There won’t be a lot of numbers in this post.
This is more of a thought-experiment than rigorous analysis. I wouldn’t consider this a formal proposal; it’s really a chance to point out that there is an alternative to RPI and why you might be interested in using it.
That alternative is Strength-of-Record (SOR), and it works like this. You look at a team’s schedule, including who they played and where the game was. Then you take a generic “good” team (I use the #10 Lax-ELO team), and you estimate how many wins they would end up with against that schedule. Finally, you compare the team’s record against what you would expect a generic “good” team to end up with.
The resulting list of “excess wins” shows the teams that have performed the best, adjusted for the strength of their schedule. That “adjusted” part is the key here. If you play a cupcake schedule, then a generic #10 ELO team is probably going to have no more than one or two losses over the course of a season. So if you want to stand-out in SOR’s eyes, you better go undefeated.
But the selling point for SOR is not that it adjusts for schedule strength; heck, RPI “adjusts” for schedule strength. It is that it adjusts for schedule strength while valuing the results of the games more than a system like RPI.
Under the RPI regime, only 25% of your score is based on your record, so if you play 14 games in the regular season, the outcome of a single game accounts for just 1.8% of your score. And since RPI heavily weights opponents’ win percentage, if you lose a game to a good team, your RPI is likely to rise. (see Johns Hopkins…)
Under SOR, that is not as true. Let’s assume you play a juggernaut who, based on ELO, would be expected to beat a top-10 team 70% of the time. If you win that game, your “excess wins” total goes up by .7. If you lose it, your total goes down by .3.
Said another way: under RPI, if you lose that game, your RPI still probably goes up. Under SOR, you get penalized for losing, but not too much because of the opponent’s strength. The result matters as much, if not more, than the opponent.
The poster child for SOR-lovers this year was High Point. The estimated baseline for a top-10 team against their schedule was 12.51 wins. In SOR parlance, we would expect a generic top-10 team to rack up 12.51 wins against their slate.
Since High Point ended up with 13 wins, their “excess wins” total is .49, good for 9th in the country. Contrast that with the Panthers’ #20 RPI.
And that is the point of SOR. Yes, High Point had an easy schedule when you look at it holistically. That 12.51 expected wins places it in the bottom 10 in terms of schedule difficulty. And they had some bad losses, which SOR penalized them for. But they won some games against really good teams (Duke, UVA) too, so SOR is able to say: “ok, you played a fairly easy schedule, but you did better than we’d expect a generic top-10 team to do against it, so you must be a top-10 team.”
The bottom line is that ranking teams doesn’t have to be a choice between valuing schedule strength and valuing wins and losses. There are methods that can combine the two and create a reasonable proxy for how good a team’s resume is. I humbly present Strength-of-Record to you as one such option.