Probabilistic Models for Bracket Picking

Papers! Links!

That will generate a bracket for you, based on a particular year’s seeding (2018?), so obviously you can transfer to a different one.

1 Like


You gotta give me a few minutes… I was submitting my brackets before linking to the papers.

1 Like

I like that paper – and it’s publicly available.

1 Like

Most of the papers I’m coming across are about the probability of a perfect bracket, which few people actually care about.

Something from last year:

This was last year’s 538 model (so it’s in its final state with 100%)

Let’s see if they’ve got 2022…


I submitted my brackets before looking at this model

So this year both the Men and Women get to call their tournaments “March Madness” and both get 68 teams in the tournament.

Looking at 538’s model, the top 4 women’s teams have a combined 83% probability of winning the tournament while the Men’s to 4 teams only have a 37% chance.

There is rounding here but the women’s to 8 have a 103% probability of winning while the top 8 men’s teams have a 59% chance.

Interestingly, last year Men’s top 4 were 56% and top 8 were 74%.

While women top 4 were 78% and top 8 were 98%.

I think this is a reflection of the bigger talent gap between the “top quartile” of women’s teams vs. the next quartile.

IMO that talent gap is less pronounced between the “top quartile” of men’s teams vs. the “bottom quartile” than the above referenced talent gap.