An easily implemented and accurate model for predicting NCAA tournament at-large bids

We extend prior research on the at-large bid decisions of the NCAA Men’s Basketball Committee, and estimate an eight-factor probit model that would have correctly identified 178 of 179 at-large teams in-sample over the 2009–2013 seasons, and correctly predicted 68 of 72 bids when used out of-sample for 2014 and 2015. Such performance is found to compare favorably against the projections of upwards of 136 experts and other methodologies over the same time span.

Predictors included in the model are all easily computed, and include the RPI ranking (using the former version of the metric), losses below 0.500 in-conference, wins against the RPI top 25, wins against the RPI second 25, games above 0.500 against the RPI second 25, games above 0.500 against teams ranked 51–100 in RPI, road wins, and being in the Pac-10/12. That Pac-10/12 membership improved model fit and predictive accuracy is consistent with prior literature on bid decisions from 1999–2008.