If you can't pay them, play them: fan preferences and own-race bias in the WNBA.

Harris, Jill ; Berri, David J.

Introduction

NBA studies on fan preferences and the racial composition of teams confirm what Becker (1957) hypothesized 70 years ago: customer and employer discrimination can be observed in the marketplace. Brown, Spiro, and Keenan (1991), Burdekin and Idson (1991), Hoang and Rascher (1991), and Bodvarsson and Partridge (2001) all found teams in the 1980s in markets with larger white populations had a larger number of white players. Burdekin, Hossfeld, and Smith (2005) examined this issue in the 1990s and found this trend persisted; moreover, as the number of white players declined over the period, the returns to white players increased.

This behavior manifests in instances of own-race bias as well. Past studies on bias in sport focused on wages. More recent studies have examined performance. Price and Wolfers (2007) find own-race bias in basketball referees making more foul calls per minute on players of different races. Parsons et al. (2007) find the same in umpires calling more strikes for pitchers of the same race. Schroffel and Magee (2012) point out that implicit own-race bias may be more common in split-second decisions than under circumstances requiring more careful consideration. Although this result has not been universally reproduced (see McCormick and Tollison, 2001) it is interesting and important.

No studies to date, that we are aware of, explore these questions in the WNBA. While the NBA and WNBA have three letters in common, they are somewhat different in organization and behavior. As the WNBA marks its 20th anniversary, an examination of the league--including customer and employer discrimination--may provide insights into the behavior of teams, players, and fans alike.

This paper investigates playing time allocation decisions by WNBA coaches over a five-year period. Prior work by Harris and Berri (2015) established decision makers in the WNBA suffer from the same types of mistakes while drafting players that their NBA counterparts do (i.e., emphasizing scoring over other performance measures). If we assume coaches want to win games, it follows they should play the most productive players--black or white--from the roster. But it is possible, given the prior work on the draft, that minutes are also allocated primarily to those who score the most. (1)

We estimate the impact on playing time of observable differences in player characteristics and performance measures. After controlling for the most reliable indicators of player talent and player characteristics, we find no statistically significant evidence that coaches favor players of their own race. We do offer some evidence of "other-race bias"; non-white players playing for non-white coaches play 4.81 minutes less. These results differ from those conducted in the NBA. However, the absence of own-race bias in this sample might be more telling than a positive result when considering the history of the NBA and development of the WNBA.

As Burdekin et al. (2005) show, the NBA sorted players based on race in the 1990s with the more skilled white players ending up in markets with relatively larger white populations. This did not happen immediately; it took years for players and teams to find each other and establish a racial "equilibrium." Could it be the WNBA is developing along these same lines, but is just a decade or two behind the NBA when it comes to the racial distribution of players and teams? Some background information on the WNBA (as it compares to the NBA) may help partially explain our findings. After this history, what follows is a brief summary of the literature involving race and basketball. Next, we estimate the minutes played per game using a production approach that controls for performance and player characteristics including race of the player and race of the coach. We include an Oaxaca decomposition of the difference in minutes played as a robustness check of our basic model. Results, discussion, and directions for future research conclude the paper.

WNBA Background

Conceived in 1996, the WNBA has experienced the typical growing pains of many relatively young sports leagues. As we saw in the early history of the NBA, teams in the WNBA have come and gone while profits and attendance have expanded and contracted. With respect to the number of teams, the league began with eight teams in 1997, expanded to 16 teams by 2000, before contracting to 12 teams by 2010. Currently the league remains with just the following teams: Atlanta Dream, Chicago Sky, Connecticut Sun, Indiana Fever, Los Angeles Sparks, Minnesota Lynx, New York Liberty, Phoenix Mercury, San Antonio Silver Stars, Seattle Storm, Tulsa Shock, and Washington Mystics.

How financially successful is the WNBA? There is a tendency to compare the league to the NBA. Websites such as Forbes.com report revenue data for the NBA. (2) Other sites report how much revenue is paid to individual players but also the limits on compensation for rookies, veterans, and all players collectively. (3) These websites for the NBA not only indicate the salaries of individual players, but they also make it clear that the NBA pays about 50% of its revenue to its players.

There are some websites that discuss WNBA revenues, but none as complete as those referencing the NBA. Still, we can piece together a picture from information that is available. For example, on the revenue side we can note the league has a $12 million per year contract with ESPN through 2022. (4) In addition, Berri and Krautmann (2013) report that average ticket prices (5) were $15 in 2011 and average per-game attendance in 2014 was 7,531 fans. This attendance mark is more than 10,000 below the NBA but quite consistent with where the NBA was at after it existed for 20 years. (6)

Based on these data, we estimate that league revenue in 2014 was around $35 million (see Table 1). This estimate ignores merchandising, sponsorships, and playoff revenue. So it underestimates the total league revenue.

