INTERRACIAL WORKPLACE COOPERATION: EVIDENCE FROM THE NBA

JOSEPH PRICE, LARS LEFGREN and HENRY TAPPEN

Using data from the National Basketball Association (NBA), we examine whetherpatterns of workplace cooperation occur disproportionately among workers of thesame race. We find that, holding constant the composition of teammates on the floor,basketball players are no more likely to complete an assist to a player of the samerace than a player of a different race. Our confidence interval allows us to rejecteven small amounts of same-race bias in passing patterns. We find some evidence ofown-race bias in situations where the outcome of a particular play is less important.Our findings suggest that high levels of interracial cooperation can occur in a settingwhere workers are operating in a highly visible setting with strong incentives to behaveefficiently. (JEL J15, J71, L23)

I. INTRODUCTION

Recent research shows that when individualsare forced to make quick decisions, they oftenexhibit same-race preferences, even if they areunwilling to admit to biased racial attitudes. Forexample, Price and Wolfers (2010) show thatNational Basketball Association (NBA) refereesare more likely to call fouls against players ofa different race than players of their own race,and Parsons et al. (2011) find that umpires aremore likely to call strikes for pitchers of theirown race. Similarly, Antonovics and Knight(2009) find that police are less likely to searchthe vehicle of someone of their own race, andDonohue and Levitt (2001) find that an increasein the number of police of a certain race isassociated with an increase in arrests of peopleof the other race.

This same-race bias could play an impor-tant role in collaboration among colleagues ina workplace. For example, managers might bemore likely to give favorable assignments tosame-race employees. Alternatively, colleaguesmay depend disproportionately on same-racecolleagues for advice or help. Collectively, such

Price: Department of Economics, Brigham Young Univer-sity, 162 FOB, Provo, UT 84602. Phone 801-422-5296,Fax 801-422-0194, E-mail joe_price@byu.edu

Lefgren: Department of Economics, Brigham Young Univer-sity, 162 FOB, Provo, UT 84602. Phone 801-422-5169,Fax 801-422-0194, E-mail lars_lefgren@byu.edu

Tappen: Department of Windows, Microsoft Corporation,1 Microsoft Way, hetappen 84/1029, Redmond, WA98052. Phone 801-422-5169, Fax 801-422-0194, E-mail:htappen@gmail.com

decisions may reduce the workplace productiv-ity and satisfaction of employees of a minor-ity race. These decisions may play a role inexplaining the extent of workplace segregation(Hellerstein and Neumark 2008). Furthermore,this bias would undermine the argument thatproductivity is higher in groups that are ethni-cally diverse (Page 2007).

In this paper, we examine the effects of groupheterogeneity on teamwork by studying specificand measurable actions within teams. In tra-ditional firm-level data, it is often difficult toobtain measures of cooperation. As a result,we use play-by-play data from the NBA. Thesedata allow us to determine for each basket, whopassed the ball and which other players wereon the court at the time. We develop a simplemodel which allows the optimal pass to dependon the particular combination of teammates onthe floor. We then test whether the pattern ofobserved assists demonstrates evidence of same-race bias.

We find no evidence that, conditional onthe set of teammates on the court, players aremore likely to pass to a teammate of their samerace. Our baseline empirical strategy controlsnon-parametrically for the joint distribution ofshot quality for all teammates on the floor.In other words, we account for differences

ABBREVIATION

NBA: National Basketball Association

1

Economic Inquiry(ISSN 0095-2583)

doi:10.1111/j.1465-7295.2011.00438.x© 2012 Western Economic Association International

2 ECONOMIC INQUIRY

in ability across teammates. Furthermore, theshooting opportunities for one teammate areallowed to depend arbitrarily on the set of otherteammates on the floor. Robustness checks, inwhich we reduce the flexibility of our empiricalspecification to increase statistical precision,yield the same substantive results. Our evidencesuggests that cross-race cooperation may notbe a problem in workplaces where employeeshave common goals and extensive experienceworking with each other.

