Just how accurate is ESPN's BPI? We decided to take a look.
Every March, college basketball fans across the country are bombarded by all sorts of statistics and analytics.
One of the favorites used by ESPN is its own metric: the Basketball Power Index, or BPI.
The BPI generates a 1-351 team ranking and is also used to predict outcomes. For every game, the BPI spits out a win probability and predicted margin of victory.
You may find yourself wondering how accurate these predictions are, so HawgsDaily decided to do some digging and answer that very question.
(It’s worth noting that the NCAA Tournament selection committee does not use the BPI or other advanced metrics, such as KenPom and Sagarin ratings. That could change in the coming years, but as it stands, the committee largely relies on RPI, a mathematical formula based on win-loss records.)
Because checking every Division I basketball result would be extremely difficult and tedious, we examined only conference games involving SEC schools. That is a sample size of 126 games.
Not surprisingly, teams given a win probability of at least 90 percent went 19-1 this season. The lone loss came on Jan. 21, when Florida was given a 92.5 percent chance to win, but lost to Vanderbilt 68-66.
On the opposite end of the spectrum, teams with a relatively low win probability (between 50 and 59.9 percent) were 6-13.
Overall, the BPI correctly picked the winner in 88 of 126 games. Here’s a complete breakdown of how SEC teams did with different win probabilities:
|80% - 89.9%
|70% - 79.9%
|60% - 69.9%
|50% - 59.9%
In order to determine the accuracy of the predicted margin of victory, HawgsDaily compared the BPI to the most dependable source of sports predictions in the history of mankind: Las Vegas.
We used the point spreads compiled on OddsShark.com for this exercise and assumed every game was -110, meaning you have to risk $110 in order to win $100.
How much money would you win or lose if you bet on every conference game in the SEC using the BPI as a guide?
For example, if Arkansas was favored by five points in the BPI and four points in Vegas, our hypothetical bet would be placed on the Razorbacks to cover the Vegas spread. Conversely, if the BPI favored Arkansas by three points, our hypothetical bet would be for the Razorbacks’ opponent to cover.
Using those parameters, an $11 bet on all 126 games would have resulted in a loss of $41.
However, some of the BPI and Vegas spreads were separated by only a fraction of a point. (The BPI’s spreads use increments of 0.1, while Vegas spreads use increments of 0.5.)
If you only bet on the 85 games in which the BPI and Vegas spreads were separated by at least one full point, you would have won $84.
At least in this small sample size, ESPN’s BPI is a solid predictor of college basketball games.