Printer Friendly

A Market Test for Ethnic Discrimination in Major League Soccer.

Introduction

Major League Soccer (MLS) has a failed history of targeting the Hispanic sports market in the United States. Although soccer is the main sport of choice for most Hispanics, Jewell and Molina (2005) finds that in the early days of MLS, attendance was significantly lower in cities that had a larger percent of Hispanic residents. The league had two teams in Hispanic markets fold in Florida and a last-minute rebrand of the Houston expansion team to appease protesting Mexican-American residents of the city (1) (Fallas, 2006). More important to this study, in 2005 the league enabled an expansion team to discriminate in its roster selection in the hopes of cashing in on the large number of Hispanic soccer fans residing in Los Angeles. To quantify the negative effects of such discrimination in Major League Soccer, this paper uses a market model similar to Mongeon (2015) and Szymanski (2000).

Previous studies of MLS's labor market include salary determination models, those that focus on "designated players," and a couple that attempt to uncover owner or fan discrimination for nationality. While Kuethe and Motamed (2010) found a wage premium for South American players, Wooten (2013) showed with updated data that the premium only existed for South American players in the top of the earnings distribution. The group as a whole was estimated to actually earn lower wages, all else being equal. The author suggested this may be due to the labor market correcting itself from previously paid wage premiums. Estimating a GLS model of salary determination, Celik and Ince-Yenilmez (2017) finds the birthplace of players to be the most important factor, with Europeans being paid significantly more and Caribbean players significantly less. Using the first year of publicly available salary data, Reilly and Witt (2007) finds no overall evidence of salary differentials across race but a premium for international players. (2) Note that these results do not necessarily imply discrimination as controls for talent in soccer (goals, assists) are less than ideal. Available measures of skill in soccer are generally dependent on the quality of a player's teammates and are all but non-existant for defenders. Therefore, coefficients on nationality or ethnic indicators in salary regressions likely containbias from omittedskill variables. The main difference in the approach taken here is that a market model is used to identify discrimination, relieving the need of quality performance statistics to measure the skill level of players.

The main identifying assumption of the market model estimate of discrimination is that each non-discriminating team pays a market price for talent regardless of ethnicity. Any deviation in performance that is explained by the ethnic makeup of a team while controlling for player salary expenditures is assumed to be due to the ethnic tastes of the teams, owners, coaches, or fans (Szymanski, 2000). The argument here is that Chivas USA and, to a lesser extent, other teams in MLS paid a premium for Hispanic talent because they valued the ethnicity of players, which is a non-productive trait. As such, the amount of talent they employed is less than what player salary expenditures would indicate, causing the team to perform worse compared to those with similar expenditures. The benefit of using this approach is that the econometrician does not need to specify or even measure what makes a player more valuable, he simply needs to know how much each player was paid.

In the market for soccer talent, there are thousands of professional teams competing globally over millions of players. Therefore, it is reasonable to assume that any MLS team must pay at least the market price for international and domestic talent with competing interests abroad. Although MLS has a structure unique among soccer leagues and places some restrictions on team rosters, neither of these factors are expected to cause a significant bias in the price paid for talent and both are discussed in the next section.

Mongeon (2015) uses a market model to test for ethnic discrimination in the National Hockey League (NHL) using game-level data but is unable to compare the results to a model using season level-data since the author only uses a single season of data. As noted by the author, aggregating the performance of teams over a season implicitly assumes that all teams have identical quality in terms of opponents. This assumption certainly does not hold for MLS teams, who are grouped into Eastern and Western conferences and play mostly intraconference games. Earning, for example, 50% of all possible points could therefore imply different levels of performance across teams depending on which conference the teams are in.

Employing a single measure of team payroll for each season provides a noisy measure of the relationship between payroll and performance when rosters are dynamic within season, as they are in MLS (Kahane, 2005). Some of the most expensive player acquisitions occur during the summer when other leagues around the world are in their offseasons. If one were to use roster and salary information from the beginning of the season as is often done in the literature, team payrolls would be greatly underestimated. For example, when David Beckham joined the LA Galaxy in the summer of 2007, his annual salary was reported to be $6.5 million according to the MLS Players' Union. The total annual guaranteed salary of the remainder of the team was approximately S2.8 million. The top left histogram in Figure 1 displays the distribution of total salary for fielded players across games in the 2007 season for the LA Galaxy. Clearly, using beginning of the season roster and salary measurements in this case would greatly underestimate the roster spending of the LA Galaxy and skew any estimated relationship between spending and performance. To demonstrate that this is not unique to the 2007 LA Galaxy, Figure 1 also contains three additional histograms displaying large variations in salary spending for New York Red Bulls, Toronto FC, and Seattle Sounders FC. Since I have 10 seasons of game-level dataavailable here, acomparison between season-level and game-level estimates is possible.

Chivas USA

In 2005, Club Deportivo Chivas USA entered MLS as the second team in the Los Angeles metropolitan area, sharing a stadium with the LA Galaxy in Carson, California. Chivas USA's name was derived from the nickname of its parent club, Club Deportivo (CD) Guadalajara, who are commonly referred to as Chivas ("goats" in Spanish) and the team's logo (3) and colors were identical to that of CD Guadalajara. The marketing angle was to attract the large portion of mostly Mexican and Mexican-American soccer fans in the Los Angeles metropolitan area with an on-field product that was an extension of the most popular team in Mexico.

CD Guadalajara is famous for discriminating in its player selection by only fielding Mexican players. The ownership desired a similar type of discrimination and made it public when one owner, Jorge Vergara, told the Los Angeles Times, "It's the Latins versus the gringos. And we're going to win.... As someone in the ministry in Mexico put it, 'We're taking the U.S. back, little by little'" (Martinez, 2005), used slogans such as "Adios Soccer, El Futbol Esta Aqui" (Goodbye Soccer, Football is here), and made statements about how Chivas USA would "teach the gringos how to play soccer" (Witz, 2014).

Rather than inform the owners of Chivas USA that employment discrimination is illegal in the United States, the league enabled the behavior by increasing the number of international players each team is allowed. Before Chivas USA played their first game, the number of senior international players allowed per team increased from 3 to 4 and teams were also granted up to three tradeable roster slots for youth international players who were 24 or younger. Additionally, Chivas USA and fellow 2005 expansion team Real Salt Lake were granted 2 additional youth international slots for their first 2 years of existence. As such, it would have been possible for Chivas USA to field a team of 9 Mexican players and only use 2 players from the United States without trading for additional youth international slots, which the team did. Any legal enforcement of employment discrimination laws for the players of a single team in MLS is complicated, as all player contracts are held by the league at large and not the individual teams. However, the team was sued for employment discriminationby an administrator (Craig v. Chivas USA Soccer LLC, 2013) and two youth coaches (Calichman v. Chivas USA Soccer LLC, 2013) shortly before being purchased back by the league.

Other Teams

Identification of discrimination may be weak using MLS data if Chivas USA were the only team in the league to exhibit ethnic preferences. However, the team is not alone in employing Hispanic players for reasons other than their skill. When the San Jose Earthquakes relocated to Houston, the team had one of the least Hispanic rosters ever fielded in MLS. In their third year in Houston, famousboxer Oscar De La Hoyabecame part owner of the team andexpressed similar preferences to those of Chivas USA,
Right now the majority of our players are Caucasian. We are looking at
different options to bring in the best players from Mexico because of
Houston having such a large Latino population (Rogers, 2008).


Although the shift in roster ethnicity was nowhere near as extreme as Chivas USA, there is evidence presented below that Houston also sacrificed performance for its preferences in roster ethnicity. A similar marketing approach to Houston's was taken early on by FC Dallas (4) who became the first professional sports team in the city to "create a separate marketing department devoted exclusively to Hispanics" (Parker, 1999) and the Chicago Fire who signed aging Mexican National team player Cuauhtemoc Blanco, partly to woo Mexican soccer fans to the stadium (Rivera, 2017). Thus, the data set covering this time period spans the continuum of ethnic preferences. The variation in roster ethnicity displayed in Figure 2 from Houston and Chivas USA alone covers almost the entire range for the league, which is useful for identifying the effect of discrimination.

The remainder of this paper is organized as follows. The next section discusses the unique aspects of the labor market in MLS that must be considered. The data is then described and summary statistics are placed within the context of the time period. The market model tests for discrimination are then presented followed by the results. Finally, a brief discussion is offered.

The MLS Labor Market

The organization of MLS is unique among soccer leagues around the world. Rather than each team operating as a separate business similar to other leagues, all teams in MLS are owned by the league, which is often referred to as a "single-entity." Those commonly referred to as team owners are, in reality, investor-operators. That is, they are each shareholders of the league and are given exclusive rights to operate and reap financial benefits from a particular team. (5) Although coaches and scouts hired by investor-operators identify and pursue players for hire, player contracts are held by the league, which is also involved in the negotiations. One of the main reasons for this structure is to prevent teams from independently engaging in bidding wars over players, something that is largely believed to have contributed to the demise of the league's predecessor, the North American Soccer League (Francis and Zheng, 2010).

