As the high school season gets going, Inside Lacrosse couldn’t be more excited to unveil our new high school team ratings formula.
While the calculations are not done the exact same way, the basic concept of the ratings will be familiar to anyone who visited LaxPower.com. The highest-rated team in the country will have its rating set at 99.9, and then all the other ratings will be relative to that team with one point of difference in ratings equivalent to a predicted one-goal difference in a game played at a neutral site.
Similar to LaxPower ratings, there will be a 10-goal difference limit for each game and home field advantage will be a factor with a value determined by the calculations.
To get a bit more technical, calculations are done by taking an ordinary least squares (OLS) regression on a system of equations where each game is an equation and team ratings are the variables. The difference in team ratings plus the value of home field advantage is set equal to the actual difference in the score for the game. The regression calculations solve for the team ratings, which minimize the difference between the actual and predicted goal differential for each game.
However, there are some differences where our ratings calculations will deviate from the LaxPower ratings. Most significantly, it calculates every team’s score — nation-wide — at once, collectively, as opposed to relying on a combination of in region and out of region calculations. The result is that the differences in ratings between top teams will be greater and, I believe, more accurately reflect the differences between the best teams in the country.
With more computing power available than ever via the cloud, there is no need to break up the calculation, and it can be done all in one, more like the way the LaxPower college ratings were computed.
Similarly, it also uses a single home field advantage value for the entire country. Perhaps it is the case that home field advantage varies depending on the region of the country, but I don’t believe that there is any good reason to assume teams of high school players react significantly differently to playing at home vs on the road in different areas.
Just like the LaxPower ratings, these ratings will take a few games to achieve a reasonable degree of accuracy. Early in the season, they will suffer from two problems: a lack of any comparison data between groups of teams that have no games linking them, and overfitting based on very few games — or data points — linking teams.
If a group of teams in one state have all played each other and a group of teams in another state have all played each other, but no teams have played any games outside that group, then there are no data points by which to compare those two groups of teams. Our formula handles those situations by making the assumption that the average rating of all independent groups of teams is the same. Obviously this is not the case, but without any game scores with which to compare the groups of teams, this assumption serves to both come as close as possible and to ensure that the tendency of mistakes is to push ratings toward the national average.
In cases where teams have only played one or two games or there are two groups of teams with only one or two game scores linking them, the ratings calculation will potentially overfit and rely too heavily on very little information. This can cause ratings to shift more dramatically early in the season if the ratings were previously relying heavily on very few games to calculate ratings for certain teams and they happened to play some of their best or worst games of the year in those few games. Unlike, for example, Elo ratings systems, which require many games for team ratings to move either up or down, regression-based ratings, like ours, can result in a team being rated as the best or worst in the country or state based on just one game.
That means we are relying on all of you to help us make sure that our database of game scores and team schedules is as accurate and complete as possible. If you think your team is rated too low or another team is rated too high, make sure we are able to base that rating on as many games as possible.