Customer-Based Discrimination in Major League
Baseball?
Craig A. Depken, II
The University of Texas at Arlington
Department of Economics
Arlington, Texas 76019-0479
Oce: (817) 272-3290
Fax: (817) 272-3145
E-mail: [email protected]
Jon M. Ford
The University of Texas at Arlington
Department of Economics
Arlington, Texas 76019-0479
Oce: (817) 272-3062
Fax: (817) 272-3145
E-mail: [email protected]
Customer-Based Discrimination in Major League
Baseball?
Abstract
We investigate the existence of customer-based discrimination in Major League Baseball
using a direct measure of customer preference, allstar balloting. Unlike other measures
used in the literature, such as trading-card prices, our data do not suer from hard to
measure monetary inuences on preference. Using allstar votes from 1992 through 1996, and
controlling for various player and team-specic attributes, we nd no evidence of customerbased discrimination. Our results imply that non-monetary based evaluations of professional
baseball fans are not inuenced by the race of the players in question.
I. Introduction
Persistent wage disparities between equally productive laborers of dierent races or gender have been a subject investigated by economists for many years. Becker (1975) shows
that wage disparities can arise from employer, co-worker or customer-based discrimination
but that, at a point of constant-returns-to-scale (CRS) technology, the former two sources
of wage disparities will be short-run phenomena. Kahn (1991a) shows that customer-based
discrimination can persist in general equilibrium even with CRS technology and nondiscriminating rms. The fact that racial and gender wage disparities have persisted in many
industries for a number of years indicates that (a) only rms that have non-CRS technologies
are able to discriminate across labor types, or (b) the true sources of discrimination have yet
to be determined, or (c) the empirical methods used to estimate the extent of discrimination
are not correct.1
This paper investigates customer-based discrimination in professional baseball using a
direct measure of customer preference, allstar balloting from Major League Baseball (MLB).
Using voting results from 1992 through 1996, and controlling for player and team-specic
attributes, we nd no customer-based discrimination against any particular racial group in
professional baseball. Our results are robust to various subsamples, right-hand side variables
and functional forms, and dier from previous studies. The dierences suggest that possible problems in estimating customer-based discrimination can dramatically alter empirical
ndings. The results are useful in showing that non-monetary-based discrimination is not
reected in the allstar balloting of Major League Baseball.
The paper is structured as follows. Section II reviews previous studies of customer-based
discrimination and discuss problems with current methods. Section III describes our data
and empirical model, and reports our ndings. Section IV oers concluding remarks.
1
II. Literature Review
The ndings of numerous prior studies show persistent discrimination in professional
sports against dierent ethnic or racial groups, using various data and models. Some studies
have found statistically signicant discrimination against blacks and/or latinos in professional
sports, while others have found discrimination against whites. Wage disparities have been
investigated in both professional basketball and baseball as well as in more service-oriented
industries.
Models which investigate professional sports typically rely upon salary information and
the very extensive and accurate productivity statistics maintained in these industries. Controlling for player and team attributes, several studies have found empirical evidence of
various forms of discrimination. Scott, Long and Somppi (1985) nd that customers of professional basketball discriminated in favor of blacks from 1978 through 1981. Schollaert and
Smith (1987) report that black players did not aect attendance at professional basketball
games in the period 1969 through 1982, whereas Kahn and Sherer (1988) concluded that
customers preferred to watch white players between 1980 through 1986.
Studies of discrimination in professional baseball include Scully (1974a), which reports
that black pitchers lower attendance, and Scully (1974b), which concludes further that black
players lower team revenues. However, these ndings are contradicted by Sommers and
Quinton (1982), which nds that blacks did not aect team revenues. These disparate
empirical ndings suggest a lack of consensus in the literature on whether customer-based
discrimination exists and what form it takes.2
One problem with trying to estimate customer-based discrimination using salary data
is that it is very dicult to control for all factors that inuence wages. The sources of
discrimination, i.e., management, co-workers, or customers, are dicult to isolate in the
empirical frameworks employed. Since wages are dependent upon the marginal revenue
2
product of the labor being investigated, one must control for (any) discrimination's eect
on marginal physical product on the value of this product at the same time. Unfortunately,
this is not an easy task. In light of this, many authors have analyzed an alternative measure
of the value placed on players by customers that does not suer from the non-customerbased inuences on wage, i.e., management and co-worker behaviors: the sports memorabilia
market. Several papers investigate the prices of trading-cards in professional baseball and
basketball and their relationship to the race of the players depicted on them.