Although our league revenue is inexact, it is sufficient for us to estimate how little of the league revenue goes to the players. There is no website that reports the salaries paid to each player. We do have the collective bargaining agreement for the WNBA, (7) which indicates that the maximum salary paid to players in 2014 was $107,500. In addition, as Table 2 indicates, the average salary in the WNBA was $75,000. (8) With 154 players in the league in 2014, total payroll in the league was $11.55 million.

Once again, our revenue estimate is likely too low. But given what we know of salaries, it seems like the WNBA is only paying--at most--33% of its revenue to its players. This is far less than the 50% paid to NBA players, but consistent with what NFL players received in the 1950s. (9) It certainly appears that WNBA players are not getting as good a deal as their male counterparts in the NBA. (10)

Despite the difference in player compensation, the leagues have something in common: roughly 80% of the players in the NBA are non-white while 83% of the players in the WNBA are non-white in our sample. In addition, approximately 30% of players in the NBA play for non-white coaches while 25% in the WNBA do. In sum, the racial composition of the WNBA is similar to the NBA; however, due to the league collective bargaining agreement and smaller league revenues, female players lack bargaining power relative to males in the sport.

Female players can compete for a spot on international team rosters to increase earnings. In fact, international play can be quite lucrative for the top players. (11) More playing time gives international scouts better information about a player's potential. Thus, it stands to reason that WNBA players may view more playing time as a substitute for higher wages, other things the same. If fans demand more playing time for whites and white players also want more minutes, coaches may award more minutes than are justified by performance.

Literature on Race and Basketball

Discrimination occurs for a variety of reasons. Becker (1957) identified a "taste for discrimination" on the part of individuals who prefer to associate with or employ members of their own race and are willing to pay a premium to avoid other races. Bodvarsson and Brastow (1999) find the race of a manager affects salaries black and white NBA players earn. Hoang and Rascher (1999) report teams with a black coach have a smaller percentage of white players. (12)

Teams with monopsony power can discriminate and pay players of different races differently. McCormick and Tollison (2001) reveal black players may have more inelastic labor supply elasticities than white players. When this is true, NBA teams can pay black players lower wages in the pursuit of higher profits. Kahn and Shah (2005) support the McCormick and Tollison story. Within groups where teams have monopsony power (i.e., players who are not free agents and not on the rookie salary scale) nonwhite players have lower salaries, compensation, and contract duration than white players. This result did not hold for older free agents. In work closely related to racial bias, Berri, Deutscher, and Galletti (2015) discovered US-born players receive more time on the court in the NBA and in the Liga ACB after controlling for performance factors. (13)

Racial bias may also emanate from the fan base. Kanazawa and Funk (2001) find Nielsen ratings for NBA games increase when white players have more time on screen. Kahn and Sherer (1998) and Hamilton (1997) show home attendance is higher if a team has more white players. In contrast, Brown, Spiro, and Keenan (1999) found the percentage of minutes played by black players did not impact home attendance. Since other work suggested more white players on a team was correlated with higher attendance, the authors suggest biased fans may settle for seeing white players on the bench. This story is retold in work by Hoang and Rascher (1999), who found a larger white population in a metropolitan area was correlated with a larger percentage of white players on a team during the 1980s. Burdekin, Hossfeld, and Smith (2005) showed this type of effect for the 1990s. Teams in mostly white metropolitan areas had more starters who were white.

Given all this, Kahn and Shah (2005) report evidence for fan discrimination and for pay discrimination based on race diminished after the 1980s. As Berri (2006) summarized, the evidence is mixed when it comes to wage discrimination in professional sports. Due to the nature of the WNBA's collective bargaining agreement, there is very little salary variability and we do not have access to data on most players' pay. Therefore, the question of racial bias must be examined with playing time on the court.

Several papers investigate the relationship between player quality and other characteristics and minutes played. Staw and Hoang (1995) conclude a player's average minutes per game during the season increase if he is drafted higher, regardless of performance. They suggest this could be due to irrational attention to sunk costs. Camerer and Weber (1999) agree with Staw and Hoang (1995) that there seems to be some evidence of this "escalation of commitment effect" in the first three years of a player's career. McCormack and Tollison (2001) examine whether the race of the coach impacts the playing time of players of the same race. They conclude black players average more minutes per game than non-blacks. However, they report no significant difference between black and white coaches for how race affects minutes played. Leeds, Leeds, and Motomura (2015) also examine escalation of commitment and find teams do not award more playing time to highly drafted players. Schroffel and Magee (2012) find NBA coaches award more minutes per game to players of their own race even after controlling for player quality using performance statistics and player fixed effects. This player-and-coach match has weaker impacts on starting rosters. They emphasize the racial bias appears to be stronger when coaches are making decisions under pressure. To date no research (we are aware of) has examined race and minutes played for the WNBA.