II. SIMILARITY AND COOPERATION

There is considerable research, empirical andtheoretical, indicating that diversity can lead toimproved economic outcomes (Alesina and LaFerrera 2005; Alesina, Spolaore, and Wacziarg2000; Hong and Page 2004). These gains fromdiversity depend on the various groups beingwilling to cooperate. This may explain whyother studies have found that increased racialdiversity is associated with lower group per-formance (Kurtulus, forthcoming; Timmerman2000). In this paper, we expand the literature onthe effects of group heterogeneity on outcomesby studying specific and measurable actionswithin teams.

Our analysis of cooperation is one formof own-race bias that has been documentedin other types of interaction including referee-player (Parsons et al. 2011; Price and Wolfers2010), employer-employee (Stoll, Raphael, andHolzer 2004), and officer-offender (Antonovicsand Knight 2009; Donohue and Levitt 2001).What distinguishes cooperation in our studyfrom these other settings is that players areworking together toward a common objective,while these other settings involve a more adver-sarial relationship. In addition, there is researchshowing that the racial composition of one’sgroup affects decisions similar to cooperation,such as willingness to provide a public good(Martinez-Vazquez, Rider, and Walker 1997),form a coalition (Brasington 1999), or increasewelfare spending (Luttmer 2001).

III. NBA DATA

Our analysis draws on play-by-play datathat we collected from espn.com for all reg-ular season and playoff games during October2002–June 2008. The data set includes an entryfor every occurrence during the game that might

be important for compiling game-level statistics.For each shot that is completed on the court, thedata provide the name of the person who shotthe basket and the person who was awarded withan assist, if the shot was assisted (58.9% are).Using the timing of substitutions in the data,we apply recursive methods to determine the 10players on court at any given time.

Thus, for each assisted shot we know whothe passer was and who the other four avail-able players would have been. For each of theseplayers we include information about their race,position, and other characteristics. The player’srace information comes from data collected byKahn and Shah (2005) and Price and Wolfers(2010), and our own coding from more recentonline photos of the players. Our racial coding isbased on a simple measure of “black” or “not-black” (we refer to the not-black category aswhite and it includes non-black Asian and His-panic players).

As part of our empirical strategy, we con-struct identifiers for each unique group of fourplayers available to receive a pass on a particu-lar play. For our primary analysis, we limit oursample to passing opportunities in which therewas at least one player of each race available toreceive the pass. This eliminates about 39.3% ofour observations for situations in which the fourplayers available to receive the pass are blackand another 0.32% of our observations whenthose four players are white. Including theseobservations in our sample would bias our esti-mates of own-race discrimination toward zero,since the passer in these situations has no choiceregarding the race of the player he can pass to.1

Our estimates using player-group fixed effectsrely on variation in the race of the passer. As aresult, we also eliminate each unique groupingof the four players in which all of the passersthat we observe for that group are the same race.This restriction eliminates an additional 11.1%of our original sample. The combination of thesetwo restrictions reduces our original sample of302,714 assisted shots to 149,059 observations.These sample restrictions leave a final samplethat includes more observations from teams witha higher fraction of white players.2 However,

1. Since our results indicate that there is no racial bias,we want to use the empirical strategy that has the best chanceof detecting racial bias if it were to exist.

2. For example, only 20.3% of the observations for theUtah Jazz are excluded (a team where 56% of the shots aremade by black players) while 81.2% of the observations forthe Detroit Pistons are excluded (a team where 92.8% of theshots are made by black players).

PRICE, LEFGREN & TAPPEN: INTERRACIAL WORKPLACE COOPERATION 3

we also run our results with the full sampleof observations and find results that are verysimilar to those of our primary sample (withcoefficients which are slightly smaller in mostcases, as expected).

Table 1 shows summary statistics for bothour restricted and unrestricted samples. The pri-mary difference between the full and restrictedsample is the fraction of shooters that are white.White players completed 37.1% of the shots andmade 20.6% of the passes in our sample.3 Sepa-rating the probability that the shooter is whitebased on the race of the passer (as done incolumns 4 and 6) provides an initial estimateof racial gaps in passing. A white passer is3.7 percentage points more likely to complete anassist to a white shooter than is a black passer.However, this simple estimate fails to captureany clumping of white players on the same team(or on the court together) or differences in thepositions they play.

