March Madness 2022: Picking Your Bracket 101

Alrighty, 2 most important things to consider when filling out your bracket:

1. Better Seeds win 71% of the time the first weekend, and at a coin flip margin after that.

Better Seeds (1-8) are 71% S/U for the first two games they play. This number has range from 6-13 games being upset in the round of 64, with an average of 10, and anywhere from 3-8 on the second round with an average closer to 5. If your bracket is feeling chalky, find a few low risk upsets to add in.

2. The best predictors of upsets are Strength of Schedule and Turnover Ratios

Since 2011, 1-4 seeds with a turnover ratio above 1 (e.g. If my defense forces 11 turnovers in a game, and my offense only had 10, my turnover ratio would be 1.1) make the sweet 16 at an 82% rate.

For 1-4 seeds with a turnover ratio below 1? A measly 28%. Last year was one of the weakest sets of 1-4 seeds in this regard, with 8 total teams on average allowing their opponents to theoretically take more shots/game than them from turnovers. They were Michigan, Florida State, Illinois, Ohio State, Purdue, Virginia, Oklahoma State, and Texas. Only 2 made the sweet 16. In any given year, no more than 2 1-4 seeds with a turnover ratio below 1 made the sweet 16. This years 4 culprits are 1s Arizona, 3e Purdue, 4s Illinois and 4m Providence

As for methodology below.

Each round has variables that go through regression based on the past 10 years brackets to try to predict the winners of each games and the probability that it happens. The natural log of the percentage confidence is than taken, with a more confident guess resulting in a lower score. The average of all scores is taken, and person with the lowest score on the Kaggle Coding competition wins. Because of this, each round has teams that might change all down-bracket results if that upset occurs if they are have better stats that indicate a higher change of winning in that particular round. As stated above, seeding matters less as the tournament progresses, and so weights of things should be different as well. As hinted at above, teams with bad turnover margains/ratios don't usually make it very far, so its not as great of an indicator for a teams success once they make the final four.

---

East Region (1+1 upsets)

Model really likes Baylor here. Of note: If Virginia Tech beats Texas, they would be favored again. Kentucky has the easiest route to the elite 8, being better than every single team. UCLA would be favored over Kentucky to get to the final 4 as well.

West Region (2+2 upsets)

Duke is the projected favorite here, and is the universal favorite for the elite 8 on the bottom half. Vermont over Arkansas has some good odds, but ucoon is favored over either. Last year Gonzaga lucked out in getting three opponents they played during the season and beat in their bracket, this year they managed to pull two that already beat them. Its a much more difficult pull.

Midwest Region (2+0 upsets)

Kansas is a sure-fire lock to the elite 8. Iowa is definitely underseeded. Providence wouldn't beat Iowa by the model if they won. Auburn would be less favored over USC if they won. LSU has better matchup vs Auburn in the elite 8 as well.

South Region (2+1)

Illinois received the cut here from the ratios mentioned above. Loyola and Michigan Favored outright over their opponents. In last year's model, this bracket had the most significant changes. Tennessee would beat Villanova, and also Houston to make the final 4.

Villanova beating Duke is the projected outcome. Last year's model would have had Kansas beating Baylor.