't accurate neither is the argument using them.

I don't really use them either. And you're right, it's not accurate, but the results looked a lot better than they did before.

 

[Teeder9]

I don't really use them either. And you're right, it's not accurate, but the results looked a lot better than they did before.

Which leads to me thinking we're really pre Bill James than post for meaningful analytics in hockey. There's things we can track, and we do but it will never be remotely close in that aspect because we can't measure what the other players on the ice without the puck are doing to lead to the puck carriers decisions. Ice time is likely equal to trust. After that I'm not sure if anything is worth much in the grand scheme of things. We're still at the point where coaches coach to not lose

 

[Mr Hockey*]

I don't really use them either. And you're right, it's not accurate, but the results looked a lot better than they did before.

why do you get into discussions about CRQoC with me?

 

[Joey Hoser]

And that still isn't accurate. Why were you 20% more likely in TO vs another arena? Was it false in TO or true? My biggest problem right now with these stats is they aren't accurate and if they aren't accurate neither is the argument using them.

It's still more accurate than not measuring things.

 

[Menzinger]

And those advantages won't be published on the Internet and certainly not on NHL.com.

The data will be proprietary to the team and I think will be heavily focused on player health and peak performance metrics.

I'd assume they're using those types of stats as well as ones we don't know about (ie they've been using player tracking camera systems). BUT they also didn't hire guys like Cam Charron because they think possession metrics/zone entries/ect are meaningless. Just this past Marlies playoff run Dubas made a comment about the team maximizing shot attempts to increase scoring chances.

The Leas stats department will need to find new stats AND new uses for older stats to stay ahead of other teams.

 

[Teeder9]

It's still more accurate than not measuring things.

Is it?

Me- There's 5 grains of sand at the beach.
You- I'd rather not guess.
Me- I'm more accurate because at least I guessed.

Is that how this works?

Don't get me wrong. i believe at some point analytics will get there. It just isn't close right now, and certainly not close enough to be the deciding factor in any debate

 

[Mr Hockey*]

It's still more accurate than not measuring things.

That is only his opinion

 

[4thline]

Which leads to me thinking we're really pre Bill James than post for meaningful analytics in hockey. There's things we can track, and we do but it will never be remotely close in that aspect because we can't measure what the other players on the ice without the puck are doing to lead to the puck carriers decisions. Ice time is likely equal to trust. After that I'm not sure if anything is worth much in the grand scheme of things. We're still at the point where coaches coach to not lose

There are definite limitations, and IMO the best way currently to overcome them and have usable data is to step beyond the statistics and apply some subjective judgement. The tech and processes are available to track and tell us the "what happened" with respect to so many on ice occurrences. The complexity of hockey limits the utility of that "what" for absolute evaluation but it does provide a very useful framework to analyze trends and find and prioritize weak problems areas, then you can go to the tape and figure out the "why"