The start of the 2013-14 college basketball season also means the public debut of the #KPI formula for basketball.
#KPIFootball (the formula is applicable for any team sport with a winner, loser and a score) debuted publicly in late August and has been going strong with weekly rankings since Labor Day. The rankings have generated meaningful data, constructive discussions and thousands of views. Basketball data will do the same and even more.
For those new to the #KPI, the formula assigns a value to every game played. This value combines opponent’s winning percentage, opponent’s opponent winning percentage, game result, scoring margin, pace of game, location of game and opponent’s #KPI ranking into one metric. The worst loss is worth approximately -1.0, the best win worth approximately +1.0 and a hypothetical tie worth zero. A season’s #KPI is the average of each game’s value. Read more about the #KPI formula here.
The best wins in 2012-13 came from No. 24 Pittsburgh (73-45 at No. 8 Georgetown, +1.06), No. 38 Illinois (85-74 at No. 5 Gonzaga, +1.03) and No. 2 Louisville (87-78 at No. 10 Memphis, +.97). Florida State’s 51-14 football win at Clemson is the best football win this season (+1.29). The average cutoff to make the NCAA Tournament was +.135 last year.
I have ten years of basketball data to rely on how accurate the formula is. It has helped me analyze and project postseason scenarios and schedule non-conference games in future seasons. The human element comes in the analysis, not the actual data collection or manipulation. There will no longer be a need for someone to arbitrarily determine what a team’s best wins or worst losses are when discussing postseason resumes late in the season. By quantifying every game, a team’s performance can be ranked by game.
The #KPI correctly projected 67 of the 68 teams in the 2013 NCAA Tournament (36 of the 37 at-large selections) strictly with math. The #KPI had Southern Miss (#KPI No. 45) in the tournament and Villanova (#KPI No. 56) outside the tournament.
There are several key differences that make the #KPI more reliable and precise than other rankings:
Each game is assigned a value: Both teams earn an independent value assigned to every game played each season. This not only allows for the most accurate assessment of each game, but allows for #KPI data to be queried in different combinations (conference only, home games only, games played on Tuesdays, games played in the central time zone, really whatever combination one can imagine)
Each game counts as ONE game: The RPI sets values of 1.4 for road wins and home losses, 0.6 wins for home wins and road losses and 1.0 for neutral wins and losses. The preset value adjusts both the numerator AND denominator (for those of you who have forgotten your fractions terminology, that’s the top number AND the bottom number). A road win or home loss is valued 233% higher than a home win or road loss. Not in the #KPI! Each game in the #KPI counts as one game when averaged for a full season #KPI number.
Margin is part of the equation: A one point game is different than a 30-point game. The margin is factored into each game’s value on a percentile basis, allowing the margin to tweak rather than dictate the final value.
Home/Away/Neutral adjustments are dictated by data, not predetermined: In order to offset advantages gained by playing games at home, on the road or at a neutral site, a value determined by the average of all games played is determined and then offset to zero. In 2012-13, the adjustment for home teams was -0.16, the adjustment for road teams was +0.15 and the adjustment for teams on a neutral floor was -0.03. These adjustments start anew each season, though have been relatively consistent over a decade. Unlike the RPI, these values are not predetermined and adjust as the season progresses.
Games vs. Non Division I teams count: Games against Non Division I teams are real games. They count in a team’s stats and record and unlike the RPI, they count in the #KPI. There is one team called Non Division I (11-338, .032 KPI #338 in 2012-13). Games count, scores count and wins and losses matter. A win is likely to be a win worth just above zero while a loss is likely to push close to -1.0. They are games after all, right?
The goal is for new data and numbers to be posted each Monday on the blog. Data may be produced twice a week late in the season. New information and anecdotes are always live on Twitter at @KPIsports.
I know the #KPI. I trust it and I swear by it. I hope you find it as reliable as I do.
This is the #KPI