Jack Costello MSc

UNIVERSITY OF SURREY

“English League transfer prices

revisited.”

By

Jack Costello

Economics MSc

Supervisor: Doctor Robert Witt

1
Jack Costello MSc

2008

i. Abstract.

The purpose of this research is to examine the role of race in the determination
of association football transfer prices. This looks at whether black player’s vis-à-
vis trade at less than their non-black equivalents using Of Least Squares method
estimation. After running a regression and testing for heteroskedasticity (using
robust-standard errors) and the functional form our empirical results suggest
there is limited evidence that black players are unequally treated concerning the
transfer prices they trade at. If anything the coefficient found suggests that black
players ceteris paribus trade 23.6 % higher than non-blacks. The regressor is
found to be not statistically significant. Other regressors are found to be
statistically significant when looking at the log transfer price. All explanatory
variables involved attribute to the 0.517 R-squared value attached to the
explained variable; log transfer price.

2
Jack Costello MSc

ii. Acknowledgements.

I would like to send out special thanks to the following people:

• My tutor, Dr Robert Witt, who has given me his time, support and
guidance.
• My friends and family, who help me remain focused and
determined to succeed.

I can confirm that this report is entirely my own work and all materials
used are accurately quoted and detailed in the reference list.

____________________________________

JACK COSTELLO

3
Jack Costello MSc

iii. Contents.
i. Abstract 2

ii. Acknowledgments 3

iii. Contents 4

1. Introduction 5

2. Literature Review 8
Reilly and Witt (1995) 8
Kahn (1993) 9
Medcalfe (2007) 11

3. Data 13
Diagram I 13
Table I 14
Diagram II 15
Diagram III 16
Diagram IV 17
Diagram V 18
Diagram VI 19
Diagram VII 20
Diagram VIII (I) 21
Data Collection 22

4. Empirical results 25
Table II 26
Diagram IX 28
Diagram X 30

Diagram VII (II) 31

5. Conclusion 33

Appendix I 35

4
Jack Costello MSc

Appendix II 35
Appendix III 36
Appendix IV 36

References 38
1. Introduction.

In a relatively recent paper Reilly and Witt (1995) examined English

football association transfers looking at whether black players were
discriminated against when referring to transfer fees in the English
football leagues. In this paper several variables were analysed as to
empirically suggest “there is little evidence of the unequal treatment of
black players in terms of the transfer prices they command.” (Reilly, B
and Witt, R. “English league transfer prices is there a racial dimension?”
Applied Economics Letters, volume 2, pp 220-222.) This paper looked at
the 1991-1992 season; the penultimate season before the introduction of
the English Premier League. The aim of this research paper is to re-
examine Reilly and Witt’s (1995) study. This thesis instead looks at the
two seasons 2005-06 and 2006-07 as to examine whether black players
are racially discriminated against when referring to reported transfer fees
and in reference to the players’ individual characteristics. Studies as of
such are still trivial in comparison to our North American counterparts
and the vastness of the work they have undertaken. This is apparent in
the study by Kahn, L. M “Discrimination in professional sports: a survey
of the literature”, Industrial and Labor Relations Review, volume 44,
1991, pp 395-418 – which acts as a review of discrimination in sports.

Through player characteristics and data which is collectable we are able

to create a log transfer price equation. A group of regressors will allow us
to analyse the role of race on the transfer price – as well of the roles of
other variables that may be significant in the model. Transfer fees allow
us a route to be able to investigate if there is discrimination in the sport. If
ceteris paribus black players trade at a value below that of their non-
black equivalents there may be evidence which should be carefully used
to suggest that discrimination is present in English football or that it
“could not be ruled out.” (Reilly, B and Witt, R. “English league transfer
prices is there a racial dimension?” Applied Economics Letters, volume

5
Jack Costello MSc

2, pp 220-222). Player’s wages may be more suitable when looking at

discrimination in English football. Unfortunately this data is currently
unavailable – it is not completely accessible meaning wages/salaries
cannot be addressed in this study. They are certainly food for thought
none the less.

Evidently as in any competitive market, the English Football League is

dynamic. So despite us throughout being able to refer to relevant papers
it must be remembered that the sport has changed and therefore so will
the variables associated with the log transfer price. For example the
arrival and notoriety of Bosman transfers (named after John Bosman,
introduced in 1995) where the European Court of Justice ruled that
“transfer fees for out of contract players were illegal where a player is
moving between one EU nation and another. From now on only players
still serving contracts with their teams could have transfer fees paid for
them.” (Football Industry Group [Link]/footballindustry/bosman)
This is one example of the ever changing transfer market which may
affect the transfer prices of players. After looking at the transfers within
this period it can be demonstrated that more are undisclosed; which may
be Clubs trying to protect their financial status and the value of player in-
case they intend to sell the player on.

The study on sport generally in this context may be particularly useful as

“many of these studies include far more extensive controls for individual
ability and performance than typical studies of discrimination that use
labor force data.” (Reilly, B and Witt, R. “English league transfer prices is
there a racial dimension?” Applied Economics Letters, volume 2, pp 220-
222.) This study may have particular relevance to the labour market and
may even be more reliable due to the extensive data being collected on
sport. Kahn (1993) nicely explains that the use of player characteristics
allows for the testing of discrimination which may not be possible in
many labour markets. Despite there being relatively extensive research
on discrimination within North American sports the research in the
English football League is limited and open to review. This is particularly
relevant in response to the ever-expanding level of player analysis and
statistical evidence available. This allows us to conduct the research to
the same critical level as those papers constructed on American sports in

6
Jack Costello MSc

studying discrimination. English sports are likely to be investigated

further; as at the moment the research is fairly diminutive.

The remainder of the paper looks at whether black players are

discriminated against when looking at log football association football
transfer prices. If black players trade at values below their non-black
equivalents than there is the possibility of discrimination being present.
The paper also looks at the prominence of other variables affecting the
regressand and therefore their statistical significance. Section II gives a
brief overview of the literature at hand – analysis and theories on
discrimination. Section three titled “Data” looks at the basic summary
statistics – including data collection methods. Chapter V is the key –
which looks at the empirical results, drawing out our findings and leading
us onto our conclusions on the study.