IV. MODEL AND EMPIRICAL STRATEGY

Before progressing to our empirical speci-fication and findings, it is helpful to outline asimple economic model of cooperation. Basket-ball involves complex offensive and defensivestrategies. For this reason, we define a simplergame that will highlight the intuition involvedand suggest an empirical strategy for identifyingpossible same-race bias in passing patterns.

Consider a game with five players. One ofthe five players is initially endowed with theball. The player then passes the ball to one ofthe four remaining players with the best shotat the basket. We define S as the set of fourplayers available for the pass. Player i ∈ S hasan opportunity for shot, the quality of which isgiven by μi . Player j ∈ S is another player inthe passer’s choice set with shot quality μj . Wedo not assume that μi and μj are independent oridentically distributed. We do, however, assumethat both are independent of the player passingthe ball. We assume that the player with the ballpasses to his teammate with the highest qualityshot.4 Given these assumptions, the probability

3. These are an overestimate of the actual fraction ofpasses and shots made by white players because our sampleis limited to observations in which there was at least oneplayer of each race available to receive the pass. Withoutthis restriction, white players would make 21.5% of theshots which is roughly in line with their representation inthe NBA.

4. It could also be true that the player takes into accountnot only shot quality and racial preference but also the

that the player with the ball passes to player igiven the set of available passing options, S, canbe written θi,S = prob(μi > μj∀j �= i ∈ S).5

This simple model suggests a tractable empir-ical specification to test the role of race inon-the-job cooperation. We can estimate thefollowing linear probability model:

1scorer=whiteS,p = θw,S + β1assister=white

S,p + εp,(1)

where 1scorer=whiteS,p is an indicator variable that

takes on a value of 1 if the player scoring thebasket is white, given the set of available scorersS, during possession p. θw,S is estimated by aset of dummy variables for every combination offour players available to receive the assist. Thiscontrols non-parametrically for the probabilitythat a white player has the best shot, takinginto account the joint distribution of shot qualityamong all players eligible to receive a pass.Thus, θw,S accounts both for the talent of everyplayer available for a pass, as well as how theplayers interact while on the floor together. Italso implicitly controls for the position of everypotential pass recipient. 1assister=white

S,p takes on avalue of 1 if the player making the assist iswhite, and εp is the residual.

Identification arises from the fact that botha white and a black player choose among thesame set of teammates when passing the ball.If white and black passers are solving the sameoptimization problem with the same constraints,β should be statistically insignificant from zero.A non-zero coefficient suggests that the race ofthe passer and potential scorers affects the pat-tern of assists and hence on-the-job cooperation.

The disadvantage of our preferred approachis that it consumes literally tens of thousandsof degrees of freedom since we include separatefixed effects for sets of players that differ onlyby a single role-player. In doing so, we discardlarge amounts of potentially useful information.To the extent that we can approximate θw,S with-out the inclusion of so many dummy variables,we will increase the precision of the estimatedsame-race bias.

preference of the coach and the social pressure of the fans.In this case we would simply reinterpret shot quality as ameasure of the total utility benefit associated with passingto a particular player.

5. As an example of this, consider the case whereμi = μ̄i + εi , where μ̄i represents fixed player ability andεi is distributed according to a Type 1 extreme valuedistribution. In this case, the probability that the ball ispassed to player i is given by prob(μi > μj∀j �= i) =exp(μ̄i )/

∑4j=1 exp(μ̄j ).

