NCAA “Point Shaving” as an Artifact of the Regression Effect and the Lack of Tie Games

Wolfers concluded that point shaving may occur in 6 % of NCAA Division I basketball games involving a team favored by more than 12 points. His analysis is flawed on two counts. First, a regression effect is introduced when the definition of a strong favorite is based on the betting market’s estimate of the winning margin. Second, the effect that a lack of tie games has on the resulting distribution of game outcomes relative to the Vegas spread is mistaken as a shift in the distribution because of point shaving. When combined, the regression effect and the lack of ties generate the observed data patterns. Numerical analysis indicates that even if the betting market makes unbiased estimates of the winning margin, an error with a standard deviation of 1.35 points is sufficient to generate the observed data pattern