7
Jack Costello MSc

2. Literature Review.

Reilly and Witt (1995)

Reilly and Witt’s paper “English League transfer prices: is there a racial
dimension?” (Applied Economics Letters, volume 2, 1995, pp 220-222) is
of paramount importance to the current paper. Using the 1991-1992
season their aim is to establish if black players trade ceteris paribus at a
lower rate than their non-black counterparts; which if found to be true
implies discrimination cannot be ruled out in the English football league.
This thesis is an update of their paper – and covers two seasons, as
opposed to one season in Reilly and Witt’s (1995). A research design is
set up – which helps to show the regressand (log transfer price equation)
on the left hand side of the econometrically based equation. A list of
regressors on the right hand side of the equation aims to explain the
value of a transfer. Productivity measures are used to formulate the
equation; with the coefficient and the standard error reported in
parenthesis. 15.3 % of the players in the study are found to be black;
averages of each of the variables help to give a basic gauge of the data
for interpretation. The average fee agreed is £348,350; with the 38.1 %
of players in the study being strikers and 18.8 % of the players full
internationals. The paper helps to show that with all else being equal
blacks trade 9 % below that of a non-black. Despite this the black
coefficient is found to be not statistically significant at the 5 % level. A t-
value of 0.75 is less than the 1.97 considered necessary to be significant
at the conventional level. The paper is cautious in its findings, and shows
that despite no discrimination being found in this form fails to rule out
other forms of discrimination all together; and heeds the future research
into player’s wages as potentially being a greater indication.

8
Jack Costello MSc

Despite the theory of discrimination being rejected; there are other

factors which are found to be significant in the paper; including
appearances last season, goals in the current season, each of the age
bands, striker, full international and many of the league status’ of the
buying and the selling clubs respectively. These contribute to the 74.7 %
of the data which is represented in explaining the log transfer price
equation. It is similar to this thesis in its numeracy of data collected; the
number of transfers indicated in the valid period is 202; 31 are black. In
comparison the current study has 230 transfers; of which 155 are white;
36 mixed race; and 39 black. This paper will act as the lynch-pin for the
current study. Despite this it is imperative to understand that the market
is dynamic and that even though the structure of the race of the player’s
are not too dissimilar we should take caution when interpreting the
differences in the datasets.

Kahn (1993)

Kahns “Discrimination in professional sports: a survey of the literature”

(Industrial and Labor Relations Review, volume 44, 1991, pp 395-418)
acts as a review of other papers relating to racial and ethnic
discrimination. It also looks at gender discrimination within sports
(tennis). The study suggests sports may be useful in other markets when
considering discrimination; they have facilities or statistics that can be
effectively analysed to test to test for its presence (as already stated).
The paper also looks at something that unfortunately this thesis was
unable to collate and analyse when determining discrimination in the
English football association. Can wages help to determine the presence
of discrimination? The paper is particularly extensive – looking at a range
of sports – some which indicate the presence of discrimination and
others which do not. Throughout comparisons are made using dummy
variables allowing statistical inferences to be made; which will be
identical methodology as used in this paper. Salary discrimination is
obviously one method which can be used as to test for discrimination
within a sport. The majority of the papers find that there is no
discrimination – that the race variable is insignificant at the conventional
level.

9
Jack Costello MSc

Despite this there are examples found which do suggest evidence of

discrimination. In baseball as an example; it was found that Latin non-
pitchers receive 42 % less than non-Latin pitchers (significant) and that
white pitchers receive ceteris paribus “13-25 %” less than there non-
white counterparts (Pascal, A and Rapping, L. “The economics of racial
discrimination in organized baseball.” Racial discrimination in economic
life, pp 119-56). Another example, against white pitchers finds a “4.6 %
discrimination coefficient against whites (significant at 10 %, 2-tailed).”
(Christiano, K. “Salary discrimination in major league baseball: the effect
of race.” Sociology of Sports Journal, Vol 3, No 2(June) pp 144-53). This
shows that each race may face discrimination and also why t-tests must
be two-tailed. In the NBA players tested using data from 1984-85 and
1985-86 season found significant discrimination coefficients “ranging
from 11% to 25% against black players.” (Rockwood, C and Asher E.
“Racial discrimination in professional basketball revisited.” American
Economist, 1976, Volume 20, No 1 (Spring) pp 59-64)

Hiring discrimination is also scrutinized with more mixed evidence being

found. This applies to those in a hiring position discriminating against
players due to their race. Another theory called positional segregation is
brought forward (black players having certain positions in sports). This
theory is similar to that of a higher proportion of black players playing as
a striker. Customer discrimination is suggested to exist. One example
states that “prejudiced white fans’ utility may depend on both winning and
the presence of white players.” (Kahn, L. M “Discrimination in
professional sports: a survey of the literature”, Industrial and Labor
Relations Review, volume 44, 1991, pp 395-418) Another study found
that “1980-86 data did find that attendance was negatively affected by
the black players, all else equal.” (Kahn, L. M “Discrimination in
professional sports: a survey of the literature”, Industrial and Labor
Relations Review, volume 44, 1991, pp 395-418) The evidence is wide-
ranging - there are examples of this being statistically significant and also
examples which reject this null hypothesis of the presence of
discrimination. The evidence suggests this varies not only from year-to-
year but also from sport-to-sport.