For example, Nardinelli and Simon (1990) analyze baseball card prices, and nd customerbased discrimination against latinos. Yet, Anderson and LaCroix (1991), using very similar
data, nd customer-based discrimination only against blacks. Stone and Warren (1996) do
not nd evidence of customer-based discrimination in the basketball trading-card market.
Most papers examining the posted price of trading-cards nd evidence of some form of
customer-based discrimination. Unfortunately, the exact nature of the discrimination seems
to be still open to debate, in that the functional form of any customer-based discrimination
varies across studies. Some nd customer-based discrimination in the estimated coecient on
only a race dummy variable (e.g., Anderson and LaCroix (1991)), while others nd customerbased discrimination in the estimated coecients on the interactions between race and other
production statistics (e.g. Nardinelli and Simon (1990) and Stone and Warren (1996)).
The use of trading-card prices suers from a dierent set of supply-side inuences on
price than does the use of wage data. These non-customer-based inuences on the price of
trading cards are perhaps even more dicult to control for than exogenous inuences on
wages. First, as most authors admit, it is dicult to control for the supply of trading-cards.
As the number of trading-cards available for sale is not perfectly inelastic, even if the stock of
trading-cards is, the estimation of trading card demand, without controlling for supply-side
eects on price, suers from an omitted variables problem. These omitted variables lead
3
to biased estimates where the direction of the bias is unknown. While the race of players
may have an inuence on the prices of trading-cards, in light of the potential biasedness and
inconsistency, carte blanche condence should not be placed in them.
A second problem with using trading-card prices is the uncertainty as to what portion
of the population is being modeled. Typically, authors use a sample of trading cards from
several years in the past. The argument for using these cards is that all players have retired
and thus there is no further inuence on the price of cards by either management, co-workers
or player productivity statistics. Therefore, any estimated race bias can be attributed to
customers. Independent variables include the career productivity statistics of players to
control for individual attributes, and estimation techniques are based upon a cross-section
of one year's production of trading-cards.3 Unfortunately, using such data almost assuredly
restricts the sample of the population whose preferences are being modeled.
For example, Nardinelli and Simon (1990) and Anderson and LaCroix (1991) estimate
cross-sectional models using current prices of baseball cards produced in the 1960 and 1977
seasons. Stone and Warren (1996) use data on basketball cards from 1977-1978 season, with
prices from 1993. As a result, only those people who have a large amount of disposable
income (typically older people), those who know the players and remember them playing
(typically older people) and those who collect baseball cards for investment purposes (again,
typically older people) are those whose preferences are most likely included in the market for
such cards. In essence, these data restrict the sample of customers that are being modeled.
If race plays a major factor in the price of these cards, it is not clear that this indicates a
society-wide racial bias towards any one group of players or not.
The inuence of investors and speculators on the prices of trading-cards is perhaps the
most dicult to control for. If individuals purchase cards for their investment value, decisions
on which cards to purchase and for how much may not reveal the purchaser's personal
4
preference for one player or another, but rather the perceived preferences of future buyers.
The use of cross-sectional data does not allow one to control for card-price volatility, which
may add to the option value of the card and thus it's desirability. It may be overly restrictive
to assume that these dierent motivations are identically distributed across all cards and
customers.
The various motivations on the part of those who purchase trading-cards and the uncertainty about which segment of society is being modeled suggest that an alternative measure of
customer preferences may be desirable. Such an alternative would seek to avoid supply-side
inuences on the measurement of the preferences of customers. Andersen (1992) suggests
looking at allstar votes. He nds racial discrimination against blacks reected in allstar
ballots from 1970 through 1986. We employ similar data from 1992 through 1996, but nd
very dierent results.
