https://marchmoggness.ptaylor.dev/

This is so cringe, I wrote it when I was like 19 years old when this was a barely working excel model. Keeping it bc it's easier than rewritting a readme.a

With March around the corner, I figured I would share a small personal project I worked on last year in hopes my network in other professional areas will get the chance to see data can be a lot of fun to work with! The attached model mentioned is available under projects on my profile for those interested.

As someone with a less-than-successful track record in my NCAA March Madness bracket pools, I entered into my 2021 season pool with a new approach.

By leveraging March Madness' renowned unpredictability, for my 2021 tournament bracket, I aimed to eradicate any bit of personal bias and leave my fate solely in the hands of tangible data.

Linked, one will find the spreadsheet I used to land 2nd in my pool and within an average top 20% of all brackets nationally across a reiteration of the model 100 times.

How the Model Works The model utilizes historical win data of every possible matchup at every seeded level in prior March Madness tournaments. The model then creates a random number for each matchup for every iteration which is fed through that matchup's historical win probability in order to return an alleged winner.

Each iteration will return entirely different results, though each run can be expected to, roughly, produce a normalized distribution of upsets and chalked winners - this, of course, becomes much more significant when scaled to 100 brackets.

Because the 2021 tournament did not yield significantly more or fewer upsets than the average collection of historical tournaments, the model was able to succeed by hedging against one inherent tendency to "predict" outcomes in a tournament that is infamously boasted as unpredictable.