10
Jack Costello MSc

The analysis of the journals within the paper on discrimination from our
American counterparts helps to demonstrate the vastness of their studies
into the subject, in comparison to the negligible research that has been
undertaken in the United Kingdom; especially on the English Football
League’s. The journal is cautious in its conclusions. Interestingly the
paper also mentions the carefulness that must be taken which applies to
each of the studies and states correlation with standard errors and race
would make regressions invalid. This is certainly worth considering for
the current study and is accordingly tested. The review is valuable in
giving different a range of studies on discrimination with evidence
showing strong and weak examples as well as the potential flaws that
can be faced. The theories offer a range of ideas which may be the
cause of discrimination faced; from salary discrimination to hiring
discrimination. The paper is certainly insightful and broad in its analysis
and theoretical support. In conclusion there are examples of
discrimination throughout the sports studies, and also examples where
the concept of discrimination is rejected. The review does give a real
indication of the bulk of research undertaken in North American sports
and the limited amount in the United Kingdom.

Medcalfe (2007)

Another paper of merit is that of Medcalfe with “English league transfer

prices: is there a racial dimension? A re-examination with new data.”
(Applied Economic Letters, Online 1 August 2007) This is an update of
the work of Reilly and Witt (1995) using data from the 2001-2002
season; a decade on. The journal aims to surpass and evolve the study
by Reilly and Witt (1995) with its inclusion of extra productivity measures
being available. The data collected shows an increase in the average
price up-to £2,934 (in £000’s) with a crucial increase in the proportion of
black players in the study up-to 45 %. Black players are shown to trade
ceteris paribus at a value of £633,000 less than their non-black
counterparts. Medcalfe concludes that there is no discrimination found.

The inclusion of the extra variables is represented by assists, pass, cross

completion, dribble completion, and tackles in the table. They are likely
to be the cause in the increase of the R-squared value which is 0.94.

11
Jack Costello MSc

Unfortunately the paper certainly has flaws. This includes its very limited
sample size in comparison to that of Reilly and Witt (1995). The study
uses the 2001-2002 season using a transfer price equation to establish if
equally productive black players command lower transfer prices. The
paper attempts to evolve the analysis using the Opta Football Yearbook
by adding extra productivity measures. This results in a particularly
impressive R-squared of 0.94. The dataset is limited by its minimal
number of observations as the Opta Football Yearbook which only deals
with Premier League affairs. This makes the analysis extremely limited
and open to criticism. Witt and Reilly’s (1995) paper uses 202
observations in comparison. Medcalfe (2007) only uses 29 observations.
Furthermore it must be considered that when looking at the explanatory
variables that only Premiership players are included; which is reflected
as 62 % of the dataset are full internationals. This certainly needs
thoughtfulness when making direct comparisons between the studies.
The analysis suggests black players command £632,977 less than non-
black players. In this sense the rejection of statistical significance is
supported by the findings of Reilly and Witt (1995) and this thesis; which
both find there is no discrimination in terms of the transfer fee. According
to each study “black players do not suffer from discrimination in transfer
prices.” (Medcalfe, S. “English league transfer prices: is there a racial
dimension? A re-examination with new data”, Applied Economic Letters,
Online 1 August 2007.)

12
Jack Costello MSc

3. Data.

The largest proportion of ethnicity is represented by white players (67.4

% or 155 players) followed by black players (17.0 % or 39 players) in
close proximity to mixed race players (15.7 % or 36 players) in the total
sample size of 230 players. Diagram I helps to illustrate the proportions
of races in the study – obviously showing white as the largest race
represented.

Diagram I. The proportion of each race represented in total sample.

White
Mixed Race
Black

White Mixed-Race Black

% of study 67.4 15.7 17.0

Below table I helps to demonstrate the means of the key variables, and
their definitions.

13
Jack Costello MSc

Table I. Means and Definitions of the variables.

Variable Mean Definition

the agreed price paid for a player in £
Price 1729
thousands.
Height 71.49 Height in inches of player.
Weight 168 weight in lb’s of player.
the number of league appearances in
Apps last season 25.91
2005/2006 season.
the number of league appearances in
Apps this season 28.67
2006/2007 season.
the number of league goals scored by in
Goals last season 3.996
2005/2006 season.
Goals current the number of league goals scored by in
4.448
season 2006/2007 season.
number of years old player is at time of
Age 25.4
transfer.
a binary variable adopting a value of 1 is the
Age<20 0.06 player is aged less than 20 years; otherwise a
value of zero is adopted.
a binary variable adopting a value of 1 is the
Age> 20 and < 25 0.443 player is aged between 20 and 25; otherwise
a value of zero is adopted.
a binary variable adopting a value of 1 is the
Age> 25 and < 30 0.396 player is aged between 26 and 30; otherwise
a value of zero is adopted.
a binary variable adopting a value of 1 is the
Age > 30 0.101 player is aged over 30 years; otherwise a
value of zero is adopted.
a binary variable adopting a value of 1 is the
Striker 0.365 player is a striker; otherwise a value of zero
is adopted.
a binary variable adopting a value of 1 is the
Full International 0.396 player is a full international; otherwise a
value of zero is adopted.
Number of previous
2.51 number of previous clubs.
clubs
Number of years at
3.19 number of years at previous club.
previous club
a binary variable adopting a value of 1 is the
White 0.674 player is white; otherwise a value of zero is
adopted.

14
Jack Costello MSc

a binary variable adopting a value of 1 is the

Mixed Race 0.157 player is mixed-race; otherwise a value of
zero is adopted.
a binary variable adopting a value of 1 is the
Black 0.170 player is black; otherwise a value of zero is
adopted.

We can see that the average transfer fee for a player in the study is
£1728 (in £000’s) which is higher than Reilly and Witt’s (1995) study and
less than Mecalfe’s (2007) This can be shown diagrammatically.

Diagram II. Differences in average fees between each of the three

studies.

Medcalfe (2007) 2934

Study

Reilly and Witt(1995) 348

Costello (2008) 1728

0 500 1000 1500 2000 2500 3000 3500

Price (in £000's)

Reilly and
Costello (2008) Medcalfe (2007)
Witt(1995)

Price (in £000's) 1728 348 2934

Not too much should be read into this – obviously several factors
influence each individual study. For example the study by Medcalfe is
only represented by Premiership transfers which are associated with a
higher transfer fee Obviously the dates the information was collected will
also play a role. Inflation and transfer inflation have occurred between
each study. In this study black players on average command over
£2,000,000 in transfer fees. This is higher than the other races average

15
Jack Costello MSc

fee in the study. Mixed race and white players trade on average below
the £2,000,000 fee.