4 ECONOMIC INQUIRY

TABLE 1Summary Statistics for Full and Restricted Samples

All Observations Black Passer White Passer

FullSample

RestrictedSample

FullSample

RestrictedSample

FullSample

RestrictedSample

(1) (2) (3) (4) (5) (6)Black shooter 0.7849 0.6290 0.7917 0.6367 0.7623 0.5985

(0.4109) (0.4831) (0.4068) (0.4801) (0.4257) (0.4902)White shooter 0.2151 0.3710 0.2083 0.3633 0.2377 0.4015

(0.4109) (0.4831) (0.4068) (0.4801) (0.4257) (0.4902)Center 0.0651 0.0593 0.0477 0.0438 0.1339 0.1226

(0.2468) (0.2406) (0.2132) (0.2046) (0.3406) (0.3280)Forward 0.2797 0.2584 0.2729 0.2529 0.3063 0.2808

(0.4488) (0.4378) (0.4455) (0.4347) (0.4609) (0.4494)Guard 0.6552 0.6823 0.6793 0.7033 0.5598 0.5965

(0.4753) (0.4656) (0.4667) (0.4568) (0.4964) (0.4906)First year with team 0.3084 0.2993 0.3177 0.3145 0.2721 0.2450

(0.4619) (0.4580) (0.4656) (0.4643) (0.4450) (0.4301)N 302,714 149,059 241,613 118,321 61,101 31,741

Notes: Standard deviations in parentheses. Information about position and first year with team refers to the passer. In therestricted sample, observations are dropped where there is no racial variation on the court.

To do so, we calculate the fraction of assistedbaskets scored by a player while on the floor,excluding baskets in which that player madethe assist. This is calculated separately for eachseason. For player i, we denote this probabilityπ̂i . This reflects the historical probability thatplayer i has the best shot. We can sum theseprobabilities across all white players on the floorat a point in time to approximate the probabilitythat a white player has the best shot at thatparticular point in time. Mathematically, this

proxy for θw,S is given by:

θ̂proxy1w,S =

∑i∈W∩S

π̂i ,(2)

where W is the set of white players and S isthe set of potential pass recipients on the court.While this measure is simple, it fails to take intoaccount any interactions between players on thecourt.

For this reason, we construct a second proxyby normalizing this measure by the propensity of

TABLE 2Factors Associated with the Probability That the Shooter Is White (Using the Restricted Sample)

(1) (2) (3) (4) (5)

Passer is white 0.0353∗∗ 0.0026 −0.0044 0.0065 0.0004(0.0122) (0.0061) (0.0056) (0.0057) (0.0040)

Black players on the court besides the passer — −0.2178∗∗ −0.2094∗∗ — —(0.0036) (0.0030)

Prediction that shooter is white (θ) — — — — 0.9029∗∗

(0.0076)Fixed effects None None Team-year Player-group None

Observations 149,059 149,059 149,059 149,059 149,059R2 0.0013 0.0729 0.0815 0.1524 0.0890

Notes: Standard errors (in parentheses) are clustered at the team-year level.The player-group fixed effects include a control for each unique combination of the four players on the court besides the

shooter. Our measure for θ comes from Equation (3). The sample is restricted to observations where there was at least oneplayer of each race to receive the pass and to player-groups for which we observe at least one passer of each race in the data.Each regression includes controls for the passer’s position and whether the passer is in his first year with the team.

∗Statistical significance at 5%; ∗∗statistical significance at 1%.

PRICE, LEFGREN & TAPPEN: INTERRACIAL WORKPLACE COOPERATION 5

all players in the choice set to score off an assist.This measure again represents a probability thata white player has the best shot but has beennormalized to lie between zero and one. Thissecond proxy is given by:

θ̂proxy2w,S =

∑i∈W∩S

π̂i

/∑j∈S

π̂j .(3)

This proxy takes into account that a player’sprobability of scoring depends not only on hisskill but the skill of the other teammates on thefloor as well. It fails to take into account, how-ever, all of the possible idiosyncratic interac-tions between teammates the way our preferredapproach does.