The rst allstar game was played in 1933 as a vehicle for augmenting the retirement fund
for professional players. Every year since, the allstar game has been played in late June or
early July in one of the MLB stadia.4 The allstar game rotates among the various stadia
every year. For the rst two allstar games, the managers shared the selection of the players
with the fans. The decision about which players were chosen for the allstar game was shifted
to the manager of each league's pennant-winning team from 1935 to 1946. Since 1947, the
manger has chosen only the pitchers and reserve position players. Fans were allowed to
select the starting players of the allstar game from 1947 to 1957. However, fan balloting
was discontinued after 1957 amidst unilateral ballot-stung on the part of Cincinnati Reds
fans. After this incident, major league managers, players, and coaches made the allstar
player selections. In 1970 fan voting was reestablished and no accusations of unilateral
ballot-stung have resurfaced since.5 Allstar balloting is now used to determine the starting
eld (non-pitcher) players for the annual allstar game between the National and American
5
Leagues in MLB.
Allstar (AS) ballot results are potentially useful in determining if race plays a factor in the
preferences of baseball customers for a number of reasons although, like many data sources,
allstar balloting suers from its own drawbacks. However, we feel that the problems of these
data are outweighed by their benets in the investigation of customer-based discrimination
in professional baseball.
The benets of using data from AS ballots are numerous. First and foremost, AS ballots are the most direct measure of customer preference for various players available to
researchers. Baseball-card prices reect noisy preferences because it is dicult to isolate
and control for supply-side as well as investor inuences on the prices of cards. These omitted variables can cause biased estimates in investigations of customer-based discrimination.
AS ballots, on the other hand, avoid supply-side issues because there are none, beyond the
number of voting fans.
AS ballots are also useful because they are a non-monetary reection of customer preference for the favorite players of the voting fans. There are no monetary benets, future or
present, to selecting a particular player over another, unlike the trading-card market. AS
ballots can thus be viewed as a direct revelation of preference where the only benet from
voting for a favored player is current utility gained if the player plays in the allstar game
that year. Therefore, this revealed preference is not complicated by monetary considerations
nor is it clouded by the supply-side issues that inuence baseball-card prices.
There are some unique aspects of AS ballots which should be illuminated. First, only
the starting players for each team are listed on each year's AS ballot. Therefore, the ballot
reects, ostensibly, the best players in each league at every position. Voting is not restricted
to any particular team or league, since all starting players are listed on every ballot, and
players are only identied by their rst initial, last name, position, team and league. The
6
allstar ballot is completed by punching holes next to the desired player's name, and while
write-in voting is allowed, no player has received enough write-in votes to warrant reporting
by MLB. Because of the manner in which the players are represented on the ballot, fans do
not have a clear indication of player's race when voting. This is unlike trading-cards, where
the race of the player would be readily apparent to the fan. We control for actual race and
also test whether perceived race, as reected in the player's last name, have dierent eects
on fan preference.
III. The Empirical Model and Results
In the end, allstar balloting oers an unique proxy for preferences of baseball fans (who
vote) and allows us to investigate any eects of assumed or actual race on the preferences of
these customers. In the process, we model the preferences of all those who participate in the
voting process, regardless of age, income or other demographic characteristics. Therefore,
any measure of racial bias will more likely reect the entire population of baseball fans,
whereas the use of baseball-card prices of players who played decades earlier may model only
a small subset of baseball fans.
In light of these arguments for using AS ballots as a reection of customer preference, we
estimate a model similar to those used in other studies that focus on the sports-card market.
We estimate a linear stochastic relationship V OTES = f (X; )+, where V OTES is a vector
of allstar votes received, X is a matrix of various player and team-specic characteristics, is a stochastic error term and is a vector of parameters to be estimated.
