How Well Can We Predict HS Basketball Games?
Applying our computer model to MHSAA Basketball
I hope you’ve all been well and that your 2023 is off to a great start.
My 2023 started off quite busy, which regretfully means I’ve had to let posting to this newsletter fall by the way side.
I apologize for this…. we’ve had some great content that I’ve teased on Twitter that I simply ran out of time to do. That’s on me, and I promise to make sure we get back on track this week and next.
So what does ‘getting back on track’ mean?
Well, we’ve got some interesting new stuff coming out the next few months; take a look at some of our planned topics:
MHSAA Basketball - Applying the computer model to boy’s basketball. We’ll have rankings for all 700 teams, analysis on the strongest districts, and more
Recruiting - A rehash of our popular series from last year, where we look at the association between talent & high school football wins
School of Choice - Where we analyze the relationship between out of district enrollments and high school football wins
A New Project - We have a new software project coming out, which I’m not quite ready to talk about. If you want to stay up to date on what we’re working on, we’ll be posting regular updates here.
If any of these interest you, stay tuned for more. And if you haven’t subscribed yet, make sure you do so by clicking the button below:
Lastly, if you like what you’ve read today (or in prior newsletters), could you do us a favor and share this with a few friends? A share from you is really the best way for us to grow our audience….. we appreciate any help we can get!
As for today’s topic, we’re turning the same computer model that you’ve seen for football to a new domain: high school basketball.
In today’s analysis, we will compare our computer model under both sports, with the hopes of understanding: just how well can we predict the outcomes of high school basketball games?
Let’s jump in…..
The Computer Model & HS Football
For those just tuning in, let’s quickly rehash how effective our computer model was at predicting high school football games.
This past season (Fall 2022), our computer model ran from Week 4 onwards (we don’t run the model Weeks 1-3 as we need data to train the model on). This means we attempted to predict the outcomes of 1,649 games.
The results were positive: we successfully predicted the winner in 1,368 of these games, meaning we had 83.0% accuracy.
We also set spreads for each game. These spreads let us go one level deeper: we can see how accurate the model is, given a certain spread. In theory, the model should be less accurate at the predicting the outcome for tighter spreads (it’s a better game, so the outcome is less certain), and more accurate at predicting the outcome for larger spreads (it’s a worse game, so the outcome is more certain).
Taking a look at the table below, this is exactly how things played out. You can see that for a spread of 0-5 points, the team favorited by our model won only 52% of the time. Against a worse opponent, where the spread was 5-10 points, that same team would win 65% of the time.
So let’s apply this framework to basketball. First up is our general accuracy: how accurate have we been?
Thus far, we’ve applied the computer model to 2,919 games. This excludes the first week and a half of the season, and certain games against out of state and home school opponents.
In those games, we’ve correctly predicted the winner 2,357 times or 80.7%. Not bad, but slightly worse than football.
As far as accuracy by spreads go, check out the table below. It’s the same one we show above for football.
As expected, accuracy of prediction increases as the spread increases. A game with a 0-5 point spread is predicted correctly only 54% of the time, while a game with a 5-10 point spread is predicted accurately 69% of the time.
One interesting observation is that for a given point spread, our model is as accurate or more accurate than football.
I know what you’re thinking: didn’t you just tell me that you’ve been less accurate predicting basketball games vs football games this year (83% vs. 80.7%)?
That’s right, we have been. But not because the model itself is less accurate: it’s because in basketball, there are more close games. For instance, thus far this basketball season, 18% of games are projected to have point spreads of 5 or less.
For football, this number is less, coming in at only 13.7%.
What does this mean? In one sentence: high school basketball has higher levels of parity between teams, and is thus harder to predict.
Said another way: there is a larger gap between the good & bad high school football teams than there is between the good & bad high school basketball teams.
That’s all for today, folks!
Stay tuned for next week Monday, where we will drop more basketball content (rankings for all 700+ teams in the State).
Also, later in the week (likely Thursday), we’ll be dropping a second basketball analysis, where we look at how tough each boy’s basketball district is rated by the computer.
If you’re interested in either, hit that subscribe button below so you don’t miss out.
For those interested in more HS football content, stay tuned… there’s more coming your way later this month.