On hockey and Elo ratings

[morehockeystats]

I already mentioned Elo ratings that I use on my website, but I realize the topic is quite new, or at least infrequent, and the forum's participants might be somewhat unfamiliar with the concept.

So I'll pile my future postings in this thread.

And for the starter - here's a brief description of how Elo ratings work:

https://morehockeystats.blogspot.com/2016/12/on-players-evaluation-part-ii-elo.html

Click to read more...

 

[morehockeystats]

And here's part III
https://morehockeystats.blogspot.com/2016/12/on-players-evaluation-part-iii-teams-elo.html

 

[Hockey Outsider]

Thanks for these articles - creative and thought-provoking.

I'm familiar with the concept of Elo ratings from chess. Alan Ryder (from hockeyanalytics.com) also did a series about team Elo ratings over time - http://hockeyanalytics.com/2016/07/elo-ratings-for-the-nhl/.

Do you have any plans of trying to introduce Elo ratings at an individual level? That would be fascinating to see (though I don't know how to even start with that). Look forward to reading the rest of the series.

 

[Doctor No]

I like this quite a bit as well - very interesting.

 

[Stat Guy]

In 2014 I tinkered with ELO, and related it to Corsi.
http://hfboards.mandatory.com/showthread.php?t=1729959&highlight=

Here is a link that discussed ELO related to Corsi focused solely on the Oilers.
http://hfboards.mandatory.com/showthread.php?t=1731745&highlight=

Here is a link that discussed ELO to rank faceoffs only.
http://hfboards.mandatory.com/showthread.php?t=1723255&highlight=

I'll see if I can dust off my work and update it.

 

[morehockeystats]

Thanks for these articles - creative and thought-provoking.

I'm familiar with the concept of Elo ratings from chess. Alan Ryder (from hockeyanalytics.com) also did a series about team Elo ratings over time - http://hockeyanalytics.com/2016/07/elo-ratings-for-the-nhl/.

Do you have any plans of trying to introduce Elo ratings at an individual level? That would be fascinating to see (though I don't know how to even start with that). Look forward to reading the rest of the series.

Thanks for the link. I saw this article earlier. The author is wrong IMHO on several points:
1) Accounting an OT win as 1-0 and a SO as a 0.5-0.5 tie.
The result must reflect the change in the crosstable. There's no other way around it.
2) Counting in home ice advantage.
Just like there is no factor for White and Black, there should not be a factor for Home and Away. And, anyways, there is 41 home and 41 away game per team, and these should cancel each other.
3) The continuity of the ratings.
My opinion is that the teams experience too much changes on one side, and too much wear and tear on the other one, as compared to individual board games' players to allow the continuity, therefore I reset the ratings at the start of each season.
4) Counting in "Importance of the game" and "Margin of victory".
Very similar to 2). Completely unnecessary, the first one even too subjective.
5) K as velocity. It's more a volatility than a velocity factor. In chess, K factor is artificially raised for new and younger players, and for shorter controls (more volatile) games.

Elo much more useful in projections over a span rather than onto a single game.

Yes, I do calculate individual Elos. The ratings for goalies are available explicitly:
http://morehockeystats.com/players/goalieratings .
The ratings for the skaters are used implicitly in "evaluation", "simulation" and "daily fantasy" rubriques of the website's "Fantasy" section.

In general, it's the daily fantasy and points season fantasy games where Elo is the easiest and the most useful.

 

[morehockeystats]

In 2014 I tinkered with ELO, and related it to Corsi.
http://hfboards.mandatory.com/showthread.php?t=1729959&highlight=

Here is a link that discussed ELO related to Corsi focused solely on the Oilers.
http://hfboards.mandatory.com/showthread.php?t=1731745&highlight=

Here is a link that discussed ELO to rank faceoffs only.
http://hfboards.mandatory.com/showthread.php?t=1723255&highlight=

I'll see if I can dust off my work and update it.

Aargh, it's Elo, not ELO.
You can use Elo for Corsi, if you can work out the expectation Corsi for a given Elo.
The Elo for faceoffs is natural and straightforward. I may add it to my website. It's not as applicable in Fantasy Hockey because you cannot know how many faceoffs will be there for each player.

 

[morehockeystats]

Here's the latest entry.
And the summary of the team Elo projections, copying from my twitter:

You can also find the Elo graphs from the start of the season in my twitter, @morehockeystats .