The list of variables and their descriptive statistics are reported in Table 1. The dependent
variable V OTES is the number of allstar votes received by each player on that year's ballot,
as reported by MLB.6 The independent variables control for individual player attributes,
attributes of a player's team, and time, which is included to control for the strike of 19947
1995 which occurred in the middle of our sample period.7
The player-specic attributes included career oensive average (COFFAV E ),8 currentyear homeruns (HRS ), career elding percentage (CFPCT ), current-year errors (ERR),
and the career games played by a player (GAMES ). We control for ethnicity by using
dummy variables (BLACK and LATINO) which equal one if the player is black (and born
in the U.S.) or the player is latino (born in Latin America), respectively.9 Because of the
special nature of the AS ballot, in which only names of the players are oered to voters, we
also control for last names which have Latin heritage. We estimate a separate model which
includes a dummy variable (LNAME ) that equals one if the player's last name is of LatinAmerican heritage. Using the LNAME dummy variable, we test for any discrimination
based upon the perceived race of a player. We control for name recognition by including
dummy variables which equal one if the player was a previous allstar (PREV AS ), and
whether the player won his respective league's Most Valuable Player award (MV P ) in the
previous year.
The team-specic variables attempt to control for any potential bias in vote totals that
may be enjoyed by some teams over the other. We proxy for attendance by including a
dummy variable which equals one if the player's team played in the previous year's playos
(PLAY ).10 We control for the expansion of the National League in 1993 by two teams
(the Florida Marlins and the Colorado Rockies) with a dummy variable (EXPAN ), because
players on these teams may receive fewer votes than other players.11 Finally, we control for
whether that year's allstar game is played in the player's home stadium (HOMESTAD).
Because our sample covers 1992-1996, it is necessary to control for time in our model.
We do so by including year dummy variables for 1992, 1993, 1995 and 1996. Therefore,
the estimated parameters of the year dummy variables compare to 1994, the year of the
players strike, which cut short the 1994 season and delayed the start of the 1995 season. We
8
would expect a negative impact on votes received after the strike, relative to those received
in 1994, if the strike had an adverse eect on overall MLB attendance. With these time
dummy variables, we proxy for the \supply" of voting fans, which will clearly aect the level
of votes received by players on the ballot.
We estimate our model using OLS on a pooled sample of 942 observations.12 Results of
our estimation are reported in Table 2. We report the estimates of three similar models.
Model 1 includes dummy variables for black and latino players. Model 2 controls only for nonwhite players, and nally, Model 3 includes a dummy variable for Latin-American surnames.
This last model is included because of the potential for perceived-race discrimination because
only last names are oered on the AS ballot. The data are tested for general heteroskedastic
using White's (1980) approach. All three models exhibit some form of heteroskedastic, with
respective White Test statistics (under the null hypothesis of homoskedasticity) of 349.00,
320.43 and 346.98. In light of this, the reported t-statistics in Table 2 have been corrected
using White's (1980) approach.
Our empirical results imply dierent eects of race on customer preference than those
found in most previous studies. Having controlled for player and team characteristics, we nd
no evidence of racial discrimination, based upon the actual race of the players involved (see
Model 1 and Model 2). Further, we nd no evidence of discrimination based upon perceived
race as reected in the last names of players (see Model 3). These results are robust to the
choice of right-hand side variables and the functional form used in the estimation.13 Unlike
other studies, we nd that customer's preferences are not aected by the race of a player,
once individual player and team characteristics have been controlled for.
The remaining estimated parameters have the expected signs and many are statistically
signicant in explaining the number of allstar votes received by players. Player attributes
such as COFFAVE, HRS, and CFPCT each have a positive and statistically signicant eect
9
on allstar votes, implying that high-quality players receive more votes by fans. Players who
have been allstars in previous years receive more votes than other players, ceteris paribus,
indicating that name recognition is important to vote receipts. Likewise, players who won
the MVP award in the previous year garner more votes. These results reinforce the intuition
that name recognition of players is important in receiving allstar votes.
Players who are older receive fewer votes (perhaps indicating some form of age discrimination) but the second-order eects of age are positive. However, players who have more
years in the major leagues receive more votes, with the second-order eects of league tenure
being negative.