Data and Empirical Model of Minutes Played

This study of minutes played per game uses player statistics from Basketball-Reference.com for WNBA teams from the 2010 to 2014 seasons. Photos of players from WNBA.com were cross-referenced with player and coach biographic sketches to categorize the players or coaches as "white" or "non-white." Although this is potentially an overly simplistic method, it is not without precedent. (14)

Summary statistics for our data are detailed in Table 3. The sample contains 513 observations on 29 variables including player characteristics and performance measures, which are used to estimate our model. A complete list of variable names and explanations is detailed in the appendix.

The average player in our sample is 6 feet tall; the tallest is 6 feet, 8 inches and the smallest is 5 feet, 4 inches. The average age is 27 years old. It appears white and nonwhite players average just more than 20 minutes per game. This is interesting if only because key variables found to be very influential on wins are significantly different between white and non-white players. These differences are included in a separate table (Table 4) with the sample means and significance levels for white and non-white players. (15) Again, all players' performance statistics are per 40 minutes played and adjusted for position as in Berri et al. (2015). (16) These adjustments can result in negative minimum values. In addition to performance measures we include player height (adjusted for position), games played, games started, age and age squared, and dummies for non-white players, non-white coaches, season, and draft position experience. (17)

Non-white players score more points, have an advantage in rebounds and steals, and do not turn the ball over as often. White players have more assists and fewer personal fouls. Prior research suggests decision makers give too much weight to scoring. For this reason, and because rebounding and steals are higher in this sample, we might expect non-white players to average more minutes per game over the sample period. The summary statistics hint at our key finding: non-white players do not

experience preferential treatment in playing time from coaches of their own race.

The empirical model is:

[MGM.sub.nt] = [[beta].sub.0] + [alpha]PRODUCTIVITYnt + [[beta].sub.1][RELHT.sub.n] + [[beta].sub.2] GAMESnt + [[beta].sub.3] GAMESSTARTEDnt + [[beta].sub.4] AGEnt + [[beta].sub.5] AGESQ + [[beta].sub.6]DRAFTEXPnt + [[beta].sub.7] NONWHITEPLAYERnt + [[beta].sub.8]NONWHITECOACHnt + [[beta].sub.9] [NONWHITEPLAYER.sub.nt] *NONWHITECOACHnt + [[beta].sub.10] SEASON + [[beta].sub.11] [FBIASW.sub.nt] + [e.sub.nt] (1)

where minutes played per game (MGM) for player n in time period t is a function of: PRODUCTIVITY = vector of player statistics including Points, Adjusted Field Goal Percentage, Rebounds, Turnover, Steals, Assists, Blocks, and Personal Fouls; DRAFT-EXP = vector of variables interacting Draft Position and Experience; NON-WHITEPLAYER and NONWHITECOACH = vector of variables either 1 for any player who is non-white and 0 otherwise and 1 for any coach who is non-white and 0 otherwise; SEASON = vector of dummy variables controlling for the seasons 2010-2014; FBIASW = vector of NONWHITEPLAYER interacted with a percent white population variable for each team's SMSA; and an error term.

The expected signs on the productivity measures are positive for points, field goal percentage, rebounds, steals, assists, and blocks. Turnovers and personal fouls should have negative signs on their coefficients. We expect minutes played to increase with the height of the player. As the saying goes, "You can't coach height." Generally, we expect the same relationship with age. A squared term is included for age to control for the diminishing impact of age over time. Games can serve as a proxy for injury, while games started signals the relative strength of a player. Therefore, as games played increase we might expect a negative impact on minutes played per game, while games started should be positively correlated with MGM.

Draft position reflects expected talent; research by Leeds et al. (2015), Staw and Hoang (1995), and Camerer and Weber (1999) shows decision makers often place too much emphasis on draft position and sunk costs. (18) That is, they likely use it to determine minutes played even after controlling for performance. By including 10 years of experience we should be able to see if and when coaches stop considering draft position when allocating time to players. (19) If bias against non-white players exists the sign on the NONWHITE coefficient should be negative; likewise, if there is own-race bias in the minutes played per game the sign on the interactive term (NONWHITEPLAYER*NONWHITECOACH) should be positive. Another interaction term captures the impact of coaching changes midseason where the race of the coach changes. We follow Schroffel and Magee (2012) by letting NONWCHANGE equal 1 for players whose coach changes from having a different race to having the same race, -1 for players whose coach changes from being of the same race to being of a different race, and 0 for other players. If own-race bias exists and a player experiences a coaching change with the race of the coach changing from different race to same race, the sign on the coefficient should be positive. Finally, if playing time is influenced by fan discrimination the sign on FBIASW should be negative.

Empirical Results

Equation 1 is estimated using both player and team fixed effects estimators and OLS for comparison. Five years of WNBA data from 2010-2014 are employed with 513 total player-year observations. Players with fewer than 10 minutes per game were dropped from our sample. Results are reported in Table 5.

