What are the best advanced stats projects?

tinyzombies · Jul 15, 2015

Doctor No · Jul 15, 2015

I do some goalie-related metrics on my site, which I describe near the bottom of this list:

http://hockeygoalies.org/stats/glossary.html

tinyzombies · Jul 15, 2015

Doctor No said:
I do some goalie-related metrics on my site, which I describe near the bottom of this list:

http://hockeygoalies.org/stats/glossary.html

Nice!

Teemu · Jul 15, 2015

http://jetsnation.ca/2015/5/26/draft-analytics-unveiling-the-prospect-cohort-success-model
http://www.theprojectionproject.com/Home/Methodology

PCS and the Projection Project for draft eligible prospects

http://hockeyviz.com/wowy/wowy.html

McCurdy's WOWY

http://ownthepuck.blogspot.ca/2015/06/impact-on-linemates-shot-generation-and.html

Impact on linemates (Galamini)

http://www.nullhypothesishockey.com/great-expectations-dcorsi/

A nice dCorsi tool

The Bad Guy · Jul 16, 2015

I don't like Corsi, I think it is arbitrary to subtract shots against from shots generated. Are there sites that attempt to isolate shots against controlling for quality of line mate and quality of opposition and the same for shots generated without converting it into Corsi?

I just think variables should be isolated before we try to create a formula unifying them in an attempted general evaluation.

tinyzombies · Jul 16, 2015

The Bad Guy said:
I don't like Corsi, I think it is arbitrary to subtract shots against from shots generated. Are there sites that attempt to isolate shots against controlling for quality of line mate and quality of opposition and the same for shots generated without converting it into Corsi?

I just think variables should be isolated before we try to create a formula unifying them in an attempted general evaluation.

Their argument is that the results haven't reflected anything more than raw Corsi would. The point is to get an approximation of reality with as little work as possible, cuz we don't have access to the advanced stuff teams do, right? And then build off that.

Micklebot · Jul 16, 2015

That passing one is cool, is there anything describing the methodology, or analysis of the results?

tinyzombies · Jul 16, 2015

Micklebot said:
That passing one is cool, is there anything describing the methodology, or analysis of the results?

Do a search for NHL Passing Project. It will turn up all the inlouwetrust.com posts.

On a separate note, found this to be helpful (as an intermediate at this) as a template for how to evaluate a player: How to use advanced stats to evaluate a player (further down the page when they get into the adv stats stuff):
http://www.todaysslapshot.com/nhl-e...ett-connolly-the-boston-bruins-forgotten-man/

Micklebot · Jul 16, 2015

tinyzombies said:
Do a search for NHL Passing Project. It will turn up all the inlouwetrust.com posts.

On a separate note, found this to be helpful (as an intermediate at this) as a template for how to evaluate a player: How to use advanced stats to evaluate a player (further down the page when they get into the adv stats stuff):
http://www.todaysslapshot.com/nhl-e...ett-connolly-the-boston-bruins-forgotten-man/

Thanks, found it myself. I had read about his passing project before, but it was much earlier on so the data was extremely limited for any of the teams that were only being tracked as ~~visitor~~ secondary teams.

The Bad Guy · Jul 16, 2015

tinyzombies said:
Their argument is that the results haven't reflected anything more than raw Corsi would. The point is to get an approximation of reality with as little work as possible, cuz we don't have access to the advanced stuff teams do, right? And then build off that.

My point is we are already collecting "total shots generated" and "total shots against" on ice. Why subtract them from one another just because when we could isolate factors in each of those variables separately which would likely tell more of a player's offensive and defensive prowess statistically?

Naturally we are not going to get the play by play breakdowns team scouting departments get, but we can still do better with what we do have.

tinyzombies · Jul 16, 2015

The Bad Guy said:
My point is we are already collecting "total shots generated" and "total shots against" on ice. Why subtract them from one another just because when we could isolate factors in each of those variables separately which would likely tell more of a player's offensive and defensive prowess statistically?

Naturally we are not going to get the play by play breakdowns team scouting departments get, but we can still do better with what we do have.

You don't have to, they have CF and CA. And you have to look at the number of events/minutes played as well, right? Not going to pretend I can get into an advanced debate, just trying to collect popular advanced stats tools.

The Bad Guy · Jul 16, 2015

tinyzombies said:
You don't have to, they have CF and CA. And you have to look at the number of events/minutes played as well, right? Not going to pretend I can get into an advanced debate, just trying to collect popular advanced stats tools.

You may not have to, but I think you better analyze the performance of a player if you do. If you analyze just CF (Which is basically total shots for, right?) and separate out effect for quality of team mate and quality of competition, then you see more of that player's personal impact on shot production in a situation. Same goes for CA (Total Shots Against, right?) but with regard to shot suppression.

Corsi (CF-CA) controlled for those factors tells you nothing really. It just states "in situation x, more or less shots than the other team were made when player alpha was on the ice." I think it would be more useful to know the break down of CF and CA, where we could actually quantitatively break down a player's talents and weaknesses.

