D-I Women’s Bracketology:
Feb. 14, 2005

Harvard’s sweep of Union this weekend moved the Crimson into first place in the ECAC standings–and down from fifth to eighth in the USCHO.com Division I Women’s Pairwise Rankings? The National Collegiate Women’s Hockey Committee implemented a criteria change before the 2003-04 season to partially correct for the cause of this huge drop–playing opponents with records so bad that even stepping onto the ice with them hurts your NCAA standing. The USCHO Pairwise Rankings still have yet to take into account this change, and as a result are slightly understating the NCAA prospects of most ECAC teams.


Why ECAC teams? Because all the ECAC teams play two games each against Vermont and Union, the two weakest teams in all of college hockey as measured by the Ratings Percentage Index (RPI), and the RPI is one of the five NCAA selection criteria used in the Pairwise Rankings. Under the RPI, a win over Union this season is worth about the same as a loss to Harvard, and a win over Vermont is worth about the same as a loss to Dartmouth.1 The recent change in the RPI under the newer criteria is that up to four of a team’s wins are dropped from the RPI calculation if those wins are worth less than the average of every other result on that team’s schedule.2 Most ECAC teams’ RPIs on USCHO are most understated relative to the newer criteria because the wins against Union and Vermont are worth less than all other wins.

The following table, as of Feb. 14, lists the RPI calculated using the newer NCAA criteria, the RPI calculated using the older pre-2004 method (the RPI still used by the USCHO Pairwise Rankings), the RPI rank using the pre-2004 method, and the difference in RPI between the two methods.3

Rk  Team             New     Old    Rk(Old) Diff
1   Minnesota        .697    .679   1       .019
2   Dartmouth        .678    .641   3       .036
3   Minnesota-Duluth .656    .644   2       .012
4   St. Lawrence     .636    .612   4       .024
5   Wisconsin        .623    .611   5       .012
6   Harvard          .612    .587   8       .025
7   New Hampshire    .612    .598   6       .014
8   Mercyhurst       .605    .597   7       .008
9   Princeton        .586    .574   9       .013
10  Providence       .552    .546   10      .006
11  Ohio State       .520    .520   11      .000
12  Brown            .516    .507   13      .008
13  Colgate          .509    .508   12      .001
14  Connecticut      .500    .500   14      .000

The following table lists what the Pairwise Rankings would be using the new NCAA criteria, as well as the current Pairwise Rankings which use the older pre-2004 criteria:

Rk      Team          Rk (Old)
1       Minnesota     1
2       UMD           2
3       Dartmouth     3
4       St. Lawrence  4
5       Wisconsin     5
6       Harvard       8
7       Mercyhurst    6
8       New Hampshire 7
9       Princeton     9
10      Providence    10
11      Ohio State    11
12      Colgate       12
13      Brown         13
14      Connecticut   14

Dartmouth and Harvard are the two teams that gain enough from the change in criteria to jump in RPI rank. Harvard is the only team that jumps in the Pairwise Rankings under the new criteria.

Bracket Prediction

The NCAA tournament has three automatic bids for conference postseason champions and five at-large bids. At present, all three conferences with automatics bids have teams in the top eight of the Pairwise Rankings, so we assume for now the teams selected will be the top eight teams in the Pairwise Rankings.

The NCAA can only seed two of the eight teams. If the season ended today, the criteria suggest Minnesota would be No. 1 and UMD No. 2. Teams are then paired geographical in order to minimize air travel with an eye towards bracket integrity. Because there are only three teams from the West Region, two must be paired with each other to avoid three flights if there exists a bracket pairing that only requires two flights. Therefore, Wisconsin is paired with UMD. Mercyhurst is over 500 miles from Dartmouth, so it must be paired with Minnesota or St. Lawrence. Superior bracket integrity results from pairing Mercyhurst with Minnesota and New Hampshire with Dartmouth. Under these pairings, each half of the bracket is balanced with the Minnesota site winner playing the St. Lawrence site winner, and the Dartmouth site winner playing the UMD site winner.

If the season ended today four weeks early, the expected bracket would be:

Mercyhurst at No. 1 Minnesota
Harvard at St. Lawrence
New Hampshire at Dartmouth
Wisconsin at No. 2 Minnesota-Duluth

1 Such nonsensical properties of the RPI are why USCHO has always endorsed ranking systems such as KRACH, in which no win can lower a team’s ranking more than any loss.

2 Another change in the RPI under the newer NCAA criteria is to drop postseason wins against teams with sufficiently poor records, but that will not be relevant in the calculation of the Pairwise Rankings until March.

3 Like the automated USCHO Pairwise Rankings, the RPI calculations above are calculated as if the season ended today. Consequently, certain games may be dropped in that RPI calculation which may not be dropped at the end of the season. USCHO intends to implement the latest NCAA criteria in the automated calculation of the Pairwise Rankings sometime in the near future.