Before discussing the results on own-race bias, the non-performance factors invite comment. Height is not significant in awarding playing time in this sample. Typically both in the NBA and the WNBA height is an advantage. Even after controlling for skill in blocking, scoring, and rebounding, decision makers tend to draft taller players and award playing time to taller players in the NBA. It could be that coaches associate height with confidence, the ability to deflect passes, or intimidate other players. Another interesting result is that games are not important but games started are very important in terms of playing time through the season. If a player starts in one more game she plays 0.34 minutes more per game. Over the course of the season this means almost 13 more minutes of playing time. The negative sign on games played could be due to the cumulative effect of essentially year-round play for many of these female athletes who participate in international leagues. In the interest of avoiding injury (and so they can play in the far more lucrative international season) players may selfselect out of additional minutes or the coaches may be protecting their investments by playing them less.

In most NBA studies, older players get more playing time initially but eventually sit the bench more. Although the coefficients on AGE and AGESQ are of the expected sign in model 1 and model 2, age is not significant in our sample. This might be because WNBA players start older in the league than their NBA counterparts in general. Virtually all players finish a college career in the US or are picked up from international league play. The average age for a rookie WNBA player is 22 and only 1% of our sample is between 18 and 21 years of age. In contrast, in the sample employed by Berri, Deutscher, and Galletti (2015), 9.8% of NBA players were in that age group.

Turning to the performance factors, points, blocks, and personal fouls are all statistically significant and of the expected signs except for blocks. None of the coefficients on draft position and years of experience are significant in any model specification (we do not include them in Table 5). It seems draft position and years of experience do not appear to be important to the coaching decision with respect to playing time. This result has some outside support in the literature (see Leeds et al., 2015). Factors influencing draft position are not always the best predictors of performance in the pro leagues. (20)

Given the performance and personal characteristics described, is there any evidence of employer discrimination? The coefficient on NONWHITE player is positive and significant. Non-white players with white coaches play 4.33 minutes more per game. The coefficient on NONWCOACH tells us that white players get about 5.16 more minutes if they play for non-white coaches. Certainly neither of these results is evidence of ownrace bias. The estimated coefficient on the interaction term is negative and significant at the 5% level. This result indicates that non-white players with non-white coaches play 4.81 fewer minutes, controlling for performance. The "other-race bias" is even more severe for players experiencing a coaching change (with change of race) midseason; non-white players in this group played 5.64 fewer minutes on average. What about fan discrimination? For a 1% increase in the white population in the team SMSA, non-white players receive 6 fewer minutes of playing time. This suggests, other things the same, that fan bias may be partially responsible for the differential in playing time.

All this suggests something quite interesting is happening in the WNBA. Coaches appear to play white players more regardless of performance. In particular, non-white coaches seem to play white players more than their counterparts, controlling for performance. So, we do not have evidence of employer discrimination (or, own-race bias). On the contrary, there is evidence of "other-race bias" when the race of the coach and player differ. In addition, it may be the case that the bias on the part of coaches is being driven, in part, by fan discrimination.

Next, we turn to economic significance. As Table 5 indicates, blocks, fouls, and points are all influential on minutes played per game. Given these predicted results, how meaningful are they? Table 6 reports how an estimated one standard deviation increase in each statistically significant performance variable impacts playing time. Points and fouls matter most: a one standard deviation increase in points translates to 2.01 more minutes of playing time. The penalty for increased personal fouls is a decrease of 1.46 minutes. A player with 2 more minutes per game could score 14 more points over the course of a season.

Such a result is consistent with previous studies of the NBA and WNBA. Scoring totals dominate player evaluation in basketball. Harris and Berri (2015) noted this for the WNBA draft. Now we see this for minutes per game as well. Players clearly have an incentive to look for their shot in both the NBA and WNBA.

Robustness Test

Labor economists often employ an alternative method for investigating the difference between two groups: the Oaxaca-Blinder Decomposition (Oaxaca, 1973; Blinder, 1973). The general idea is that an observed difference in return to performance between two groups can be decomposed into explained and unexplained parts. The unexplained portion of the difference can then be related to bias. Table 7 reports the decomposition for white and non-white players. Non-white players play 1.34 minutes more than white players after controlling for performance. The first part of the difference reflects the mean increase in playing time for non-white players if they had the same characteristics as the white players. This means that 1.17 of the 1.34 additional minutes played (about 65%) can be accounted for by performance and personal characteristics of the players in our sample. The remaining 35% of the differential is not explained by player endowments. This playing time gap could possibly be the result of bias on the part of the coaches. The decomposition supports our empirical findings in two ways. First, it shows the significant difference in playing time afforded to players grouped by race. Second, it reveals that only a portion of this difference can be attributed to differences in player endowments. While we cannot be certain that the unexplained portion of the playing time is the result of discrimination, we can report that something other than player performance and personal characteristics is responsible for the difference.

Concluding Observations

This inquiry into employer discrimination in the WNBA finds no evidence of ownrace bias on the part of coaches in the 2010-2014 seasons. On the contrary, non-white coaches play their non-white players (on average) 4.81 minutes less per game than white players. This differential can be partially explained by differences in player ability. Is this "other-race bias" coming from the fans? The sign and significance of our population variable interacted with the player race variable (FBIASW) suggests some fan discrimination may exist in the WNBA.

