Inline Power Rankings

What the number means

A team’s IPI is its strength in goals against an average Division I team. A team at +2.0 would be expected to beat an average D1 team by about two goals; a team at −3.0 would be expected to lose to one by three. Two teams’ numbers read straight across: a +1.5 team against a −0.5 team is a two-goal favourite, which in college inline hockey means it wins about three times out of four.

How it works, in plain English

Every game of the season is one piece of evidence: the home team’s rating minus the away team’s should equal the goal margin. The model finds the one set of ratings that fits all of those games at once, as closely as possible. That is the whole idea. A few details make it work in practice:

• Blowouts are capped at eight goals, inline hockey’s mercy rule: many tournaments end a game at an eight-goal lead, so eight is as far ahead as a team can get, and a 12–0 win counts the same as 8–0. An overtime or shootout result counts as half a goal.
• Teams start from their division’s level. Before a game is played, a D2 team is assumed to be about a goal and a half behind a D1 team, a D3 team three goals, and an AA team two and a half goals behind its school’s top team. Results then move each team away from that starting point, more the more games it plays. The games between divisions (AA squads against D-teams, D2 against D3) tell the model how far apart the divisions really are each season.
• Nationals games count double in the final ranking: they are the games every team is built for, against the best fields.
• Only teams that played are ranked. A team entered for the season that never took the floor, an alumni side, or an Affiliate team is left out, games included.
• The ranking is rebuilt every night during the season, from scratch, so a result changes every rating it touches.

How accurate is it?

We test it the honest way: replaying each season in order. Before every event day, the model is fitted on the games played so far and picks that day’s winners; then, fitted on the whole regular season, it picks the Nationals winners. Over 8 seasons (2018-2019 to 2026-2027) it has picked right in 77.4% of regular-season games and 68.0% at Nationals. The table compares it with the other rankings we built and tested on the same games. Log loss scores the win chances a model gives, not just its picks (lower is better).

This season (2022-2023): Tier-points matrix 76.7% · InlineStats Power Index (IPI) 78.3% · Elo rating 76.3% · at Nationals: Tier-points matrix 63.5% · InlineStats Power Index (IPI) 64.3% · Elo rating 64.3%.

Why it beats the other ways of ranking

• Standings and win-loss records ignore who you played and by how much. A 10–2 team in a weak region and a 7–5 team in a strong one look like the first is better; the games between regions usually say otherwise. Picking the better record gets 72.7% of games right, four or five points fewer.
• Elo ratings (chess-style, used by many sports sites) move one game at a time, so a team’s number depends on when it played whom, and a win over a team that turns out to be strong is credited too little. Elo also uses the margin only as a nudge. Fitting the whole season at once fixes both, and gives better-calibrated win chances.
• A tier-points matrix (the points-table system used in college inline hockey) awards fixed points for beating a team from a given tier. The tiers are set in advance, so a season in which D2 is unusually strong, or a division’s top half is really its top three, is scored wrong all year. The IPI lets the games say how good each division is.
• Most “power rankings” on the internet are a panel’s opinion. This one is a calculation you can check: the same inputs always give the same ranking, and its accuracy is measured on games it hadn’t seen.

Reading it well

• Early in a season it is weak. On the opening weekend it is right about 60% of the time, on the second 71%, then 76–80% from the third weekend on. That is why a season’s ranking only appears after three event weekends.
• Nationals are hard to predict for everyone. Single-elimination hockey is noisy: at Nationals every ranking, this one included, sits within a few points of simply picking the better record.
• The champion isn’t forced to #1. The trophy marks who won; the number says how strong the team was over the whole season, Nationals counted double. A team that dominated the regular season and lost a one-goal final can still rate higher than the champion, and usually should.
• A rating is only as good as the data. Scores come from the NCRHA’s published results; a missing or wrong score shifts the teams it touches. Corrections are welcome through Contact Us.