The single-entity structure of MLS may seem to create a problem here as the identifying assumption of the empirical model is that the amount of talent employed by each team is perfectly represented by player salaries, which are assumed to be set by an open market. However, while the league maybe successful in preventing bidding wars amongst its own teams over individual players, it has no control over the bids of teams outside of MLS, of which there are thousands who employ players of a similar skill range. Even if the league attempts to use its structure to lower player salaries, it is dwarfed in size by the sheer number of global competitors and practically has no market power to do so. So while MLS as a single-entity may have market power in the media and product markets as described in Kesenne (2015), there should be no expectation that it has power in the player market. (6)

The courts agreed with this assessment when the players sued the league claiming antitrust violations as "[i]t found that players had failed to prove...that the relevant geographic market is the United States and that the relevant product market is limited to Division I professional soccer players" (Fraser v. Major League Soccer, 2002). During the proceedings, league representatives claimed that their players had come from and transferred to 67 different leagues. Given that a typical league has around 20 teams, this conservatively puts the potential competition for MLS players at that time around 1,340 teams if we only count each country's top division. Therefore, paying anything less than the market price for talent is not sustainable as all players have a plethora of outside options. Lastly, previous literature has also used team wage bills in MLS as a measure of team productivity (Coates et al., 2016).

Similar to other North American sport leagues, MLS does control total spending on player salaries by employing a salary cap, which is usually binding for all teams. The cap is typically increased every year but has historically been relatively meager compared to the spending of top tier teams in Europe. A typical player in the English Premier League earned an annual salary on par with what an entire team in MLS would spend on its roster during the time period under study (between $1.9 and $3.9 million). Note that this does not imply that MLS was able to underpay talent, but rather that it employed less talent than the English Premier League.

Designated Players

Beginning in 2007, MLS allowed each team to sign up to 2 players whose salaries would not fully count against the salary cap. These players would be referred to as "designated players" or DPs, and the rights to sign a DP could be traded to other teams. The rule was commonly referred to as the "Beckham Rule" as it was created to allow the LA Galaxy to sign David Beckham for $6.5 million a year when the salary cap was only $2.1 million for their entire roster. (7)

A third DP roster slot was added in 2010 along with a tax that had to be paid to the league for its use. All taxes collected in this manner are split amongst the remaining teams in the league who do not use all three DP slots. These funds are referred to as "allocation money" and discussed in the appendix. In 2012, two additional DP categories were created, one for players between the ages of 21 and 23, and one for players under the age of 21. Only $200,000 and $150,000 of these "young DP" salaries count against the cap respectively. In a recent paper, Coates et al. (2016) investigate how the salary inequality within teams exacerbated by DP contracts affected team performance. Table 1 is a recreation of that from Coates et al. (2016) with data from the 2014 season and additional DP categories added.

The limitations on a team's ability to sign many high-salary players restricts the amount of talent any team may acquire but does not affect the price paid for talent. The tax on a third DP, however, does inflate the price paid for talent by teams once a third DP is employed. If the global market for talent is competitive, then the only way a team would be willing to pay a tax for a third DP on top of his market salary is if that player had more value to the team than the rest of the market.

This additional value may be from increased revenue generated from hiring a globally recognized name as the LA Galaxy had signing David Beckham or the New York Red Bulls had signing the former captain of team France, Thierry Henry. Additional value may also come from the player connecting with a particular ethnic market, as the signing of Mexican national team player Cuauhtemoc Blanco did for Chicago Fire (Rivera, 2017). Since the tax may represent the payment teams are willing to make above and beyond the market value of the player for reasons other than skill, I include all luxury taxes paid in the payrolls of teams. (8) However, the taxes are small relative to total salary spending of teams that employ three DPs and results are almost identical if they are not included.

Home-Grown/Generation Adidas

Youth players that are developed by an individual team's academy for at least 80 days may be signed directly by that team without entering the league through a draft. Such players may otherwise elect to enter the MLS SuperDraft [sic] whereby they may be drafted by any team before the start of each season. Whether the player is allocated to a senior roster spot or reserve roster spot, he must be paid at least the league minimum for that designation and may earn a maximum salary of the minimum plus a predetermined amount ($125,000 in 2016). Up to two homegrown players may earn above the league minimum and not count against the team's budget.

Although there does exist an opportunity for teams to acquire high levels of talent through their academies while paying belowmarket rates through the homegrown mechanism, this possibility is likely not a concern here. If a high-ability player was offered a low salary by his "home" team, he could always opt for the draft to negotiate a higher salary or sell his services on the open market abroad. Additionally, home grown players typically earn the league minimum and do not earn significant playing time in their first year of play. Nevertheless, this study controls for whether players were acquired in this manner in the following analysis.

The league also has a partnership with Adidas to develop young talent in the United States through the Generation Adidas project (Project 40 sponsored by Nike before 2006). Generation Adidas players sign a contract with the league and are entered into the Super-Draft. Adidas sponsors the project and the league pays the players' salaries, which do not count against their team's salary cap. The small number of players chosen are subjectively determined to be the best available talent that may be considering playing in college. The program provides financial incentive for the players to become professional athletes instead of attending college, thereby making all who sign with Generation Adidas ineligible for NCAA soccer. (9) Players are considered "graduated" from the Generation Adidas program when they earn significant playing time with their teams. From that point on, their salaries count against the cap. (10) Although it has been suggested that the league slightly overpays Generation Adidas players to outbid potential foreign interests (Eisenmenger, 2010), the players are small in number and spread over all the teams in the league. Furthermore, there does not appear to be a systematic bias in teams that sign them. For robustness, this study also controls for whether players signed Generation Adidas contracts.

There are additional interesting and unique features of MLS's labor market that are not expected to alter the price per unit talent paid for players. For the sake of brevity, these are relegated to the appendix for the interested reader.

Data

Much of the data used in this study is no longer publicly available from official sources and so the majority of game data and roster information come from soccerstats.us, which originally derived the data from the league's website. The data was cross-checked for accuracy with archived versions of the old official website of the league, mlsnet.com. Salary information is from the MLS Players' Union and player birthplaces are available on the league's current website, mlssoccer. com. All dollar figures are converted to 2014 dollars using the Consumer Price Index for All Urban Consumers from the Bureau of Labor Statistics. For game-level analysis, all games involving players whose salary data are missing are dropped. This is the case for approximately 8% of the games, reducing the sample from 2,529 to 2,331 games.

A player in the data sample is determined to be Hispanic if he was born in a Spanish speaking country or born in the United States with a last name that is derived from a Spanish speaking country. (11) The decision to separate players in this way is derived from the famous quote from Chivas USA owner Jorge Vergara, who acquired majority ownership ofChivas USA in 2012 and, according to an anti- discrimination complaint filed against the team, told his staff, If you don't speak Spanish, you can go work for the Galaxy, unless you speak Chinese, which is not even a language. (Calichman v. Chivas USA Soccer LLC, 2013)

Summary statistics of player data separated by ethnicity are provided in Table 2. Only players for which salary data are available are considered in the table. Hispanic players make up approximately 22% of the players in the sample. They are, on average, paid more and have slightly more games played, goals, and assists. They also slightly outperform on per game measures of goals and assists but none of the differences are statistically significant.

The vast majority of fielded teams are 37% or less Hispanic in any given year. Figure 2 plots the percent of players fielded in aseason for each team against that team's percentage of points earned for the season. Chivas USA fielded a roster that was majority Hispanic in half the seasons it played and claims the five most Hispanic teams in the sample. For the league as a whole, the percentage of points does not appear to correlate with the percent of Hispanic players fielded. However, for Chivas USA and, to a lesser extent, Houston Dynamo there is a clear negative relationship between the percent of Hispanic players fielded and performance.

There was a great variation in the amount of discrimination Chivas USA employed across time as can be seen in Figure 2. In the team's inaugural year, the percent of players fielded that were Hispanic was 77% and the team earned less than 19% of the points possible, the worst season in the history of MLS at the time. After the embarrassing inaugural season, Chivas USA ownership relaxed their discrimination in player selection and put more weight on winning. The fraction of fielded players that were Hispanic dropped to 24% by 2008 and varied mostly within range of the rest of the league through 2012. During this time, the team made the playoffs in four consecutive years.

At the end of the 2012 season, Jorge Vergara and his wife at the time, Angelica Fuentes Tellez, became the sole investor-operators of Chivas USA buying out the shares of the other investors. In an effort to get "back to their roots," Vergara promptly fired American coach Robin Fraser in favor of Mexican coach Jose Luis Sanchez Sosa, who told the media, I feel that Chivas USA, at some point, lost that flavor and technique with the ball that is emphasized in Mexico and Latin America. We must re-establish that part and combine it with the MLS style of play. (Mahoney, 2012).

The team indeed returned to their roots both on and off field as it jettisoned most of its non-Hispanicplayers, coaches, and staff. Once again, performance suffered and Chivas US A earned less than 26% of the possible points and were in last place in the Western Conference (second to last overall). The team was soon thereafter sued for employment discrimination by former coaches Dan Calichman and Teddy Chronopoulos (Calichman v. Chivas USA Enterprises LLC, 2013) and by former human resources executive Cynthia Craig (Craigv. Chivas USA Soccer LLC, 2013). Prior to the 2014 season, the league bought out Vergara and Fuentes and announced that the team would cease operations at the end of the season.