Discrimination can reflect tastes and preferences as in Becker (1957) and Kahn (2012), or follow from statistical reasons as in Arrow (1973) or Phelps (1972). Our model's performance variables are likely picking up most of the underlying statistical discrimination in which coaches might engage. As Berri et al. (2015) point out, coaches may consider other factors about their players that researchers cannot measure. Factors they might consider include perceived levels of confidence, deflecting passes, and intimidation. These are unobservable and probably influence the minutes played per game. However, given the breakdown of abilities along racial lines it seems apparent that some coaches--non-white coaches--are taking great pains to not display preferential treatment of non-white players. This could result in fewer wins for their teams over the course of a full season. If this behavior is playing to fan discrimination, it has implications beyond won-loss records. Students of history know history often repeats itself. If it is true that WNBA players and teams are starting to sort themselves along racial lines (as studies indicate the NBA has done), a great deal can be learned about the institutional effects of this behavior in the coming years. One additional path for future research could be to investigate the role of sex in the playing time decision. Half of the WNBA coaches in the 2010-2014 seasons were male. Do these coaches make playing time decisions differently than their female counterparts? Finally, in a world where race, bias, and performance of job-related duties is constantly in the news, it seems clear we can still learn much from the behavior of decision makers in sports.

References

Arrow, K. J. (1973). The theory of discrimination. In O. Ashenfelter & A. Rees (Eds.), Discrimination in labor markets (pp. 3-33). Princeton, NJ: Princeton University Press.

Becker, G. S. (1957). The economics of discrimination. Chicago, IL: University of Chicago Press.

Berri, D. J. (2005). Economics and the National Basketball Association: Surveying the literature at the tip-off. In J. Fizel (Ed.), The handbook of sports economics research (pp. 21-48). London, UK: M.E. Sharpe, Inc.

Berri, D. J. (2010). Measuring performance in the National Basketball Association. Working paper.

Berri, D. J., & Krautmann, A. C. (2013). Understanding the WNBA on and off the court. In M. Leeds & E. Samkova Leeds (Eds.), Handbook on the economics of women in sports (pp. 132-155). New York, NY: Springer.

Berri, D. J., Brook, S., & Fenn, A. (2011). From college to the pros: Predicting the NBA amateur player draft. Journal of Productivity Analysis, 35, 25-35.

Berri, D. J., Brook, S., Fenn, A., Frick, B., & Vicente-Mayoral, R. (2005). The short supply of tall people: Explaining competitive imbalance in the National Basketball Association. Journal of Economic Issues, 39, 1029-1041.

Berri, D. J., Deutscher, C., & Galletti, A. (2015). Born in the USA: National origin effects on time allocation in US and Spanish professional basketball. National Institute Economic Review, 232, R41-50.

Berri, D. J., & Schmidt, M. B. (2010). Stumbling on wins: Two economists explore the pitfalls on the road to victory in professional sports. Princeton, NJ: Financial Times Press.

Berri, D. J., & Simmons, R. (2009). Catching a draft: On the process of selecting quarterbacks in the National Football League amateur draft. Journal of Productivity Analysis, 35, 37-49.

Berri, D. J., Schmidt, M. B., & Brook, S. L. (2006). The wages of wins: Taking measure of the many myths in modern sport. Palo Alto, CA: Stanford University Press.

Berri, D. J., Brook, S. L., & Schmidt, M. B. (2007). Does one simply need to score to score? International Journal of Sport Finance, 2, 190-205.

Blinder, A. S. (1973). Wage discrimination: Reduced form and structural estimates. The Journal of Human Resources, 8, 436-455.

Bodvarsson, O. B., & Brastow, R. T. (1999). A test of employer discrimination in the NBA. Contemporary Economic Policy, 17, 243-255.

Bodvarsson, O. B., & Partridge, M. D. (2001). A supply and demand model of co-worker, employer, and customer discrimination. Labour Economics, 8, 389-416.

Brown, E., Spiro, R., & Keenan, D. (1991). Wage and nonwage discrimination in professional basketball: Do fans affect it? American Journal of Economics and Sociology, 50, 333-345.

Burdekin, R. C. K., Hossfeld, R., & Smith, J. K. (2005). Are NBA fans becoming indifferent to race? Journal of Sports Economics, 6, 144-159.

Burdekin, R. C., & Idson, T. L. (1991). Customer preferences, attendance and the racial structure of professional basketball teams. Applied Economics, 23, 179-186.

Camerer, D. F., & Weber, R. A. (1999). The econometrics and behavioral economics of escaltion of commitment: A re-examination of Staw and Hoang's NBA data. Journal of Economic Behavior and Organization, 39, 59-82.