Players who were on a playo team in the previous season receive more votes, reecting
both name recognition and the home-team bias that would accompany the higher attendance
rates of more successful teams. Players on expansion teams do not receive fewer votes, nor
do players on for the team hosting the allstar game that year receive more votes.
The estimated coecients on the time dummy variables suggest some rather interesting
implications. All estimated coecients are negative and statistically signicant, indicating
that in each of these years there were fewer allstar votes cast, for any player, relative to 1994,
the year of the player's strike. As one would expect, there was a large drop-o in allstar
balloting in 1995 and 1996, corresponding to reduced voter participation after the strike.
Interestingly enough, however, the coecient on the 1992 dummy variable is also negative.
In total, the estimates of these three models imply that, regardless how race is modeled,
either as actual or perceived, no customer-based discrimination is apparent in MLB allstar
balloting. These results dier from most other studies of customer-based discrimination in
trading-card markets.
In comparison to the other models in the literature, we estimate the model specied
by Anderson and LaCroix (1991), using allstar ballots as the dependent variable instead
10
of baseball-card price. Our qualitative results do not dier from theirs except for the race
variables. In particular, we nd no evidence of racial discrimination using the model specied
by Anderson and LaCroix (1991) when AS votes is the left-hand side variable. This indicates
that (a) there is something unique about the sports memorabilia market in which customerbased discrimination is more evident than in allstar balloting, (b) the omitted supply-side
variables which inuence trading-card prices cause a considerable bias in the estimates, or (c)
baseball-card prices should be modeled across time rather than in a purely cross-sectional
environment. Finally, the results found in other studies may reect the preferences of a
particular segment of baseball fans, perhaps those who are older and have more disposable
income, whereas our data reects the preferences of all voting baseball fans, regardless of
demographics, age and income.
IV. Conclusions
This paper has added to the empirical literature investigating customer-based discrimination by introducing a dierent measure of customer preference. Instead of using the prices
of sports trading-cards, or the wages paid directly to players, we employ a direct revelation of
customer preferences in professional baseball as measured by allstar balloting. Using allstar
ballots of Major League Baseball from 1992-1996, we investigate the eect of a player's race
on the votes received by that player. After controlling for various player and team attributes
we nd no evidence of customer-based discrimination in allstar balloting. Furthermore, in
estimating the models used by other authors, in which they nd a positive and statistically
signicant level of customer-based racial discrimination, we still nd no discrimination in
allstar balloting.
Potential problems in using baseball-card prices include an omitted variables problem
which may bias the estimates obtained such that the null hypothesis of no customer-based
11
discrimination is falsely rejected. On the other hand, there may be some specic attribute
about the trading-card market which causes customer-based discrimination to be more important. This may indeed be the case if the preferences being modeled are those of a very
specic subset of the population which participates in the trading-card markets previously
modeled. Further, replication of functional forms and identical right-hand side variables
indicates no racial discrimination in AS balloting, our data used to measure preference is
distinctly dierent from those used in other studies.
The conclusions presented in this study are robust to various subsamples, independent
variable-sets, and functional forms. This robustness lends credence to the results and gives
reason to reexamine the previous approaches to estimating customer-based discrimination.
Finally, the results presented here are useful in showing that non-monetary-based discrimination does not seem to be reected in the preferences of professional baseball fans.
12
References
Andersen, Torben (1992), \Race Discrimination by Major League Baseball Fans: Evidence
from All-Star Voting Data," unpublished manuscript, Red Deer College.
Anderson, Torben and Sumner J. LaCroix (1991), \Customer Racial Discrimination in
Major League Baseball," Economic Inquiry, 29, pp. 665-677.
Becker, Gary (1975), The Economics of Discrimination, 2nd edition, University of Chicago
Press: Chicago.
Kahn, Lawrence M. (1991a), \Customer Discrimination and Armative Action," Economic
Inquiry, 29, pp. 555-571.
Kahn, Lawrence M. (1991b), \Discrimination in Professional Sports: A Survey of the Literature," Industrial and Labor Relations Review, 44, pp. 395-418.