Model

The basic intuition behind a market model is that relative salary spending should by and large control for the amount of talent present on a team. If there is a taste for discrimination as is claimed here, the teams would be willing to pay a premium for Hispanic playersbeyond what their skills merit. (12) Therefore, any amount of explanatory power the ethnic makeup of a team has on performance when salaries are controlled for is assumed to be due to discrimination in player employment.

Note that if the players employed by Chivas USA were of superior skill as the team claimed, the test would not find evidence of discrimination even if the team was an outlier in its ethnic makeup. If this were the case, the model would find that Chivas USA spent relatively more money to purchase better players who just happened to be mostly Hispanic and performed well relative to most other teams. The key assumption here is that the market for soccer talent is competitive so that salaries reflect the level of talent on a team, which in turn explains much of the team's performance. However, for MLS particularly, there are factors outside of roster spending to consider when determining a team's performance. So long as there are no omitted variables that are correlated with the ethnic makeup of the team and performance, the coefficient on team ethnicity will still identify discrimination when salaries are controlled for.

An additional variable that uniquely explains performance in MLS is the length of time a team has been with the league. Eight of the nineteen teams in the sample were not in the league in every year. (13) Being an expansion team introduces many additional hurdles that existing teams do not face. Each expansion team must build a team of players that have likely never played together and will be coached and managed by a new organization that also has not developed the chemistry that comes with years of experience working together. For this reason, the marginal dollar spent by newer teams in the league may not go as far in terms of performance relative to more established teams.

The season-level relationship estimated here is

[mathematical expression not reproducible] (1)

where [ppg.sub.jt] is points per game earned by team j in year t and [X.sub.jt] includes the natural log of salary spending (including any luxury taxes paid), age and age-squared of team, and the number of homegrown players, Generation Adidas players, and total players employed. All measures are relative to the league average in year t. A separate indicator for a team's expansion year, total distance traveled, along with a squared term are also included. The total number of players used in a season is included in Szymanski (2000) to account for the negative effect of player turnover within a season and is particularly important to control for here. If Chivas USA's poor performance in 2013 was due to the large player turnover and not the overpayment of the Hispanic players, then the results may mistakenly be biased towards crediting ethnicity for poor performance. Team indicators ([T.sub.j]) are employed to control for any team-specific factors not explicitly measured here and the variable of interest, [H.sub.jt], is the percent of fielded players that are Hispanic relative to league average in year t. The error term [[euro].sub.jt] is assumed to be distributed normally and robust standard errors are estimated.

The procedure described above based on Szymanski (2000) uses season-level measures to identify the effects of wage discrimination on performance. As pointed out in Kahane (2005), Mongeon (2015), and in this paper above, using season-level data may introduce errors in the estimation of parameters. Such errors arise when team rosters are dynamic within season or when teams do not have identical schedules, both of which are concerns for MLS. The available data allows me to overcome these issues and estimate the relationship between team inputs and performance at the game-level. Rather than use a single measure of roster spending, this study measures the amount of annualized guaranteed salary of all players fielded for each game.

Using game-level data also overcomes the bias that exists with season-level data when an expensive player becomes injured and misses many games. In a season-level analysis, the salary of an injured star would be recorded, but the team would likely perform worse than one would expect since the team is unable to field the expensive player. Counting the player as if he were with the team all season would bias the relationship between spending and performance to zero. The same bias exists if a star is transferred to an MLS team during summer and the econometrician uses total salaries, as is done here, or beginning of season roster spending, as is commonly done in the literature.

By using a game-level analysis, an injured expensive player (or a transferred expensive player) will not have his salary recorded for games that he does not play. The variation in talent available across game seen in Figure 1 is used to estimate the coefficient on payroll, which is then more efficiently estimated. Lastly, by using game-level data and measuring control variables as the difference between teams, the strength of opponent is controlled for and the assumption of equally challenging schedules across team need not be made.

Since the outcome of a single game is discrete and ordinal from any team's perspective (win, lose, or tie), an ordered multinomial probit model is estimated. Previous literature has estimated ordered probit models using soccer data to construct measures of competitive balance (Koning, 2000) and to test the efficiency of betting markets and knowledge of tipsters (Goddard and Asimakopoulos, 2004; Forrest and Simmons, 2000a,b; Kuypers, 2000). Since the focus here is on the effect of ethnicity on game outcome controlling for other factors and not maximizing the overall fit of the model, past performance measures employed in some of the previous literature are not utilized here as they will pollute the coefficients on the variables of interest with multicollinearity.

The result of each game played between home team i and away team j is assumed to depend on the latent variable representing the strength of the home team relative to the away team [y.sub.jt]*. which is modeled as a linear combination of game-specific factors and an error term that is normally distributed [[euro].sub.jt]. In the analysis that follows, the error terms are clustered at the team-pair level. For example, all games between the Columbus Crew and the LA Galaxy are assumed to have error terms drawn from a normal distribution with the same variance regardless of where the game was played.

[mathematical expression not reproducible] (2)

The latent difference in strength variable [y.sub.jt]* then translates to game outcome by

[mathematical expression not reproducible] (3)

where [y.sub.jt] = 1 if the home team wins, [y.sub.jt] = 0 if the game ends in a tie and [y.sub.jt] = -1 if the home team loses. Game-level measurements in [X.sub.jt] include the natural log of guaranteed salaries (including luxury taxes) for fielded players, age of the team along with a squared term, variables for the fraction of fielded players that are Hispanic, number of homegrown players and Generation Adidas players, and are constructed as the difference of the home team's measure from the awayteam's. This study also controls for the relative total number of players used for the entire season, which is not agame-level variable but remains significantinthe game-level analysis. The distance traveled by the away team along with a squared term is employed, as well as indicators for the home or away team being an expansion team and for the game being significant for either team (team has not yet clinched nor been eliminated from playoffs). Finally, fixed effects for the home (a.) and away team (a.) are also estimated. Coefficients are estimated using Stata's oprobit command, which maximizes the log-likelihood function of observed outcomes, and marginal effects are reported in the next section. For comparison to the season-level analysis, a game-level OLS model is also estimated treating relative points earned by the home team as a continuous variable.

Previous literature has allowed for the possibility of both manager and fan discrimination to exist in soccer labor markets. Preston and Szymanski (2000), Pedace (2008), and Wilson and Ying (2003) include attendance measures in their models to test whether or not fans respond to the ethnicity or nationality of players. The studies find little to no evidence of fan discrimination using data from the English Football Association in the case of Preston and Szymanski (2000) or the top five European leagues (English Premier League, La Liga, German Bundesliga, Serie A, Ligue 1) according to Wilson and Ying (2003). Only Pedace (2008) finds any evidence of fan preferences for nationality and then only in favor of South American players for fans of the English Premier League. Unfortunately, attendance data for MLS is notoriously noisy as often teams reported tickets distributed instead of sold or actual attendance (Baxter, 2016). Moreover, distributed tickets includes those given away for no charge to charities or local companies and are more likely to makeup a large portion of tickets when attendance is low. Given that the policy of reporting tickets distributed, sold, or actual attendance varies across season and team and is largely unknown, extracting useful information from attendance data free of bias would not be a straightforward venture. Therefore, separating the source of discrimination by owner and fan is out of the scope of this study.

Summary statistics for season- and game-level variables are displayed in Table 3. Compared to the average MLS team, Chivas USA paid less in salaries, employed more Hispanic players, more players overall, slightly fewer home-grown and Generation Adidas players, and travelled more distance. The team also earned fewer points than the average team.

Results

The estimated coefficients of Equation 1 and marginal effects of the probit model (Equations 2 and 3) are displayed in Table 4. For the season-level estimates in the first column, player turnover is found to be negatively related to a team's performance as expected. Expansion teams perform worse on average than more established teams holding all else constant. Total distance travelled over a season is estimated to have a positive effect that diminishes with distance. This may be due to geographically isolated teams playing half their games at home where other teams must travel great distances to play. Although the coefficients on payroll and Hispanic are of the expected signs, neither one is significant. No significant effect is found by employing more homegrown or Generation Adidas players. The relative age of a team also has no significance once an expansion indicator is employed (it is significant if the expansion indicator is excluded).

When game-level data are utilized, controlling for strength of opponent and allowing for more detailed measurements, a positive significant relationship is estimated between payroll and relative points per game in Column 2. The coefficient on Hispanic is also significant and implies that a 0.01 increase in the fraction of Hispanic players a team fields relative to the opponent decreases the relative points earned in that game by approximately 0.008, holding payroll and other variables constant.

To put this in perspective, Chivas USA's fraction of fielded players who were Hispanic was, on average, 0.271 greater than the league average. For a 34-game season, the OLS model in Column 2 would predict a loss of 7.5 points a season, which is approximately equal to changing 2 wins and a tie into three losses.

