A model-based approach to football strategy.

April 1, 2004


Team-Specific Home Field Advantage

During the 2001 NFL regular season, home teams had a record of 136-112. Home teams improved to 148-107-1 in 2002, and 157-99 in 2003. This translates into a combined win percentage of 58% for those three seasons. It seems clear that playing at home is an advantage.

This advantage ostensibly increases during the playoffs: Home playoff teams were 21-9 from 2001 through 2003 (70%), and 102-38 (73%) from 2000 through 2003. However, this apparent increase is presumably due to "reverse causality": In general in the playoffs, the team with the better record gets to play at home.

An article at TwoMinuteWarning presents evidence that home field advantage increases in the final weeks of the season. We share their suspicion that this finding is a result of sampling variation, rather than a genuine phenomenon.

The question we wish to address here is whether some teams are particularly strong at home, or particularly weak on the road, or both. In other words, whether there is "team-specific" home field advantage. At first glance, it certainly appears that there is. For example, in 2003 the Seattle Seahawks were 8-0 at home and 2-6 on the road, generating considerable discussion of the reason for the large disparity. On the other hand, even if home field advantage were identical for all teams, in any particular season some teams would do much better at home than on the road, just due to random variation. So the first thing we should ask is whether having six more home wins than road wins is really such a surprising event.

How much difference in home-versus-road records would we expect among teams just from random variation? To try to answer this question, we simulated a large number of NFL seasons, assuming that (1) all teams are identical, and (2) the home team wins with probability 0.58. The variable we are interested in is "excess home-field wins" (EHFW), defined as a team's home wins minus its road wins. For example, if a team is 6-2 at home and 4-4 on the road, EHFW equals 2, whereas if a team is 2-6 at home and 3-5 on the road, EHFW equals -1. For Seattle in 2003, EHFW was 6.

It turns out that under the assumptions made above, in 18% of seasons there will be a team with at least 6 more home wins than road wins. In 53% of seasons, there will be a team with at least 5 more home wins than road wins. So, Seattle's 2003 home-versus-road disparity would be a bit unusual due to chance alone, but not extremely so.

One caveat regarding this analysis is suggested by the observation that if a team's overall record is 16-0 or 0-16, EHFW is necessarily zero. Because we have assumed that the teams are identical, the simulation presumably generates too many teams with records near .500, and so may overstate the likelihood of large values of EHFW. This bias is probably not large, though, because the assumption of identical teams actually generates more variation in overall records than one might think. Indeed, even with identical teams, in 70% of seasons there will be a team with at least 12 wins.

So, even if all teams possess a home field advantage to an identical degree, in any given season we would expect to see a team whose record is much better at home than on the road. This suggests that merely on the basis of a large EHFW, we shouldn't conclude that a team is particularly strong at home relative to the road. In other words, team-specific home field advantage may be a mirage.

Scatter Plot

There is another analysis that we can do to shed more light on this question. If home field advantage is somthing teams are endowed with equally, then EHFW will still vary substantially among teams in a given year, but a team's EHFW in one year should be uncorrelated with its EHFW in subsequent years. On the other hand, if it's really true that some teams are unusually good at home relative to on the road, we would expect that disparity to persist from one year to the next. We can check to see if a team's EHFW in 2002 is a useful predictor of its EHFW in 2003. Each point in the scatter plot at left represents an NFL team. The horizontal axis shows its EHFW in 2002, and the vertical axis shows its EHFW in 2003. (The uppermost point in the scatter plot is Seattle, which actually won one fewer game at home than on the road in 2002.) Visually it's clear that there is no relationship here, but for those who want a more formal analysis, the slope coefficient of the least-squares regression of 2003 EHFW on 2002 EHFW is 0.0436, with a standard error of 0.2435.

In summary, home field advantage is real, but there is scant evidence that it varies significantly among teams.

Copyright © 2004 by William S. Krasker