Kahn, Lawrence M. and Peter D. Sherer (1988), \Racial Dierences in Professional Basketball Players' Compensation," Journal of Labor Economics, 6, pp. 40-61.
Nardinelli, Clark and Curtis Simon (1990), \Customer Racial Discrimination in the Market
for Memorabilia: The Case of Baseball," Quarterly Journal of Economics, 105, pp.
575-596.
Schollaert, Paul T. and Donald Hugh Smith (1987), \Team Racial Composition and Sports
Attendance," Sociological Quarterly, 28, pp. 71-87.
Scully, Gerald W. (1974a), \Discrimination: The Case of Baseball," in R. Noll, ed., Government and the Sports Business, Brookings: Washington, D.C., pp. 221-273.
Scully, Gerald W. (1974b), \Pay and Performance in Major League Baseball," The American Economic Review, 64, pp. 915-930.
13
Scott, Frank A., Jr., James E. Long and Ken Somppi (1985), \Salary vs. Marginal Revenue
Product Under Monopsony and Competition: The Case of Professional Basketball,"
The Atlantic Economic Journal, 13, pp. 50-59.
Sommers, Paul M. and Noel Quinton (1982), \Pay and Performance in Major League
Baseball: The Case of the First Family of Free Agents," Journal of Human Resources,
17, pp. 426-436.
Stone, Eric W. and Ronald S. Warren, Jr. (1996), \Customer Discrimination in Professional Basketball: Evidence from the Trading-Card Market," Unpublished Manuscript,
University of Georgia.
White, H. (1980), \A Heteroscedasticity-Consistent Covariance Matrix Estimator and Direct Test for Heteroscedasticity," Econometrica, 48, pp. 817-838.
14
Notes
Here we include such problems as specication error, omitted variables, and data mismeasurement as potential sources of econometric problems. Any or all of these would lead
to estimation results which are suspect.
1
2
See Kahn (1991b) for a survey of literature on discrimination existing up to that point.
The prices of trading-cards are taken at one particular time, whereas the cards were
actually produced several years before. The prices thus reect \equilibrium" prices, obtained
from price guides. No study has controlled for the number of cards in existence at the time
of production, nor at the time the prices were obtained. This is only one potential problem
with using trading-card prices.
3
There was no allstar game held in 1945 because of war-time conditions. There were two
allstar games played each year from 1959 through 1962.
4
It is basically understood that all team-fans stu the ballot box for their given players.
The hope is that these ballot stung eorts even out for the most part.
5
The allstar balloting results are published each year in most newspapers. Our data
were collected from various issues of the Dallas Morning News. In every year except 1993,
vote totals for all players on the ballot were reported. In 1993, Major League Baseball only
reported the top eight vote totals for each position.
6
Player statistics were obtained from various sources, including Total Baseball (1994),
Complete Baseball, (1994 and 1995), and various mass-produced baseball cards. Data on
MVP awards, previous allstar selection and home stadia were obtained from Major League
Baseball at http://www.majorleaguebaseball.com.
7
15
8
Career oensive average is calculated as
+ 3 TRIPLES + 4 HRS + BB + SB (1)
COFFAV E = 1 SINGLES + 2 DOUBLESAB
+ BB
and is arguably a better measure of oensive performance than either batting average or
slugging percentage. See Anderson and LaCroix (1991) for a further discussion of this statistic.
9
We determine latino ethnicity based upon place of birth, skin-color notwithstanding.
Of course, one may suspect that these variables may also somewhat proxy for playerrecognition.
10
Because the majority of the starting players on the expansion teams were obtained in
the expansion draft (in which incumbent teams are allowed to \protect" certain key players
while other, ostensibly lesser quality, players are left \unprotected"), one might expect these
lesser quality players on the ballot to receive fewer votes.
11
The observations not included are rookie players (i.e., those players in their rst year in
MLB) listed on a given year's AS ballot. These observations are dropped because no career
statistics are available.
12
In particular, we also included interactions between race and productivity statistics and
between race and year dummy variables. The estimated coecients on these variables were
not statistically signicant.