 

[morehockeystats]

Part V - Goaltender's Elo
Here's the status as of yesterday

:

 

[morehockeystats]

Elo graphs for teams:

 

[morehockeystats]

Also, I tried to make points projections for the season (top-30):

 

[tfong]

Isn't ELO only supposed to work on an individual level (ie 1v1 in chess) and attempts to capture team elo with individuals make it inaccurate?

 

[morehockeystats]

Isn't ELO only supposed to work on an individual level (ie 1v1 in chess) and attempts to capture team elo with individuals make it inaccurate?

It's neither ELO, nor elo.

Elo is a metric that 1) gravitates to the center, 2) reciprocally predicts the results according to history.

Just like a team is not a monolith, a single player is not a robot. As I said earlier, even a "Black" Elo and a "White" Elo are not distinguished. However, the teams play in mostly similar lineups throughout the year; the lineup changes vs the Elo changes may produce an interesting WAR-like insight.

By resetting the Elo at the start of each season and by working over a very long stretch (82 games), I think Elo has enough accuracy on team level too; an article on individual Elo measurement in hockey is coming up in a couple of days.

 

[morehockeystats]

an article on individual Elo measurement in hockey is coming up in a couple of days.

And it's here:

https://morehockeystats.blogspot.com/2017/01/on-players-evaluation-part-vi-skaters.html

I'm ahead of GRRM! I'm done with Part VI.

 

[ekcut]

And it's here:

https://morehockeystats.blogspot.com/2017/01/on-players-evaluation-part-vi-skaters.html

I'm ahead of GRRM! I'm done with Part VI.

So what you are saying with very big words, is Conner McDavid has the most assists in the league?

 

[morehockeystats]

So what you are saying with very big words, is Conner McDavid has the most assists in the league?

But Nicklas Backstrom is expected to score more assists per game, still.

 

[morehockeystats]

Didn't have time to post it yesterday, so belated (before the 8-7 carnage last night) tables for teams, players and season graphs will follow now...

 

[morehockeystats]

http://morehockeystats.com/teams/elo

 

[morehockeystats]

Points:

http://morehockeystats.com/fantasy/evaluation

 

[morehockeystats]

Here's the latest installment in my blog:
Part VII and Final

 

[sanitysrequiem]

I'm curious, how do you calculate your expected points/game?

 

[Jag68Sid87]

I already mentioned Elo ratings that I use on my website, but I realize the topic is quite new, or at least infrequent, and the forum's participants might be somewhat unfamiliar with the concept.

So I'll pile my future postings in this thread.

And for the starter - here's a brief description of how Elo ratings work:

https://morehockeystats.blogspot.com/2016/12/on-players-evaluation-part-ii-elo.html

I would probably rate "Evil Woman" as ELO's best song, but it's close.

 

[morehockeystats]

I would probably rate "Evil Woman" as ELO's best song, but it's close.

Elo is the last name of Arpad Elo, the inventor of the system, not an acronym.

 

[morehockeystats]

I'm curious, how do you calculate your expected points/game?

points as player's points - expected goals + expected assists

 

[morehockeystats]

points as player's points - expected goals + expected assists

I will give an example.
Let's take Sydney Crosby and Alex Ovechkin.
At the end of 2015, given their history Ovechkin had GOAL rating of 2206 and Crosby had 2117 giving them expected goal per game at 0.66 and 0.46 respectively (not accounting for the opponent), and in ASSIST the ratings were 2002 and 2145 which translate to 0.26 and 0.69 assists per game.

This season Crosby has been scoring more goals and Ovechkin has been handing out more assists than usual. Thus, their ratings became in goals: Ovechkin 2164, Crosby 2155, exp. goals 0.58 and 0.56, in assists: Ovechkin 2031, Crosby 2156, exp. assists 0.35 and 0.73.

Now, the aforementioned expected values are against an "average" opponent that has AGAINST ratings of 2000 in goals and in assists. If we take their performances against MTL whose ratings against in these categories are 2004 and 2005 (I should make the spread wider, I think, so that MTL is at about 2015-2020 and Colorado is at 1950 rather than 1980) respectively the expected numbers go down:
Ovechkin - goals from 0.58 to 0.57, assists from 0.35 to 0.33
Crosby - goals from 0.56 to 0.55, assists from 0.73 to 0.72