Using the second of these two proxies forthe probability that a white player has thebest shot, we supplement our primary findingsby estimating linear probability models of thefollowing form.6

1scorer=whiteS,p = α0 + α1θ

proxy2w,S(4)

+ β1assister=whiteS,p + εp

If our proxy performs well, we would expectour estimate of α1 to be close to one. Under thenull hypothesis of no same-race bias, we wouldstill expect β to be close to zero.7

V. FINDINGS

As a starting point, we estimate a simpleregression model in which we analyze the rela-tionship between the passer’s race and the raceof the shooter. All of the regressions we describein this section include controls for the passer’sposition and whether he is in his first year withthe team. In this very simple regression, wefind that the probability that the shooter is blackincreases by about 3.53 percentage points whenthe passer is white. This is similar to the rawdifference that we observed in Table 1.

This simple regression fails to take intoaccount the fact that there is racial clumpingof players in the NBA with some teams having

6. We estimate all the regressions using logit (Table 3)and conditional logit (Table 2) regressions. The marginaleffects and standard errors are nearly identical to those usinglinear probability models.

7. In all specifications, we cluster correct the standarderrors at the team level. This takes into account that passingdecisions may not be independent across assist opportunities.For example, a particular player may consistently look topass to a certain teammate for reasons independent of therace of the two players.

many white players and others having few. InTable 2, when we control for the number ofblack players (besides the passer) on the court,we find that this estimated racial difference dis-appears completely (with an insignificant pointestimate of 0.26 percentage points). We alsoobtain a very small estimate of racial bias whenwe include team-year fixed effects (−0.44 per-centage points).8

Ideally, we want to control for the probabil-ity that a white player has the best shot, tak-ing into account the joint distribution of shotquality among all players eligible for a pass.We do this by including a fixed effect foreach four-player combination on the court asidefrom the passer. We find that the probabilitythat the shooter is white increases by 0.65 per-centage points relative to a baseline probabilityof 37.1 percentage points (or a 1.8% increase)and is not statistically significant. Even at theupper end of the 95% confidence interval, awhite passer is only 1.8 percentage points morelikely to pass to a white teammate than a blackpasser.9

Our player-group fixed-effects model has thedisadvantage of consuming thousands of degreesof freedom since we include separate fixedeffects for sets of players that differ by onlyone person. In column 5, we provide the resultsof an alternative specification in which weapproximate the probability that a white playerhas the best shot (θw,S) as shown in Equation(3). As expected, we find that the coefficient onour proxy of θw,S is close to one, indicating thatour constructed measure effectively captures theprobability that a white player has the best shot(it is even closer to one when we use the fullsample in Table 3).

8. If we were to include the team-year fixed effectswithout controls for the number of black players on thecourt, we would obtain a large negative coefficient (−5.37percentage points). This simply reflects the mechanical biasthat, on a given team, a white passer will have fewer whiteplayers than black players to pass to. This is similar to theproblem that Guryan, Kroft, and Notowidigdo (2009) discussin estimating peer effects in golf.

9. While we find no effect of same-race bias in passingpatterns, an apparent effect would not necessarily indicateon-court animus. Instead, a high incidence of within racepassing might indicate familiarity between players of thesame race. In other words, a white player might be happy tomake a pass to a black player in position to score but insteadpasses to a white teammate because he knows better wherethat teammate will be on the floor. In this case, race-basedsocial investments off the court translate into low levels ofcross-racial cooperation off the court. While the proximatecause of the racial bias differs, the ultimate consequence isthe same.

6 ECONOMIC INQUIRY

TABLE 3Factors Associated with the Probability That the Shooter Is White (Using the Full Sample)

(1) (2) (3) (4) (5)

Passer is white 0.0288 0.0016 −0.0023 0.0035 0.0002(0.0170) (0.0038) (0.0036) (0.0033) (0.0022)

Black players on the court besides the passer — −0.2524∗∗ −0.2456∗∗ — —(0.0017) (0.0022)

Prediction that shooter is white (θ) — — — — 1.0105∗∗

(0.0023)Fixed effects None None Team-year Player-group None

Observations 302,714 302,714 302,714 302,714 302,714R2 0.0008 0.2670 0.2721 0.4230 0.2847

Notes: Standard errors (in parentheses) are clustered at the team-year level.The player-group fixed effects include a control for each unique combination of the four players on the court besides the

shooter. Our measure for θ comes from Equation (3). Each regression includes controls for the passer’s position and whetherthe passer is in his first year with the team.

