Predicting NBA Road Team Attendance: Do Marginal Fans Come to See the Home Team Lose?

How many of these fans aren’t just here for the Suns?

By John Ezekowitz and Alex Koenig

Before the start of the latest NBA season, NBA superfan and ESPN columnist Bill Simmons wrote a column ranking all NBA teams in order of “must-see.” Simmons has his personal definition of what makes a visiting team a “must-see,” but that got us thinking: what makes fans come out to see a road team in the NBA?

There is a huge disparity in road attendance for NBA teams. Last year, the Cavs and Lakers played to completely full arenas on the road, while the lowly 76ers and Nets drew under 80 percent attendance on average. Now one might say that this question is fatuous: of course teams like the Lakers and Celtics will draw more fans on the road; they have wider fan bases and more super stars. But using a dataset of a decade of attendance data and a technique called fixed effects regression, we control for “team identity” factors in order to tease out what really determines NBA road attendance.

First, we compiled road attendance, data from 2000-2010- as a percentage of capacity, not a raw number. We did not have the time or the data to do a game-by-game breakdown, and thus we used season averages. While this limits some of our explanatory power, it does not change the conclusions since NBA teams generally play every team on the road at least once each season. Additionally, before 2008, the database did not have percentages, simply raw attendance numbers. As such, we had to manually calculate averages based on arena capacity. We were also able to control for teams that switched arenas or cities.

We also compiled a comprehensive basketball and demographic profile for each team in the dataset. This included the team’s record, its offensive and defensive ratings, and its pace. We also incorporated the number of all-stars on the team. For this, we used participation in the previous year’s All-Star Game. While All-Star Game selections are somewhat arbitrary, the starting lineups are determined by fan voting, and are thus should be a good reflection of the casual fan’s interest in specific players and teams. Because All-Stars are not a continuous variable, we standardized it so that it had a normal distribution. We also controlled for the presence of “superstars,” which we defined as being on the first or second All-NBA team in the previous season. On the demographic side, we added variables for the total population in a team’s metro area (transformed by the log transformation for normality), and the number of other pro sports teams in the metro area.

The Results:

To control for the aforementioned “team brand name” effects and also for attendance trends over the years, we used a technique known as fixed-effects regression. This allows us to identify effects that vary across individual teams, but not over time (the brand effects), and over time but not over teams (the year effects). Additionally, fixed effects also should capture any attendance differences based on ticket pricing. This is because ticket pricing stays relatively constant over the course of the year for a specific team (although ticket giveaways may distort this slightly).

Our main independent variable was winning percentage. As you can see from the chart below, Winning Percentage is not linearly related to road attendance. To fix this and satisfy the assumptions that underlie fixed-effects regressions, we transformed Winning Percentage to the power of 3/2.


Two quick notes: the outlier in the upper left of the graph is the 2002-2003 Washington Wizards, also known as the team that Michael Jordan came out of retirement to play for. Additionally, many teams have nominal attendance capacities that can be exceeded. This explains the teams that have “over full capacity” road attendance.

The results of the regression are summarized below:

As you can see from above table, winning percentage is a statistically significant predictor of road attendance. Because powerwl is not linear, we can’t make linear coefficient estimates; increasing a team’s winning percentage by 5 percent will have a different effect on their road attendance depending on where the team was before (i.e. at .350 winning percentage vs. increasing from a .60 winning percentage). To better illustrate the effects, we present this graph, which shows the estimated effect of winning percentage holding all other variables in the regression constant at their means.

If anything, this graph under-estimates the true road attendance of teams with higher winning percentages (above .600) because those teams almost invariably had at least one All-Star. A team with two All-Stars, for instance, would be expected to have 1 percent higher road attendance than an identical team with zero All-Stars.

These results at first seem intuitive: of course better teams with have better road attendance. But the implications are much more interesting. Because we are using fixed effects regression, we are actually getting a great picture of the behavior of the “marginal fan.” These marginal fans, ones who do not regularly attend NBA games, account for the differences in road attendance percentages of teams. Because winning percentage is positively correlated with road attendance, these marginal fans appear to come out in larger numbers as the home team’s chances to win decrease. We’ve also controlled for the “brand name effect,” so it’s not just that better teams have larger fan bases in road cities: the “marginal fans” really do appear to come out to see the home team lose more often than not.

Of course, what we have done here is limited by our use of season average data. Dave Berri at Wages of Wins has done studies showing the presence of a “superstar externality.” Berri and his coauthor Martin Schmidt find that the presence of superstars also has a large effect on road attendance. Another study by Paul Holmes (not linked as it is a PDF) found similar results, controlling for scheduling variables at the individual game level. Here we find that winning percentage has a larger effect than the presence of All-Stars or superstars. This may be because we do not use single game data, but also could be because we use a much larger dataset (the Holmes paper only used four seasons of data). Whatever the reason for the different magnitudes of effects, it is still clear that road attendance in the NBA depends not on team style, but rather on winning and star power.

About the author

harvardsports

View all posts

1 Comment

Leave a Reply

Your email address will not be published. Required fields are marked *