Login

***artermis*** · 02-07-2019, 11:43 PM

We hear so much about how "random" and nonsensical the results that come out of STHS are, but I've been told in the past that STHS's has the ability to produce NHL-like probabilities and I wanted to put that to the test, but I'm lazy so I took data from only the past 5 playoffs for both the SHL and NHL.

Numbers and stuff: Spreadsheet here

This isn't so much as to make money as it is to start a conversation. First of all I only used point % as I figure that's the simplest number to use to represent a team's perceived skill.

So, my numbers have shown me things:

The most important of which being:
- Both the NHL and SHL produce very similar odds of having the underdog win, in fact the NHL's odds are higher

- The SHL's point %s are much higher than that of the NHL, and thus the differences are greater, likely thanks to a longer schedule involving more teams
- Thanks to this, we have some distorted numbers, in that the NHL's biggest upset (NSH over CHI) is an 11 % difference in comparison to the SHL's 21% difference in WKP's defeat of WIN.
- On the other hand, the NHL's biggest win by the favorite is 18.3% (Happened three times) and the SHL's is 31% (EDM over LAP)
- Two of the past 5 stanley cup winners were underdogs, whereas SHL's past 3 were underdogs (granted, conferences render point % less accurate)
- Both the average and median of the differences for both leagues are nearly 0, so I figure that was basically pointless

Now these points seem to suggest that both league's are very similar and that STHS is doing a solid job, however one main difference are series involving a difference of 10% or greater.

- In these scenarios, an NHL underdog can expect to win 10% of the time, but an SHL team can expect to win 30% of the time.
- This could be accounted for SHL's larger difference in general, so it too may fall within the limits of normality

If someone wants access to the sheet to their own numbers, hit me up., I really only did the simple stuff.

ml002 · 02-07-2019, 11:57 PM

Credit to Copenhagen, Wasty, FlappyGiraffe, InciteHysteria, and caltroit_red_flames
[Image: MM4nqx6.png]

Tomasnz · 02-08-2019, 04:48 AM

Good start and solid starting methodology.

Definitely the level of difference not affecting the chance of an upset seems to be the main difference in the two leagues.

Something I might try is to start an ELO ranking of the SHL teams and then compare ELO difference in terms of upsets across the two leagues (I think fivethirtyeight maintain an ELO ranking for the NHL teams)

Avakael · 02-08-2019, 05:00 AM

02-07-2019, 11:57 PMml002 Wrote:

This isn't even funny, it's just a depiction of my actual reaction to this post. Goddamn it.

*Leppish* · 02-08-2019, 08:37 AM

sths succ

Daco · 02-08-2019, 08:50 AM

02-08-2019, 04:48 AMTomasnz Wrote: Good start and solid starting methodology.

Definitely the level of difference not affecting the chance of an upset seems to be the main difference in the two leagues.

Something I might try is to start an ELO ranking of the SHL teams and then compare ELO difference in terms of upsets across the two leagues (I think fivethirtyeight maintain an ELO ranking for the NHL teams)

I thought ELO was unique to chess, does it actually work in ranking other sports? Same type of scale where like 2500 is grandmaster level?

GCool · 02-08-2019, 09:23 AM

02-08-2019, 08:50 AMDaco Wrote:
02-08-2019, 04:48 AMTomasnz Wrote: Good start and solid starting methodology.

Definitely the level of difference not affecting the chance of an upset seems to be the main difference in the two leagues.

Something I might try is to start an ELO ranking of the SHL teams and then compare ELO difference in terms of upsets across the two leagues (I think fivethirtyeight maintain an ELO ranking for the NHL teams)

I thought ELO was unique to chess, does it actually work in ranking other sports? Same type of scale where like 2500 is grandmaster level?

Derivatives of Elo are used all over, actually. Esports like CSGO/Overwatch/League of Legends use it, I think at least one version of Elo was used for one of the big tennis tours for a while ... could be wrong there, though.

I'd be careful with using Elo across multiple seasons, simply because sites like fivethirtyeight will not adjust or deflate high Elos for anomalous seasons, like if a team was tanking to get a high draft pick. Granted, that action is speculation in and of itself--but you know what I mean. Context is lost in a sea of numbers.

SlashACM · 02-08-2019, 10:43 AM

ELO is a good band

Tomasnz · 02-08-2019, 11:23 AM

02-08-2019, 09:23 AMGCool Wrote:
02-08-2019, 08:50 AMDaco Wrote: I thought ELO was unique to chess, does it actually work in ranking other sports? Same type of scale where like 2500 is grandmaster level?

Derivatives of Elo are used all over, actually. Esports like CSGO/Overwatch/League of Legends use it, I think at least one version of Elo was used for one of the big tennis tours for a while ... could be wrong there, though.

I'd be careful with using Elo across multiple seasons, simply because sites like fivethirtyeight will not adjust or deflate high Elos for anomalous seasons, like if a team was tanking to get a high draft pick. Granted, that action is speculation in and of itself--but you know what I mean. Context is lost in a sea of numbers.

I think if a team is trying to bad and tank then by the end of their season the elo will adjust.

Likewise season start is always when ELO is the most questionable as so much might have changed .But as a good starting point as any as teams that did well in last year's regular season are probably going to do well again.

GCool · 02-08-2019, 11:36 AM

02-08-2019, 11:23 AMTomasnz Wrote:
02-08-2019, 09:23 AMGCool Wrote: Derivatives of Elo are used all over, actually. Esports like CSGO/Overwatch/League of Legends use it, I think at least one version of Elo was used for one of the big tennis tours for a while ... could be wrong there, though.

I'd be careful with using Elo across multiple seasons, simply because sites like fivethirtyeight will not adjust or deflate high Elos for anomalous seasons, like if a team was tanking to get a high draft pick. Granted, that action is speculation in and of itself--but you know what I mean. Context is lost in a sea of numbers.

I think if a team is trying to bad and tank then by the end of their season the elo will adjust.

Likewise season start is always when ELO is the most questionable as so much might have changed .But as a good starting point as any as teams that did well in last year's regular season are probably going to do well again.

Ahh yeah, I guess I hadn't thought of it that way. It'd be interesting how Elo would shift, if at all, post-expansion.

Tomasnz · 02-08-2019, 11:57 AM

02-08-2019, 11:36 AMGCool Wrote:
02-08-2019, 11:23 AMTomasnz Wrote: I think if a team is trying to bad and tank then by the end of their season the elo will adjust.

Likewise season start is always when ELO is the most questionable as so much might have changed .But as a good starting point as any as teams that did well in last year's regular season are probably going to do well again.

Ahh yeah, I guess I hadn't thought of it that way. It'd be interesting how Elo would shift, if at all, post-expansion.

Just thinking how to do it..

If I set each team at 1000 for start of season 45, then ran the schedule through to give a new update for each team at regular season end. Use that as a S46 score adding each expansion in at 1000.

...Welp .that's the easy part. Not sure my Excel skills are quite up for this... I guess I could macro it..

Navigation

Extra Menu

About us