After a week of Madness, let’s re-visit our match-up prediction model to see what went well and what could have gone better.
What went well – Upsets accurately predicted
Our match-up model can claim success on calling a number of upsets (according to tournament seeding or Vegas odds makers) over the opening weekend, including:
- 7 Wichita State over No. 2 Kansas
- 10 Ohio State over No. 7 VCU
- 6 Butler over No. 11 Texas
Seven upsets were predicted (excluding No. 8 vs No. 9 matchups), resulting in a 43% success rate – which is good since the average win probability for some of these underdogs was significantly lower.
To further understand and improve our model, analysis of its shortcomings is critical. Let us analyze two games in particular –an upset which occurred that wasn’t predicted, and a predicted upset that didn’t materialize.
Upset We Did Not Predict – No. 13 Georgia State over No. 3 .Baylor
This game was a head-scratcher from a match-up perspective, and it’s easy to see why our model did not pick the winner. Of the six match-up attributes most consistent for teams that beat Baylor, Georgia State matched only three during the regular season (see table below). Even more surprisingly, of these three matching attributes, Georgia State did not play to these strengths during the game – particularly in terms of their defensive efficiency and defensive 2-point percentage (2P%).
So, how did Georgia State win? They shot 43% from a 3-point percentage (3P%), which was very uncharacteristic from them during the regular season (which saw them in the bottom 20th percentile). Their 3-point percentage from this single game would have put them in the 99thpercentile!
This speaks to the variation in the data that sits below the surface when looking at the regular season averages…and the tournament isn’t won or lost on averages. All it takes is one game to unexpectedly advance (or be sent home) from the tournament. If these games were played best two out of three, we would expect Baylor to prevail.
Profile of Teams who beat Baylor in the Regular Season
Predicted Upset That Did Not Materialize – No. 12 Seed Buffalo
In our previous post we outlined why our model predicted that Buffalo, a 12th seed, would beat 5th seed West Virginia. The table below outlines the profile of teams that beat West Virginia during the regular season, and whether Buffalo matched their regular season play during the tournament game.
In summary, Buffalo was not the same defensive team against West Virginia they had been during the regular season. They could not force West Virginia to attempt more three-point shots, and allowed them to score from the 2-point line. So what happened? First, they played against much stiffer competition in West Virginia then they were accustomed to during the regular season. The model accounts for this, however, so another explanation is needed. The truth is, Buffalo just had an uncharacteristically bad game at the wrong time.
Our Sweet 16 Picks
Below are our Sweet 16 picks. Since many of the remaining teams were high performing during the regular season, we expect our model will fare better. For the next round, our model predicts two underdogs (by tournament seeding or Vegas odds) to advance:
- 7 Michigan State over No. 3 Oklahoma
- 3 Notre Dame (who is actually an underdog) over No. 7 Wichita State
We can’t wait to see what these teams (and our model) will do this weekend!
Many thanks to both Dave Caughman, Strategist and Drew Yao, Senior Operations Research Consultant for assisting with conducting the analysis