[Quick Hit] Updated 2021 March Madness Predictions – Sweet 16
Because of the break in the action, I took the time to update my college basketball rankings in order to make my March Madness predictions for the sweet 16. This article will be short and straight to the point and will avoid my usual mathematical details. First, I’ll show you my updated team rankings and ratings and discuss differences from last time. After that, I’ll repeat what I did in my March madness machine learning article in order to estimate the probabilities of each remaining team making the Elite 8, Final 4, Championship game and ultimately winning the tournament.
This tournament truly has felt mad. In a few weeks, when all is said and done, I’ll wrap up the college basketball season with a quick statistical analysis where I measure the degree of madness using the Poisson binomial distribution (I know, it sounds like I made that up by smashing together two other distributions, but I promise it exists and is precisely what we will need).
For now, though, here is the bottom line up front. Gonzaga is even more heavily favored (~60% now) to win the whole thing than before – my updated model thinks we previously underrated Gonzaga. The most underrated team award doesn’t go to Loyola-Chicago, in fact it goes to the entire Pac-12 conference. Houston has seen the largest increase in their championship chances thanks to their entire region melting away thanks to Buddy and the Boeheims. Oral Roberts Cinderella story is overwhelmingly unlikely to continue. For the details, read on to the following sections.
March Madness Sweet 16 Team Rankings
Using the same technique to rank and rate teams as my previous March Madness machine learning article, I updated my rankings taking into account the tournament games already played. Because of the effect of hot streaks and recent games being better predictors of the future, when adding in the tournament games to my model I weighted them by a factor of 3. To the model, this looks like each of the tournament games had been played three times with the same outcome. To us, this means the tournament games played last week mean more in picking the remaining winners than the games played at the beginning of the season. As they should.
Here are our new and updated rankings. The “Rating Change” Column is the difference between the team’s new rating and their pre-tournament rating.
| Team | Rating | Rating Change |
|---|---|---|
| Gonzaga | 28.3 | 1.9 |
| Baylor | 22.2 | 0.2 |
| USC | 20.1 | 4.8 |
| Houston | 19.9 | 0.7 |
| Michigan | 19.6 | -1.7 |
| Alabama | 17.8 | 0.6 |
| Villanova | 17.6 | 0.2 |
| Oregon | 17.4 | 4.6 |
| Loyola-Chicago | 17 | 2.4 |
| Arkansas | 16.3 | 0.8 |
| UCLA | 16.2 | 4.4 |
| Florida St. | 15.8 | 1.2 |
| Creighton | 15.3 | -0.8 |
| Syracuse | 12.1 | 0.9 |
| Oregon St. | 11.3 | 5.2 |
| Oral Roberts | 3.7 | 3.4 |
These ratings are what we’ll use in our simulations next section. You can visualize the teams with the largest increase in rating in the following image.
Very interestingly, Michigan saw the largest decrease in team rating even though they won two straight games fairly convincingly by 16 and 8 points. If you’ve been following the media narrative, though, this decline in rating is almost certainly due to my model souring significantly on the Big 10 overall. Michigan’s past wins against some of the top Big 10 teams now don’t look quite as good as they did before the first two rounds of the tournament.
On the other hand, USC, Oregon, Oregon St., and UCLA are skyrocketing due to a very similar conference adjustment. The Pac 12 teams have been out-performing their predictions by about 4.5 points. That is a huge adjustment.
March Madness 2021 Probabilities – Sweet 16
Just like my last article, I simulated the entire tournament starting at the sweet 16 using Monte Carlo simulation with the team ratings shown in the above table and chart. The following table can be sorted by any of the columns. The columns corresponding to a given round give the estimated probability that a team makes it that far. For instance, Gonzaga’s 59.5 value under the ‘Championship’ column indicates we think there is about a 60% chance that Gonzaga wins the title. Alabama’s 27.3 under the ‘F4’ column indicates roughly a 27% chance we think Alabama makes the final four.
| Team | Rating | E8 | F4 | Finals | Championship |
|---|---|---|---|---|---|
| Gonzaga | 28.3 | 93.8 | 80.8 | 71.4 | 59.5 |
| Creighton | 15.3 | 6.2 | 2.2 | 0.8 | 0.3 |
| USC | 20.1 | 62.3 | 12.2 | 7.4 | 3.9 |
| Oregon | 17.4 | 37.7 | 4.7 | 2.3 | 0.8 |
| Michigan | 19.6 | 67.6 | 41.7 | 9.4 | 4.8 |
| Florida St. | 15.8 | 32.4 | 14.6 | 1.7 | 0.5 |
| UCLA | 16.2 | 42.0 | 16.4 | 2.2 | 0.8 |
| Alabama | 17.8 | 58.0 | 27.3 | 4.9 | 2.0 |
| Baylor | 22.2 | 70.4 | 54.8 | 37.6 | 13.7 |
| Villanova | 17.6 | 29.6 | 17.4 | 8.3 | 1.6 |
| Arkansas | 16.3 | 92.9 | 27.6 | 11.7 | 2.0 |
| Oral Roberts | 3.7 | 7.1 | 0.2 | 0.0 | 0.0 |
| Loyola-Chicago | 17.0 | 75.5 | 32.5 | 12.1 | 2.3 |
| Oregon St. | 11.3 | 24.5 | 5.1 | 1.0 | 0.1 |
| Syracuse | 12.1 | 17.7 | 6.2 | 1.2 | 0.1 |
| Houston | 19.9 | 82.3 | 56.3 | 28.2 | 7.6 |
This won’t be as long as the last one, but I’ll make a few observations. First, Gonzaga somehow was underrated by our model before the tournament started. They’ve picked up a point or two in rating and the quality of potential opponents has really thinned out. The net result is that we now estimate Gonzaga has roughly a 60% chance of winning the whole thing. Nuts.
Baylor has the second best chances at winning the whole thing, roughly 13%. This is only up about 3% from our estimated pre-tournament odds. Houston – not the other 1 seed Michigan – has the third best Championship odds. This effect is partly because our model now sees Houston as about a half point better than Michigan on a neural court. However, this effect is more attributable to how easy their respective regions are. More likely than not, Houston’s final 4 berth requires only beating Syracuse (which may be a taller task than we think) and Loyola Chicago. Michigan still has their 2 seed around.
Incredibly, our model absolutely loves the USC Trojans. This shouldn’t be surprising, really, they beat Drake by 16 and Kansas by over 30 so far. They haven’t been in a close game yet. Though my model doesn’t know it, they have the likely 2nd pick in the upcoming NBA draft on their roster. (An interesting side note, 5 of the first 9 NBA prospects to get picked according to this source are still in the tournament). What is actually quite surprising is that our model now thinks that USC is the third best team in the tournament, behind only Baylor and Gonzaga. Unfortunately, though they match up with conference rival Oregon in the Sweet 16 game I am probably looking forward to the most, the winner there has to play the juggernaut Zags.
There are some other interesting storylines one can pull from these numbers, but I’ll stop there and let you look for yourself. Happy March Madness season.