Coates D., & Oguntimein, B. (2010). The length and success of NBA careers: Does college production predict professional outcomes? International Journal of Sport Finance, 5, 4-26.

Gourieroux, C., Monfort, A., & Trognon, A. (1984). Pseudo-maximum likelihood methods: Applications to Poisson models. Econmetrica, 52, 701-720.

Harris, J., & Berri, D. J. (2015). Predicting the WNBA Draft: What matters most? International Journal of Sport Finance, 10, 299-309.

Hollinger, J. (2003). Pro basketball prospectus: 2003-2004. Washington, DC: Brassey's Inc.

Humphreys, B. (2000). Equal pay on the hardwood: The earnings gap between NCAA Division I basketball coaches. Journal of Sports Economics, 1, 299-307.

Kahn, L. M., & Sherer, P. D. (1988). Racial differences in professional basketball players' compensation. Journal of Labor Economics, 6, 40-61.

Leeds, D., Leeds, M., & Motomura, A. (2015). Are sunk costs irrelevant? Evidence from playing time in the National Basketball Association. Economic Inquiry 53, 1305-1316.

Marlowe, C. M., Schneider, S. L., & Nelson, C. E. (1996). Gender and attractiveness biases in hiring decisions: Are more experienced managers less biased? Journal of Applied Psychology, 81, 11-21.

Massey, C., & Thaler, R. H. (2006). The loser's curse: Overconfidence vs. market efficiency in the National Football League Draft. NBER Working Paper, (W11270).

McCormick, R. E., & Tollison, R. D. (2001). Why do black basketball players work more for less money? Journal of Economic Behavior and Organization, 44, 201-219.

Oaxaca, R. (1973). Male-female wage differentials in urban labor markets. Internationl Economic Review, 14, 693-709.

Oliver, D. (2004). Basketball on paper. Washington, DC: Brassey's Inc.

Phelps, E. S. (1972). The statistical theory of racism and sexism. The American Economic Review, 64, 659-661.

Price, J., & Wolfers, J. (2010). Racial discrimination among NBA referees. The Quarterly Journal of Economics, 125, 1859-1887.

Robst, J., Van Gilder, J., Coates, C., & Berri, D. J. (2011). Skin tone and wages: Evidence from NBA free agents. Journal of Sports Economics, 12, 143-156.

Schroeffel, J. L., & Magee, C. S. (2012). Own-race bias among NBA coaches. Journal of Sports Economics, 13, 130-151.

Staw, M. M., & Hoang, H. (1995). Sunk costs in the NBA: Why draft order affects playing time and survival in professional basketball. Administrative Science Quarterly, 40, 474-494.

Endnotes

(1) Winning may not be most important to some coaches; profit or some other variable may be the focus for the coach. However, in the WNBA, losing coaches are quickly dispatched. Given the league's relatively "poor" status compared to the NBA, win maximization seems to be a reasonable behavioral assumption.

(2) "The Business of Basketball," Forbes. Retrieved from http://www.forbes.com/nba-valuations/

(3) Larry Coon's website details the NBA's collective bargaining (http://www.cbafaq.com/ salarycap.htm). In addition, various websites report salaries for individual players (see http://espn.go.com/nba/salaries; http://hoopshype.com/salaries/; http://www.basketball-reference.com/contracts/).

(4) Details on this contract were reported by http://www.sportsbusinessdaily.com/Daily/ Issues/2013/03/28/Media/WNBA.aspx. We should also note that viewership on ESPN2 averaged 240,000 over 19 games (up from the previous year) and views on NBA TV increased over this time period as well (53,000 over 34 games).

(5) Over the 2012-2013 seasons it appears average ticket prices increased by several dollars. We do not have average ticket prices for 2014, so the revenue estimates we provide are likely too low. See http://www.sportsbusinessdaily.com/Daily/Issues/2014/08/19/Research-and-Ratings/ WNBA-gate.aspx

(6) See http://www.apbr.org/attendance.html. The NBA did not average more than 7,500 fans per game until the 1969-70 season, or more than 20 years after it began operations in the late 1940s.

(7) See http://wnbpa-uploads.s3.amazonaws.com/docs/WNBA%20CBA%202014-2021Final.pdf

(8) This was reported in the Dallas Morning News. Retrieved from http://www.pressreader.com/ usa/the-dallas-morning-news/20150726/282364038377418/TextView

(9) Stefan Szymanski reports that NFL players received 32% of league revenue in the 1950s while Major League Baseball players only received 17%. Soccernomics. Retrieved from http://www.soccernomics- agency.com/?p=639; last updated October 10, 2014.

(10) The WNBA reports that only six of its teams are profitable. It is important to note, though, that the NBA also claimed as recently as 2011 that many of its teams were not profitable as well. In fact, owners consistently argue in professional sports that teams are not profitable. See http://www.huffingtonpost.com/david-berri/think-the-wnba-is-in-trouble-lets-talk-nba-history_b_10279354.html

(11) Diana Taurasi was paid $1 million by her Russian team to sit out the 2014-2015 season with the Phoenix Mercury, for example.