Diagram III. Transfer price of each race and the average of sample
population.

2500

2000

1500

Price

1000

500

0
Mean Mean White Mean Mixed Mean Black

Mean Mean Mean

Mean
White Mixed Black
Price 1729 1662 1668 2042

This is an early indication that there is no discrimination in the market.

The evidence seems to suggest that black players on the whole are
trading at higher values. Caution must be taken when making statements
of this variety – there may be other factors affecting the size of the fee
(investigated later). Diagram IV illustrates a pattern which seems to be
relatively simultaneously occurring through the dataset. By putting each
transfer in ascending order due to the size of the transfer fee we can look
at the movements in each races curve relative to each other. The three
curves rarely divulge away from another race which suggests that the
fees are not dissimilar from each other. In other words the diagram is
supportive that there is no discrimination within the market. If anything

16
Jack Costello MSc

than for a large proportion of the diagram the black players have the
highest value of the transfer. This insinuates there may be a case to
answer for non-black players being discriminated against; which has
been seen before – in Christiano (1986). In this study it is put forward
that white pitchers in baseball are discriminated against. This also helps
explain why we must use two-tailed t-tests in the study.

Diagram IV. All data each races transfer fee in ascending order.

20,000
White
18,000 Mixed
Black
16,000

14,000
Transfer value (in £000's)

12,000

10,000

8,000

6,000

4,000

2,000

0
1 6 11 16 21 26 31 36
Player's value in ascending order

Interestingly the dataset clearly exhibits two outliers; the transfers of

Darren Bent (black) from Charlton Athletic to Tottenham Hotspurs for
£16,500,000 and Michael Carrick (white) from Tottenham Hotspurs to
Manchester United for £18,600,000. A clearer representation can be
shown when these outliers are removed. Once more the correlation
between the white; mixed race; and black players seems to be relatively
close and suggests no discrimination in the market to the naked eye
(empirical tests etc later in paper).

17
Jack Costello MSc

Diagram V. Each races transfer fee in ascending order – excluding

outliers.

8,000
White
Mixed
7,000
Black

6,000
Transfer value (in £000's)

5,000

4,000

3,000

2,000

1,000

0
1 6 11 16 21 26 31 36
Player's value in ascending order

The average height of a player is 71.49 inches and weight 168 pounds
which is generally consistent across the races (Appendix I, II, and III)
Both league appearances and league goals in the penultimate season
are less than the current season. The former from 25.91 being increased
to 28.67 and the latter from 3.996 to 4.448. You would expect the number
of league appearances and the transfer fee to work in a positive direction
– so that the more games played in the previous season the greater
value of the player. This is supported by supported by each of the
variables carrying a positive coefficient found in the empirical results.
This is fairly intuitive when considering an employee prefers to get the
most out of a player. In the workplace this could be aligned to an
employer preferring the employee to be at work as many days as
possible. The Club gets more out of the players who play the most
games – they have the most value added.

18
Jack Costello MSc

A large proportion of the players are full internationals; represented by

two out of 5 players with the average player earning 8.5 caps. Over half
of the black players in the study (53.9 %) are full internationals which is
higher than both mixed race and the white race players – which may be
one of the indications as to why they trade at a higher value.

Diagram VI. Comparison of players in each race who represent their

country respectively.

Full International
60

50

40
% of players

30

20

10

0
Mean Mean White Mean Mixed Mean Black
Race

Mean White Mixed Black

Full International 39.6 36.8 38.9 53.9

The diagram shows that black players are represented by the greatest
number of international players proportionately. This is approximately 15
% greater than the mixed race and 17 % compared to the white players
in the study. Similar to the analysis of signalling and screening in the
employment market – an international career acts as a signal of quality
to potential employers – the Club. This is shown as international players
on average trade at £2,530 and non-international player’s trade at a
value of £1,203 (both in £000’s) (Diagram VII). This may be largely
contributory towards black players trading at a higher value than their
non-black counterparts. It is intuitive to believe that with the black race
having more internationals – this may be contributing to their higher
associated transfer fees when comparing the white and mixed-race
players.

19
Jack Costello MSc

Diagram VII. Average values of international players versus non-

internationals.

Average Value (in £000's)

Non -
Average value (in £000's)

1,203
International

International 2,530

International status

The average age of a player in the sample is 25.4 years old ranging from
a minimum of 17 years to a maximum of 38 years. The age group most
represented in the sample is 20 years old up-to and including 25 years
old players representing 44.3 % of the sample, with 26 to 30 year olds
representing 39.6 %. The two opposite age bands in the sample, the
under 20’s and over 30’s represent 6 % and 10.1 % respectively. It is
also worth noting there are no mixed race players in the under 20s
category compared to 12 white race transfers. This can be investigated
further when looking at the data econometrically and may show to have
relevance when looking at the statistical significance of the variables.
Currently we cannot read too much into this – but upon establishing the
OLS estimation we shall be able to delve deeper. We approach this in
further detail in “econometric results.”

The average number of previous clubs is 2.51 with 3.19 years spent at
this club. Interestingly on average black players have more previous
clubs and spend less time at this club which suggests they are more like

20
Jack Costello MSc

to be transferred ceteris paribus. In theory this may imply that they are in
higher demand – once more this can be analysed when looking at the
inferences in the empirical results. Just over one in three of the players in
the study are strikers (as a hunch we expect strikers to command a
higher fee). It’s worth noting that with the black players its closer to half
of the sample; 18 out of the 39 are strikers (which may contribute to the
higher transfer fee associated with black players). This is in comparison
to white players who contribute to just under a third at 51 out of 155
players in the study. Strikers as expected are found to trade at higher
values than there non-striking equivalents which can be found in the
empirical results section (associated with a positive coefficient). Mixed
race players also proportionality have less strikers compared to looking
at the black-players. This may act as another link as to why black players
on average seem to trade at higher prices.

