|Ivars Peterson's MathTrek|
January 19, 1998
One of the attractive features of spectator sports is the uncertainty of the outcome. Even when one team is overwhelmingly favored to win, the underdog may still come through with a surprising victory.
Nonetheless, the ability to pick winners can be of considerable interest, especially to gamblers, who bet on the outcomes of games, and to the oddsmakers who, in effect, determine the playing field for the bettors.
In the current issue of Chance, statistician Hal S. Stern of Iowa State University in Ames takes a look at what sort of simple information may be helpful for identifying winning teams, though not necessarily for making bets that beat the spread (or odds). "The question of primary interest is what proportion of game outcomes could be correctly predicted by an intelligent observer," Stern says.
Stern focuses mainly on U.S. professional sports, though the analysis can be easily applied to other sports, as long as the right sorts of data are available.
One simple prediction rule is to pick that team that plays at home. "This rule ought not to predict very well because it completely ignores the relative ability of the teams that are competing," Stern remarks.
Nonetheless, the evidence supports the existence of a home-field advantage (see table), especially in basketball. Moreover, the home-field advantage for college sports appears to be slightly larger than for professional sports, Stern says.
Additional analysis indicates that playing on one's home field rather than at a neutral site is worth about 3 points in football, 4.5 points in basketball, and 0.25 run in baseball.
Oddsmakers employed by sports betting establishments make a living by forecasting the outcomes of games, though their goal is not so much to make accurate predictions as to set the odds (point spread). When they succeed, the proceeds from losing bets pay off the winning bets, with a small percentage going to the betting establishment.
Oddsmakers' predictions generally prove to be a superior guide for identifying winning teams. Inbaseball, however, the oddsmakers do only slightly better than always picking the home team. Oddsmakers do somewhat better predicting basketball and football outcomes. College games prove to be more predictable than professional contests in the same sport.
It's also possible to apply simple statistical techniques (such as the method of least squares) to the win-loss records or margins of victory in previous games of the participating teams. In effect, the methods provide an estimate of the ability of each team.
In football, Stern's rudimentary statistical approach does nearly as well as the experts and considerably better than the strategy of always picking the home team, particularly when scores are used. A similar pattern occurs in the other sports.
"How well can we predict sports outcomes?" Sterns asks. "The answer depends on the sport."
"Baseball appears to be the most random sport," he concludes. "The best prediction approaches are just a bit better than using coin flips to predict."
In basketball and football, the prediction accuracy can reach 75 percent, but that still leaves plenty of uncertainty. "One might argue that if things were any more predictable than that it would be difficult to convince people to pay for the privilege of watching the games!" Stern notes.
Copyright 1998 by Ivars Peterson
Stern, Hal S. 1997. How accurately can sports outcomes be predicted? Chance 10(No. 4):19-23.
Comments are welcome. Please send messages to Ivars Peterson at email@example.com.