(12) The authors find the difference between teams with white coaches and teams with black coaches is not statistically significant.

(13) The Liga ACB is the Spanish professional men's basketball league.

(14) See Robst et al. (2011) for details of their study using skin tone to explore race and wages.

(15) When adjusting for position played the average for the players in that position is subtracted first, then the average for all players is added back in. Therefore it is possible to have a negative minimum value in these performance measures after they are position adjusted.

(16) Position bias is overcome by calculating a position adjusted value for each metric. Each player's per-minute performance with respect to points, rebounds, steals, blocked shots, assists, and turnovers is determined. Then, the average per-minute accumulation at each position in our data set is subtracted. The average value of the statistic across all positions is added back in. After these steps, the result is multiplied by 40 minutes (the length of a college game), to return the player's per 40 minutes production of each statistic.

(17) Games played and games started are both correlated with minutes played per game. Games are included as a proxy for injury. Games started are included as a signal about the relative strength of a player on the team roster. We do not include players with less than 10 minutes played per game. This cut-off (theoretically) should eliminate so-called "mop-up" players from our sample.

(18) These studies conclude that decision makers--even after controlling for performance--overemphasize draft position and scoring while making playing time decisions. In other words, in spite of evidence to the contrary they will play a player more often than she might deserve based on observed performance variables. The exception is Leeds et al. (2015). They find teams do not award more playing time to highly drafted players.

(19) As draft position is "worse" the higher the number, the expected sign on the draft experience variables is negative.

(20) See Berri, Brook, and Fenn (2011) and Harris and Berri (2015) for examples. A referee suggested taking the logs of draft position to better fit the data. We ran the model with logged draft variables; there was no change in results. Appendix Variable Description Variable Label Height (relative) HT Age AGE Games G Games Started GS Free Throw Percentage FTPER Rebounds REB Assists AST Steals STL Blocks BLK Turnover Percentage TOPER Fouls PF Points PTS Field Goal Percentage ADJFG (adjusted) Non-white player NONWHITE PLAYER Non-white coach NONWCOACH Coaching Change NONWCHANGE Draft position DRAFT Minutes per game MGM Height Squared HTSQ Age squared AGESQ Draft Experience DFT1 Home Attendance HOMEATT Percent Team White PCTTEAMWHT All-Stars STARS Stadium STAD Population POP Population white POPW Fan Bias FBIASW Variable Description Description Height (relative) Height--avg height for position + avg overall Age Age of player Games Games played Games Started Games started Free Throw Percentage Free throws made/free throws attempted Rebounds Total offensive + defensive re-bounds Assists Total assists Steals Total steals Blocks Total blocks Turnover Percentage Turnovers/(Turnovers+Field Goal at-tempts + 0.44Free Throw attempts)*100 Fouls Personal Fouls Points Total points Field Goal Percentage [(Points-Free Throws made)/ (adjusted) Field Goal Attempts) ] Non-white player Dummy variable = 1 if non-white player Non-white coach Dummy variable = 1 if non-white coach Coaching Change Dummy = 1 if midseason coaching change with change to same race, =-1 if change to different race Draft position Position drafted 1-36 Minutes per game Minutes played/games played Height Squared Height squared Age squared Age squared Draft Experience Dummy variable = 1 if one year experi-ence*DRAFT Home Attendance Average attendance per season/ arena capacity Percent Team White White players on roster /total players on roster All-Stars The number of All-Stars on team Stadium The stadium or arena capacity Population The population of a team SMSA Population white The percentage white population in a team SMSA Fan Bias Interaction term = POPW*NONWHITEPLAYER

Jill Harris [1] and David J. Berri [2]

[1] United States Air Force Academy

[2] Southern Utah University

Jill Harris, PhD, is an assistant professor of economics in the Department of Economics and Geosciences. Her research interests include the nature and behavior of the National Collegiate Athletic Association, women in sport, and non-compliance behavior.

