Favorite-Longshot Bias
The favorite-longshot bias occurs when betting-implied win probabilities are too low for favorites and too high for longshots. Risk-seeking and attempts to recover losses are proposed explanations. The passage also describes how differing base rates affected the calibration of sports commentators.
THE IDEA
Domain 5: Calibration in Sports
In no other domain are predictions more ubiquitous – or arguably less important – than in sports. Commentators, fans, bookies, and bettors all have a reason for assessing the probability of a win, the chances of the champion surviving the first round, and the likelihood of setting a new world record. Sports events have been a major source of data for studies of predictive judgment, but few of these studies have used expert judges. An exception to this rule is the study of horserace odds by economists interested in the rationality of the aggregate market. Odds for horse races shift as a result of the money placed by large numbers of individual investors, so the unbiasedness of such odds is a test of the rationality of market forces. Many studies have revealed a small but systematic bias such that betting-implied win probabilities and the shares wagered on favorites (high-probability winners) are too low whereas those for longshots (low-probability winners) are too high (termed the favorite-longshot bias for obvious reasons; e.g., Snyder, 1978; Thaler & Ziemba, 1988). This has been explained by general risk-seeking in gamblers (the pleasure of the long-shot) and by loss aversion driving gamblers to try to turn their losses into gains by a big win. It has even been tied to an “overconfidence” bias, though the observed pattern is actually one of underextremity.
We know of one cross-national study of expert commentators (television and newspaper sports reporters) that offers a clear test case for the direct support account. Ayton and Onkal (1996) asked media commentators in England and Turkey to give probabilistic predictions for a set of football (soccer) Cup Matches in their own countries. The natural target hypothesis is a home team win. In fact, there was a marked home field advantage in the Turkish cup games and little home field advantage in the English cup games; in other words, the base rate for a home team win was high (80%) in Turkey and moderate (close to 50%) in England. The commentators in both countries were moderately (and equally) good at discriminating between winning and losing teams (α = 1.07) leading to a situation in which the direct support account implies good calibration for the English commentators and substantial underprediction for the Turkish commentators. As the top lines in Fig. 39.9 attest, this pattern emerged very clearly.
When to use it
Evaluating betting strategies in sports or financial markets to avoid systematic biases.
Improving the accuracy of expert forecasts by adjusting for base rates and known biases.
Understanding consumer behavior in lottery or gambling contexts where longshots are overvalued.
Designing decision support systems that account for calibration biases in probabilistic judgments.
FROM THEORY TO EVERYDAY LIFE
In practice
Favorite-Longshot Bias in Horse Racing
In the pattern described here, betting prices imply a lower win probability for favorites than their observed win rate and a higher probability for longshots. Expressed as payout odds, this means favorites offer relatively generous returns and longshots offer relatively poor returns. Pooling many bettors’ judgments does not automatically remove this bias.
Cross-National Commentator Calibration
Ayton and Onkal (1996) studied football commentators in England and Turkey predicting cup matches. In Turkey, home team win base rate was 80%, while in England it was 50%. Commentators in both countries showed similar discrimination ability, but the Turkish commentators consistently underpredicted home team wins, demonstrating how base rates affect calibration.
A MOMENT TO REFLECT
Put it to work.
How does the favorite-longshot bias affect your own predictions in sports or other domains where probabilities are involved?
Can you recall a time when you overvalued a low-probability event (longshot) or undervalued a high-probability event (favorite)? What drove that decision?
How might understanding base rates improve your calibration when making predictions under uncertainty?
Often confused with
References
- Study on horse race odds bias — Snyder (1978)
- Parimutuel betting markets — Thaler, R. H., & Ziemba, W. T. (1988)
- Effects of base rates on calibration of probability judgments in soccer — Ayton, P., & Onkal, D. (1996)
CONTINUE THE CONVERSATION
Source texts
- Heuristics and Biases The Psychology of Intuitive Judgment (Thomas Gilovich, Dale Griffin, Daniel Kahneman)