Alternatives to the Current Standings

[WKU-Perrone-76]

With the updated RPI, here is the average ranking position of each team using their position in my system, the Win-Loss % ranking system, the NHL Point system, and Andy's RPI.

Attachments:

[WKU-Perrone-76]

I've updated everything after the Chicago tournament today, when Andy gets a chance, can you guys send me the updated RPI standings?

[hiller 87]

The standings aren't updated right, JMU is really 9-2-1 and Nebraska is 0-4. I'll get the RPI soon though

[WKU-Perrone-76]

Okay, cool. Who were JMU's losses to? So far I have their loss to GVSU and their OT-L to Kent. In my formulas I don't add their OT-Ls to the Loss category so I have them 9-1-1 with 11 games played total. If you have 11 games played total for them with a record of 9-2-1 then we're on the same page, I believe.

[hiller 87]

That's what I've got too. Nebraska is 0-4 as well. Other than that everything looks good.

[Zigmister]

i fixed that error

[Zigmister]

Ok good people! Please read carefully, I’ve put a lot of thought into the following words, and I’m even dedicating a class project to determining a ranking system. And I still don’t think we should have rankings at all!

Data sets to work with:

2012, 2011: Complete Data that should be used to test formula results against historical outcomes.
2013: Incoming Data to test formulas and provide fair match ups at Nationals.
2010-2004: Limited incomplete results to use as examples. (GVSU & MSU are mostly complete)
These results are compiled in publicly available Google Drive Spreadsheets, and could be transferred into a MySQL database if needed. Most likely, they will be transferred. The data is open and I could make available the ability for coders to design queries to produce their own Standings. If somebody inquires.

Point: I think we should be using the two completed seasons (2011 and 2012) to determine how to rank teams.

Problems Identified:

The NCDA is a small, divided league, mostly concentrated in the Midwest with 25 Member Teams. There isn’t a set or required amount of games per season, nor required opponents, and in many ways the Nationals weekend is the only league matches a Team could get all year. In order to be a fair representation, Rankings needs to take in effect the following:

• Isolated, satellite teams on the periphery (Nebraska, Louisiana, New York)
  
• Teams that play each other many times over the Season (MI, KY, East Coast, Chicago)
  
• Dominating programs in close proximity to each other (Michigan)
  
• Some teams may have 20 game seasons, some may have 4 game or fewer seasons.
  
• Teams may not play at all during one season, then pick up play in the next.
  
• Results are not explicitly cumulative. For example, a tournament may have 10 matches played, but they might not be reliably listed chronologically, in an hour by hour way. This is a problem for strength of schedule calculations. (Must include a special rule that all games on the same date have the same inherent SoS or Ratings Change for that day.)
  
• Using a margin of victory bonus is not reliable because of:
  - Inaccurate reporting of the actual final point total of the match.
  - Point differential is not an accurate display of team strength in the sport of Dodgeball. A more reliable statistic would rely on throws, catches, team catches and the other types of actual gameplay, but obviously this is unfeasible for a large ranking scale.
  
• Retaining the ability to change the formula on a year to year basis, adapting to the needs of the League.
  
• Overtime can gain more prominence because of stallball tactics.
  

Teams can’t be too penalized for lack of play because of distance issues; we’ll just lose those teams and hinder expansion. Teams should also be rewarded for their monetary commitment to playing games and traveling. Ultimately we should decide our own fate, but use the elements of history to best develop a system that meets the unique needs of our brand of dodgeball.


Nationals:

Any ranking system we come up with is ultimately designed for the Nationals event. Considering Nationals, the Event consists of two portions, pool play and a tournament bracket.

Pool play is used by many teams to be a considerable amount of that team’s matches during the course of the Season. The goal has always been to give teams games against teams they wouldn’t normally meet because of geographic/funding/travel difficulties. Pool play matches should be scheduled to give a competing team:
1. A fun and valuable experience (matches relating to hard, even, and easy levels of play).
2. Matches against opponents they might not have played before (because of the sheer amount of games played in the Season, some Teams can’t be guaranteed never-played-before teams)

Based off the outcome of the Regular Season and Nationals Pool Play results, the Rankings determine fair seeding for the Tournament bracket.

Conventional tournament seeding dictates high seeds against low seeds, but this assumes that the seed has a reliable “predictive accuracy”. The IRB rankings explanations explains: “If, over a period of time, the system tends to be good at predicting which side will win each match, then we can be confident that it is presenting an accurate and reliable picture of current strength, and responding appropriately to changes in form.” A reliable predictive accuracy is what we should be focused on creating, in regards to a dodgeball ranking system.


Other Misc (Radical Radical) Ideas:

• Dismissing Standings all together for pool play, teams sort out their own matchups, pick up style.
  
• Dismissing pool play, teams play a bunch of fun based Rulesets instead of NCDA rules. Sunday remains using the classic NCDA Tournament bracket.
  
• Only using the last 10 matches (or variation of) to determine Standings
  
• Including last year’s Nationals results in the pool play calculation, along with the regular season results.
  
• Using a weighted average of all the different systems instead of a equal average idea that was presented by Perrone.
  

********

Options detailed thus far.
UK, if you'd like to submit your data, i'd appreciate it. I'm starting to code these formulas.

W/L Percentage (WLP): Win over loss percentage. [W / L]

NHL System: Overtime losses are rewarded. [W=2, L=0, OTL=1]. Tie breakers: Better WLP, Total matches played, date of league entry.

