Who’s the boss?

The finals of the NCAA’s field hockey conferences will begin this week. The Conference champions will secure one of the guaranteed spots in the NCAA Division I Field Hockey Championship starting on 17 November and be joined by eight ‘at-large’ spots selected by the NCAA on 5 November.

The NCAA Field Hockey Competition operates differently to hockey competitions in many other parts of the world. There are ten conferences in Division I, and each conference has 7-9 teams. In addition, there are three independent teams. This leads to a total of 82 teams in the competition. Each team will play the other teams in its conference once to determine the Conference standings. The teams play a further 7-12 games against Division I teams. The additional games together with the conference games determine a team’s overall record.

The competition format makes it difficult to rank the best field hockey teams in the NCAA. Specifically, the quantity of wins may not indicate the best team – comparing teams based on wins, win percentage or the points derived from wins may not be fair as the quality of the opposition might differ significantly between two teams. Therefore, the Strength of Schedule is often considered together with the team’s win percentage when comparing teams.

One way to do this is using the Rating Percentage Index (RPI). The RPI is a metric used to rank field hockey teams by the NCAA. The RPI has three components with the following weightings:

• Win Percentage (25%)

• Opponents’ Win Percentage (50%)

• Opponents' Opponents' Win Percentage (25%)

The Win Percentage is the number of wins divided by the number of games played.

The Opponents’ Win Percentage is an average of the opponents’ win percentages excluding any games where the teams have played each other.

The Opponents' Opponents' Win Percentage is an average of the win percentage for teams that the opponent has played against.

The RPI heavily weights the Strength of Schedule. It tends to benefit teams from strong conferences and adversely impact teams from less strong conferences. This is one of the reasons that it may not be the best predictor of results when trying to determine who will do well at the Conference or NCAA Championships.

Several other rating systems exist, particularly for NCAA Basketball (e.g., Sagarin Ratings, Ken Pom), and there are other rating systems for NCAA field hockey (e.g., Kevin Pauga Index (KPI) and FieldHockeyCorner).

Instead of relying on these rating systems, we built our own, straight-forward statistical model and updated each week. Our thesis was that effectively predicting goal differences would enable us to predict the winners of future matches. The most recent iteration of the model used data from the 693 regular season games in 2023 to predict goal differences (aka margin of victory) based on the teams playing and whether there is a home field advantage. We tested the model in the latter third of the season. It performed much better than relying solely on RPI or win percentages to predict winners. We have used the latest iteration of the model to help guide our assessment of the likely winners for the ten conferences (see below).

There appears to be a small home field advantage for field hockey teams in the NCAA. We estimate an advantage of 0.35 goals per game for the home team. The home field advantage is also evident from the distribution of goal differences (home team goals less away team goals) from the 2023 regular season is shown in Figure 1. The histogram has an unusual shape with more positive goal differences due to the home field advantage and a gap in the middle of the histogram. The latter occurs because drawn games are not possible in NCAA field hockey.

Figure 1: Histogram of the goal difference for the home team (or team two) from the 2023 NCAA regular season. A negative number indicates how many goals the home team was behind. Data includes neutral venues, at these venues the goal difference is measured as team two’s goals less team one’s goals.

Picking winners

Disclaimer: The results below are for entertainment purposes only.

There are six conferences where, according to our model, the number one seed appears to be a strong favourite to become Conference Champion:

• Atlantic 10: Saint Joseph’s

• Big East: Liberty

• Big Ten: Northwestern

• Ivy: Harvard

• North East Conference (NEC): Fairfield

• Mid-American Conference (MAC): Miami (OH)

There are four conferences where the model suggests there will be a close contest between two teams to become Conference Champion. Each conferences’ two major favourites are:

• ACC: North Carolina and Duke

• America East: UMass Lowell and UAlbany

• CAA: Monmouth and Drexel

• Patriot: American and Lafayette

The ‘at large’ contenders

The NCAA's Division I Field Hockey Committee has discretion to use several criteria in selecting at-large teams:

• RPI

• head-to-head competition

• results against common opponents

• late season performance (i.e., the last seven games)

• significant wins and losses

• KPI

Based on these criteria, it is possible to speculate on which candidates would likely be selected as 'at large' candidates using the RPI and KPI rankings. For instance, if the first seeds are unable win their Conference, then the following teams are strong candidates for ‘at large’ selection:

• Northwestern (Big 10)

• North Carolina (ACC)

• Liberty (Big East)

• Harvard (Ivy)

• Saint Joseph's (Atlantic 10)

• UMass Lowell (America East)

Excluding the first seeds, we consider the next best candidates for an 'at large' spot come from the Atlantic Coast and Big 10 conferences:

• Iowa (Big 10)

• Duke (ACC)

• Virginia (ACC)

• Maryland (Big 10)

There are several other good candidates for an 'at large' spot, which coincidentally come from the Atlantic Coast and Big 10 conferences. Whether the following teams are selected will depend on how many spots are available.

• Rutgers (Big 10)

• Boston College (ACC)

• Syracuse (ACC)

• Ohio State (Big 10)

• Louisville (ACC)

Improving the model

The model used for this analysis was deliberately straight-forward. While the model demonstrates improved predictive power compared to the RPI baseline, it is clear further improvements can be achieved through additional data and more advanced analytic techniques.