Although teams potentially have the ability to underpay their homegrown players and field more quality per dollar spent, the coefficient on homegrown is significantly negative indicating that the use of homegrown players hurts performance. The likely explanation for this is that the homegrown system was in its infancy at the time and teams fielding homegrown players likely did so due to injuries and star players missing while they played for their national teams, not because the homegrown players were winning starting lineup spots. The coefficients on home and away expansion conform to expectations as expansion teams perform worse on average. With game-level variation, the coefficient on team age has the expected positive sign even when controlling for expansion years. A significant game for the home team (one where the team is not yet in the playoffs nor eliminated) is found to improve performance for that team but the same cannot be said for away teams.

The last 3 columns of Table 4 display the marginal effects estimated from the ordered probit model, Equations 2 and 3. The magnitudes listed are the marginal effect of increasing each independent variable (measured as the difference between home and away team) in a game with two otherwise identical teams, when the game is significant for both teams and the away team travelled the median distance (1,547 km - LA to Seattle). That is, all relative variables of differences between team-level characteristics are set to zero when calculating the marginal effects.

As expected, an increase in the relative payroll of a team increases the probability of winning and decreases the probability of tying or losing. Controlling for payrolls, a team that is more Hispanic is predicted to have a lower probability of winning and a higher probability of tying or losing. Roster turnover is associated with significantly less wins and more losses and ties. The age of the team is again estimated to have a significant effect independent of the expansion indicator; more established teams are estimated to be more likely to win and less likely to tie or lose. If a home or away team is in its expansion year, it is more likely to tie or lose and less likely to win. If the game is significant for the home team, it is more likely to end with a home win and less likely to end in a tie or with a home loss. The significance of the game to the away team does not seem to significantly affect the game's outcome.

The marginal effect of an increase in the fraction of Hispanic players fielded is estimated to decrease the probability of a win and increase the probability of a tie or loss, holding all else constant. To measure the effect this is predicted to have on Chivas USA, the predicted probabilities of each game outcome is calculated twice, once with the actual data and once when the percent of Hispanic players fielded by Chivas USA is replaced by the annual league average in each game. The difference in the points per game earned for each season implied by these two estimates is then taken as the estimated effect of discrimination on performance. The estimates are displayed in Table 5 and Figure 3. The estimated impact suggests that around 6 to 7 points a season were lost due to discrimination in the two seasons where Chivas USA discriminated the most (2005 and 2013). Losing 6 points is equivalent to changing 2 wins to losses while losing 7 is equivalent to changing 2 wins and 1 tie to 3 losses. The lowest amount given up in 2008 was 0.60 points, when Chivas USA ethnic composition was approximately average. Overall, Chivas USA is estimated to have sacrificed a little under 34 points due to discrimination in its 10-year history.

The market models estimated here depend on the assumption that payroll largely captures the amount of talent on each team and thus explains the majority of performance in the league. Although the dependent variables utilized here differ from what previous literature employs, a back-of-the-envelope calculation indicates that the coefficients estimated here are on par with those results. Szymanski (2000), Preston and Szymanski (2000), and Pedace (2008) all employ the log odds of rank as their dependent variable. That is, they use ln[rank/(93--rank)] since there are 92 teams in the 4 divisions of their samples. Taking the payroll coefficients from the preferred ordered probit model of this study, the marginal effect of a unit increase in log relative payroll between two teams playing a game is calculated to be worth 0.109 points. (14) For a 34 game season, this comes out to 3.706 points in total. This amount of additional points is typically enough to increase rank in the MLS sample used here by 2 or 3 positions. (15) For a middle of the league team ranked 8th in a 16-team league, this translates to a decrease in the log odds of rank that is between 0.474 and 0.724, meaning an improvement in rank since lower numbers imply higher ranks. (16) The corresponding figures from previous literature are a decrease between 0.368 and 0.691 in Szymanski (2000), 0.737 in Preston and Szymanski (2000), and 0.211 to 1.267 in Pedace (2008). Thus, the results found here are completely in line with those found in previous studies.

Robustness

To test the robustness of the results, several additional specifications of the model are estimated. The stated preferences of owner Jorge Vergara indicated that the Spanish language should serve as the line of demarcation when determining which players are preferred. However, the preference of Chivas USA, and potentially the league, may have actually been in favor of a broader group, Latin Americans. This demographic includes Portuguese speaking residents of Brazil as well as residents of Guyana, Suriname, and French Guiana. If the discrimination in MLS were actually in favor of Latin players, then the strength of the coefficients and fit of the model would be stronger if the definition of the favored group were changed to reflect this preference. To test the robustness of the results, the models' coefficients are estimated a second time with the definition of Hispanic expanded to include players who were born in any country in South America or Portugal. Table 6 displays the results when the expanded definition of Hispanicis used. While the qualitative results still hold, the coefficients on Hispanic are noticeably smaller in magnitude and less significant. This is taken as evidence that the Spanish language definition of Hispanic is the correct one to use for testing ethnic discrimination in MLS for this time period.

Pedace (2008) and Wilson and Ying (2003) find evidence of discrimination in the European leagues based on nationality instead of ethnicity or race while Wooten (2013), Reilly and Witt (2007), and Celik and Ince-Yenilmez (2017) estimate salary differentials for international players in MLS. As a robustness check, the variable Hispanic is replaced with variables on the birthplace of the players to test whether the results found here are due to correlation between Hispanic and the geographic origins of players. Tables 7 and 8 display the results for OLS and ordered probit models respectively. The first three columns of Table 7 find no evidence of discrimination for or against players of Latin American, African, or Asian origin. There is, however, a significant positive coefficient for players of European origin in Column 4. However, I would not conclude that European players face discrimination in MLS from this result. Note that in this specification, the coefficient on payroll is no longer significant. When all geographies are controlled for simultaneously in Column 5, neither payroll nor any of the geographies enter the equation with significance. However, in Column 6 when European is omitted, payroll once again is significant. It seems in this case that the variable European is serving as a proxy for payroll as many of the most expensive designated players in MLS are European. This would explain why payroll loses significance whenever European is included and is consistent with the literature. The scatterplots displayed in Figure 4 display the high correlation between European and payroll as well as a near zero correlation for the other geographic variables and payroll. Qualitatively similar results when an ordered probit is estimated are displayed in Table 8, both with and without employing a variable for European origin. (17)

Discussion

This paper tests for discrimination in MLS based on player ethnicities with particular focus on Chivas USA. Using game-level data, the marginal effect of a team employing a greater fraction of Hispanic players is found to decrease the probability of that team winning and increase the probability of it tying or losing when talent is controlled for via salary spending. Particular focus in given to Chivas USA, who explicitly discriminated in its player and employee selection along ethnic lines. The model predicts that the team gave up to 7 points (equivalent to changing 2 wins and a tie to 3 losses) in the years it discriminated the most and 34 points over its 10-year history.

A concern with using data that has been aggregated at the team-season level is that the model will implicitly assume that each team had an equally challenging schedule, which is not the case for MLS (Mongeon, 2015; Goodfriend, 1992). Additionally, the salary measures will be noisy in the presence of rosters that are dynamic within season. This is particularly true for MLS where expensive "designatedplayers" are often purchased midseason from foreign leagues. Given theamountofdata available, itispossible to compare estimates using season-level data as is done in Szymanski (2000) to those using game-level data (Mongeon, 2015). The results of the season-level model in Table 4 do not find significant relationships between a team's payroll or ethnic makeup and its performance for MLS. However, when estimating the effects using game-level data, both factors are found to significantly affect performance in the expected manner. This is true for both the OLS model and the ordered probit model.

The seminal theory on discrimination in competitive labor markets presented in Becker (1957) predicts that when discriminatory tastes vary across firms, only those that do not discriminate will survive. Of course, MLS is a single-entity and not a collection of fully independent teams. However, it is interesting to note that the losses incurred by Chivas USA were enough that the league found it optimal to buy out their "owners" and sell the rights to the second Los Angeles team to other investors in order to maximize the value of the team. Of course, it may also be profit maximizing to discriminate in this way if the taste for discrimination comes from customers (Becker, 1957; Burdekin and Idson, 1991). (18) The issue of whether or not customer discrimination was also a factor of Chivas USA's outcomes could be tested with accurate attendance data as is done in Preston and Szymanski (2000), Pedace (2008), and Wilson and Ying (2003). However, publicly available MLS attendance data is notoriously unreliable (Baxter, 2016) and would require a method to extract useful variation from the often upwardly biased numbers.

Although the research here agrees with previous literature in finding a performance cost in player discrimination, it must be noted that there have historically existed teams that are very successful, even when discriminating in favor of a particular group. Successful teams that have, or have had, a policy to only field domestic players include Deportivo Saprissa (Costa Rica), El Nacional (Ecuador), Athletic Bilbao (Spain), and of course Chivas USA's parent club, CD. Guadalajara (Duerr, 2017). Other forms of player discrimination that seemingly have not resulted in a team's poor performance involve divisions along sectarian lines; Celtic and Rangers, both of which play in Glasgow, Scotland, historically only fielded Catholics and Protestants respectively while being the top two clubs in Scotland by any measure. The same sectarian discrimination previously existed without a cost to performance for Hibernian and Heart of Midlothian in Edinburgh, Scotland. A fruitful area of future research would be to reconcile the empirical findings discussed here and in previous literature with the success that many of the most discriminatory teams have achieved.

