After the first day, my model was feeling pretty good about itself. It correctly picked 13/16 games, had the Loyola-Chicago upset, and had me in the top 6% in the Google-Kaggle competition.
After the second day? Well…
| Team 1 | Team 2 | Predicted Winner (%, odds) | Implied Point Spread | Actual Winner |
|---|---|---|---|---|
| Kansas (1) | Penn (16) | Kansas (94.2%, 16.2:1) | -17 | Kansas |
| Villanova (1) | Radford (16) | Villanova (98.2%, 54.6:1) | -22.7 | Villanova |
| Virginia (1) | UMBC (16) | Virginia (99.3%, 141.9:1) | -24.6 | UMBC |
| Xavier (1) | Texas Southern (16) | Xavier (99%, 99:1) | -24.1 | Xavier |
| Cincinnati (2) | Georgia St. (15) | Cincinnati (97%, 32.3:1) | -20.7 | Cincinnati |
| Duke (2) | Iona (15) | Duke (97.8%, 44.5:1) | -22 | Duke |
| North Carolina (2) | Lipscomb (15) | North Carolina (96.1%, 24.6:1) | -19.4 | North Carolina |
| Purdue (2) | Cal St. Fullerton (15) | Purdue (97.1%, 33.5:1) | -20.9 | Purdue |
| Michigan (3) | Montana (14) | Michigan (87.6%, 7.1:1) | -11.5 | Michigan |
| Michigan St. (3) | Bucknell (14) | Michigan St. (93%, 13.3:1) | -15.7 | Michigan St. |
| Tennessee (3) | Wright St. (14) | Tennessee (94.7%, 17.9:1) | -17.6 | Tennessee |
| Texas Tech (3) | Stephen F. Austin (14) | Texas Tech (94.4%, 16.9:1) | -17.3 | Texas Tech |
| Arizona (4) | Buffalo (13) | Arizona (76.9%, 3.3:1) | -7.5 | Buffalo |
| Auburn (4) | College of Charleston (13) | Auburn (93.2%, 13.7:1) | -15.9 | Auburn |
| Gonzaga (4) | UNC Greensboro (13) | Gonzaga (89.3%, 8.3:1) | -12.6 | Gonzaga |
| Wichita St. (4) | Marshall (13) | Wichita St. (87.9%, 7.3:1) | -11.7 | Marshall |
| Clemson (5) | New Mexico St. (12) | Clemson (77.9%, 3.5:1) | -7.8 | Clemson |
| Kentucky (5) | Davidson (12) | Kentucky (69.4%, 2.3:1) | -5.8 | Kentucky |
| Ohio St. (5) | South Dakota St. (12) | Ohio St. (83.4%, 5:1) | -9.5 | Ohio St. |
| West Virginia (5) | Murray St. (12) | West Virginia (74.6%, 2.9:1) | -7 | West Virginia |
| Florida (6) | St. Bonaventure (11) | Florida (73.1%, 2.7:1) | -6.6 | Florida |
| Houston (6) | San Diego St. (11) | Houston (75.1%, 3:1) | -7.1 | Houston |
| Miami FL (6) | Loyola Chicago (11) | Loyola Chicago (50.2%, 1:1) | 0 | Loyola Chicago |
| TCU (6) | Syracuse (11) | TCU (68%, 2.1:1) | -5.4 | Syracuse |
| Arkansas (7) | Butler (10) | Butler (56.8%, 1.3:1) | -2.2 | Butler |
| Nevada (7) | Texas (10) | Nevada (63.7%, 1.8:1) | -4.3 | Nevada |
| Rhode Island (7) | Oklahoma (10) | Oklahoma (54.2%, 1.2:1) | -1.4 | Rhode Island |
| Texas A&M (7) | Providence (10) | Texas A&M (67.5%, 2.1:1) | -5.3 | Texas A&M |
| Creighton (8) | Kansas St. (9) | Creighton (56.7%, 1.3:1) | -2.2 | Kansas St. |
| Missouri (8) | Florida St. (9) | Florida St. (52.2%, 1.1:1) | -0.7 | Florida St. |
| Seton Hall (8) | North Carolina St. (9) | Seton Hall (55.7%, 1.3:1) | -1.9 | Seton Hall |
| Virginia Tech (8) | Alabama (9) | Virginia Tech (55.4%, 1.2:1) | -1.8 | Alabama |
While the earlier games pushed my Kaggle ranking to the top 3%, the evening upsets dropped me to top 40%. The model was correct 78% of the time when just considering win-no win for each game.
On the plus side, the model did call Florida St. and Butler to upset their opponents, but it wasn’t super confident that either would pull it off. It also saw that Kentucky and Arizona were in for tough games, with Arizona losing heavily to Buffalo.
It’s two biggest misses were Marshall over Wichita St. and, of course, UMBC over Virginia. UMBC pulled off the upset of the century, taking out Virginia and wrecking millions of brackets single-handedly. My model was as confident in a Virginia win as you could be (99.3%), and Vegas basically agreed (VA was -20).
As for the second round:
| Team 1 | Team 2 | Predicted Winner (%, odds) | Implied Point Spread |
|---|---|---|---|
| Kansas (1) | Seton Hall (8) | Kansas (57%, 1.3:1) | -2.3 |
| Villanova (1) | Alabama (9) | Villanova (87.1%, 6.8:1) | -11.2 |
| Xavier (1) | Florida St. (9) | Xavier (71.8%, 2.5:1) | -6.3 |
| Tennessee (3) | Loyola Chicago (11) | Tennessee (70%, 2.3:1) | -5.9 |
| Clemson (5) | Auburn (4) | Auburn (59.1%, 1.4:1) | -3 |
| Kentucky (5) | Buffalo (13) | Kentucky (77.8%, 3.5:1) | -7.7 |
| Ohio St. (5) | Gonzaga (4) | Gonzaga (58.8%, 1.4:1) | -2.9 |
| West Virginia (5) | Marshall (13) | West Virginia (89.4%, 8.4:1) | -12.7 |
| Florida (6) | Texas Tech (3) | Texas Tech (59%, 1.4:1) | -2.9 |
| Houston (6) | Michigan (3) | Michigan (65.8%, 1.9:1) | -4.9 |
| Nevada (7) | Cincinnati (2) | Cincinnati (74.9%, 3:1) | -7 |
| Rhode Island (7) | Duke (2) | Duke (86.8%, 6.6:1) | -11 |
| Texas A&M (7) | North Carolina (2) | North Carolina (73.1%, 2.7:1) | -6.6 |
| Butler (10) | Purdue (2) | Purdue (67.3%, 2.1:1) | -5.3 |
| Syracuse (11) | Michigan St. (3) | Michigan St. (81.8%, 4.5:1) | -8.9 |
| UMBC (16) | Kansas St. (9) | Kansas St. (94.4%, 16.9:1) | -17.3 |
There are a bunch of close games this weeked, starting with #1 Kansas versus #8 Seton Hall. The model has Kansas as a slight favorite, only giving 2.3 points. This is even tighter than Vegas, which has Kansas -4.5.
Only one matchup has the favorite greater than 90% to win, and that is Kansas St. However, they are playing…UMBC. We saw how well that worked out for Virginia.
Reminder: you can find all of the model’s predictions for every possible matchup here. These are based on all 68 teams that were selected for the tournament this year.