Author Topic: 2013-14 D3hockey Computer Rankings Discussion  (Read 10496 times)

Offline Steve Wiitala

  • Starter
  • ***
  • Posts: 281
  • Karma: +10/-0
  • D3hockey.com Computer Rankings Specialist
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #15 on: December 08, 2013, 01:13:05 pm »
Joe,

Canton shouldn't count, but there was evidence last year that somebody didn't know that at the NCAA level.

I've watched Williams on NSN feeds a couple of times - they are an excellent team - I'd put them ahead of Amherst who I've seen a couple of times.  Trinity is way too high IMHO.  The big problem is still, of course, the insularity of the NESCAC schedule.  I gone through the data entry process yet for my rankings, but I think the results will be interesting (but still not definitive) this week.  There will be a bit more clarity, and a lot of movement (I think).
D3hockey.com computer rankings specialist

Offline NorthernLite

  • Junior Varsity
  • *
  • Posts: 21
  • Karma: +0/-0
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #16 on: December 09, 2013, 02:59:40 pm »
Are overall statistics (i.e. Team off/def, PP/PK,) considered in the computer formula ?
These are stats that I like to refer to when assessing strengths and weaknesses of teams that I know very little about.

Offline Steve Wiitala

  • Starter
  • ***
  • Posts: 281
  • Karma: +10/-0
  • D3hockey.com Computer Rankings Specialist
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #17 on: December 09, 2013, 03:49:19 pm »
I don't use anything but game results. Win%, EWP, OW, OOW, and OOOWP are the metrics I use, and all of those are simply based on who wins/loses each game. Ultimately that's the "goal" of the game.   

As an aside, I think there are some interesting issues regarding some of the traditional statistics.  I'm trying to put together a better way of measuring PK/PP stats, but it's still brewing in my addled brain.

A good example of what bugs me about those stats is the way partial penalties are handled.  Last night I was watching Colorado  play Vancouver.  17 secs into a PP, Colorado is called for a slash.  I totally agree that Colorado should have been charged with a failed power play - they had an opportunity for a PP goal and didn't get one, and possibly Vancouver can be credited with a PK (I'm still thinking about that one, all they had to do was have a guy get the stick chopped out of his hands).  After the 4X4 expired, Vancouver had a 17 sec power play and was charged with a failed power play, and Colorado got credit for a successful PK those spillover PPs after overlapping minors should be treated differently.  (At the other extreme, when a team gets a major power play and doesn't score, they are 0 for 1, which doesn't measure their failure well enough to suit me either.  In fact, last year one thing that Norwich did very poorly was the 5 min PP - if you don't score on one of those, you ought to be charged with 0 for 2.5, it seems to me.)

It would also be interesting to see what kind of a better statistic we could come up with to measure offensive effort besides shots on goal - you see some games where one team seems to dominate the play, but they don't end up with that many more shots than the other team. 
D3hockey.com computer rankings specialist

Offline Matthew Webb

  • Moderator
  • Starter
  • *****
  • Posts: 408
  • Karma: +13/-1
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #18 on: December 09, 2013, 04:11:22 pm »

As an aside, I think there are some interesting issues regarding some of the traditional statistics.  I'm trying to put together a better way of measuring PK/PP stats, but it's still brewing in my addled brain.

For years I've thought goals per unit of time would be a far better representation than the straight conversion % is. 
"It is hard to free fools from the chains they revere" - Voltaire

Offline joecct

  • Second-stringer
  • **
  • Posts: 94
  • Karma: +5/-0
  • NCAA Hockey in Maryland!!
    • View Profile
    • Facebook
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #19 on: December 09, 2013, 04:51:17 pm »

As an aside, I think there are some interesting issues regarding some of the traditional statistics.  I'm trying to put together a better way of measuring PK/PP stats, but it's still brewing in my addled brain.

For years I've thought goals per unit of time would be a far better representation than the straight conversion % is.
If the goal is scoring a PPG, does it matter if you get it in the first 10 seconds or the last 10 seconds of a PP?  Is it like saying that a 10 goal win is better than an OT win?

BTW, I'm not as sophisticated as Steve, so I just use Win%, OWP, and OOWP, which gets factored into the unused RPI, which is how I rank teams.
Clarkson College of Technology '77 & '78
2014 NCAA Women's National Collegiate Hockey Champions

Your team was KAI'd last night.