Endnotes

(1) The San Jose Earthquakes relocated to Houston and were originally announced to be called "Houston 1836" on January 1, 2006. The date was said to refer to the year the year Houston was founded and the logo contained a silhouette of General Sam Houston. On January 29, an op-ed was penned by a local professor of history claiming that the name and logo would be "divisive" and "alienate" Mexican fans since 1836 was the year Texas became independent of Mexico (Ramos, 2006). Shortly thereafter, Harris County Commissioner Sylvia Garcia petitioned the owners and the league to make a change. The logo was completely redesigned and the new name, Houston Dynamo, was announced on February 25, a mere 10 days before the team's first game.

(2) The authors do find some evidence that black players are rewarded differentially for age and U.S. citizenship. One caveat to note, however, is that race was determined here from photos and not from players self-identifying or from a league database.

(3) The only discrepancy in the logos are that the MLS team has "Club Deportivo Chivas USA" written in place of "Club Deportivo Guadalajara S.A. DE C.V."

(4) In that year, FC Dallas was named the Dallas Burn.

(5) In the leagues early years, there were a couple investor-operator groups (Anschutz Entertainment Group and Hunt Sports Group, LLC) that managed multiple teams.

(6) Twomey and Monks (2011) shows that the percent of estimated revenue MLS spends on player salaries is less than that of the four major sports leagues in the United States as well as some top-tier leagues in Europe and Japan. This is viewed as indirect evidence consistent with the hypothesis that MLS has monopsony power over the player market. It is also, however, consistent with the fact that the league simply hires much less talent than the best leagues in Europe (and Japan) where the world's top talent is employed.

(7) See Coates et al. (2016), Jewell (2015), and Lawson et al. (2008) for a discussion on the initial effects of the DP rule on demand for MLS.

(8) On the other hand, players may be willing to receive less to play in the United States if they view the city they play in as an amenable location (Rosen, 1986), in which case the player will suffer the tax burden and the total amount paid for the player (salary plus tax) may be equal to his true market value. Results are almost unchanged when the luxury taxes are not included in roster spending.

(9) Players are provided funds to return to college should their professional careers end while in the program.

(10) In the cases where the graduation date is not known, I consider all players graduated after 5 years, which is the limit for players to be in the program.

(11) On rare occasions, there will be a US-born player whose last name is not Spanish that is still designated as Hispanic. This occurs when evidence is found that one of the player's parents is Hispanic. This decision was made since Chivas USA reportedly surveyed its player pool on the ethnicity of their parents, indicating that this may be one manner in which discrimination took place (Calichman v. Chivas USA Soccer LLC, 2013).

(12) Discrimination in salaries could also be modeled as Chivas USA underpaying non-Hispanic talent. Given the global competitive market for soccer talent, this is not feasible as players have a plethora of teams to sell their services to and would not agree to play for less than their market values.

(13) The San Jose Earthquakes were relocated to Houston in 2006 and returned to the league in San Jose as a separate team in 2008. The Houston Dynamo is treated as a continuation of the San Jose Earthquakes that began play in 1996 since all personnel were simply relocated to Houston and the second MLS iteration of the San Jose Earthquakes is treated as a new team in 2008.

(14) E[[DELTA]points] = -0.031 x 0 points -0.008 x 1 point + 0.039 x 3 points = 0.109 points

(15) The median number of points separating teams that are 2 ranks apart in the sample used here is 3. The corresponding median number of points for teams that are 3 ranks apart is 5. Therefore, the marginal impact of increasing the relative log of salaries for a team would translate roughly to an increase of 2 to 3 ranks in MLS according to the ordered probit coefficients.

(16) The median number of teams in the league across the 10 years of the sample used here is 15.5, which is rounded to 16. If a 15 team league is assumed instead, the conversion of the payroll coefficient to change in log odds of rank becomes a decrease of 0.788, which is still well in line with previous estimates.

(17) The ordered probit results are also robust to a change in the outcome variable from one that categorizes outcome as a win, loss, or tie for the home team to one that categorizes outcome according to the goal differential (home score minus away score). There are 12 potential outcomes in the sample ranging from a goal differential of-5 to 6. When replacing point differential for goal differential in the OLS game-level model, the coefficients on payroll and Hispanic still have the expected sign but only that on payroll remains significant. The tables are not included for brevity but are available upon request.

(18) I would like to thank an anonymous referee for pointing out this fact and providing a reference.

(19) This last qualifier is important since MLS periodically has players that defect from Cuba to join the league.

(20) All 3 Canadian teams entered the league during the sample period under study here starting with Toronto FC in 2007. Vancouver Whitecaps entered the league in 2011 and Montreal Impact was added the following season in 2012.

(21) The CONCACAF Champions League is a separate league consisting of the best teams in the CONCACAF region. Additional allocation money is given to teams who participate in this league as it has many fixtures during the end of the MLS season when teams are competing for playoff position. Historically, this increase in intense, meaningful games that require traveling long distances has stretched MLS rosters too thin for teams to remain competitive in both leagues.

References

Baxter, K. (2016). MLS Math Doesn't Always Add Up. Los Angeles Times. Retrieved from http:// www.latimes.com/sports/soccer/la-sp-soccer-baxter-20161029-story.html

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

Burdekin, R. C. and Idson, T. L. (1991). Customer preferences, attendance and the racial structure of professional basketball teams. Applied Economics 23 (1), 179-186.

Celik, O. B. and Ince-Yenilmez, M. (2017). Salary differences under the salary cap in Major League Soccer. International Journal of Sports Science & Coaching 12 (5), 623-634.

Coates, D., Frick, B., and Jewell, T. (2016). Superstar salaries and soccer success: The impact of designated players in Major League Soccer. Journal of Sports Economics 17 (7), 716-735.

Duerr, G. M. (2017). Civic integration or ethnic segregation? Models of ethnic and civic nationalism in club football/soccer. Soccer & Society 18 (2-3), 204-217.

Eisenmenger, L. (2010). McCabe Explains Generation Adidas. ussoccerplayers.com. [website] Retrieved from http://ussoccerplayers.com/2010/01/mccabe-explains-generation-adidas.html

Fallas, B. (2006). Goodbye 1836, hello Houston Dynamo. Houston Chronicle. Retrieved from http://www.chron.com/sports/dynamo/article/ Goodbye-1836-hello-Houston-Dynamo-1634753.phpt

Forrest, D. and Simmons, R. (2000a). Focus on sport: Making up the results: The work of the football pools panel, 1963-1997. Journal of the Royal Statistical Society: Series D (The Statistician) 49 (2), 253-260.

Forrest, D. and Simmons, R. (2000b). Forecasting sport: the behaviour and performance of football tipsters. International Journal of Forecasting 16 (3), 317-331.

Francis, J. and Zheng, C. (2010). Learning vicariously from failure: The case of Major League Soccer and the collapse of the North American Soccer League. Group & Organization Management 35 (5), 542-571.

Fraser v. Major League Soccer (2002). 284 f.3d 47 (1st cir. 2002).

Goddard, J. and I. Asimakopoulos (2004). Forecasting football results and the efficiency of fixed-odds betting. Journal of Forecasting 23 (1), 51-66.

Goodfriend, M. (1992). Information-aggregation bias. The American Economic Review 82 (3), 508-519.

Jewell, R. T. (2015). The effect of marquee players on sports demand the case of us Major League Soccer. Journal of Sports Economics 18 (3), 239-252.

Jewell, R. T. and Molina, D. J. (2005). An evaluation of the relationship between hispanics and Major League Soccer. Journal of Sports Economics 6 (2), 160-177

Kahane, L. H. (2005). Production efficiency and discriminatory hiring practices in the National Hockey League: a stochastic frontier approach. Review of Industrial Organization 27 (1), 47-71.

Kesenne, S. (2015). The single entity status of a sports league. Journal of Sports Economics 16 (8), 811-818.

Koning, R. H. (2000). Balance in competition in Dutch soccer. Journal of the Royal Statistical Society: Series D (The Statistician) 49 (3), 419-431.

Kuethe, T. H. and Motamed, M. (2010). Returns to stardom: evidence from us Major League Soccer. Journal of Sports Economics 11 (5), 567-579.

Kuypers, T. (2000). Information and efficiency: An empirical study of a fixed odds betting market. Applied Economics 32 (11), 1353-1363.

Lawson, R. A., Sheehan, K., and Stephenson, E. F. (2008). Vend it like Beckham: David Beckham's effect on MLS ticket sales. International Journal of Sport Finance 3 (4), 189.