(obviously there are other factors to control for as well, but I think quality of competition and line mate help isolate a players actual contribution)

So anyways back to my original question, are there any sites that in depth analyze CF and CA separately? (especially with regard to quality of line mate and competition)

mkwong268 · Jul 16, 2015

The Bad Guy said:
You may not have to, but I think you better analyze the performance of a player if you do. If you analyze just CF (Which is basically total shots for, right?) and separate out effect for quality of team mate and quality of competition, then you see more of that player's personal impact on shot production in a situation. Same goes for CA (Total Shots Against, right?) but with regard to shot suppression.

Corsi (CF-CA) controlled for those factors tells you nothing really. It just states "in situation x, more or less shots than the other team were made when player alpha was on the ice." I think it would be more useful to know the break down of CF and CA, where we could actually quantitatively break down a player's talents and weaknesses.

(obviously there are other factors to control for as well, but I think quality of competition and line mate help isolate a players actual contribution)

So anyways back to my original question, are there any sites that in depth analyze CF and CA separately? (especially with regard to quality of line mate and competition)

Did you mean something like this? http://www.puckalytics.com/superwowy.html

You can look up CF and CA separately of different line combinations and strength situations.

tinyzombies · Jul 16, 2015

mkwong268 said:
Did you mean something like this? http://www.puckalytics.com/superwowy.html

You can look up CF and CA separately of different line combinations and strength situations.

That's why I posted the Brett Connolly thing below, because you can run thru a series of tools to get a picture of a player like they did.

But I've also read that those extras like score-adjusted, zone starts, QOC, quality of linemates, etc. did not make a significant change over the long haul from using raw Corsi. What that means in relation to what you're saying, I don't know.

I don't even know if I have all the tools these dudes use regularly, and I am not 100% up on the ones I have collected.

charlie1 · Jul 16, 2015

faceoffs.net

charlie1 · Jul 16, 2015

tinyzombies said:
That's why I posted the Brett Connolly thing below, because you can run thru a series of tools to get a picture of a player like they did.

But I've also read that those extras like score-adjusted, zone starts, QOC, quality of linemates, etc. did not make a significant change over the long haul from using raw Corsi. What that means in relation to what you're saying, I don't know.

I don't even know if I have all the tools these dudes use regularly, and I am not 100% up on the ones I have collected.

I've been hearing this lately and it's definitely wrong. Zone starts affect a player's Corsi, A LOT.

Here are some tweets on this by @stat_sam from a few months ago. There is a clear correlation between ZS% and CF%:

Micklebot · Jul 16, 2015

charlie1 said:
I've been hearing this lately and it's definitely wrong. Zone starts affect a player's Corsi, A LOT.

Here are some tweets on this by @stat_sam from a few months ago. There is a clear correlation between ZS% and CF%:

David Johnson posted a reply to that string of tweets,

http://hockeyanalysis.com/2015/03/21/zone-starts-and-impact-on-players-statistics/

in short, he suggests that the effect being show isn't zone starts on corsi, but rather teams that are playing well get more OZ starts, and better Corsi. If Ovechkin or Phaneuf get 60-70% OZ starts, there's a good chance it's because the team is playing well. If they get 40%, odds are their team is not doing well. It's playing well, not zone starts, that results in higher corsi. Corsi and zone starts are both byproducts of good play but not the driving forces themselves on one another.

It all boils down to correlation does not equal causality.

tinyzombies · Jul 17, 2015

Another tool (I think there's a tool for it out there?) is Expected Goals, Expected Shooting Percentage... Expected ____?

I saw the Hamilton trade analyzed with all the other stuff and how he's not the same player away from Bergeron and Chara, but wouldn't you analyze the trade by using the "Expected" tools, cuz that's where his value lies?

Here's the link: http://xtrahockeystats.com/teams.php?startseason=20142015&endseason=20142015&viewtype=byseason&display=20s&startgame=938&endgame=1230

EDIT: dCorsi does this.

Micklebot · Jul 17, 2015

Shot charts:

http://www.sportingcharts.com/nhl/icetrack/

http://somekindofninja.com/nhl/

http://www.nicetimeonice.com/ is a good one for finding info on specific games too, it pretty much links to the info on other sites, but is a good jump point to get the data you're looking for.

tinyzombies · Jul 17, 2015

Micklebot said:
Shot charts:

http://www.sportingcharts.com/nhl/icetrack/

http://somekindofninja.com/nhl/

http://www.nicetimeonice.com/ is a good one for finding info on specific games too, it pretty much links to the info on other sites, but is a good jump point to get the data you're looking for.

Ok, I've updated the first post with all the suggested links so far. Keep em coming.

tinyzombies · Jul 17, 2015

The Bad Guy said:
You may not have to, but I think you better analyze the performance of a player if you do. If you analyze just CF (Which is basically total shots for, right?) and separate out effect for quality of team mate and quality of competition, then you see more of that player's personal impact on shot production in a situation. Same goes for CA (Total Shots Against, right?) but with regard to shot suppression.