Offline Steve Wiitala

  • Starter
  • ***
  • Posts: 281
  • Karma: +10/-0
  • D3hockey.com Computer Rankings Specialist
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #20 on: December 09, 2013, 06:16:53 pm »
I agree partially with both you on the power play,  but a 17 second power play that is a gift from the opposition killing their own penalty shouldn't be counted against you and not scoring once on major should count more than not scoring on 17 second overlap penalty, too.  Also - scoring a shorthanded goal should be given bonus points.
D3hockey.com computer rankings specialist

Offline Matthew Webb

  • Moderator
  • Starter
  • *****
  • Posts: 408
  • Karma: +13/-1
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #21 on: December 09, 2013, 11:28:28 pm »
If the goal is scoring a PPG, does it matter if you get it in the first 10 seconds or the last 10 seconds of a PP?  Is it like saying that a 10 goal win is better than an OT win?
[/quote]

In terms of measurable power play efficiency I'd say it could make quite a difference.  If one team is scoring a PP goal per every 3.1 minutes of opportunity time and another scoring once per every 4.8 minutes, the first team clearly has a more effective power play though that wouldn't necessarily be as obvious using the standard system.

Could easily compute PK efficiency in the same manner.


"It is hard to free fools from the chains they revere" - Voltaire

Offline llama

  • Junior Varsity
  • *
  • Posts: 25
  • Karma: +0/-0
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #22 on: January 16, 2014, 02:36:22 pm »
Any chance Prof's Computer Rankings are on the way soon?  Lots of hockey has been played since the last edition....curious to see where his model has everyone ranked these days.

Offline joecct

  • Second-stringer
  • **
  • Posts: 94
  • Karma: +5/-0
  • NCAA Hockey in Maryland!!
    • View Profile
    • Facebook
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #23 on: January 19, 2014, 12:28:23 pm »
Here's my RPI rankings which include games vs. Canton but exclude games vs. the D-II's

Code: [Select]
Current Last Wk Name RPI
1 5 St. Norbert 0.6009
2 7 Williams 0.5976
3 2 Adrian 0.5974
4 1 Wisconsin-Stevens Point 0.5957
5 3 Plattsburgh 0.5941
6 9 Babson 0.5902
7 11 Trinity 0.5843
8 4 Norwich 0.5841
9 6 Massachusetts-Boston 0.5749
10 14 Geneseo 0.5719
11 13 Wisconsin-Eau Claire 0.5711
12 10 St. Thomas 0.5704
13 8 New England College 0.5679
14 16 Wisconsin-River Falls 0.5626
15 12 Utica 0.5590

Have at it.
Clarkson College of Technology '77 & '78
2014 NCAA Women's National Collegiate Hockey Champions

Your team was KAI'd last night.

Offline Matthew Webb

  • Moderator
  • Starter
  • *****
  • Posts: 408
  • Karma: +13/-1
    • View Profile
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #24 on: March 07, 2015, 03:07:10 am »
Steve,

If you still happen to check in here...

Since no one wants to release these things publicly, I wrote a simple desktop app that will calculate adjusted SOS to within a meaningless amount of thousandths of a point or so based on this weekend's results. In doing so, I noticed something interesting. Trust me that the math is right per what's listed in the NCAA manual, but...

We already know they average the averages which is an abomination in and of itself, but based on my backtesting they also don't appear to directly take into account the amount of times an opponent is played in a season.

Meaning, if a team plays another that is 23-2-2 five times it does not count as (5/#of games played) % toward SOS, but rather counts as (1/number of opponents played) %.

I'm not the stats doc but this seems ridiculous at the cursory level. If my explanation makes sense...is my take on it seeming ridiculous correct?
« Last Edit: March 07, 2015, 03:15:03 am by Matthew Webb »
"It is hard to free fools from the chains they revere" - Voltaire

Offline joecct

  • Second-stringer
  • **
  • Posts: 94
  • Karma: +5/-0
  • NCAA Hockey in Maryland!!
    • View Profile
    • Facebook
Re: 2013-14 D3hockey Computer Rankings Discussion
« Reply #25 on: March 08, 2015, 01:33:38 pm »
Matt

I'm not Steve, but I noticed the same thing.

Williams is 16-8-2  and 1-1 vs. Amherst.
Amherst is 19-4-2 and 1-1 vs. Williams

Therefore Williams's record when compared to Amherst is 15-7-2 and Amherst, compared to Williams is 18-3-2.  There is only one instance of the opponent when calculating OWP and OOWP (not two or more).  FYI, my spreadsheet would have Amherst to Williams as 19-3-2 and 18-4-2.  That distorts the SoS.

« Last Edit: March 08, 2015, 02:47:08 pm by joecct »
Clarkson College of Technology '77 & '78
2014 NCAA Women's National Collegiate Hockey Champions

Your team was KAI'd last night.