∗Statistical significance at 5%; ∗∗statistical significance at 1%.

Our test of own-race bias is based on whetherdeviations from this predicated probability areinfluenced by the race of the passer. We findno evidence of a preference by players to passto players of their own race. Our estimate inthe first column indicates that the probabilitythat a white player receives an assist is only0.04 percentage points higher when the passer iswhite than when the passer is black. In addition,the standard errors are about a third smaller thanin our fixed-effects model, providing an evensmaller upper bound estimate of the amount ofown-race bias.

VI. ROBUSTNESS CHECKS

One concern with all specifications is that wehave data on only completed assists. Levels ofracial cooperation could also affect the numberof unassisted baskets as well as passing patternsthat do not lead to assists. Fortunately, we areable to test the hypothesis that the probability ofan unassisted basket depends on the racial com-position of a player’s teammates. In particular,we estimate a linear probability model in whichthe dependent variable is the probability that acompleted shot was assisted.

In our preferred specifications, we control forthe race and position of the scorer as well asthe fraction of that player’s baskets that wereassisted over the course of the entire season.This takes into account that some players havehigher unconditional probabilities of scoringunassisted baskets. We also include a fixed effectfor the set of other players who are on the courtat a given time. The variable of interest is the

interaction between the white shooter variableand the number of black players on the court.

Under the racial animus hypothesis, the coef-ficient should be significantly positive. In allspecifications, the coefficient is close to zero.Once we control for either team-year or team-mate combination fixed effects, the coefficient isstatistically insignificant. Overall, these specifi-cations, which are available in supporting infor-mation in the online version of this article, areconsistent with high levels of cross-racial coop-eration.

We cannot, however, observe passes thatwere made that did not lead to shots and poten-tial assists that were not converted. While thisdata limitation could affect the apparent magni-tude of same-race bias, it does not affect the signof the coefficient. Suppose a player systemati-cally passes to teammates of his own race, eventhough they have worse shots than teammatesof another race. In this case, assists betweenplayers of the same race will be more com-mon than assists between players of differingraces. This difference in assists will be lessthan the difference in attempted assists, how-ever, because passes made for race-based rea-sons will be less likely to lead to convertedbaskets. When players choose to score aloneor make a pass to a teammate out of positionto score instead of assisting to a teammate ofa differing race, this also increases the relativefrequency of same-race assists. Thus data limi-tations may affect the magnitude of the observedsame-race bias, but our procedure still provides avalid test for the existence of same-race passingpreferences.

PRICE, LEFGREN & TAPPEN: INTERRACIAL WORKPLACE COOPERATION 7

An additional possibility is that passers dis-criminate differentially against the worst play-ers of the other race. If this is true, we wouldexpect lower levels of interracial cooperationwhen the same race players are better shootersthan other race players. To test this hypothe-sis, we re-estimate our preferred specificationbut include an interaction between our whitepasser indicator variable and the difference infield goal percentage between the potential whiteand black pass recipients. The coefficient on thisinteraction and the white passer variable are bothinsignificant, suggesting that racial differencesin shooting percentage do not appear to affectthe level of own race bias.

Recent research by Parsons et al. (2011)indicates that racial bias on the part of baseballumpires is price sensitive. The umpires exhibitmore bias in situations where there is lessscrutiny of their decision or when the outcomeof their decision is less important. We examinewhether own-race bias in passing is influencedby the importance of the situation by focusingon passes during the fourth quarter and splittingour sample based on the score margin at the startof the quarter.