Diagram VIII. Proportions of each race who are strikers.

60

50
% of full internationals

40

30

20

10

0
Mean Mean White Mean Mixed Mean Black
Race

Mean Mean Mean

Mean
White Mixed Black
Full International 39.6 36.8 38.9 53.9

This conveniently moves us onto Kahn’s (1993) theory of the positional

segregation. This is where individual races are picked to play in certain
positions. In this example it is likely to be black players in the forward
position. This is also aligned with black players scoring more in their

21
Jack Costello MSc

current season and previous season in comparison to the white race –

which is to be expected with more playing in the centre forward position
(Appendix I, II, and III) This may also be further allied to less black
players playing in those positions exerting “leadership and critical
thinking.” (Kahn, L. M “Discrimination in professional sports: a survey of
the literature”, Industrial and Labor Relations Review, volume 44, 1991,
pp 395-418) Much research has been undertaken on this matter
including on baseball, basketball, American football and hockey. A great
example is the study of Madison and Landers (1976) which helped to
show “blacks are more likely than whites to move from central to non-
central positions from college to pro.” (Madison, D and Landers D.
“Racial Discrimination in Football: A Test of the “stacking” playing
positions hypothesis.” Social Problems in Athletics; Essays on the
Sociology of Sports; pp 151-56). In our example this could be explained
as why black players play as forwards due to its non-central role. Once
more caution should be taken when making this statement as it is
certainly twinned with other reasons.

The division of the buying and selling clubs has played a substantial role
when looking at previous papers. It is certainly worth looking at this. This
is looked at in further detail in the “empirical results” section. The
summary statistics are useful in working out the data we are working
with. Simple analysis such as the mean, maximum, minimum, and
standard deviation from each race (Appendix I, II, and III) help us to get
an ambience of the data and interpret some of the differences amongst
the race variables. So far it is difficult to suggest discrimination in the
English football association and the racial dimension with caution would
be rejected.

Data collection.

The data that is collected in this study allows us to make empirical

estimations as to evaluate the factors (explanatory variables) which
contribute to the log transfer price equation and their significance in the
model. The data used in this study is taken from the Sky Sports Football
Yearbook 2006-2007 for season I (the 2005-2006 season) and the Sky
Sports Football Yearbook 2007-2008 for season II (the 2006-2007

22
Jack Costello MSc

season). This was accompanied by the use of the PFA Footballers’

Who’s Who for the players’ characteristics which were collected. The
transfers themselves are collected from each of the Sky Sports Football
Handbooks.

Instead of one seasons transfers being collected and analysed it was

decided to use two; as this gave a more satisfactory amount of
observations suitable for analysis (230 valid transfers in the sample).
After each transfer was stated as valid each of the player’s personal
characteristics and statistics were collected accordingly. The validity of
each transfer is of crucial importance to the study. Transfers must be
from any of the 92 to English Football League Clubs to one of the other
92 Football English League Clubs. Transfers including lower Leagues,
teams from Scotland and abroad are excluded. Additionally this only
involves transfers that have fees published in the Skysports Football
Yearbook within the designated time frame, and therefore does not
include those transfers in the book outside this time frame. For the time
of the season outside the time frame of the Skysports handbook the
study uses the Guardian newspaper and its publishing’s for consistency.
The dates between which the transfers are valid are 19 th August 2005
until the 10th August 2007 (including the close season). Dates before and
after this are not under consideration and are therefore considered
invalid. Free transfers are invalid in the study; they will offer bias as they
do not represent the true value of the player (including Bosman
transfers). Nominal transfers are invalid as they do not carry an official
value. Undisclosed transfer fees are also invalid as they do not carry an
official fee. Furthermore loan transfers are excluded. This is unfortunate
as many of these transfers have a general consensus on the reported
fee which cannot be used due to the bias which would be generated.

The player’s characteristics – for example age, height, weight, position,

number of previous clubs, and numbers of years at previous clubs were
collected from the PFA’s footballer’s who’s who. The Skysports football
yearbook for its corresponding year was used to collect productivity
measures, such as appearances for the current and previous season,
and the goals scored in the current and previous season respectively.
Furthermore this handbook was used to ascertain the international status

23
Jack Costello MSc

of the player and the number of international caps held. Finally the race
variable – which can either take white race, mixed race, or black in race,
is taken from the evidence provided on Clubs individual websites. A
dummy variable is used once more to allow for inferences and empirical
results from the race variable. The decision not to evolve the data; such
as done by Medcalfe (2007) was decided against; largely due to the
limiting affect on the amount of observations which could be collected as
a result.

24
Jack Costello MSc

Empirical Results.

The econometric results are of paramount to the thesis on the whole;

and effectively answer the question when looking at whether there is a
racial dimension when looking at English football association transfers.
The final regression will also allow us to see which variables affect the
transfer price of a player by looking at the significance of the variables
when contributing to the log transfer price of the player. We can analyse
each variable and its significance to the transfer fee. This is what the
remainder of the chapter aims to finalize. The summary section as below
gives the variable coefficients at its standard errors in parenthesis along
with reported diagnostics.

25
Jack Costello MSc

Table II. Summary of the log transfer price regression.

Constant 6.380(0.409)
Appearances last season 0.010(0.007)
Appearances current season 0.003(0.008)
Goals last season 0.001(0.021)
Goals current season 0.008(0.019)
Age < 20 0.388(0.398)
Age > 20 but < 25 0.761(0.241)*
Age > 25 but < 30 0.580(0.271)*
Striker 0.390(0.195)*
Full International 0.303(0.148)*
Number of previous clubs 0.045(0.060)

Division of selling club

Championship 0.072(0.156)*
League One 0.236(0.220)
League Two 2.926(0.445)*

Division of buying club

Championship 1.073(0.278)*
League One 1.123(0.252)*
League Two 1.866(0.156)*

Black 0.236(0.18)

Ramsey Reset Model 0.288

Goldfeld-Quant Test 2.341*
Breusch-Pagan ϰ217 6.011*
R-squared 0.517
Observations 230

Coefficients with standard errors stated in parenthesis.

