The 2024-25 College Basketball Model(s)
Introducing game predictions and regularly updated Elo ratings and rankings for men's college basketball.
While the NFL may dominate the sports scene during the Thanksgiving holiday, I have always considered Thanksgiving to be the real kick-off of the college basketball season. By this point, teams have a few games under their belt and the college basketball schedule is jam packed with awesome matchups and compelling tournaments such as the Maui Invitational, Players Era Thanksgiving Festival, and Battle 4 Atlantis. This is when college basketball really begins to grab our attention.
Earlier this month, I released my pre-season Elo ratings and projections for the college basketball season. If you missed it, you can check it out here:
Now that we have had some data to go off of, I will introduce my updated Elo ratings and rankings, which will be kept regularly updated (at lease twice-weekly) throughout the college basketball season. I also will introduce my single game forecast model which, unlike my college football model, is based strictly off Elo ratings with adjustments for home court advantage.
What Goes Into the Elo Updates?
Each team has a current Elo rating, which began as their preseason Elo, and adjusts after every game based on whether or not they won or lost, were home or away, how much of a “surprise” (if any) the result was, and the point differential. Elo is a closed system in which every point of Elo gained by the winner comes directly from the loser’s Elo rating. Prior to a game, the home team (assuming the game is played on the team’s home court), is given a boost in Elo based on their home court advantage estimation from KenPom. The exact Elo gain given to the home team is:
The first term is the team’s KenPom home court advantage, expressed as points favoring the home team.1 Following the game, a forecast delta is found based on how “surprising” the outcome was based upon the Elo difference between the winner and loser. The margin of victory is also considered, and Elo adjustment is more drastic the greater the point differential (in a positive direction for the winner and negative for the loser). Elo is adjusted by a factor - the ‘K-factor’ - which determines how sensitive the change in Elo is to recent outcomes which, after some trial and error, I set at 33. The higher the factor K, the more sensitive the adjustment is to recent outcomes. These factors are combined to produce an Elo point adjustment for each team following the game, the exact formula is below:
The second and third terms represent the forecast delta, with the third term the difference between the win probability of the winner and loser. I found that adding the win probability difference to the team Elo difference proved more accurate than Elo difference alone. The formula to calculate the probability of a team winning based on Elo ratings is as follows:
What the full formula above spits out is the Elo point adjustment, which is the point value added to the winner’s rating and subtracted from the loser’s rating. Once this adjustment is made to each team, their new Elo rating is determined.2
Current team Elo ratings can be found anytime under the College Basketball tab on this newsletter’s homepage, along with conference ratings and rankings based on average team Elo. As I said, I plan to keep these updated regularly and will be targeting twice-weekly updates. Here are the current Elo ratings:
Game Forecasts
We can use these Elo ratings to forecast game outcomes. As shown above, one can calculate the probability of a team winning or losing based upon Elo scores, with the Elo difference being the Elo for the team of interest minus the Elo for their opposing team. A projected point differential, in relation to the winning team, can also be found via the formula below:
These two calculations make up my single-game forecast model, which can again be found anytime under the College Basketball tab on the homepage, and currently shows forecasts for the games scheduled on the current day and the next day. On the prediction table, you will see an “Excitement Index” column. This is a quantification of how exciting the game is expected to be relative to the other games in the coming days. It is calculated by creating an index based upon how close the projected point spread is relative to the other projected point spreads of upcoming games.
I hope you enjoy these ratings and forecasts, I will be referencing them many times in posts and articles throughout the college basketball season, so be sure to familiarize yourself with them!
College basketball is in full swing, let the madness begin!
Games which are played on a neutral court have no home court advantage adjustment and are given a home court Elo boost of 0.
Significant portions of this formula is derived from the now-defunct (sigh) Elo models of FiveThirtyEight. Credit should be given there for a large amount of the research and mathematics that went into this.