Rather than pick an arbitrary cutoff, we findthe estimated coefficient and confidence intervalfor each possible cutoff (i.e., games in whichone team is up by at least X points at thestart of the fourth quarter). The solid line inFigure 1 provides the estimated effect of own-race bias and the shaded area provides the95% confidence interval. Figure 1A is based onestimates using the player-group fixed effectsand Figure 1B is based on estimates using ourtheta-proxy model. While the precision of ourestimates is much tighter when using the theta-proxy model, both figures provide suggestiveevidence that the amount of own-race biasincreases when the importance of making aparticular pass is lower.10 In fact, when usingthe theta-proxy approach, we find statisticallysignificant levels of own-race bias when using

10. To further isolate the effect of the game’s importanceon racial bias, we ran the same regressions as in column(3) of Tables 2 and 3, but split the sample based on whetheror not the team performing the assist is still in playoffcontention. We find that the passer’s race coefficient isinsignificant and similar in magnitude in both specifications,suggesting that the importance of the game itself does notappear to affect racial bias. Since this is a measure of thegame’s overall importance, our finding does not conflict withthe in-game bias we observe as score differentials increase.These regressions are found in supporting information in theonline version of this article.

FIGURE 1Change in Estimated Bias Based on AbsoluteValue of Score Difference at the Start of theFourth Quarter. (A) Player-group fixed-effect

estimate; (B) Theta-proxy estimate

A

B

Notes: The 4th quarter margin refers to the absolutevalue of the difference in the two scores at the start ofthe fourth quarter. The dark line provides the estimatedcoefficient of the same-race bias for games in which theblow-out margin was at least as large as the numberindicated on the x-axis. The shaded region provides the 95%confidence interval.

a cutoff of a score margin between 11 and 17points at start of the fourth quarter.

The prior results suggest that discriminationmay be more likely when it is unlikely to affectthe outcome of a game. It may also be possi-ble that discrimination is practiced most by starplayers whose employment in the NBA is secureeven if they exhibit discriminatory behavior thatadversely affects performance.11,12 Marginal

11. Any adverse performance effects may still bereflected in their compensation, however.

12. Analogously, Becker (1971) points out that employ-ment discrimination is more likely to occur in firms withmarket power than in competitive firms. In competitive set-tings, non-discriminating firms will expand and drive outdiscriminating ones. In the case of a monopoly, however, adiscriminating firm can exist indefinitely, though with lowerprofits.

8 ECONOMIC INQUIRY

players have little leeway to exhibit discrimina-tory behavior, since doing so may reduce theireffectiveness and cost them their jobs. We testthis hypothesis by running separate regressionsfor players who averaged more than 30 min-utes per game and players who averaged fewerthan 10 minutes per game. We believe that thesetwo groups represent inframarginal and marginalplayers, respectively. In results not shown, wefind that both of these groups exhibit insignifi-cant amounts of own-race bias. The estimates forthe two groups are also insignificantly differentfrom each other.13

VII. INTERPRETATION AND CONCLUSION

Assist patterns in the NBA exhibit very littleevidence of same-race bias. More specifically,given a particular set of players on the court,a white passer is no more likely to pass to awhite teammate than a black passer. While thisresult is interesting, it is important to note whythe high degree of interracial cooperation maybe specific to the NBA.

Our model can be thought of as the final nodein a more complex game in which players arematched to teams, and coaches decide whichplayers are on the floor and which plays to run.To the extent that some players exhibit strongsame race preferences, they may be matchedto teams with a high frequency of same-raceteammates. With sufficiently high levels of racialsegregation, the economic penalty of same racepreferences could be quite small. Furthermore,conditional upon the team roster, coaches maychoose personnel combinations in which effi-cient passing choices occur. Such optimizingbehavior does not affect the consistency of ourresults. It does imply, however, that our resultsonly reflect the level of same-race bias in situ-ations that occur in equilibrium.14 The averagelevel of same race bias may be higher.

In many workplace environments, the effectof same-race bias may not be immediatelyapparent or have little impact on the actorsinvolved. The NBA differs from such instancesin that player behavior is closely observed bycoaches, owners, and many thousands of fans,and the result of poor interracial cooperation

13. The results are found in supporting information inthe online version of this article.

14. Charles and Guryan (2008) examine how equilib-rium levels of discrimination vary with the taste for dis-crimination and the number of black workers.

may have an immediate effect on the outcomeof the game. Also, to the extent that play-ers derive utility from winning, they have animmediate incentive to engage in interracialcooperation.

