My Model Monday: Biggest March Madness Upsets

This year’s NCAA tournament brought some crazy upsets, including the first-ever 9 vs. 11 Elite Eight matchup and first ever 16 seed to win a game, which is being talked about as one of the biggest upsets in sports history. In today’s My Model Monday, I see if the numbers back up that claim, or if there are other dogs that give UMBC a run for their money, and see if there is anything we can learn from the teams that won.

Methodology

I’ll be looking at all NCAA Tournament games from 2001 to 2018 and trying to identify the biggest upsets using Vegas Spreads, our Model 284 spreads, and final scores. I am not focusing on a team’s seed for this, as they are chosen subjectively by the selection committee.

To rank these upsets, I take an average of the Vegas Spread and 284 Spread (the predicted spread from our model) to get each game’s “Avg Spread.” As one view, the biggest upset could simply be the team that lost with the highest Avg Spread. As an alternative view, I’m defining a team’s “Upset Value” as the difference between their Avg Spread and actual margin of victory (MOV). For example, Virginia was favored by 20.5 points over UMBC and our 284 spread had them favored by 15.4 points, making their average spread 18 points. They lost by 20 points, so their “Upset Value” is 18 + 20 = 38. The table below displays all teams that lost with an Avg Spread of at least 5 points (63 games in total):

YRTeam1Team2Upset
Value
Avg
Spread
Vegas
Spread
284
Spread
MOV
20181-virginia16-umcb38.0-18.0-20.5-15.4+20
20182-unc7-texas-am28.6-7.6-7.5-7.7+21
20184-arizona13-buffalo27.9-6.9-9-4.8+21
20084-vandy13-siena27.0-6.0-6.5-5.6+21
20162-mich-st15-mid-tenn25.0-16-16.5-15.6+9
20184-gonzaga9-florida-st22.2-7.2-6-8.5+15
20111-duke5-arizona21.8-5.8-8-3.5+16
20122-missouri15-norfolk-st21.7-19.7-21-18.4+2
20103-g-town14-ohio21.6-7.6-13.5-1.7+14
20163-west-va14-sfa21.6-7.6-7-8.3+14
20164-california13-hawaii19.9-8.9-5.5-12.3+11
20031-kentucky3-marquette19.6-5.6-7-4.2+14
20143-duke14-mercer19.4-12.4-13-11.8+7
20133-new-mex14-harvard18.3-12.3-10.5-14.0+6
20111-kansas11-vcu17.5-7.5-10.5-4.5+10
20141-florida7-uconn16.7-6.7-7-6.4+10
20096-west-va11-dayton16.4-8.4-9-7.8+8
20132-g-town15-fgcu16.3-6.3-13+0.5+10
20012-iowa-st15-hampton16.2-15.2-17.5-12.9+1
20132-miami-fl3-marquette16.0-6.0-5.5-6.5+10
20055-alabama12-wi-milwaukee15.3-5.3-4.5-6.0+10
20062-texas4-lsu15.2-5.2-3-7.4+10
20012-unc7-penn-st14.8-6.8-9.5-4.1+8
20184-wich-st13-marshall14.2-8.2-13.5-2.8+6
20021-cincy8-ucla14.1-10.1-9-11.2+4
20165-baylor12-yale14.1-10.1-5.5-14.7+4
20122-duke15-lehigh14.0-9.0-11-7.1+5
20172-duke7-south-car14.0-7.0-7-7.1+7
20181-xavier9-florida-st13.4-8.4-5.5-11.2+5
20026-gonzaga11-wyoming13.3-6.3-6.5-6.2+7
20161-virginia10-syracuse13.3-7.3-8-6.6+6
20065-nevada12-montana13.1-5.1-5-5.2+8
20151-villanova8-nc-state12.5-9.5-9.5-9.5+3
20054-syracuse13-vermont11.8-8.8-9-8.7+3
20153-iowa-st14-uab11.8-10.8-13.5-8.1+1
20053-kansas14-bucknell11.1-10.1-14-6.2+1
20065-pitt13-bradley11.0-5.0-5-4.9+6
20072-wisc7-unlv11.0-5.0-6-4.0+6
20024-usc13-nc-wilm10.7-6.7-11-2.4+4
20183-mich-st11-syracuse10.6-8.6-10-7.2+2
20138-nc-state9-temple10.5-6.5-4.5-8.4+4
20084-uconn13-san-diego9.9-8.9-11.5-6.2+1
20101-kansas9-northern-ia9.9-7.9-11-4.7+2
20165-purdue12-little-rock9.5-7.5-8-7.1+2
20014-indiana13-kent-state9.4-5.4-6-4.7+4
20025-florida12-creighton9.4-8.4-10.5-6.2+1
20032-kansas3-syracuse9.4-6.4-5.5-7.2+3
20064-kansas13-bradley9.4-5.4-7-3.8+4
20145-vcu12-sfa9.0-7.0-6-8.0+2
20134-kansas-st13-la-salle8.8-6.8-6-7.6+2
20182-cincy7-nevada8.7-6.7-8.5-5.0+2
20114-louisville13-morehead-st8.6-7.6-9-6.2+1
20143-syracuse11-dayton8.6-6.6-8.5-4.8+2
20153-baylor14-georgia-st8.5-7.5-8.5-6.5+1
20041-kentucky9-uab8.3-7.3-10.5-4.2+1
20041-stanford8-alabama8.3-5.3-8-2.5+3
20061-uconn11-g-mason8.2-6.2-8-4.4+2
20142-kansas10-stanford8.1-5.1-6.5-3.7+3
20021-duke5-indiana7.6-6.6-13.5+0.3+1
20172-arizona11-xavier7.6-5.6-7-4.2+2
20146-ohio-st11-dayton7.2-6.2-6.5-5.8+1
20035-miss-st12-butler6.2-5.2-7.5-3.0+1
20108-texas9-wake-forest6.2-5.2-5.5-4.8+1