David J. Berri, PhD, is a professor of economics in the Department of Economics and Finance. His research focuses on the economics of sport, specifically the topics of consumer demand, competitive balance, worker productivity, and women in sport. Table 1. WNBA League Revenue Estimate in 2014 Revenue Factors Revenue Television Revenue $12,000,000 Average Attendance 7,578 Average Ticket Price $15 Gate Revenue per game $113,670 Total Gate Revenue for 204 $23,188,680 Regular Season Games TOTAL REVENUE $35,188,680 Table 2. WNBA Payroll Costs in 2014 Salary Factors Average Salary $75,000 Number of Players 154 Total Payroll $11,550,000 Payroll as Percent of Revenue 32.8% Table 3. Descriptive Statistics for Dependent and Independent Variables (2010-2014) Variable Obs. Mean Std. Dev. Min Max HT 513 72.355 2.149 64 80 AGE 513 27.634 4.290 20 42 G 513 28.593 7.657 2 34 GS 513 15.632 13.587 0 34 FTPER 512 0.756 0.125 0 1 REBOUNDS 513 6.535 1.777 0.466 13.627 ASSISTS 513 3.281 1.140 -0.363 8.094 STEALS 513 1.554 0.640 -0.093 4.396 BLOCKS 513 0.747 0.570 -0.884 5.371 TOPER 513 17.843 6.464 0 50.911 FOULS 513 3.956 1.342 1.222 9.671 POINTS 513 14.261 4.876 -0.699 30.712 ADJFG 513 0.461 0.080 0 0.739 NONWpl 512 0.828 0.378 0 1 NONWco 513 0.250 0.433 0 1 DRAFT 432 7.729 8.711 0 36 MGM 513 20.815 7.854 2 35 NONWPLxNONWCO 512 0.205 0.404 0 1 FBIASW 513 0.502 0.236 0 0.816 DFT1 513 7.646 8.612 0 36 DFT2 513 6.938 8.563 0 36 DFT3 513 6.007 8.300 0 36 DFT4 513 5.093 7.847 0 35 DFT5 513 4.403 7.622 0 35 DFT6 513 3.685 7.271 0 35 DFT7 513 2.840 6.527 0 35 DFT8 513 2.019 5.484 0 35 DFT9 513 1.556 5.030 0 36 DFT10 513 1.164 4.177 0 35 Table 4. Average Performance Statistics by Race and Test of Significant Difference Non-White White Players Points 14.440 13.387 Adj. Field Goal % 0.456 0.485 Free Throw % 0.753 0.774 Rebounds 6.618 6.091 Turnover % 17.423 19.916 Steals 1.584 1.412 Assists 3.189 3.743 Blocked Shots 0.750 0.731 Personal Fouls 3.99 3.767 Minutes per game 20.87 20.541 Statistical Difference Points [not equal to] at 5% sig Adj. Field Goal % [not equal to] at 5% sig Free Throw % cannot reject Ho Rebounds [not equal to] at 5% sig Turnover % [not equal to] at 5% sig Steals [not equal to] at 5% sig Assists [not equal to] at 5% sig Blocked Shots cannot reject Ho Personal Fouls [not equal to] at 5% sig Minutes per game cannot reject Ho Note: All statistics are position adjusted. (Ho = no difference in means) Table 5. Estimation of Equation (1) Dependent Variable = MGM (minutes played per game) Variable Model 1 Model 2 Model 3 PTS 0.412 ** 0.403 *** 0.402 *** REB 0.060 0.037 0.220 * AST 0.040 0.067 0.317 * STL -0.269 -0.277 0.287 BLK -1.063 ** -0.718 * -0.741 * PF -1.137 *** -1.120 *** -0.867 *** TOPER 0.077 0.86 -0.018 FTPER 0.895 1.076 3.497 * ADJFG 7.735 2.207 *** 3.088 HT 5.084 4.888 1.827 AGE 0.244 0.211 -0.219 AGESQ -0.007 -0.006 0.004 G -0.046 -0.042 0.003 GS 0.339 *** 0.338 *** 0.374 *** NONW PLAYER 4.336 * 4.076 * 0.126 NONW COACH 5.159 *** 4.534 ** 1.915 * NONWPL *NONWCO -4.814 ** -4.842 ** -0.626 FBIASW -6.600 * -6.587 * -0.063 NONWCHANGE -5.644 *** -5.340 ** -6.377 ** PLAYER FE YES YES NO TEAM FE NO YES YES R-sq 0.721 R-sq 0.720 R-sq 0.765 Notes: Models use data from 2010-2014; N=503 on 154 players. Model 1 uses player fixed effects only, Model 2 uses both player and team fixed effects, and Model 3 is an OLS model. In the sample, 56 players experienced coaching changes involving a change in the race of the coach between years, while 40 players experienced a coaching change midyear where the race of the coach changed from non-white to white or white to non-white. The midyear changes are captured by the NONWCHANGE variable. * denotes significance at 10%, ** 5%, and *** 1%. Table 6. The Impact of a One Standard Deviation Increase in Statistically Significant Performance Variables (2010-2014) Variable # min per game player gains POINTS 2.01 BLOCKS -0.42 PERSONAL FOULS -1.46 Note: N=503. If a player has a one standard deviation increase in performance, on average, minutes per game played will be impacted by the levels reported above. Scoring points and avoiding fouls are the most important performance actions for players in the sample. Table 7. Oaxaca-Blinder Decomposition Results WNBA 2010-2014 Seasons Independent Var. Coefficient Standard Error z-statistic White 19.93 *** 1.103 18.08 *** Non-White 21.27 *** 0.461 46.18 *** Difference -1.336 1.195 -1.12 Explained -0.872 1.638 -0.53 Unexplained -0.464 0.542 -0.26 Obs. White 88 1.369 -0.34 Obs. Non-White 415 Note: *** significant at 1% level