13
16
Variable
VOTES
BLACK
LATINO
NWHITE
LNAME
COFFAVE
HRS
CFPCT
ERR
GAMES
AGE
AGE2
YRS
YRS2
MVP
Denition
Mean Std. Dev.
All-star votes (in thousands)
522.00 584.37
Equals 1 if Black, 0 otherwise
0.29
0.45
Equals 1 if Latino, 0 otherwise
0.18
0.38
Equals 1 if Nonwhite, 0 otherwise
0.47
0.50
Equals 1 if Latino name, 0 otherwise
0.20
0.40
Career oensive average
0.49
0.07
Current season home runs
12.80
10.94
Career elding percentage
0.96
0.03
Current season errors
7.83
6.10
Career games played
829.33 555.03
Age of player
29.39
3.80
Age squared
878.37 232.23
Years player has played
7.66
4.06
Years squared
75.11
78.30
Equals 1 if player was previous MVP,
0.01
0.10
0 otherwise
PREVAS
Equals 1 if player was a previous All-star,
0.20
0.40
0 otherwise
PLAY
Equals 1 if player played in previous playos,
0.12
0.33
0 otherwise
HOMESTAD Equals 1 if AS game played in player's home stadium, 0.03
0.16
0 otherwise
EXPAN
Equals 1 if player is on an expansion team,
0.06
0.23
0 otherwise
YR92
Equals 1 if year is 1992, 0 otherwise
0.21
0.41
YR93
Equals 1 if year is 1993, 0 otherwise
0.12
0.32
YR95
Equals 1 if year is 1995, 0 otherwise
0.23
0.42
YR96
Equals 1 if year is 1996, 0 otherwise
0.23
0.42
Table 1: Variable Names, Denitions, Means, and Standard Deviations.
Variable
BLACK
NWHITE
Model 1
-4.233
(-0.12)
-3.712
(-0.11)
|
LNAME
|
LATINO
Model 2
|
Model 3
|
|
|
-4.019
(-0.14)
|
|
-41.034
(-1.20)
INTERCEPT 5808.190 5809.090 5957.110
(4.47)
(4.53)
(4.49)
COFFAVE
449.928 449.360 424.627
(1.99)
(2.04)
(1.97)
HRS
8.854
8.854
8.922
(4.82)
(4.87)
(4.92)
CFPCT
498.109 498.011 483.600
(1.77)
(1.77)
(1.70)
ERR
1.474
1.476
1.426
(0.64)
(0.64)
(0.62)
GAMES
0.753
0.753
0.751
(5.99)
(5.99)
(6.02)
AGE
-399.306 -399.338 -406.274
(-4.34)
(-4.38)
(-4.33)
AGE2
5.950
5.950
6.031
(3.82)
(3.83)
(3.82)
YRS
77.328 77.357 80.758
(3.36)
(3.43)
(3.48)
YRS2
-7.104
-7.104
-7.202
(-5.56)
(-5.62)
(-5.61)
MVP
644.642 644.522 635.529
(2.09)
(2.10)
(2.07)
PREVAS
289.932 289.956 291.412
(5.39)
(5.43)
(5.44)
PLAY
248.679 248.676 247.773
(4.79)
(4.80)
(4.76)
HOMESTAD -62.600 -62.5627 -62.466
(-0.74)
(-0.74)
(-0.74)
EXPAN
-52.675 -52.666 -51.171
(-1.14)
(-1.14)
(-1.11)
YR92
-430.179 -430.192 -431.658
(-7.38)
(-7.40)
(-7.40)
YR93
-184.773 -184.789 -186.292
(-2.76)
(-2.77)
(-2.79)
YR95
-503.478 -503.483 -504.179
(-9.13)
(-9.15)
(-9.14)
YR96
-557.560 -557.564 -557.994
(-9.65)
(-9.67)
(-9.67)
R2
0.457
0.457
0.457
F
38.803
40.890
41.009
N
942
942
942
Table 2: Regression Results (Asymptotic t-Statistics reported in parentheses)
indicates signicance at the 0.05 level in a two-tailed test.
19