This coming Sunday is my favorite Sunday of the year: “Selection Sunday”, and that means that I’m tweaking my yearly NCAA Auto-Picker! It’s currently using generic teams, but all that will change when the committee announces their picks.
What is it?
It’s a webpage that auto-generates a bracket for the NCAA Division I Men’s Basketball Tournament (also known as “March Madness”).
Why did you make this?
Picking a mediocre bracket is easy! For each game, pick the team with the lowest number (the lower seed, or the “better team”) as the winner. In later rounds, you might have equally-seeded teams up against one another (two #1 seeds might meet in the Final Four, for example). For those, you can either flip a coin or watch ESPN to get the conventional wisdom. However…every year there are teams that buck the conventional wisdom and beat teams with lower seeds. This makes it very very difficult to pick a GREAT bracket – one that will win your office tournament. I SUCK at picking great brackets, so I made this page to help figure out which upsets might be likely. Plus, I’m lazy…
How does it work?
Using a statistical snapshot of all of the modern tournaments, I can tell how often a lower-seeded team is upset by a higher-seeded team. In a geeky, Dungeons and Dragons kind of way, I roll a virtual die for each game, and based on the statistical history, choose a winner.
However, some upsets (#16 beating #1, for example) or some head-to-head matchups have NEVER happened! For those, I need to assign a probability that an upset will happen. I do this based on the seed difference. If a #16 seed plays a #1 seed, this is:
Difference = (HighSeed – LowSeed) = 16 – 1 = 15
You can see the probabilities used on the right-hand side of each bracket. These are fixed – although I would be willing to change them given some constructive commentary.
Hey! Your stupid page picked “insert lame team name here“!!!
Yup, it did. And next time, phrase your answer in the form of a question! Again, this is all based on statistics. If you don’t like how it picks, pick a new bracket (refresh your page) & it will pick another one for you. Alternatively you could program your own picker page with a better algorithm…or send me some email at scottkuma+picker [at] gmail.com with ideas, constructive criticism, etc.
You should probably know that every pick made by the page is logged…and each year I do some post-tournament statistical analysis to improve the page.
UPDATE!! — I’ve noticed in past years that in cases where there isn’t a lot of historical data, that it can become VERY easy for a poor seed to go a long way in the tournament. For example, a 9 seed has played a 5 seed only once in the third round, and it won that game. That means that in my pick system, that 9 seed would beat the 5 seed 99 times out of 100…not realistic at all.
What I’ve done this year to attempt to counter this type of “nerfing” is to go back to my static table (on the right side of the page) when there are fewer than 5 historical examples for each match-up. In my testing, that seems to make things more realistic. I will play with the threshold to try to optimize the formula.
I lost my office pool because I used your page!
Ok, if you use my page as your basis for betting your rent/mortgage payment, or any significant amount of money, I’m not responsible, and YOU ARE INSANE!! Besides, you probably would have lost anyways.
However, should you win a ton of money, I’d appreciate throwing a pittance my way…