Ultimately, our findings do not imply thatefficient interracial cooperation occurs through-out the economy. They do imply, however, thatinterracial cooperation can occur in equilibriumwhen incentives are well aligned for efficientcooperation. Firms may want to consider howthey can alter incentives to promote efficientcooperation among a diverse workforce.

REFERENCES

Alesina, A., and E. La Ferrera. “Ethnic Diversity and Eco-nomic Performance.” Journal of Economic Literature,43, 2005, 762–800.

Alesina, A., E. Spolaore, and R. Wacziarg. “Economic Inte-gration and Political Disintegration.” American Eco-nomic Review, 90, 2000, 1276–96.

Antonovics, K., and B. Knight. “A New Look at Racial Pro-filing: Evidence from the Boston Police Department.”Review of Economics and Statistics, 91, 2009, 163–75.

Becker, G. S. The Economics of Discrimination. Chicago,IL: University of Chicago Press, 1971.

Brasington, D. “Joint Provision of Public Goods: The Con-solidaton of School Districts.” Journal of Public Eco-nomics, 73, 1999, 373–93.

Charles, K., and J. Guryan. “Prejudice and Wages: AnEmpirical Assessment of Becker’s The Economics ofDiscrimination.” Journal of Political Economy, 116,2008, 773–809.

Donohue, J., and S. Levitt. “The Impact of Race on Policingand Arrests.” Journal of Law and Economics, 44, 2001,367–94.

Guryan, J., K. Kroft, and M. Notowidigdo. “Peer Effects inthe Workplace: Evidence from Random Groupings inProfessional Golf Tournaments.” American EconomicJournal: Applied Economics, 1, 2009, 34–68.

Hellerstein, J. K., and D. Neumark. “Workplace Segregationin the United States: Race, Ethnicity, and Skill.” TheReview of Economics and Statistics, 90, 2008, 459–77.

Hong, L., and S. Page. “Groups of Diverse Problem SolversCan Outperform Groups of High-Ability ProblemSolvers.” Proceedings of the National Academy ofSciences, 101, 2004, 16385–89.

Kahn, L., and M. Shah. “Race, Compensation and ContractLength in the NBA: 2001–2.” Industrial Relations, 44,2005, 444–62.

Kurtulus, F. “What Types of Diversity Benefit Workers?Empirical Evidence on the Effects of Co-Worker Dis-similarity on the Performance of Employees.” Indus-trial Relations, forthcoming.

Luttmer, E. “Group Loyalty and the Taste of Redistri-bution.” Journal of Political Economy, 109, 2001,500–28.

Martinez-Vazquez, J., M. Rider, and M. B. Walker. “Raceand the Structure of School Districts in the UnitedStates.” Journal of Urban Economics, 41(2), 1997,281–300.

Page, S. The Difference: How the Power of Diversity CreatesBetter Groups, Firms, Schools, and Societies. Prince-ton, NJ: Princeton University Press, 2007.

PRICE, LEFGREN & TAPPEN: INTERRACIAL WORKPLACE COOPERATION 9

Parsons, C., S. Johan, Y. Michael, and H. Daniel. “StrikeThree: Umpires’ Demand for Discrimination.” Ameri-can Economic Review, 101, 2011, 1410–35.

Price, J., and J. Wolfers. “Racial Discrimination amongNBA Referees.” Quarterly Journal of Economics, 125,2010, 1859–87.

Stoll, M., S. Raphael, and H. Holzer. “Black Job Applicantsand the Hiring Officer’s Race.” Industrial and LaborRelations Review, 57, 2004, 267–87.

Timmerman, T. “Racial Diversity, Age Diversity, Inter-dependence, and Team Performance.” Small GroupResearch, 31, 2000, 592–606.

SUPPORTING INFORMATION

Additional Supporting Information may be found in theonline version of this article.

Please note: Wiley-Blackwell are not responsible for thecontent or functionality of any supporting materials suppliedby the authors. Any queries (other than missing material)should be directed to the corresponding author for the article.