Kansas wins the TDG simulation NCAA Tournament

Kansas emerged as the champion of the TDG NCAA Tournament simulation. They defeated Duke in the championship game.

Now that the simulation is complete, a few words as to how we did it. First and foremost, zipsofakron was our game master, and Minnesota’s destiny was his original idea. As a consequence, none of the following applies to the Gophers’ run through the tournament, though I personally agree with all of it minus the lost to Kansas. #Robbed

First we had to get a tournament bracket. To do so, we used The Bracket Project’s Bracket Matrix and took the consensus average. Again because zipsofakron was our game master, there were a few changes to make it so that Minnesota ended up against Texas in the play in game, and to make Auburn a 5 seed. We did not care about geographic balance, Duke never leaving North Carolina in early round games, or other aspects that the NCAA selection committee always cares about.

Second, we got the underlying win probability via KenPom.com. That had a few implications for the simulation. Most notably, Kansas is an absolute wrecking crew of a team, and so bludgeons all of its opponents. Next, the seven and ten seeds were unusually strong relative to historical brackets. KenPom LOVES Texas Tech and Arizona, though why is not super clear. The system always put major emphasis on the Big Ten being the best conference in the country. As a result, the Big Ten cleaned house in the early rounds. GopherNation personally overrode the system allowing Iowa to get to the Elite Eight because we continue to hate Iowa even in a simulation world.

Finally, since every simulation needs to have some kind of variance, we elected to run each matchup 1000 times with some error in each of their offensive and defensive efficiencies. That meant in any given match-up, it was technically possible for an upset. Well, except Kansas. To get Kansas losing in this simulation required some very implausible assumptions about how efficient the Jayhawks’ opponent could be in a match-up. We did not update KenPom probabilities after each game for simplicity. A more complete simulation would have included new efficiency numbers after every game. In the sad event that we have to do something like this again, we plan to add that feature.

Should you be interested in perusing the (quite sloppy) simulation code, the program and data are here.