ELO and the VHL part 3

[Enorama 2,038]

Part 1

Part 2

 

Make sure you read through each of the previous ELO posts to better understand this one (or, at the very least, the first one so you know the basic idea behind ELO). 

 

First, an update with K = 32:

 

Seattle is no longer absolute trash, Riga is on the tail end of a U bend in their graph, and Helsinki is flirting with the idea of coming back to earth. Oh and Vancouver is sucking again, but we all knew that was coming.

 

I thought it would be interesting this week to demonstrate what happens when we play with the aforementioned K value. First off, here's a look with K = 1:

 

Just the blandest graph you've ever seen. It might be neater to see as we zoom in further, but I can't be arsed, because it still likely won't show us too much. It would likely just represent the current standings, more or less.

 

What's more interesting is if we bump up the K value to something like 64, like so:

 

This shows us something more akin to power rankings, as ELO is more prone to fluctuations and is biased to more recent results. As we can see, Seattle is well on their way up, while Vancouver has fallen to be the 'coldest' team in the league, while Helsinki is no longer nearly as far away from the crowd as they once were, with both Riga and New York nipping on their heels after their respective upswings. Of course, as I bump the K value further and further upwards, the jumps in ELO from individual games skyrockets, so we could play with higher and higher numbers to make ELO more and more biased to recent results.

 

Something I am interested in and am looking at doing for next season is incorporating a sort of dynamic K value, affected by margin of victory. I've seen this done before, where OT games result in very little movement, wins by 1 goal slightly more, wins by 2 slightly more, and so on. We'd love to see the bump that Moscow got after their latest 7-0 victory against Vancouver, and how far Vancouver would have fallen because of it. What I'm currently debating is whether it should be a linear growth or an exponential growth (should a 3 goal victory be worth 50% more than a 2 goal victory, or should it be higher or lower?).

 

One last thing, for @Nykonax's sake, here's what happens at K = 10000:

 

417 words

[Nykonax 1,564]

nice