You know how you used to tell your teacher you’d never use math in real life? Well this is where you use math in real life. For basketball. As Archimedes intended.
The Power Rankings (also known as “The Model”) are built on the structure of the Heal Points, partly to feel familiar but mostly because the Heal Points are a fantastic system for ranking teams that don’t necessarily play each other or even have opponents in common.
I’ve made a Top 10 version of the rankings for the Bangor Daily News, and you can see the full rankings on my website, along with so much more.
High school basketball power rankings
So how does it work?
The process starts with games. Instead of the binary won/loss method you’re used to, the Model takes the score of the game and applies both a home court adjustment and a Pythagorean expectation, a formula Bill James created for baseball that determines how many games a team “should” have won based on their runs scored and allowed.
But instead of using this for a full season, the Model uses it for each and every game. So when the Oceanside boys beat Medomak Valley 64-59 at home, that gave them a win percentage of 66 percent, which basically means that based solely on that result (and nothing else) you would expect Oceanside to win a rematch on a neutral floor 66 percent of the time.
But, when the Narraguagus girls beat Shead 41-36 at home — the same 5-point margin — that translated to a win percentage of 78.9 percent.
Why the difference? Well, the more points are scored in a game, the less secure a 5-point lead becomes. A 5-point lead with two minutes left in a 20-15 junior varsity game is pretty much game over. But a 5-point lead with two minutes left in a 100-95 game is nothing.
The Model then takes that percentage and applies it to the opponent’s preliminary index. So if your opponent is worth 10 points and you got a win percentage of 75 percent, you’ll get 7.5 points. If you got 10 percent (i.e. you lost), you’d still get 1 point.
Yes, you can gain points by losing. In fact, the Model doesn’t care if you won the game. A win-loss record is informative, but you can learn a lot more about a team by their point differential and the quality of their opponents. Beating Our Sisters of Mercy doesn’t tell you anything. But losing by 2 points to a good team? That tells you a whole lot.
The preliminary index is similarly calculated, with some differences. Instead of just getting 40 points for beating a Class A opponent, the preliminary index gained is a percentage of, well, not that exactly, but something close to that. Beyond that, there’s a number of small adjustments I’ve added over the year to dial in the rankings. These power rankings are updated daily.
Once the Power Rankings have been calculated, the Model then applies another adjustment to make a prediction of every game on the schedule, both in the form of a percentage and point spread (which involves another adjustment). These predictions are correct more than 80 percent of the time and something like 99 percent of the time when a team is given a greater than 85 percent chance of winning. The Brunswick girls blowing out Greely the other day? The Model had the Dragons winning that by nearly 20 when everyone else in the state had it pegged as a close game.
These predictions are then used to calculate the odds a team will make the tournament, the likelihood they’ll be able to get a particular seed, whether or not they’ve clinched a tournament spot, and the odds they’ll take home the Gold Ball once the tournament starts. These predictions start sometime after Christmas and are obsessively watched by a large chunk of the coaches in the state who really, really want to know if they’re in the tournament yet.
If this all sounds excessively complicated, it’s because it is. But it’s also fascinating and a great thing to argue about when the junior varsity game turns into a blowout.