Measuring tourney predictability by deviation from high-seed dominance
Last year’s dance was the maddest tournament of the 64-team era. It capped the craziest four-year stretch in 29 years. How can I tell that? Because I devised a simple statistic to measure the madness. It’s called the “Madometer.”
The Madometer works by measuring the seed-position differences between actual winners and perfect high-seed success or failure throughout the dance. If the higher seed advanced in all 63 games (perfect sanity), the cumulative seed value of the winners would be 203. If the lower seed always advanced (sheer madness), their cumulative seed value would be 868. The difference between the two—665—is the predictability range.
Let’s take a closer look at the 2013 dance. If you added up the seed positions of all the teams that advanced through the tourney, the number would come to 341, certainly closer to perfect seed dominance (203 positions), but still 138 positions toward madness along the 665-point predictability range. That works out to a Madometer reading of 20.8%. To put that number in context, the average tournament over the 29 years of the 64-team era has deviated from by-the-seed results by 14.3%.
The 20.8% Madometer reading makes 2013 the most unpredictable tourney since the field expanded to 64 teams. Before last year, the craziest dance occurred in 2011, when tourney advancers deviated from perfect high-seed dominance by 19.8%. You have to go all the way back to 1986 for third most unpredictable tournament, when the Madometer hit 18.8%.
You could pass it off as coincidence that the two wildest dances have occurred in the last three years. But the fact is, we’ve seen four straight dances that were well above average madness. 2010 and 2012 saw 17.1% Madometer readings, tying them for the eighth craziest tournament out of 29. Take a look at how the last four years stack up against the other 25 dances on the Madometer:
Since 1985, the 64-team tourney has ranged from 4.1% “John-Woodenesque” high-seed predictability in 2007 to 2013’s 20.8% “Jimmy Chitwood-like” darkhorse craziness. (Chitwood’s the guy who famously sank the last-second jumper that gave little-known Hickory High its upset win in Hoosiers.)
Looking at the Madometer readings, it becomes abundantly clear how unique the 2007 dance was for its by-the-seed results. The tourney deviated from perfect seed supremacy by a scant 4.1 percent (27 of 665 possible low-to-high-seed differences). And just two years later, after the only dance where all four top seeds reached the Final Four, the 2009 tourney logged the second-most predictable Madometer reading.
Then, after three years of yawn-inducing dances, the tournament turned wildly unpredictable. The closest four-year period to the unpredictability we’ve seen most recently is 1999-2002. Over that time, the average Madometer reading was 17.0%; since 2010, the gauge has averaged 18.7%. To put that number in perspective, before 2011, only one tournament had a Madometer reading higher than the average of the last four (18.8% in 1986).
Measuring tourney predictability by number of upsets
If all of that just hurt your head, there’s a more basic way to measure tourney madness. You simply tally the number of upsets. I define an upset as any game in which a longshot knocks off a favorite that’s seeded four or more seed positions above it. So a 10 beating a seven in round one isn’t an upset; but a five seed beating a top seed in the Elite Eight is.
By this definition, there were 11 upsets in 2013. That’s slightly more than the average of 8.7, but it falls two shockers short of the record. The 2011 tourney saw 13 upsets, tied for the highest total with the 1985, 1986, 1990 and 2002 dances. These are the 11 upsets from last year:
- 15 seed Florida Gulf Coast over 2 seed Georgetown in round 1
- 14 seed Harvard over 3 seed New Mexico in round 1
- 13 seed La Salle over 4 seed Kansas State in round 1
- 12 seed Oregon over 5 seed Oklahoma State in round 1
- 12 seed California over 5 seed UNLV in round 1
- 12 seed Mississippi over 5 seed Wisconsin in round 1
- 11 seed Minnesota over 6 seed UCLA in round 1
- 15 seed Florida Gulf Coast over 7 seed San Diego State in round 2
- 12 seed Oregon over 4 seed Saint Louis in round 2
- 9 seed Wichita State over 1 seed Gonzaga in round 2
- 9 seed Wichita State over 2 seed Ohio State in the Elite Eight
While there weren’t as many upsets in 2013 as 2011, many of them were historically unlikely. Florida Gulf Coast added to the recent spate of 15 seeds knocking off two seeds. Remember that Norfolk State and Lehigh accomplished that feat only a year earlier. So while only seven of 116 15 seeds have shocked two seeds in the 64-team era, three of them have done it in the last two years—another sign of the growing madness of the dance. But Florida Gulf Coast did every 15 seed one better. They actually sprung the first second-round surprise as well, beating San Diego State to reach the Sweet 16. That wasn’t the only unlikely upset in 2013. Wichita State became the only nine seed in the modern era to reach the Final Four when they stunned Ohio State in the Elite Eight.
Since 1985, the average tournament has seen 8.7 upsets. The three years before 2010 were remarkably upset free, with just three Cinderellas in 2007, eight in 2008 and five in 2009. Take a look at the rest of the upset totals throughout the 29 years of the modern era:
Will 2014 continue the streak of mad dances?
The 2009 tourney marked the sixth dance in seven years of below average upsets. So 2010-13 constituted a big trend reversal. Will we see more of the same in 2014? Before the 2010 dance, I stumbled on a more concrete way to assess whether a dance would be unpredictable or not.
I compared the Pythagorean values of the 20 most efficient teams in every tourney field since 2004 to their “Chalkometer” value, essentially the opposite of the Madometer reading. For instance, if 2007 was 4.1% crazy, that means it was 95.9% sane. Then I plotted the Pythag and Chalk values along comparable scales. What I found was that the years when the 20 elite teams were the most efficient, the tournaments played out more predictably. And when the top 20 teams were relatively less efficient, the tournaments were crazier. Take a look at this comparison of the Top 20 Pythag and “Chalkometer” values:
The year of the best “top 20,” according to kenpom.com, was 2007. That was also the year when the dance played out at its most chalky. Conversely, the three years when the top 20 were the weakest—2011, 2012 and 2013—correlated with the biggest period of unpredictability in the modern era. Perhaps it’s easier to see the correlation between elite team efficiency and tourney predictability by comparingthe Pythag and Chalkometer rankings of the last ten years:
The five most efficient Top 20 seasons (see the blue rankings) yielded the first, fifth fourth, third and eighth most chalky tournaments respectively. Meanwhile the five least efficient Top 20 seasons yielded the sixth (tie), ninth, tenth, sixth (tie) and second most predictable tourneys. Only 2006 and 2009 are really out of whack. But when you look at the raw Pythag values, they’re not all that surprising. The top 20 in 2006 were much weaker than any of the dances between 2004 and 2008—and very close to the quality of the 2009 elite. The difference in average Pythag was just .0032 (.9369 to .9337).
Of course, the correlation between the efficiency of elite squads and tourney chalkiness isn’t ironclad. But there’s certainly something to it. And that doesn’t portend a very by-the-seed tourney in 2014. As of December 29, the average Pythag of the 20 most efficient squads stands at .9147. That’s what the orange circle in the line chart above signifies. It’s dramatically lower than the average efficiency of any of the last ten years, a full .0117 weaker than 2012, when the elite squads were at their weakest. With this year’s best so bad, I can’t imagine anything but a continuation of the recent string of upset-prone dances.