05-02-2018, 11:53 AM
(This post was last modified: 05-02-2018, 11:58 AM by BostonCard.)
Yes, Hulk; it is the fact that they play in the SEC. At an extreme, consider a team that loses every game against teams with a perfect records. They would get 0 for their record and .5 for their opponent's records (let's leave out the opponent's opponent's record). In contrast, a team with a perfect record against winless teams gets .25 for their own record and 0 for their opponents.
The only real virtue of RPI is that it is fairly easy to calculate and transparent. But it is a bad model for determining how good a team is.
If I were creating a simple system, for every win, you add the opponent's winning percentage and divide by a tally of your losses against 1 - the opponent's winning percentage (that is, their losing percentage). Consider the following team:
That gives you an effective record of 1.7 points for wins versus 1 point for losses, a modified RPI of .629. You could further this by doing a second order calculation.
The conventional RPI would give the team the following
record: 3-2, 0.6 (worth 25%)
opp record: 27-23, 0.54 (worth 50%)
opp opp record (depends, but probably will cluster around 0.5, worth 25%)
overall RPI .545
BC
The only real virtue of RPI is that it is fairly easy to calculate and transparent. But it is a bad model for determining how good a team is.
If I were creating a simple system, for every win, you add the opponent's winning percentage and divide by a tally of your losses against 1 - the opponent's winning percentage (that is, their losing percentage). Consider the following team:
| result | opponent record | points |
| W | 7-3 | +0.7 |
| W | 6-4 | +0.6 |
| L | 8-2 | -0.2 |
| W | 4-6 | +0.4 |
| L | 2-8 | -0.8 |
That gives you an effective record of 1.7 points for wins versus 1 point for losses, a modified RPI of .629. You could further this by doing a second order calculation.
The conventional RPI would give the team the following
record: 3-2, 0.6 (worth 25%)
opp record: 27-23, 0.54 (worth 50%)
opp opp record (depends, but probably will cluster around 0.5, worth 25%)
overall RPI .545
BC