The Lieblich System: Strength of schedule based on NCAA Basketball RPI.
Formula: [(Team’s WLP * .25) + (OWLP * .5) + (OOWLP *.25)]

The Perrone System: Quality win component similar to the BCS.
Basic formula: [(Wins+OT-Ls)-(Losses)] / Total Games Played
[ {
(W, OWLP = 1.0 WLP) = 4 points
(W, OWLP = .75-.999 WLP) = 3.5 points
(W, OWLP = .5-.749) = 3.25 points
(W, OWLP = .25-.499) = 3 points
(W, OWLP = .001-.249) = 2.75 points
(W, OWLP =  0 = 2.5) points
+
(OTL, OWLP = 1.0 WLP) = 1.05 points
(OTL, OWLP = .75-.999)= .975 points
(OTL, OWLP = .5-.749) = .95 points
(OTL, OWLP = .25-.499) = .925 points
(OTL, OWLP = .001-.249) = .9 points
(OTL, OWLP = 0) = .875 points
} - {
(L, OWLP = 1) = -1.25 points
(L, OWLP = .75-.999) = -1.375 points
(L, OWLP = .5-.749) = -1.5 points
(L, OWLP = .25-.499) = -1.625 points
(L, OWLP = .001-.249) = -1.75 points
(L, OWLP = 0) = -2 points
}
] / Total Games Played

Gonzalez System: A Ratings Exchange system modeled after the IRB.
Formula: RatingsChange = .1 * (RatingsGap) + GameResultIndex (W=1, L=-1)

[hiller 87]

Ok so I've actually been thinking about this, specifically regarding Nationals pool play. I think we should still have pool play, and that it should be kept in it's current format (3 games against schools that you don't normally get to play.) However, this obviously makes for a tough schedule for Nationals.

Using this year's current standings and adding NSULA, WIU, and UWP as the 18th, 19th, and 20th teams, I came up with this:

Pool 1:
UK

Pool 2:
Kent

Pool 3:
JMU

Pool 4:
MSU

Pool 5:
GVSU


Now SVSU should theoretically slide into Pool 5 since they're the six seed. But the problem of making sure you get different regions in the pools makes me make the regions like this:

1: UK, DePaul, OSU, UWP
2: Kent, CMU, WKU, VCU
3: JMU, SVSU, BGSU, WIU
4: MSU, MAD, TU, NSULA
5: GVSU, UMD, Miami, Nebraska

What I did was basically try to do the standings in order, but if there was a conflict of regions I slid the team down one notch. This isn't a problem unique to college dodgeball though, the NCAA Basketball tournament deals with the same problem with conferences.

The best solution I can think of is:

Polls.



Just kidding.

[WKU-Perrone-76]

Polls. lol. No. Remember when that was a legitimate thing? We thought that was a good idea??

When WKU hosted Nationals 2011, the way we did pools were based on rank in region:

Pool A: 1st IL, 2nd MI, 3rd KY, 4th OH, 5th IL
Pool B: 1st MI, 2nd KY, 3rd OH, 4th IL
Pool C: 1st KY, 2nd OH, 3rd IL, 4th MI
Pool D: 1st OH, 2nd IL, 3rd MI, 4th KY

And at that time the rankings were determined by number of wins, then by win percentage. So the pools ended up looking like this:

Pool A: WIU, MSU, UofL, Miami, Moody
Pool B: CMU, UK, Kent, UWP
Pool C: WKU, BGSU, Kansas, SVSU
Pool D: OSU, Depaul, GVSU, NSULA

I think pool play is important in determining, or at least having at least some weight in, the determination of rankings for bracket play on Sunday. Like you said, some teams may not get to play before Nationals and for them to be put at the bottom might not be fair. But this is just something we'll have to play around with in order to come to a conclusion everyone can agree with.

[WKU-Perrone-76]

Here are the updated rankings I have.

Attachments:

[hiller 87]

obviously a little thrown off with UWP's win percentage but other than that it looks good!

[mtrip8]

can someone refresh me on the way we have calculated the seeds for day 2 of nationals at the past nationals as in what percentage of it is determined by regular season and what percent is day 1 of nationals?

[WKU-Perrone-76]

Since I've been in the league, here have been the methods for seeding calculations.

GVSU - April 2009 - Record from Saturday play put you into either the 3-0, 2-1, 1-2, or 0-3 group, the "tie breaker" was a vote during the captain's meeting.

BGSU - April 2010 - Record from Saturday play put you into either the 3-0, 2-1, 1-2, or 0-3 group, the "tie breaker" was the amount of points you scored minus the amount of points scored against you. Highest point differential gave you a higher seed within your 3-0, 2-1, 1-2, 0-3 group.

WKU - April 2011 - We had gone to the NHL Style Point system at this time. Each win from Saturday gave you two points, each OT loss gave you 1 point, a loss was no points. Same point system for regular season as well. The seeding for Sunday was determined by 80% of the Saturday play and 20% of the Regular Season points. In the event of a tie, the win-loss percentage was used as a tie breaker.

SVSU - April 2012 - Same as 2011

[kentuckybrown]

Saturday: Will be based on your wins for the regular season and what schools are in your state or who you play all the time. (Rewards those with a lot of games played)

Sunday: This year what we are tinkering with is having a percentage of your points be Saturday and the rest is a combination of reg season wins/RPI/SOS/WL Ratio and etc. Nothing is set as of today as we are still working on fair formulas to reward teams. (Rewards teams who do well on Saturday and have quality wins throughout regular season)