Mahoney, R. (2012). Chivas USA goes back to the past. Soccer America. [website] Retrieved from http://www.socceramerica.com/article/49646/chivas-usa-goes-back-to-the-past. html?print

Martinez, A. (2005). Soccer vs. Futbol, Galaxy vs. Goats--L.A.'s Ethnicity as Sport. Los Angeles Times. Retrieved from http://articles.latimes.com/2005/apr/20/opinion/oe-martinez20

Mongeon, K. (2015). A market test for ethnic discrimination in the National Hockey League a game-level panel data approach. Journal of Sports Economics 16 (5), 460-481.

Parker, K. (1999). Action in the Community. AdAge. [website]. Retrieved from adage.com/article/american-demographics/action-community/43442

Pedace, R. (2008). Earnings, performance, and nationality discrimination in a highly com- petitive labor market as an analysis of the English professional soccer league. Journal of Sports Economics 9 (2), 115-140.

Preston, I. and Szymanski, S. (2000). Racial discrimination in English football. Scottish Journal of Political Economy 47 (4), 342-363.

Ramos, R. (2006). Kicking around Houston 1836: Soccer team sends the wrong message to Latinos. Houston Chronicle. Retrieved from http://www.chron.com/opinion/outlook/article/ Kicking-around-Houston-1836-1884308.php

Reilly, B. and Witt, R. (2007). The determinants of base pay and the role of race in Major League Soccer: Evidence from the 2007 league season. School of Economics Discussion Papers 1907, (pp. 1-27). University of Surrey.

Rivera, G. (2017). Looking back at the Fire's last big signing in 2007, Cuauhtemoc Blanco. Chicago Fire Confidential. [website]. Retrieved from http://www.chicagonow.com/fire-confidential/2017/05/looking-back-at-the-fires-last-big-signing-in-2007-cuauhte-moc-blanco/

Rogers, M. (2008). De La Hoya brings heavyweight class to Houston. Yahoo! Sports. Retrieved from https://www.yahoo.com/news/la-hoya-brings-heavyweight-class-035900227--mls.html

Rosen, S. (1986). The theory of equalizing differences. In Ashenfelter, O., and Layard, R. (Eds Handbook of labor economics Vol.1. (pp. 641-692). Amsterdam: Elsevier/North Holland.

Szymanski, S. (2000). A market test for discrimination in the English professional soccer leagues. Journal of Political Economy 108 (3), 590-603.

Twomey, J. and Monks, J. (2011). Monopsony and salary suppression: The case of Major League Soccer in the United States. The American Economist 56 (1), 20-28.

Wilson, D. P. and Ying, Y.-H. (2003). Nationality preferences for labour in the international football industry. Applied Economics 35 (14), 1551-1559.

Witz, B. (2014). A Soccer Team Struggles to Remake Its Identity: Chivas U.S.A. Trying to Escape M.L.S. Doldrums. The New York Times. Retrieved from https://www.nytimes.com/2014/ 09/05/sports/soccer/chivas-usa-trying-to-escape-mls-doldrums.html?_r=

Wooten, J. J. (2013). Can ranking nationalities explain the salary discrepancies in Major League Soccer? (Doctoral dissertation). Washington State Universit. Retrieved from http://ses.wsu.edu/wpcontent/uploads/2014/09/WPVersionl.pdf

Appendix A. MLS Labor Market Details

Other unique features of the MLS labor market that are not expected to alter the salary-talent relationship are discussed below.

International Roster Slots

Unlike many European leagues, MLS restricts the number of international players each team may employ. As mentioned in the previous section, in 2005 each team in MLS was granted 7 roster slots (4 senior and 3 youth), which were tradable between teams, that couldhave been filled by international players save Chivas USA and RSL, both of which could have up to 9 until the beginning of the 2007 season. In 2008, each team in the league was allowed 8 international slots and no distinction was thenceforth made for age of the players.

For teams in the United States, an international player is one who is not a citizen, does not hold a green card, and does not haverefugeestatus. (19) ForteamslocatedinCanada, Canadian and U.S. players alike are both treated as domestic. Canadian teams have an additional restriction that at least 3 of their players must be Canadian citizens. (20) Given the large amount of talent supplied to professional soccer leagues (both domestic and international) and the weak restriction on number of domestic players each MLS team must employ, the constraint is not expected to alter the price paid per unit talent.

Allocation Money

So called "allocation money" may be used to buy down players' salaries to fit under the cap or reduce a player's salary to be below the DP threshold. In addition to the potential shared revenue generated from the third DP tax, teams are given allocation money from the league whenever they miss the playoffs in a previous season, lose a player on a transfer to a foreign team, or are involved in the CONCACAF Champions League. (21) The effect of the allocation money is to essentially increase the salary cap for receiving teams. Although it increases the amount of talent teams are able to afford, it does not impact the price per unit of talent paid.

Allocated Players

The player personnel department of MLS creates a list of players that, should they enter the league, must be allocated according to the ranking order of teams, which is set by taking the reverse order of the team's standings similar to how the draft order is determined. Players on the allocated ranking list are comprised of Select U.S. Men's and Youth National Team players and players that transferred out of MLS garnering transfer fees of at least $500,000. Any expansion team is automatically ranked first for player allocation and whenever a team signs an allocated player, it is moved to the bottom of the order. The purpose of the allocated players list appears to be one of keeping some sort of popularity parity in the league when it comes to American or former star MLS players.

The allocation process can affect the total price paid (but not the salary) for talent if teams have to make trades or pay to change their place in the rank order to sign a player on the allocation list. However, such trades or signings in the league are very rare. Additionally, the strictness with which these rules are enforced are often questioned as the captain of Team U.S.A., Clint Dempsey, was signed by Seattle Sounders S.C. in August of 2013 when Seattle was not atop the ranking order for allocated players and did not make a trade to obtain him.

Craig Kerr (1)

(1) California Polytechnic University, Pomona

Craig Kerr is an associate professor of economics at California State Polytechnic University in Pomona and an economist at Phillips, Fractor & Company in Pasadena. His research in sports focuses mostly on labor market discrimination and incentives.

doi.org/10.32731/IJSR141.022019.02
Table 1. The Designated Player Rule and the Salary Cap

Year  Salary Cap  1st       2nd       3rd

2005  $1.9 M      NA        NA        NA
2006  $2.0 M      NA        NA        NA
2007  $2.1 M      $400,000  $325,000  NA
2008  $2.3 M      $400,000  $325,000  NA
2009  $2.3 M      $415,000  $335,000  NA
2010  $2.6 M      $335,000  $335,000  $335,000+$250,000 tax
2011  $2.7 M      $335,000  $335,000  $335,000+$250,000 tax
2012  $2.8 M      $350,000  $350,000  $335,000+$250,000 tax
2013  $2.9 M      $368,750  $368,750  $368,750+$250,000 tax
2014  $3.1 M      $387,500  $387,500  $387,500+$150,000 tax

Year  21-23 yr old  under 21

2005  NA            NA
2006  NA            NA
2007  NA            NA
2008  NA            NA
2009  NA            NA
2010  NA            NA
2011  NA            NA
2012  $200,000      $150,000
2013  $200,000      $150,000
2014  $200,000      $150,000

M=million, NA= not applicable.

Table 2. Summary Statistics by Ethnicity

Variable              Non-Hispanic        Hispanic
                        (n = 3887)        (n=1076)

games                      14.900          15.126
                          (11.203)        (10.781)
goals                       1.250           1.560
                           (2.638)         (2.841)
assists                     1.164           1.339
                           (2.051)         (2.325)
goals per game              0.069           0.085
                           (0.122)         (0.128)
assists per game            0.064           0.073
                           (0.097)         (0.104)
guaranteed salary    $133,322        $156,166
                    ($410,678)      ($342,006)

Figures listed are averages with standard deviations in parentheses.
Unit of observation is the player-year and salary figures are in 2014
dollars.

Table 3. Summary Statistics

                         Season-Level           Game-Level
                         Full Sample   Chivas   Full Sample  Chivas

log(ppg)                   0.28         0.062
points ([dagger])         (0.259)      (0.349)     0.690      -0.572
                                                  (2.462)     (2.543)
% home win                                        47.790      47.157
% home loss                                       24.796      28.094
% tie                                             27.413      24.749
ln(payroll) ([dagger])     0.000       -0.190      0.024      -0.355
                          (0.407)      (0.245)    (0.659)     (0.582)
Hispanic ([dagger])        0.000        0.271      0.002       0.259
                          (0.131)      (0.178)    (0.197)     (0.219)
player count ([dagger])    0.000        1.577      0.045       1.786
                          (2.400)      (2.89)     (3.45)      (3.789)
Adidas ([dagger])          0.000       -0.035     -0.001      -0.019
                          (0.047)      (0.030)    (8.153)     (0.092)
home grown ([dagger])      0.000       -0.006     -0.001      -0.007
                          (0.044)      (0.046)    (0.061)     (0.045)
team age ([dagger])        0.000       -4.923     -0.001      -4.629
                          (5.506)      (1.833)    (8.153)     (5.901)
distance (1000's of km)   28.491       33.059      1.771       2.041
                          (5.174)      (2.924)    (1.117)     (1.285)
expansion                  0.057        0.100
                          (0.233)      (0.316)
home expansion                                     0.058       0.070
                                                  (0.234)     (0.256)
away expansion                                     0.054       0.067
                                                  (0.225)     (0.250)
N                        157           10       2331         299
Years                     10           10         10          10
Teams                     19            1         19          19
Games                                           2331         299

Figures listed are averages with standard deviations in parentheses.
The number of observations for season level data is less than that of
Teams x Years due to expansion teams that began play after 2005.
([dagger]) Season-level measures are relative to league average. Game-
levelmeasures are in the form of home team relative to away team. Note
that player count and team age vary only across season for each team,
and not across game.

