Facebook Twitter

2016 NCAA Basketball Model

We have again made minor changes to our NCAA Basketball model for this year's NCAA Tournament to improve overall accuracy. The 2016 College Basketball Ratings and Odds now include recruiting class ratings and players drafted by the NBA as additional inputs for calculating a team's rating.

The Base Model

From 2012-2014, ProFootballLogic used a sort of "fair" model for rating college basketball teams, in that teams were judged only by the current season's results. Every game's result was assigned a probability of occurring based on the competing teams' ratings, the location, and the final score difference. Then each game's odds were multiplied together, and an optimization was run to find the most likely combination of every team's rating, the extent of home court advantage, and the extent of random variation in a game. Then the difference between 2 teams' ratings functioned as essentially a point spread for predicting any possible matchup.

Essentially what that model did was provide the best ratings possible under 4 conditions: that a game's final score is the only data used, that a team's quality stays constant over the course of a season, that point differential acts linearly and is more predictive than game outcome (W/L), and that factors outside the current season are not considered.

The first condition is used simply because in a sport with 351 teams, compiling and analyzing further game stats or injury data would be incredibly time intensive. While we are certain that a game's final score is more predictive than simply which team won, we have not fully explored the 2nd and 3rd conditions and may in the future, but in general are confident that they are not far off reality. So to this point, the final condition is the one we have worked on improving to achieve better accuracy. It might seem like around 30 game results for each team would make preseason expectations irrelevant, but in a division with 351 teams that aren't very well connecting by those games, they don't.

In 2015, we added the past 2 seasons of a team's rating to the model. Because a team's quality in prior years, especially the most recent year, is highly predictive of it's current quality, a factor was added for the odds that a team's rating would vary a certain amount from the preseason expectation based on a regression using the prior years' ratings. This factor is then multiplied in with all other odds in the optimization, functioning like a Bayesian prior.

For more detailed information on the base model, see last year's article on the 2015 NCAA Basketball Model.

2016 Additions

To make the preseason expectations even more accurate, in 2016 we have added a team's rating from 3 years ago, recruiting class ratings from 247sports.com, and the NBA Draft position of every team's players as additional factors in determining a team's preseason rating. This was done by optimizing weight factors for each variable in producing accurate preseason ratings for past seasons.

In all, we now use the past 3 years of a team's rating and the past 4 recruiting classes and NBA drafts. The result are preseason ratings that rival common published top 25 preseason rankings, but include every team rather than just 25.

The primary factor remains the team's rating from the previous season, which alone can make last year's best team's preseason rating about 26 points higher than the very worst team. Team ratings from 2 seasons ago can only provide a difference of a couple points, while from 3 seasons ago will be less than a point.

The past 4 recruiting classes are all valued somewhat similarly, slightly peaking at the 3rd most recent class. While recruiting classes prior to the most recent one may seem redundant since their quality could also show up in past year's team ratings, their relevance can be attributed to things like increased playing time for those players over the years. Overall, the quality of each recruiting class can vary the preseason rating of team by around 3 points at most.

The value of past NBA draft picks follows an interesting but somewhat predictable patten. Having lost several players to the most recent NBA draft can decrease a team's preseason rating around 4 points, depending on how high they were drafted. But having players drafted 2 drafts ago has a negligible impact since most of that expected decline has already shown up in the team's rating from last season. However, having players drafted in prior drafts can actually increase a team's preseason slightly by a point or 2, presumably because it is just a sign of a strong program even beyond what past team ratings and recruiting classes can tell us.

Overall, we expect that the new additions may increase the accuracy of certain teams' preseason ratings by several points, which could in turn move their final pre-tournament ratings by a point or 2. It's not a huge change, but over the course of a long tournament it could noticeably change odds. All told, our team ratings and tournament odds routinely beat most media outlets or pundits, stack up well against even the most sophisticated ratings, and closely align with betting odds.

Recent Articles
Grading NFL Franchises   -   9/28/17
Tom Brady Is Not The G.O.A.T.   -   8/3/17
NFL Census 2016   -   4/19/17
Do NCAA Basketball Teams Get Hot?   -   3/15/17
How Good Are Pro Bowl Teams?   -   1/28/17
Wk 17 Playoff Scenarios 2016   -   1/1/17
Rating Field Goal Kickers   -   8/28/16
Rating NFL Quarterbacks   -   6/20/16
2016 NCAA Basketball Model   -   3/16/16
NFL Injury Rate Analysis   -   2/22/16
Fundamental Articles
Site Summary and Features   -   7/5/12
The ProFootballLogic Method   -   7/5/12
Stats Explained   -   7/5/12
How Variation Affects Outcomes   -   7/5/12
Ratings Explained   -   7/5/12
General Play Type Analysis   -   7/5/12
London and Home Field Disadvantage   -   8/21/12
Ranking College Conferences   -   1/11/13
Draft Position and Player Quality   -   4/25/13
The Changing Landscape of the NFL   -   5/21/13
College Yearly Rating Regression   -   8/28/13
How College Basketball Rankings Fail   -   2/28/14
Franchise Tag Position Problems   -   3/7/14
Developing a World Cup Model   -   6/9/14
Optimizing College Playoff Selection   -   11/26/14
The Science of Football Deflation   -   1/27/15
2015 College Football Model   -   9/3/15
2015 NFL Team Ratings Model   -   10/20/15
NFL Injury Rate Analysis   -   2/22/16
2016 NCAA Basketball Model   -   3/16/16
How Good Are Pro Bowl Teams?   -   1/28/17
Do NCAA Basketball Teams Get Hot?   -   3/15/17
NFL Census 2016   -   4/19/17
Tom Brady Is Not The G.O.A.T.   -   8/3/17
Grading NFL Franchises   -   9/28/17