The Point Spread as a Reference Point: Evidence from the National Basketball Association

TL;DR

Using NBA regular-season point-spread and outcome data, the paper finds that heavy favorites fail to cover the spread more often than slight favorites, and heavy underdogs also lose by more than expected. The authors argue these patterns are more consistent with underdogs using the point spread as a reference point than with point-shaving, which they deem unlikely in the NBA because of high salaries.

Key findings

• Heavy favorites (at least 14 points) were significantly less likely to cover than lesser favorites (13.5 points or less); the kernel density for heavy favorites peaks at NFP<0 while lesser favorites peak near NFP=0, and the test of equal densities yields p-value = 0 (n=984 heavy, n=25,755 lesser).
• The paper rejects the null that NFP distributions are equal for heavy and slight favorites; it reports p-value close to zero / p=0 for the comparisons.
• Heavy underdogs also fail to cover and lose by more than predicted at an unexpectedly high rate, a right-tail finding the authors say favorite-only explanations do not account for.
• The authors conclude NBA point-shaving is unlikely because high salaries create prohibitively high disincentives, and propose reference points: underdogs may pursue a “moral victory” by beating the spread when winning is out of reach.

Methods

• Data: NBA regular season games from 1993-1994 through 2018-2019, with point spreads and outcomes from NBA Odds and Scores on Kaggle; games without posted spreads and zero-point-spread games were omitted, leaving 26,739 regular season games (the text also reports N = 19,303 in one place).
• Computed Net Favored Points (NFP) = favored team’s score − underdog team’s score + point spread, then estimated kernel densities of NFP for heavy vs slight favorites at thresholds such as 14 and 10 points.
• Compared distributions using nonparametric/bootstrap tests of equality, with confidence bands, testing H0: f(NFP)=g(NFP) for all NFP against H1: f(NFP)≠g(NFP) for some NFP.

Limitations

• Stated: the reference-point explanation for underdog risky strategies is not directly tested; the paper says “Whether this occurs in fact is the subject for further research.”
• Evident: the analysis is observational and cannot directly distinguish point-shaving, coaching strategy, bettor bias, or reference-point behavior; the paper’s causal language about the point spread as a reference point is an interpretation of distributional patterns.
• Evident: the sample omits games without posted spreads and zero-spread games, and the text reports inconsistent sample sizes (N = 19,303 vs. 26,739), which may affect generalizability.