* denotes those coefficients which are statistically significant at the 5 %
level.

26
Jack Costello MSc

Heteroskedasticity is tested for in the regression to allow us to make

suitable inferences. We use two tests: the Goldfeld-Quant test and the
Breusch Pagan statistic. Both suggest that the regression suffers from
heteroskedasticity. As a result we use robust-standard errors which are
stated in parenthesis. This allows us to use standard t-tests without any
fear of the inferences that are made suffering from bias.
Heteroskedasticity is where the error variances across the terms are not
constant. The Goldfeld Quandt test works by ordering the observations in
numerical order and than splitting the sample into three. In our sample
the middle 80 observations are removed. Through comparing the sum of
the squared residuals of each sample an F-statistic of 2.341 is found.
Therefore we reject the Null hypothesis of homoskedasticity in favour of
the alternative – heteroskedasticity. The Breusch Pagan statistic again
tests for the presence of heteroskedasticity. After obtaining the OLS
residuals and using the R-squared for the unrestricted and restricted
model an F-statistic is found to be 6.011. This supports the Goldfeld
Quant test; we decide to fail to reject the presence of heteroskedasticity
– which backs up the use of robust standard errors. The Ramsey reset
model is used to test the functional form of the regression. The Reset
adds “polynomials in the OLS fitted values to include in an expanded
regression.” (Introductory Econometrics A modern approach, J.M.
Wooldbridge, A Modern Approach 2E, Thomson South Western). We fail
to reject the Null hypothesis and therefore conclude that the functional
form is satisfactory. This is because the F-statistic is less than the
conventional critical value.

Once the tests above have been concluded we are in a position to

interpret the variables. Robust standard errors are used to ensure that
the test statistics are applicable and there is no bias. The appearances of
the current season, and previous season as well as the current goals
and goals from the previous season are not significant at the 5 % level.
All of them carry positive coefficients which mean that the more
appearances and the more goals that are scored each season have a

27
Jack Costello MSc

positive effect on the log transfer price – this is to be expected. The most
significant of these variables is a1; which is the appearances from the
previous season (which may be surprising) although this may be when a
reputation was enhanced for example and their value as an asset
actually increased. This is supported by the findings in Reilly and Witt
(1995) which finds this regressor statistically significant at the
conventional level. It is pleasing to see that all four of these variables
have positive coefficients (more appearances and goals the higher the
log transfer fee) as this is a realistic. A fee is likely to increase with the
more appearances played as this is value added (same applies to league
goals scored).

Diagram IX. To show correlation between League appearances in

2005/2006 season and transfer fees.

10,000

8,000
Transfer fee (in 000's)

6,000
value

4,000

2,000

0
0 10 20 30 40 50
League appearances 2005/2006

The diagram supports the positive coefficient. We see that as the amount
of appearances increase that the transfer fee increases. The plot slopes
from left to right upwards. The number of previous clubs is attached with
a coefficient of -0.045 which is not statistically significant at the 5 % level.
This suggests a player with more previous clubs is attached with a lower
log transfer fee and that with each additional club ceteris paribus the
players log value is decreased by 4.5 %. Reilly and Witt (1995) support
the negative coefficient although they find a slightly higher value –
reducing fees by 7.8 % and the variable is found to be statistically
significant too. Medcalfe (2007) also finds a negative coefficient of
£628.184 (although this is not statistically significant). This may be linked

28
Jack Costello MSc

to a player’s reputation. The Club buying the player may have reduced
demand for the player as the player offers less loyalty. They would fear
losing the player due to his characteristics.

The age variables were separated using three separate partitions. This
was achieved through the use of dummy variables. For the under 20
variable if the player was under 20 than they take a binary variable of 1
and if 20 or over they take a binary variable of 0. If 20 up to and including
25 in the next variable; the player takes a dummy variable of one if fits
into this group and if outside this section than takes a value of 0. The
same applies to the final group; where if fits into this group than takes a
value of 1 and if outside the group takes a value of 0. This allows
interpretation to be relatively straightforward explanation due to the
explained variable being a logarithm. Two of the three variables are
statistically significant; which are the > 20 and < 25 group and the > 25
and below < 30 group. These two have an effect on the log price transfer
and are considered to be statistically significant. Unsurprisingly age has
a role to play on the log transfer price. Two of the three categories are
found to be statistically significant. Players aged between 20 and 25
trade at approximately 76.1 % more than those not in this age band
overall.

The striker term is not statistically significant at the 5 % value but is at

lower levels of significance – namely 4 % (allowing us to read into the
variable and its coefficient). Once more this is a dummy variable which
takes a binary value of 1 if the player is a striker and a value of 0 if the
player is a non-striker. The coefficient is positive which helps us to
suggest that a striker transfer for fees which are above those of other
players which are non-strikers (not surprising – this is the hunch that I
would take). The coefficient of 0.390 suggests that striker’s trade at 39 %
higher than those players who are not strikers. This is because the key
component of football is clearly scoring goals; where obviously strikers
will be most likely to do. They have the most value added in terms of a
position within a football team and therefore transfer at more relative to
those who are non-strikers.

29
Jack Costello MSc

Diagram X. Strikers average fee compared to non-strikers.