Corsi (CF-CA) controlled for those factors tells you nothing really. It just states "in situation x, more or less shots than the other team were made when player alpha was on the ice." I think it would be more useful to know the break down of CF and CA, where we could actually quantitatively break down a player's talents and weaknesses.

(obviously there are other factors to control for as well, but I think quality of competition and line mate help isolate a players actual contribution)

So anyways back to my original question, are there any sites that in depth analyze CF and CA separately? (especially with regard to quality of line mate and competition)

Are you talking about dCorsi?

http://nhlnumbers.com/2014/7/19/dcorsi-introductions

silverfish · Jul 17, 2015

Micklebot said:
David Johnson posted a reply to that string of tweets,

http://hockeyanalysis.com/2015/03/21/zone-starts-and-impact-on-players-statistics/

in short, he suggests that the effect being show isn't zone starts on corsi, but rather teams that are playing well get more OZ starts, and better Corsi. If Ovechkin or Phaneuf get 60-70% OZ starts, there's a good chance it's because the team is playing well. If they get 40%, odds are their team is not doing well. It's playing well, not zone starts, that results in higher corsi. Corsi and zone starts are both byproducts of good play but not the driving forces themselves on one another.

It all boils down to correlation does not equal causality.

I've been doing some of my own analysis on zone starts, and I've come to the conclusion that the further out you go for relZS, the more it effects relCF%. The charts below are forwards and D for the 14/15 season, players who played at least 350 minutes all 5v5.

silverfish said:
More zone start work (put my new stat on pause because it's not working even though it totally should be and that's pissing me off )

I talk a lot about the signal and the noise of zone starts, so I figured I'd put it on a graph, since everyone in this thread loves those.

The Noise: Players who have zone starts from -5 through +5

The Noise Removed: Players who have zone starts below -5 or above +5

The true signal: Players who have zone starts below -10 or above +10

The R^2 for both the noise removed and the signal for zone starts and relative corsi are stronger than the R^2 for Corsi, Fenwick, and shots are to win percentage. So if you believe in possession metrics, why don't you believe in zone starts?

http://www.pensionplanpuppets.com/2013/7/10/4508094/what-statistics-are-meaningful-in-a-given-season-corsi-fenwick-PDO-hits-fights-blocked-shots

If you believe then you can achieve

This, and a million more of my rants can be found on the NYR section in the Advanced Stats thread.

tinyzombies · Jul 17, 2015

silverfish said:
This, and a million more of my rants can be found on the NYR section in the Advanced Stats thread.

Do you know of any tools to add to the list in the first post?

silverfish · Jul 17, 2015

tinyzombies said:
Do you know of any tools to add to the list in the first post?

No, unfortunately. I stick to War-On-Ice, download their .csv's and do my own analysis for the most part.

If you're interested in that, you can head on over to the blog I just launched - http://nerdhockey.com - but I wouldn't really recommend it as a prominent tool :laugh:

Micklebot · Jul 17, 2015

silverfish said:
I've been doing some of my own analysis on zone starts, and I've come to the conclusion that the further out you go for relZS, the more it effects relCF%. The charts below are forwards and D for the 14/15 season, players who played at least 350 minutes all 5v5.

This, and a million more of my rants can be found on the NYR section in the Advanced Stats thread.

There are a couple of points to make here:

1. I don't think anyone will say zone starts don't help. What people do tend to suggest is the effect is exaggerated by many.

2. Roughly half of the NHL falls into what you're calling noise and only 20% in what you call the true signal. While there is a correlation, that isn't causation. Other factors could be the cause. Some players are deployed defensively because the add nothing offensively, and vice versa. A guy like Gaustad is there purely for Defense, and doesn't even try to score. The task set upon him by the coach is shut the other team down until our scorers get on the ice.

3. Further to point two, guys who get OZ starts tend to be the offensive guys, so it stands to reason they would outshoot the opposition. Guys deployed defensively tend to be there to shut down scoring chances. The tend to be the Grybas, Gorges and Smids of the world. Sure, they might get better corsi by being deployed in the OZ, but if you swapped their zone starts with the Karlssons of the world, it would just lessen the correlation you're seeing between rel OZ starts and Corsi.

I guess what I'm saying is zone starts to me seem to be a pretty minor factor. It's certainly going to help, but all the other uncontrolled variable are likely a lot more important.

What are the best advanced stats projects?

Registered User

Registered User

Registered User

Caffeine Free Since 1919

Registered Tool

Registered User

Moderator

Registered User

Moderator

Registered Tool

Registered User

Registered Tool

Registered User

Registered User

It's all McDonald's

It's all McDonald's

Moderator

Registered User

Moderator

Registered User

Registered User

got perma'd

Registered User

got perma'd

Moderator

Ad

Ad

Ad