I have industriously criticized the RPI in the last week, and since then, there has been plenty of good discussion on the topic. Matt Norlander had this post on how to rank teams, combining several rankings systems. Seth Davis joined Norlander on the CBS podcast to discuss the RPI among other things, after writing about the RPI earlier in the week. Oh, and of course, there was John Gasaway’s tremendous piece on the history of the RPI and the man who developed it.
I thought Gasaway had a great observation.
The paradox of Van Valkenburg’s handiwork is that by formulating a metric that eschews any knowledge whatsoever of points scored, he in fact created the perfect conditions under which scoring margin, adjusted for strength of schedule, could and has become a surpassingly powerful evaluative tool. But those of us who use that tool should acknowledge the analytic free ride we’ve been given, courtesy of Van Valkenburg. The stats we’re able to deploy in 2012 are great at evaluating teams — but the only teams we’ve ever been able to evaluate have been ones striving to look good to Jim Van Valkenburg, and not to us.
One of the criticisms offered by RPI proponents is that accounting for margin of victory encourages running up the score. While I think this is largely overstated as a sportsmanship concern, and the true opportunities to “run up the score” are few and far between once we get past the largely disparate matchups early in the non-conference, it is true that teams are not incentivized to maintain the level of their leads late. Teams optimize winning, whereas if they knew margin matters, they may work harder to maintain that 18 point lead late, rather than put in the walk-on with 90 seconds left.
I’ve got a proposal, though, that would maintain that “striving to look good to Van Valkenburg” motive, while still accounting for scoring margin. I personally don’t believe that we should just select teams based on efficiency stats or scoring margin. Teams that win key games should be rewarded, even if our efficiency numbers say they should be a few slots below the team they beat.
My issue with the RPI is that it is a bad system for measuring which wins were actually quality. As I said previously, it’s like the whisper game, and you’re looking several iterations down to determine what is a good win. And despite what the committee says, wins over Top 20, Top 50, etc. RPI teams matter.
So my proposal is this. Use the efficiency stats, and much like Norlander’s post above, use as many quality ones as possible to minimize outliers. Then use those rankings, and not the RPI, to determine which wins are good, which losses are bad, and everywhere in between. The result is that the teams that are seeded higher than their efficiency rankings will have earned those spots by proving they can beat the best teams, and not through some three-level shell game.
Because my industriousness is overrated, I only looked at the Pomeroy ratings for this project, but I went through and assigned points for wins, and negative points for losses. Points were assigned in an exponential fashion, so that Kentucky was worth more than a win over Missouri, while teams ranked 60th and 75th would be worth basically the same for a win, and any win against a team ranked below #230 was lumped together for tournament teams, and largely worthless. Similarly, bad losses were way more costly.
The specific weights I used could certainly be adjusted to reflect whatever value (Elite Wins, Avoiding Bad Losses, etc) you wanted to emphasis, and it would probably move teams a seed line or two. However, I think these reflect a pretty good bracket, and it combines efficiency and actual wins and avoiding bad losses. The teams don’t have an incentive to run up scores despite the use of efficiency stats. Running up the score would only benefit your future opponents, not you.
So what does this week’s bracket look like, with our RPI-like emphasis on key wins and avoiding bad losses, laid over Pomeroy’s efficiency numbers? Here it is. This is a straight bracket using the results this year, and from their you could make adjustments, like Illinois’ recent poor form dropping them out, New Mexico’s early bad loss counting for less and bumping them up, or moving numerous teams slightly with personnel changes or injuries. The CAA should like this method, by the way, because it says that both VCU and Drexel should be in the running for at-large spots despite the conference’s low RPI standing.
#1 seeds: Kentucky*, Syracuse*, Michigan State*, Duke*
#2 seeds: Kansas*, Missouri, North Carolina, Ohio State
#3 seeds: Michigan, Baylor, Indiana, Marquette
#4 seeds: Murray State*, Temple*, Wichita State*, Louisville
#5 seeds: Wisconsin, Georgetown, Florida, UNLV*
#6 seeds: Kansas State, Creighton, Florida State, San Diego State
#7 seeds: Middle Tennessee State*, New Mexico, Harvard*, Virginia
#8 seeds: Gonzaga*, Notre Dame, St. Louis, California*
#9 seeds: Iowa State, St. Mary’s, Southern Miss, Vanderbilt
#10 seeds: Memphis*, Oral Roberts*, Purdue, Drexel*
#11 seeds: BYU, Virginia Commonwealth, Seton Hall, Washington
#12 seeds: Alabama, Illinois, Oregon, Cincinnati
#13 seeds: Iona*, Mississippi State, Connecticut versus Northwestern, Texas versus NC State
#14 seeds: Belmont*, Davidson*, Long Beach State*, Nevada*
#15 seeds: Akron*, Cleveland State*, Bucknell*, Wagner*
#16 seeds: Norfolk State*, Weber State*, Vermont*+, UNC-Asheville*+, UT-Arlington*+, Miss. Valley State*+
LAST 6 AT-LARGES OUT: Minnesota, West Virginia, Arizona, Colorado State, Xavier, Miami-FL
[photo via Getty]