2,000

1,900

1,800
Average fee (in
000's) 1,700

1,600

1,500

1,400
Striker Non-striker
Average fee (in £000's) 1,921 1,617

The full international variable carries a positive coefficient. The higher

performing players and therefore those who carry a higher fee are more
likely to have international caps. This is effectively like the labour market;
where employers look for signals in quality. In this scenario the employer
is the buying club and the player represents the employee. Once more a
dummy variable is adopted; with any player with at least one
international cap given a binary value of 1 and those with 0 international
caps (not internationals) carry a value of 0. This is statistically significant
at the 4 % level and has a t-value of 1.92 (not at the 5 % level).
Furthermore the coefficient implies players who hold international status
carry a 30 % fee higher than those ceteris paribus that do not. This can
be linked to black players trading at a fee higher than there non-black
counterparts – more black players are internationals than whites and
mixed which is likely to contribute black players trading at potentially
more. This is something that could be used in future research. Diagram
VII indicates the extra value at which international players trade at (used
earlier in the study).

30
Jack Costello MSc

Diagram VII. Average values of international players versus non-

internationals.

Average Value (in £000's)

Non -
Average value (in £000's)

1,203
International

International 2,530

International status

Particular interest can be taken in the dummy variables which look at the
divisional status of the buying and selling Clubs; which is applicable to
the Championship, League One and League Two. This contributes six
variables; of which four are said to be statistically significant when
analysing t-statistics at the 5 % level and therefore is likely to contribute
towards the R-squared in a relatively large way. Again these are dummy
variables and can be explained using one as an example. When looking
at the division of the Selling Club; say for example a Championship Club;
if the selling club is from the Championship than they take a binary value
of one, and if the selling Club is not from the Championship than they
take a value of zero. The League two results are statistically significant at
the 5 % level. This also carries a high coefficient value of – 2.926 which
suggests that clubs selling a player from this league are likely to receive
a significantly smaller fee ceteris paribus to those in the parallel higher
Leagues. This is relatively intuitive as the players sold at this level (the
lowest) should be of the lowest quality as it is the lowest League in the
English hierarchal constabulary and is therefore likely to be aligned to
the lowest log transfer fees. When looking at the division of the buying
club all three of the variables are statistically significant (5 % level once
more). Each carry negative terms, which means each has a negative

31
Jack Costello MSc

effect on the transfer fee. This is likely to be due to the Premier League
and the power that it holds over other Clubs due to their followings, and
their relatively large turnovers; meaning that a large proportion of the
transfer fees that are paid are from the English Premier League. This
once more could be used in future research.

The R-squared is 0.517 and therefore represents 51.7 % of the data.

This is smaller than Reilly and Witt (1995) which can be taken as a point
of reference rather than be considered to be directly comparable and
may be due to two variables being missed in this study; whether the
transfer was by a tribunal and the log (receipts) variable, with the latter
being highly statistically significant and therefore will contribute to the R-
squared. The adjusted R-squared is less than the R-squared - this takes
account of the number of variables which are being used. Furthermore
this is also considerably less than the 0.94 R-squared found in Medcalfe
(2007) – this can again be associated with the inclusion of some key
variables (although it is offset by minimal amount of observations in the
survey). The R-squared is not considered unreasonable.

Finally we come to the black variable - the most important variable in the
study. If the player is black than a 1 is adopted and if the player is non-
black (white or mixed race) than a binary variable of zero is adopted. The
black variable is not statistically significant at the 5% level, meaning that
the Null Hypothesis of black players trading at prices ceteris paribus
trade below that of white players is rejected. If anything they appear to
trade at a higher price (the black coefficient is positive, suggesting black
players have a positive effect on the transfer price which is a
associated). This may be driven by black players in this study being
more likely to be strikers which is aligned with a higher log transfer fee
(could be analysed in another study) and also that more black players
has international caps (explained before). The 0.2360661 coefficient
suggests that black players trade at a higher fee than white and mixed
race players due to the positive coefficient which is attached. The
coefficient suggests black player’s trade at 23.6 % higher than their non-
black equivalent. The t-value of 1.31 means that the coefficient is not
statistically significant (the t-value is below the 1.97 required). Therefore
we reject the Null Hypothesis of black players trading at less than there

32
Jack Costello MSc

non-black equivalents. If anything we suggest at a lower significance

level that black players actually trade at higher amounts than their non
black counterpart (but statistically significant at the conventional level).

Conclusions.

Our study aims to examine the role of race in the determination of

association football transfer prices. We examine whether ceteris paribus
black players trade at lower values than their non-black equivalents. If
this is found to be true there is evidence to suggest that we cannot rule
out discrimination in the market. In effect this paper is an update of the
paper formed by Reilly and Witt (1995). OLS estimation is the
methodology used to test this hypothesis. After the collection of data,
consistently using the Skysports football yearbook and PFA’s player’s
who’s who the analysis could begin. Each of the players attributes and
statistics were collected from a common source to ensure there was no
bias in the study. Upon the data collection being completed it was
possible to give some basic data analysis; which seemed to originally
suggest that there was little evidence of black players being
discriminated against. The study adversely found the raw data suggested
black players traded for a higher fee than their non-black equivalents.
This hypothesis testing could be pushed further forward by the
introduction of the analysis of the player’s wages – unfortunately this
data was not available. This may give a better indication of discrimination
in the market.

The OLS estimation was formed and after finding the presence of
heteroskedasticty through the reported diagnostics of the Goldfeld-
Quandt test and the Breusch Pagan statistic, robust standard errors were
implemented as to deter bias and allow inferences to be fairly made. The
functional form was found to be satisfactory at the conventional level
using the Ramsey reset test. This allowed us to look at the coefficients,
their standard errors, and as a result the inferences that could be made
of the explained variable – the log transfer price. Of primary importance
was the black variable. It was found that black players trade at 23.6 %
higher than their non-black counterparts. The black variable was found
not statistically significant at the conventional 5% level. Therefore it could

33
Jack Costello MSc

be concluded that there is no evidence to suggest that black players are

discriminated again in football association league prices. This is
supported by the parallel papers of Reilly and Witt (1995) and Medcalfe
(2007) despite their coefficients being negative.

