For these standings we use the revamped Gonzalez system, but I also included the time honored Champ system for ease of reference. There’s currently a discussion and vote going in the Executive Board for alterations to our NCDA Ranking Algorithm, which has been used for seeding the Nationals bracket for 2016 and 2015 (plus a proto system used for 2014). The accuracy for the Nationals bracket for the two years was 93.75%. The predictive accuracy for the new Gonzalez system is currently 80.16% (196 technical upsets / 988 ranked matches). To briefly sum the incoming ratings with the current…
UK has the most Champ points currently. In Gonzalez they have climbed from #9 at the start of the season to #4 with a few quality wins. OSU’s losses have dropped their rating from #3 down to #6. VCU is still climbing with more wins. #4-8 ratings are super close and competitive. Same goes for #13-18. One solid win or bigger upset in either of these brackets could allow a team to jump many places.
Rank | Gonzalez Rating | Team | Champ Pts |
1 | 54.977 | GVSU | 6 |
2 | 50.318 | CMU | 2 |
3 | 48.124 | MSU | 6 |
4 | 45.344 | UK | 7 |
5 | 45.146 | Kent | 4 |
6 | 45.018 | OSU | 4 |
7 | 44.668 | SVSU | 1 |
8 | 44.448 | Towson | 6 |
9 | 43.817 | JMU | — |
10 | 41.216 | VCU | 4 |
11 | 41.047 | PSU | — |
12 | 40.459 | UNT | — |
13 | 38.865 | UMD | 2 |
14 | 38.719 | WKU | 2 |
15 | 38.364 | BW | — |
16 | 38.277 | UVA | — |
17 | 38.252 | Akron | — |
18 | 38.062 | UWP | 2 |
19 | 37.579 | BGSU | 0 |
20 | 36.773 | DePaul | 3 |
21 | 36.367 | UNL | 2 |
22 | 36.057 | SU | 0 |
23 | 32.321 | Ohio | — |
Short Explanation of the Gonzalez system
Gonzalez is similar to ELO and is a rating exchange system. If the high seed has a rating of 50 and the low seed a rating of 45, the Ratings Gap between these two ratings is 5.
The Gonzalez Exchange will be .500 if the high seed wins, which we count as the predicted result. The teams trade points equal to the Gonzalez Exchange: the winner gains the exchange, and the loser gives up their points. If the low seed wins, the low seed gains 1.500 in a technical upset. The low seed gains more points for defeating a tougher opponent.
The Rating Exchange is a scale from 0 to 2. The exchange directly corresponds to 0 to 10 Gonzalez rating points, but there are two types.
A predicted result will have an exchange from 0 to 1.
A technical upset will have an exchange from 1 to 2.
Other Variables
The minimum exchange is 0.010 and the maximum exchange is 2. A very strong team (50) could play a much weaker team (38). This Rating Gap is 12, causing the exchange to be -0.200. Without the minimum value, a weaker team could schedule hard opponents and raise their ranking by losing. With a minimum, every match at least counts for something, even if the value of the match is very small. The value of lopsided matches should also help prevent these from occurring more than occasionally.
Ratings are displayed to the thousandth. BGSU currently has a rating of 37.579, but we use many more decimal points (defined by whatever Google Sheets manages). BGSU is technically 37.5788529268111.
The best guarantee to raise a Rating is to play opponents of similar strength. A team with a rating of 41 versus another rating of 40 is going to be a pretty even exchange of either 0.900 or 1.100. Put together 10 wins of these similar strength matches and that team’s Rating will be about 10 points higher. That’s going from #12 to #2 in 10 matches. The lower a team is rated, the easier it to gain ground with quality wins or big upsets.
Home court advantage gives +3 to the host. At the start of the 2016 season, OSU was 47.691 and #3. When OSU hosted the Buckeye Invite, they went into the day with a 50.691 rating, enough to put them as a #2 ranked team facing their opponents.
Margin of Victory does not matter. Running the score up on a team doesn’t matter. The original inspiration for this system, World Rugby, uses a margin of victory modifier and adjusts the Rating Exchange just like home court advantage does. We don’t keep this variable simply because our record of points scored within the game is just not accurate. Some aren’t even recorded. Margin of victory never proved to be consistent representation of the match, so we go with a simpler Win/Loss approach. We also have running clocks and games that devolve into fun mix it up points.
Overtime halves the Rating Exchange. We enhance the Win/Loss variable by keeping track of Overtime results. If a team is able to top another in regulation time, they’ll get more out of the Exchange. The UWP/DePaul OT game at Nebraska this weekend predicted that Platteville would gain 0.694 with a win. Because DePaul was able to push the game into Overtime, UWP only gained 0.347 on the OTW. DePaul lost fewer points for keeping the game close, and UWP failed to gain points for not finishing up the game in regulation.
Gonzalez is a rolling system. It is very good at determining and adjusting ratings for teams that don’t play as often. A team that only plays 6 games a season against any caliber of team will have a pretty accurate rating compared to a team that can play a 20 match season. Because we don’t have a set schedule or minimum amount of games needed to be played per season, adapting to this issue is very important for any system we adopt.
That game from two years ago doesn’t matter to your rating as much as might you think it does. Technically speaking, a game affects ratings forever once played, just with a smaller and smaller weight that gradually diminishes to statistically nothing over time. Ratings at the start of the season are mostly about who was good last season. As teams play opponents, ratings will be adjusted accordingly based on a team’s win record and strength of schedule. Games toward the end of the season will count more. And for games played at Nationals, the exchange is doubled. It’s not impossible that a #12 going into Nationals can get a #3 seed and change their stars.
Specific Changes for 2017
This year, teams keep 75% of their rating from the previous season, which matches the data reported by Member Teams in the Summer 2016 off-season Proposal vote. On average, teams retained about 75% of their roster. This causes a minor hit to predictability (about 1-2%), but better reflects real world “losses” on rosters.
After cutting 25% off every rating, we need to revert that loss percentage back to the Mean so incoming teams aren’t ranked higher than pretty much everyone. The incoming team’s initial rating is 40. For the other 25%, we’ll take 25% of the Mean Rating at the end of the previous season. This helps prevent a huge League rating spread, that gap between the lowest rating and the highest rating. Currently the League Rating Spread is 22.656. A lower spread allows lower ranked teams to climb the rankings by winning less games.
For example, middle of the pack WKU was ranked #14/23 with 38.467 at the end of the 2016 Season, so we’ll keep 75% of that. Then we’ll add 25% of the end 2016 Season Mean, 41.488. (38.467 * .75) + (41.488 * .25) = 39.222. WKU starts 2017 with a 39.222 Gonzalez Rating.