What’s the biggest upset?

By the Upset Value metric, UMBC blows everyone else out of the water. What is unique about that game is the fact that the UMBC won by 20 points, while most of the other 1, 2, and 3 seeds only lost by a couple points. In terms of Upset Value, the three top games are all from 2018, further evidence at just how crazy this year’s tournament has been. If we rank games by the biggest Avg Spread, Norfolk State > Missouri tops the list, followed closely UMBC > Virginia, and then Middle Tennessee > Michigan State.

What can we learn from these teams?

Okay, so now we have defined all these huge upsets – is there anything special about the teams that played in those games? Do they all have similar characteristics? Were they all lacking in the same area?

Conferences – as you might expect, 56 of 63 teams that were upset (89%) were from power conferences. For teams that won these games, only 18 of 63 winners (29%) were from a power conference, so generally, these are mid-majors. Kansas (6 times) and Duke (5 times) are the only teams that make the upset list more than twice, which is likely somewhat driven by the fact that they are big favorites more often.

Used to Playing in Close Games – looking at all teams to play in the tournament since 2001, the 63 teams that were upset ranked in the 81st percentile (on average) in margin of victory, while the teams that won these games ranked in the 40th percentile. So, the teams that pulled these upsets had more experience playing in close games, and a little adversity didn’t put them out of their comfort zone.

Not Guard-Oriented – we measure how much of a team’s scoring comes from their lead guards, and the 63 teams that were upset ranked in the 40th percentile here, on average, meaning that they got more scoring from other positions. We have seen this factor time and again when it comes to tournament success: having elite scoring guards who can get you a bucket in crunch time is a huge advantage.

Offensive Rebounding – teams that lost were typically good offensive rebounders, ranking in the 60th percentile on average, while the teams that lost only ranked in the 39th percentile. I don’t have data to back this up, but offensive rebounds seem like they could vary quite a bit from game-to-game, and thus being heavily reliant on them isn’t ideal for a single-elimination tournament.

Opposite Shooting Styles – teams that pulled these upsets were not typically great shooting teams, but they were above average in terms of 3-point attempts. Conversely, the teams that lost ranked in the 68th percentile of true shooting percentage, but were not typically high-volume 3-point shooting teams. Once you digest those styles, it is easy to see how they are the recipe for an upset.

Against the Spread – teams that pulled these upsets ranked in the 62nd percentile in terms of against the spread record, on average, suggesting that they had been somewhat underrated by betting markets all year.