Several other variables in the study are found to be statistically

significant and are attributed to the reasonable 51.7 % R-squared which
can be found. This may also contribute as to why black players appear to
trade at a higher price than non-players. The striker coefficient which
suggests that strikers trade at values 39 % higher than their non-black
equivalent – which is statistically significant at the 4 % level. Furthermore
proportionally black players have the highest percentage of strikers in the
study. Incidentally players with international status all affect the log
transfer price with a positive coefficient implying international player’s
trade at 30.3 % higher than their non-international counterparts
(statistically significant at the 4 % level). In tandem with this is that black
players in comparison to non-black players have a higher percentage
representing that are internationals (the international coefficient is
significant at the 4% level). International status and striker status can be
suggested to help contribute to black players trading at higher price vis-
à-vis non-black players. The proportions of black players in each are
high. Aligned with each being statistically significant at the 4 % level
there is no doubt that the blacks higher trading price can be partly
attributed to this.

In conclusion we find that there is no evidence to suggest that black

players are discriminated against when testing against the determination
of race in the determination of association football transfer prices. The
variable is not statically significant at the conventional 5% level.

34
Jack Costello MSc

Appendix I. All data’s summary statistics.

Standard
Mean Maximum Minimum
Deviation
Age 25.374 38 17 3.753
< 20 0.057 46 0 n/a
>20 and <25 0.443 1 0 n/a
>25 and <30 0.396 1 0 n/a
Season 1.517 2 1 n/a
Height 71.491 78 64 2.304
Weight 166.283 224 10 21.398
Striker 2.974 1 0 n/a
Goals1 3.996 23 0 4.884
Goals2 4.448 30 0 5.515
Apps1 25.909 46 0 12.603
Apps2 28.665 46 0 10.577
Number of
2.509 8 1 1.489
previous clubs
Number of years
3.191 11 1 1.824
at previous club
Value (in £000's) 1,728 18,600 20 2382

Appendix II. White player’s summary statistics.

Standard
Mean Maximum Minimum
Deviation
Age 25.374 38 17 3.882
< 20 0.077 1 0 n/a
>20 and <25 0.413 1 0 n/a
>25 and <30 0.413 1 0 n/a
Season 1.471 2 1 n/a
Height 71.329 78 64 2.302
Weight 166.581 203 67 15.512
Striker 0.329 1 0 n/a
Goals1 3.884 23 0 4.937
Goals2 4.090 30 0 5.371
Apps1 25.858 46 0 12.728
Apps2 28.187 46 0 11.366
Number of
2.406 7 1 1.371
previous clubs
Number of years
3.394 11 1 1.972
at previous club
Value (in £000's) 1,662 18,600 20 2341

35
Jack Costello MSc

Appendix III. Mixed race player’s summary statistics.

Standard
Mean Maximum Minimum Deviation
Age 24.583 31 20 2.750
< 20 0 n/a
>20 and <25 0.639 n/a
>25 and <30 0.306 n/a
Season 1.556 n/a
Height 71.361 76 66 2.404
Weight 165.639 196 140 12.481
Striker 0.417 n/a
Goals1 3.556 10 0 2.903
Goals2 6.333 22 0 6.646
Apps1 24.750 46 5 11.423
Apps2 30.917 44 15 7.617
Number of
previous clubs 2.722 7 1 1.542
Number of years
at previous club 2.639 7 1 1.496
Value (in £000's) 1668 7,000 50 2001

Appendix IV. Black player’s summary statistics.

Standard
Mean Maximum Minimum
Deviation
Age 26.103 35 19 3.966
< 20 0.026 N/A
>20 and <25 0.385 N/A
>25 and <30 0.410 N/A
Season 1.667 N/A
Height 72.256 75 68 2.112
Weight 165.692 224 10 40.492
Striker 0.462 N/A
Goals1 4.846 21 0 6.020
Goals2 4.128 21 0 4.658
Apps1 27.179 45 0 13.326
Apps2 28.487 43 9 9.564
Number of
2.000 8 1 1.849
previous clubs
Number of years
2.897 6 1 1.294
at previous club
Value (in £000's) 2,042 16,500 75 2856

36
Jack Costello MSc

Reference List.

Journal articles & text books

The PFA Footballers’ Who’s Who 2006-2007, Barry J. Hugman
Skysports Football Yearbook 2007-2008, Glenda Rollin and Jack Rollin
Skysports Football Yearbook 2006-2007, Grenda Rollin and Jack Rollin
Introductory Econometrics – A modern approach 2E. J. M. Wooldridge
Thomson South Western

Websites
Fo otb all Ind us try Group
www. liv. ac .uk /foo tba lli nd us try/b os ma n
Eac h of the 9 2 Foot ba ll Le ag ue Clu bs a nd th eir o wn offic i al
webs it es (rac e ide ntif ic atio n)

Bibliography
Medcalfe, S. “English league transfer prices: is there a racial dimension?
A re-examination with new data”, Applied Economic Letters, Online 1
August 2007.
Kahn, L. M “Discrimination in professional sports: a survey of the
literature”, Industrial and Labor Relations Review, volume 44, 1991, pp
395-418
Reilly, B and Witt, R. “English league transfer prices is there a racial
dimension?” Applied Economics Letters, volume 2, pp 220-222.
Hill, R and Spellman, W. “Pay discrimination in baseball: data from the
seventies.” Industrial Relations: A journal of economy and society,
volume 23, issue 1, pp 103-112.
Pascal, A and Rapping, L. “The economics of racial discrimination in
organized baseball.” Racial discrimination in economic life, pp 119-56
Christiano, K. “Salary discrimination in major league baseball: the effect
of race.” Sociology of Sports Journal, Vol 3, No 2(June) pp 144-53.
Rockwood, C and Asher E. “Racial discrimination in professional
basketball revisited.” American Economist, 1976, Volume 20, No 1
(Spring) pp 59-64
Madison, D and Landers D. “Racial Discrimination in Football: A Test of
the “stacking” playing positions hypothesis.” Social Problems in Athletics;
Essays on the Sociology of Sports; pp 151-56.

37