Table 4. Estimation Results

                                              OLS
                             Season-Level     Game-Level
                             (1)              (2)
                             Points per Game  Home-Away Points

payroll ([dagger])             0.052             0.204 (*)
                              (0.100)           (0.106)
Hispanic ([dagger])           -0.311            -0.814 (**)
player count ([dagger])       (0.284)           (0.362)
                              -0.050 (***)      -0.081 (***)
                              (0.009)           (0.017)
team age ([dagger])            0.015             0.101 (***)
                              (0.016)           (0.038)
team age squared ([dagger])   -0.001             0.000
                              (0.002)           (0.001)
Adidas ([dagger])             -0.187            -0.121
                              (0.512)           (0.585)
home grown ([dagger])         -0.320            -2.127 (**)
                              (0.721)           (0.978)
distance                       0.103 (*)         0.184
                              (0.053)           (0.206)
distance squared              -0.002 (**)       -0.037
                              (0.001)           (0.050)
home significant                                 0.524 (**)
                                                (0.257)
away significant                                -0.116
                                                (0.248)
expansion                     -0.266 (***)
                              (0.101)
home expansion                                  -0.421 (*)
                                                (0.235)
away expansion                                   0.806 (***)
                                                (0.206)
constant                      -0.046            -0.307
                              (0.772)           (0.418)
N                            157              2331
R-squared                      0.394             0.067

                                        Ordered Probit
                                        Game-Level
                                        (3)
                                 Loss              Tie

payroll ([dagger])             -0.031 (**)         -0.008 (**)
                               (0.015)             (0.004)
Hispanic ([dagger])             0.127 (***)         0.032 (**)
player count ([dagger])        (0.048)             (0.013)
                                0.012 (***)         0.003 (***)
                               (0.002)             (0.001)
team age ([dagger])            -0.014 (***)        -0.003 (***)
                               (0.004)             (0.001)
team age squared ([dagger])
Adidas ([dagger])               0.014               0.004
                               (0.080)             (0.020)
home grown ([dagger])           0.294 (*)           0.074 (*)
                               (0.160)             (0.040)
distance                       -0.010              -0.002
                               (0.008)             (0.002)
distance squared
home significant               -0.078 (*)          -0.012 (***)
                               (0.045)             (0.004)
away significant                0.017               0.005
                               (0.031)             (0.009)
expansion
home expansion                  0.063 (*)           0.011 (**)
                               (0.035)             (0.004)
away expansion                 -0.109 (*) (***)    -0.048 (***)
                               (0.025)             (0.017)
constant
N                            2331                2331
R-squared

                                      Ordered Probit
                                      Game-Level
                                        (3)
                                        Win

payroll ([dagger])                      0.039 (**)
                                       (0.018)
Hispanic ([dagger])                    -0.159 (***)
player count ([dagger])                (0.060)
                                       -0.015 (***)
                                       (0.003)
team age ([dagger])                     0.017 (***)
                                       (0.005)
team age squared ([dagger])
Adidas ([dagger])                      -0.018
                                       (0.100)
home grown ([dagger])                  -0.367 (*)
                                       (0.199)
distance                                0.012
                                       (0.011)
distance squared
home significant                        0.090 (*)
                                       (0.048)
away significant                       -0.022
                                       (0.040)
expansion
home expansion                         -0.073 (*)
                                       (0.039)
away expansion                          0.157 (***)
                                       (0.041)
constant
N                                    2331
R-squared

All coefficients listed are marginal effects. Standard errors are in
parentheses and are robust standard errors for the first two columns.
Standard errors in Columns 4 through 6 are clustered at the team-pair
level. (*) p < 0.10,  (**) p < 0.05, (***) p<0.01. ([dagger]) Season-
level independent variables are relative to league average and payroll
is the natural log of the relative measure. Game-level measures are all
in the form of home team relative to away team and player count and age
measures vary only across season for each team, not across game.

Table 5. Estimated Impact of Discrimination on Performance for Chivas
USA

                                 2005  2006    2007   2008   2009

Points Per Game (fitted)         0.71   1.25   1.16   1.27   0.97
Points Per Game (hypothetical)   0.90   1.39   1.27   1.29   1.03
Difference                      -0.19  -0.14  -0.11  -0.02  -0.06
Games                           32     32     30     30     30
Total Loss of Points            -6.08  -4.48  -3.30  -0.60  -1.80

                                 2010   2011   2012   2013   2014

Points Per Game (fitted)         0.98   1.35   1.29   0.83   0.92
Points Per Game (hypothetical)   1.10   1.41   1.34   1.03   1.02
Difference                      -0.12  -0.06  -0.05  -0.20  -0.10
Games                           30     34     34     34     34
Total Loss of Points            -3.60  -2.04  -1.70  -6.80  -3.40

Fitted values are calculated using the coefficients of the multinomial
logit estimation. Hypothetical values are the fitted percentage of
points earned when all Chivas USA values for percent Hispanic players
fielded are replaced with that season's league-wide average.

Table 6. Estimation Results: Latin America & Portugal Categorized
Hispanic

                             OLS
                             Season-Level      Game-Level
                             (1)               (2)
                              Points per Game  Home-Away Points

payroll ([dagger])             0.055              0.203 (*)
                              (0.081)            (0.106)
Hispanic ([dagger])            0.046             -0.588 (*)
player count ([dagger])       (0.281)            (0.340)
                              -0.043 (***)       -0.080 (***)
                              (0.008)            (0.017)
team age ([dagger])            0.004              0.102 (***)
                              (0.014)            (0.038)
team age squared ([dagger])   -0.001              0.000
                              (0.002)            (0.001)
Adidas ([dagger])              0.032             -0.163
                              (0.401)            (0.586)
home grown ([dagger])         -0.379             -2.175 (**)
                              (0.619)            (0.978)
distance                       0.093 (*)          0.187
                              (0.049)            (0.206)
distance squared              -0.002 (**)        -0.038
                              (0.001)            (0.050)
home significant                                  0.519 (**)
                                                 (0.257)
away significant                                 -0.111
                                                 (0.248)
expansion                     -0.268 (***)
                              (0.102)
home expansion                                   -0.394 (*)
                                                 (0.235)
away expansion                                    0.788 (***)
                                                 (0.206)
constant                      -1.010             -0.311
                              (0.722)            (0.417)
N                            157               2331
R-squared                      0.40               0.066

                                         Ordered Probit
                                         Game-Level
                                         (3)
                           Loss            Tie               Win

payroll ([dagger])         -0.031 (**)     -0.008 (**)      0.039 (**)
                           (0.015)         (0.004)         (0.018)
Hispanic ([dagger])         0.087 (*)       0.022 (*)      -0.109 (*)
player count ([dagger])    (0.045)         (0.012)         (0.056)
                            0.011 (***)     0.003 (***)    -0.014 (***)
                           (0.002)         (0.001)         (0.003)
team age ([dagger])        -0.014 (***)    -0.003 (***)     0.017 (***)
                           (0.004)         (0.001)         (0.005)
team age
squared ([dagger])

Adidas ([dagger])           0.019           0.005          -0.024
                           (0.080)         (0.020)         (0.100)
home grown ([dagger])       0.308 (*)       0.077 (*)      -0.385 (*)
                           (0.162)         (0.041)         (0.201)
distance                   -0.010          -0.003           0.012
                           (0.008)         (0.002)         (0.011)
distance squared
home significant           -0.078 (*)      -0.012 (***)     0.089 (*)
                           (0.045)         (0.004)         (0.048)
away significant            0.016           0.005          -0.021
                           (0.031)         (0.009)         (0.040)
expansion
home expansion              0.059 (*)       0.010 (**)     -0.069 (*)
                           (0.035)         (0.004)         (0.039)
away expansion             -0.107 (***)    -0.046 (***)     0.153 (***)
                           (0.025)         (0.017)         (0.042)
constant
N                        2331            2331            2331
R-squared

All coefficients listed are marginal effects. Standard errors are in
parentheses and are robust standard errors for the first two columns.
Standard errors in Columns 4 through 6 are clustered at the team-pair
level. (*)p<0.10, (**) p < 0.05, (***)p <0.01. ([dagger]) Season-level
independent variables are relative to league average and payroll is the
natural log of the relative measure. Game-level measures are all in the
form of home team relative to away team and player count and age
measures vary only across season for each team, not across game.
Hispanic includes players with Hispanic last names or born in Latin
America, Spain, or Portugal.

