Tips for your NCAA Brackets [Data Geek style]

March 15, 2011

by Scott Sambucci


Quick tips for your NCAA tournament brackets, grounded in academic research and economic theory of course:

1. Look at no more than six cues or factors for each decision, and three should do the trick.  Lamenting and pouring over more information and factors doesn’t lead to any increased accuracy in predictions. In fact, using only three factors can work just as well because using six or more leads to overconfidence.  That’s probably why the guy down the hall who’s watched every college game since November thinks he’s going to win your office pool – “I love my bracket!!”  You know that guy… He usually comes in last.  Isn’t is wonderful?

[Source:  "Effects of amount of information on judgment accuracy and confidence." by Claire I. Tsai, Joshua Klayman, and Reid Hastie.]

2. Look for similar characteristics on each team, not differences. Turns out that people disregard components that the alternatives share and focus on the components that distinguish them.  “This team has a great offense or that team has a big center!”

3. Be conservative. People tend to prefer a small chance at a big gain, which is why you’ll see lots of brackets crash and burn with too many #12 and #13 seed picks.  “I’ve got Belmont in the Elite 8!”

[Source: "Prospect Theory." Daniel Kahneman and Amos Tversky.]

3.  Just do the opposite of me.  I’m already preparing the menu for my sixth year running of losing to my wife in our head-t0-head bracket challenge.  We bet a nicely prepared home-cooked dinner.  I’m thinking Italian this year…

[Source: My miserable gambling history]


James March 16, 2011 at 7:58 am

Nice stuff!

Reflecting on the first paper, I'd like to see an experiment where subjects must decide how many and which clues to view before making their prediction. Suppose the payoff for predicting the winner was $100 if correct, $0 if incorrect, MINUS $1 per each clue requested before making the bet. Also experimental conditions could be varied such that the subject is allowed to buy a next clue of their own selection or one that would be selected at random. This would measure the value a subject places on marginal information, how well the subject can select an optimal set of clues, and the confidence and accuracy that come with buying additional information.

As a data vendor, you might not want the world to know that people tend to pay too much for additional infomation. Or perhaps it's the other way around, and people tend not to spend enough on additional information or not to select the right kind of information. I'm not sure what would turn up.

Best of luck on your bracket!

Scott Sambucci March 16, 2011 at 8:07 am

Hi James – I like the idea of paying for additional information. At some point, the subject would better see the diminishing returns for purchasing additional data points.

Yes, as an analytics provider, people assume that we tell folks- "Hey – the more data the better!" Though… Our housing analytics are unique and it is pretty rare that we even attempt to sell a data package that includes all 400+ statistics that we calculate every week. Most customers purchase a more focused package of stats based on their modeling applications.

For our recently launched Forward Valuation Model, we cut our input variables from the initial list of 400 to about 40.

As always, great to hear from you – thanks for reading!

Comments on this entry are closed.

Additional comments powered by BackType

Previous post:

Next post: