In early May 2007, a study gained publicity that claimed NBA referees altered their habits of calling fouls based on the racial makeup of the offending players. Maria Rainier recently helped write an article about changing gender discrimination. In keeping with that theme, I felt that it would be appropriate to continue the discussion with racial bias.
The NBA is a $4.0 billion dollar a year enterprise (source). 30 teams play basketball 82 games a year for 48 minutes a game. Everything is quantified using statisticians. The study linked above looks at all games played from 1991-2004. Needless to say, there is a lot of data to filter through. What this means, however, is that any significant results are very trustworthy due to the accuracy in the data as well as the sheer volume of the data collected.
Nobody here is intending to suggest that there is a deliberate bias found in any of the results. There is no claim that there is a conspiracy by the higher-ups in the NBA to “cater” to the fans or that the referees specifically target minorities to call fouls. Anybody who watches an NBA game can tell you that a handful of calls throughout the game can usually go either way. The referees need to make a decision in a split-second, right or wrong. It is at this exact moment, however, where underlying prejudices may be revealed. They may be so minor or latent, also, that they can only be revealed over the vast amount of data collected.
The authors Joseph Price of BYU and Justin Wolfer of UPenn use very clever methods to determine racial bias. Because the NBA (in order to protect the identities of their referees) does not reveal who makes each individual call per game, the authors simply create a variable representing the racial makeup of the refereeing crew. Naively regressing fouls on racial composition does not work well because of some very simple objections. White players, on average, are taller than black players and play different positions. On average, then, they should be called for more fouls. Also, some have argued that European players (who tend to be white) play a different style of play which cannot be compared adequately. The authors control for this by including a variable (called fixed effects) for every single player who played between 1991-2004. Thus, any question about differences in play-style are captured by this variable.
The same argument can be used to control for variables such as referees (some referees may just call more fouls on average than others), location (home vs. away), attendance (fouls may be called more often under certain circumstances) as well as a myriad of other variables (such as year, differences in foul calling across years as the game has evolved). The authors find a significant effect on the racial makeup of the referees versus the player called for the foul. For an organization such as the NBA that prides itself on the utmost accountability and objectivity of its referees, this seems puzzling.
When confronted with evidence such as this for discrimination, it is important to try to determine what is occurring. It is difficult to say how this bias is emerging, but it is clear from the evidence that there is a significant “split-second” bias emerging from the data. How is it possible to counteract this underlying prejudice? Any suggestions?
P.S. The paper is being edited to be included in a forthcoming edition of the QJE. Also, earlier studies have attempted to estimate the amount of racism amongst fans of sports (and the NBA) with limited success. At the risk of editorializing, I would like to point out that these studies produced weak results.