Table 7. OLS Results with Country of Origin Variables

                                (1)           (2)
                                PPG           PPG

payroll ([dagger])               0.204 (*)       0.197 (*)
                             (0406)             (0.106)
Latin ([dagger])                -0.580
                                (0.404)
African ([dagger])                              -0.813
                                                (0.643)
Asian ([dagger])
European ([dagger])
player count ([dagger])
                                -0.080 (***)    -0.078 (***)
                                (0.017)         (0.017)
team age ([dagger])              0.108 (***)     0.103 (***)
                                (0.038)         (0.037)
team age squared ([dagger])      0.000           0.000
                                (0.001)         (0.001)
Adidas ([dagger])               -0.137          -0.129
                                (0.585)         (0.586)
home grown ([dagger])           -2.315 (**)     -2.299 (**)
                                (0.986)         (0.985)
distance                         0.189           0.195
                                (0.206)         (0.206)
distance squared                -0.038          -0.040
                                (0.050)         (0.050)
home significant                 0.518 (**)      0.497 (*)
                                (0.257)         (0.257)
away significant                -0.110          -0.078
                                (0.248)         (0.247)
home expansion                  -0.400 (*)      -0.398 (*)
                                (0.235)         (0.236)
away expansion                   0.797 (***)     0.791 (***)
                                (0.206)         (0.207)
constant                        -0.317          -0.320
                                (0.417)         (0.416)
N                             2331            2331
R-squared                        0.066           0.066

                                 (3)           (4)
                                 PPG           PPG

payroll ([dagger])              0.195 (*)       0.145
                               (0.106)         (0.109)
Latin ([dagger])
African ([dagger])
Asian ([dagger])                0.550
                               (1.753)
European ([dagger])                             1.063 (**)
player count ([dagger])                        (0.536)
                               -0.080 (***)    -0.082 (***)
                               (0.017)         (0.017)
team age ([dagger])             0.103 (***)     0.105 (***)
                               (0.038)         (0.037)
team age squared ([dagger])     0.000           0.000
                               (0.001)         (0.001)
Adidas ([dagger])              -0.126          -0.110
                               (0.586)         (0.585)
home grown ([dagger])          -2.169 (**)     -2.258 (**)
                               (0.981)         (0.981)
distance                        0.192           0.189
                               (0.206)         (0.206)
distance squared               -0.039          -0.039
                               (0.050)         (0.050)
home significant                0.502 (*)       0.515 (**)
                               (0.256)         (0.256)
away significant               -0.093          -0.098
                               (0.247)         (0.246)
home expansion                 -0.389          -0.403 (*)
                               (0.237)         (0.236)
away expansion                  0.787 (***)     0.809 (***)
                               (0.208)         (0.205)
constant                       -0.306          -0.326
                               (0.417)         (0.417)
N                            2331            2331
R-squared                       0.065           0.067

                                (5)             (6)
                                PPG             PPG

payroll ([dagger])              0.164           0.207 (*)
                               (0.109)         (0.106)
Latin ([dagger])               -0.511          -0.636
                               (0.415)         (0.408)
African ([dagger])             -0.775          -0.916
                               (0.651)         (0.645)
Asian ([dagger])                0.065           0.136
                               (1.766)         (1.764)
European ([dagger])             0.861
player count ([dagger])        (0.550)
                               -0.080 (***)    -0.078 (***)
                               (0.017)         (0.017)
team age ([dagger])             0.110 (***)     0.109 (***)
                               (0.038)         (0.038)
team age squared ([dagger])     0.000           0.000
                               (0.001)         (0.001)
Adidas ([dagger])              -0.127          -0.143
                               (0.585)         (0.585)
home grown ([dagger])          -2.463 (**)     -2.450 (**)
                               (0.998)         (0.997)
distance                        0.189           0.192
                               (0.206)         (0.206)
distance squared               -0.039          -0.039
                               (0.050)         (0.050)
home significant                0.526 (**)      0.517 (**)
                               (0.257)         (0.258)
away significant               -0.100          -0.096
                               (0.247)         (0.247)
home expansion                 -0.410 (*)      -0.404 (*)
                               (0.235)         (0.235)
away expansion                  0.806 (***)     0.793 (***)
                               (0.206)         (0.207)
constant                       -0.344          -0.332
                               (0.417)         (0.417)
N                            2331            2331
R-squared                       0.068           0.067

Robust standard errors are in parentheses. (*) p<0.10, (**) p<0.05,
(***) p<0.01. ([dagger]) Measures are all in the form of home team
relative to away team and player count and age measures vary only
across season for each team, not across game.

Table 8. Ordered Probit Marginal Effects with Country of Origin
Variables

                                All Regions
                            (1)             (2)
                            Loss            Tie

payroll ([dagger])         -0.025          -0.006
                           (0.015)         (0.004)
Latin ([dagger])            0.075           0.018
                           (0.052)         (0.013)
African ([dagger])          0.105           0.026
                           (0.100)         (0.024)
Asian ([dagger])            0.018           0.004
European ([dagger])        (0.252)         (0.062)
                           -0.137 (*)      -0.033 (*)
player count ([dagger])    (0.081)         (0.020)
                            0.011 (***)     0.003 (***)
                           (0.002)         (0.001)
team age ([dagger])        -0.015 (***)    -0.004 (***)
                           (0.004)         (0.001)
Adidas ([dagger])           0.012           0.003
                           (0.080)         (0.020)
home grown ([dagger])       0.348 (**)      0.085 (**)
                           (0.161)         (0.040)
distance                   -0.010          -0.002
                           (0.008)         (0.002)
home significant           -0.078 (*)      -0.011 (***)
                           (0.045)         (0.004)
away significant            0.015           0.004
                           (0.031)         (0.009)
home expansion              0.061 (*)       0.010 (**)
                           (0.035)         (0.004)
away expansion             -0.110 (***)    -0.047 (***)
                           (0.025)         (0.017)
N                        2331            2331

                          All Regions      No Europe
                            (3)            (4)
                            Win            Loss

payroll ([dagger])          0.031          -0.031 (**)
                           (0.019)         (0.014)
Latin ([dagger])           -0.093           0.095 (*)
                           (0.065)         (0.053)
African ([dagger])         -0.130           0.128
                           (0.124)         (0.097)
Asian ([dagger])           -0.022           0.008
European ([dagger])        (0.314)         (0.252)
                            0.170 (*)
player count ([dagger])    (0.100)
                           -0.014 (***)     0.011 (***)
                           (0.003)         (0.002)
team age ([dagger])         0.019 (***)    -0.015 (***)
                           (0.004)         (0.004)
Adidas ([dagger])          -0.016           0.015
                           (0.099)         (0.079)
home grown ([dagger])      -0.433 (**)      0.346 (**)
                           (0.198)         (0.161)
distance                    0.012          -0.010
                           (0.011)         (0.008)
home significant            0.089 (*)      -0.076 (*)
                           (0.047)         (0.044)
away significant           -0.019           0.014
                           (0.039)         (0.031)
home expansion             -0.072 (*)       0.060 (*)
                           (0.039)         (0.035)
away expansion              0.156 (***)    -0.107 (***)
                           (0.041)         (0.025)
N                        2331            2331

                                    No Europe
                            (5)              (6)
                            Tie              Win

payroll ([dagger])         -0.008 (**)      0.039 (**)
                           (0.004)         (0.018)
Latin ([dagger])            0.023 (*)      -0.118 (*)
                           (0.013)         (0.066)
African ([dagger])          0.031          -0.159
                           (0.024)         (0.121)
Asian ([dagger])            0.002          -0.009
European ([dagger])        (0.062)         (0.314)
player count ([dagger])
                            0.003 (***)    -0.014 (***)
                           (0.001)         (0.003)
team age ([dagger])        -0.004 (***)     0.018 (***)
                           (0.001)         (0.004)
Adidas ([dagger])           0.004          -0.019
                           (0.020)         (0.099)
home grown ([dagger])       0.085 (**)     -0.431 (**)
                           (0.040)         (0.199)
distance                   -0.003           0.013
                           (0.002)         (0.011)
home significant           -0.011 (***)     0.088 (*)
                           (0.004)         (0.047)
away significant            0.004          -0.018
                           (0.009)         (0.039)
home expansion              0.010 (**)     -0.071 (*)
                           (0.004)         (0.039)
away expansion             -0.046 (***)     0.153 (***)
                           (0.017)         (0.042)
N                        2331            2331

All coefficients listed are marginal effects. Standard errors are in
parentheses and are clustered at the team-pair level. (*) p < 0.10,
(**) p < 0.05, (***) p < 0.01. ([dagger]) Measures are all in the form
of home team relative to awayteam and player count and age measures
vary only across season for each team, not across game.
COPYRIGHT 2019 Fitness Information Technology Inc.
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 2019 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Author:Kerr, Craig
Publication:International Journal of Sport Finance
Article Type:Report
Date:Feb 1, 2019
Words:13329
Previous Article:Factors that Shape the Demand for International Football Games Across Different Age Groups.
Next Article:Jumping in the Pool: What Determines Which Players the NBA Considers in the Draft?
Topics:

Terms of use | Privacy policy | Copyright © 2020 Farlex, Inc. | Feedback | For webmasters